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">
10 <div class="doc_title">
11 LLVM Language Reference Manual
15 <li><a href="#abstract">Abstract</a></li>
16 <li><a href="#introduction">Introduction</a></li>
17 <li><a href="#identifiers">Identifiers</a></li>
18 <li><a href="#typesystem">Type System</a>
20 <li><a href="#t_primitive">Primitive Types</a>
22 <li><a href="#t_classifications">Type Classifications</a></li>
24 <li><a href="#t_derived">Derived Types</a>
26 <li><a href="#t_array" >Array Type</a></li>
27 <li><a href="#t_function">Function Type</a></li>
28 <li><a href="#t_pointer">Pointer Type</a></li>
29 <li><a href="#t_struct" >Structure Type</a></li>
30 <!-- <li><a href="#t_packed" >Packed Type</a> -->
33 <li><a href="#highlevel">High Level Structure</a>
35 <li><a href="#modulestructure">Module Structure</a></li>
36 <li><a href="#globalvars">Global Variables</a></li>
37 <li><a href="#functionstructure">Function Structure</a></li>
39 <li><a href="#instref">Instruction Reference</a>
41 <li><a href="#terminators">Terminator Instructions</a>
43 <li><a href="#i_ret" >'<tt>ret</tt>' Instruction</a></li>
44 <li><a href="#i_br" >'<tt>br</tt>' Instruction</a></li>
45 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
46 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
47 <li><a href="#i_unwind" >'<tt>unwind</tt>' Instruction</a></li>
49 <li><a href="#binaryops">Binary Operations</a>
51 <li><a href="#i_add" >'<tt>add</tt>' Instruction</a></li>
52 <li><a href="#i_sub" >'<tt>sub</tt>' Instruction</a></li>
53 <li><a href="#i_mul" >'<tt>mul</tt>' Instruction</a></li>
54 <li><a href="#i_div" >'<tt>div</tt>' Instruction</a></li>
55 <li><a href="#i_rem" >'<tt>rem</tt>' Instruction</a></li>
56 <li><a href="#i_setcc">'<tt>set<i>cc</i></tt>' Instructions</a></li>
58 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
60 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
61 <li><a href="#i_or" >'<tt>or</tt>' Instruction</a></li>
62 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
63 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
64 <li><a href="#i_shr">'<tt>shr</tt>' Instruction</a></li>
66 <li><a href="#memoryops">Memory Access Operations</a>
68 <li><a href="#i_malloc" >'<tt>malloc</tt>' Instruction</a></li>
69 <li><a href="#i_free" >'<tt>free</tt>' Instruction</a></li>
70 <li><a href="#i_alloca" >'<tt>alloca</tt>' Instruction</a></li>
71 <li><a href="#i_load" >'<tt>load</tt>' Instruction</a></li>
72 <li><a href="#i_store" >'<tt>store</tt>' Instruction</a></li>
73 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
75 <li><a href="#otherops">Other Operations</a>
77 <li><a href="#i_phi" >'<tt>phi</tt>' Instruction</a></li>
78 <li><a href="#i_cast">'<tt>cast .. to</tt>' Instruction</a></li>
79 <li><a href="#i_call" >'<tt>call</tt>' Instruction</a></li>
80 <li><a href="#i_vanext">'<tt>vanext</tt>' Instruction</a></li>
81 <li><a href="#i_vaarg" >'<tt>vaarg</tt>' Instruction</a></li>
84 <li><a href="#intrinsics">Intrinsic Functions</a>
86 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
88 <li><a href="#i_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
89 <li><a href="#i_va_end" >'<tt>llvm.va_end</tt>' Intrinsic</a></li>
90 <li><a href="#i_va_copy" >'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
96 <div class="doc_text">
97 <p><b>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a> and <A href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></b><p>
100 <!-- *********************************************************************** -->
101 <div class="doc_section">
102 <a name="abstract">Abstract
104 <!-- *********************************************************************** -->
106 <div class="doc_text">
108 <p>This document is a reference manual for the LLVM assembly language. LLVM is
109 an SSA based representation that provides type safety, low-level operations,
110 flexibility, and the capability of representing 'all' high-level languages
111 cleanly. It is the common code representation used throughout all phases of the
112 LLVM compilation strategy.</p>
116 <!-- *********************************************************************** -->
117 <div class="doc_section">
118 <a name="introduction">Introduction</a>
120 <!-- *********************************************************************** -->
122 <div class="doc_text">
124 <p>The LLVM code representation is designed to be used in three different forms:
125 as an in-memory compiler IR, as an on-disk bytecode representation (suitable for
126 fast loading by a Just-In-Time compiler), and as a human readable assembly
127 language representation. This allows LLVM to provide a powerful intermediate
128 representation for efficient compiler transformations and analysis, while
129 providing a natural means to debug and visualize the transformations. The three
130 different forms of LLVM are all equivalent. This document describes the human
131 readable representation and notation.</p>
133 <p>The LLVM representation aims to be a light-weight and low-level while being
134 expressive, typed, and extensible at the same time. It aims to be a "universal
135 IR" of sorts, by being at a low enough level that high-level ideas may be
136 cleanly mapped to it (similar to how microprocessors are "universal IR's",
137 allowing many source languages to be mapped to them). By providing type
138 information, LLVM can be used as the target of optimizations: for example,
139 through pointer analysis, it can be proven that a C automatic variable is never
140 accessed outside of the current function... allowing it to be promoted to a
141 simple SSA value instead of a memory location.</p>
145 <!-- _______________________________________________________________________ -->
146 <div class="doc_subsubsection">
147 <a name="wellformed">Well-Formedness</a>
150 <div class="doc_text">
152 <p>It is important to note that this document describes 'well formed' LLVM
153 assembly language. There is a difference between what the parser accepts and
154 what is considered 'well formed'. For example, the following instruction is
155 syntactically okay, but not well formed:</p>
158 %x = <a href="#i_add">add</a> int 1, %x
161 <p>...because the definition of <tt>%x</tt> does not dominate all of its uses.
162 The LLVM infrastructure provides a verification pass that may be used to verify
163 that an LLVM module is well formed. This pass is automatically run by the
164 parser after parsing input assembly, and by the optimizer before it outputs
165 bytecode. The violations pointed out by the verifier pass indicate bugs in
166 transformation passes or input to the parser.</p>
168 <!-- Describe the typesetting conventions here. -->
172 <!-- *********************************************************************** -->
173 <div class="doc_section">
174 <a name="identifiers">Identifiers</a>
176 <!-- *********************************************************************** -->
178 <div class="doc_text">
180 <p>LLVM uses three different forms of identifiers, for different purposes:</p>
184 <li>Numeric constants are represented as you would expect: 12, -3 123.421,
185 etc. Floating point constants have an optional hexidecimal notation.</li>
187 <li>Named values are represented as a string of characters with a '%' prefix.
188 For example, %foo, %DivisionByZero, %a.really.long.identifier. The actual
189 regular expression used is '<tt>%[a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
190 Identifiers which require other characters in their names can be surrounded
191 with quotes. In this way, anything except a <tt>"</tt> character can be used
194 <li>Unnamed values are represented as an unsigned numeric value with a '%'
195 prefix. For example, %12, %2, %44.</li>
199 <p>LLVM requires the values start with a '%' sign for two reasons: Compilers
200 don't need to worry about name clashes with reserved words, and the set of
201 reserved words may be expanded in the future without penalty. Additionally,
202 unnamed identifiers allow a compiler to quickly come up with a temporary
203 variable without having to avoid symbol table conflicts.</p>
205 <p>Reserved words in LLVM are very similar to reserved words in other languages.
206 There are keywords for different opcodes ('<tt><a href="#i_add">add</a></tt>',
207 '<tt><a href="#i_cast">cast</a></tt>', '<tt><a href="#i_ret">ret</a></tt>',
208 etc...), for primitive type names ('<tt><a href="#t_void">void</a></tt>',
209 '<tt><a href="#t_uint">uint</a></tt>', etc...), and others. These reserved
210 words cannot conflict with variable names, because none of them start with a '%'
213 <p>Here is an example of LLVM code to multiply the integer variable
214 '<tt>%X</tt>' by 8:</p>
219 %result = <a href="#i_mul">mul</a> uint %X, 8
222 <p>After strength reduction:</p>
225 %result = <a href="#i_shl">shl</a> uint %X, ubyte 3
228 <p>And the hard way:</p>
231 <a href="#i_add">add</a> uint %X, %X <i>; yields {uint}:%0</i>
232 <a href="#i_add">add</a> uint %0, %0 <i>; yields {uint}:%1</i>
233 %result = <a href="#i_add">add</a> uint %1, %1
236 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several important
237 lexical features of LLVM:</p>
240 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
243 <li>Unnamed temporaries are created when the result of a computation is not
244 assigned to a named value.</li>
246 <li>Unnamed temporaries are numbered sequentially</li>
249 <p>...and it also show a convention that we follow in this document. When
250 demonstrating instructions, we will follow an instruction with a comment that
251 defines the type and name of value produced. Comments are shown in italic
254 <p>The one non-intuitive notation for constants is the optional hexidecimal form
255 of floating point constants. For example, the form '<tt>double
256 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
257 4.5e+15</tt>' which is also supported by the parser. The only time hexadecimal
258 floating point constants are useful (and the only time that they are generated
259 by the disassembler) is when an FP constant has to be emitted that is not
260 representable as a decimal floating point number exactly. For example, NaN's,
261 infinities, and other special cases are represented in their IEEE hexadecimal
262 format so that assembly and disassembly do not cause any bits to change in the
267 <!-- *********************************************************************** -->
268 <div class="doc_section">
269 <a name="typesystem">Type System</a>
271 <!-- *********************************************************************** -->
273 <div class="doc_text">
275 <p>The LLVM type system is one of the most important features of the
276 intermediate representation. Being typed enables a number of optimizations to
277 be performed on the IR directly, without having to do extra analyses on the side
278 before the transformation. A strong type system makes it easier to read the
279 generated code and enables novel analyses and transformations that are not
280 feasible to perform on normal three address code representations.</p>
282 <!-- The written form for the type system was heavily influenced by the
283 syntactic problems with types in the C language<sup><a
284 href="#rw_stroustrup">1</a></sup>.<p> -->
288 <!-- ======================================================================= -->
289 <div class="doc_subsection">
290 <a name="t_primitive">Primitive Types</a>
293 <div class="doc_text">
295 <p>The primitive types are the fundemental building blocks of the LLVM system.
296 The current set of primitive types are as follows:</p>
299 <table border="0" align="center">
303 <table border="1" cellspacing="0" cellpadding="4" align="center">
304 <tr><td><tt>void</tt></td> <td>No value</td></tr>
305 <tr><td><tt>ubyte</tt></td> <td>Unsigned 8 bit value</td></tr>
306 <tr><td><tt>ushort</tt></td><td>Unsigned 16 bit value</td></tr>
307 <tr><td><tt>uint</tt></td> <td>Unsigned 32 bit value</td></tr>
308 <tr><td><tt>ulong</tt></td> <td>Unsigned 64 bit value</td></tr>
309 <tr><td><tt>float</tt></td> <td>32 bit floating point value</td></tr>
310 <tr><td><tt>label</tt></td> <td>Branch destination</td></tr>
315 <table border="1" cellspacing="0" cellpadding="4" align=center">
316 <tr><td><tt>bool</tt></td> <td>True or False value</td></tr>
317 <tr><td><tt>sbyte</tt></td> <td>Signed 8 bit value</td></tr>
318 <tr><td><tt>short</tt></td> <td>Signed 16 bit value</td></tr>
319 <tr><td><tt>int</tt></td> <td>Signed 32 bit value</td></tr>
320 <tr><td><tt>long</tt></td> <td>Signed 64 bit value</td></tr>
321 <tr><td><tt>double</tt></td><td>64 bit floating point value</td></tr>
331 <!-- _______________________________________________________________________ -->
332 <div class="doc_subsubsection">
333 <a name="t_classifications">Type Classifications</a>
336 <div class="doc_text">
338 <p>These different primitive types fall into a few useful classifications:</p>
341 <table border="1" cellspacing="0" cellpadding="4" align="center">
343 <td><a name="t_signed">signed</td>
344 <td><tt>sbyte, short, int, long, float, double</tt></td>
347 <td><a name="t_unsigned">unsigned</td>
348 <td><tt>ubyte, ushort, uint, ulong</tt></td>
351 <td><a name="t_integer">integer</td>
352 <td><tt>ubyte, sbyte, ushort, short, uint, int, ulong, long</tt></td>
355 <td><a name="t_integral">integral</td>
356 <td><tt>bool, ubyte, sbyte, ushort, short, uint, int, ulong, long</tt></td>
359 <td><a name="t_floating">floating point</td>
360 <td><tt>float, double</tt></td>
363 <td><a name="t_firstclass">first class</td>
364 <td><tt>bool, ubyte, sbyte, ushort, short,<br>
365 uint, int, ulong, long, float, double,
366 <a href="#t_pointer">pointer</a></tt></td>
371 <p>The <a href="#t_firstclass">first class</a> types are perhaps the most
372 important. Values of these types are the only ones which can be produced by
373 instructions, passed as arguments, or used as operands to instructions. This
374 means that all structures and arrays must be manipulated either by pointer or by
379 <!-- ======================================================================= -->
380 <div class="doc_subsection">
381 <a name="t_derived">Derived Types</a>
384 <div class="doc_text">
386 <p>The real power in LLVM comes from the derived types in the system. This is
387 what allows a programmer to represent arrays, functions, pointers, and other
388 useful types. Note that these derived types may be recursive: For example, it
389 is possible to have a two dimensional array.</p>
393 <!-- _______________________________________________________________________ -->
394 <div class="doc_subsubsection">
395 <a name="t_array">Array Type</a>
398 <div class="doc_text">
402 <p>The array type is a very simple derived type that arranges elements
403 sequentially in memory. The array type requires a size (number of elements) and
404 an underlying data type.</p>
409 [<# elements> x <elementtype>]
412 <p>The number of elements is a constant integer value, elementtype may be any
413 type with a size.</p>
418 <tt>[40 x int ]</tt>: Array of 40 integer values.<br>
419 <tt>[41 x int ]</tt>: Array of 41 integer values.<br>
420 <tt>[40 x uint]</tt>: Array of 40 unsigned integer values.<p>
423 <p>Here are some examples of multidimensional arrays:</p>
426 <table border="0" cellpadding="0" cellspacing="0">
428 <td><tt>[3 x [4 x int]]</tt></td>
429 <td>: 3x4 array integer values.</td>
432 <td><tt>[12 x [10 x float]]</tt></td>
433 <td>: 12x10 array of single precision floating point values.</td>
436 <td><tt>[2 x [3 x [4 x uint]]]</tt></td>
437 <td>: 2x3x4 array of unsigned integer values.</td>
444 <!-- _______________________________________________________________________ -->
445 <div class="doc_subsubsection">
446 <a name="t_function">Function Type</a>
449 <div class="doc_text">
453 <p>The function type can be thought of as a function signature. It consists of
454 a return type and a list of formal parameter types. Function types are usually
455 used when to build virtual function tables (which are structures of pointers to
456 functions), for indirect function calls, and when defining a function.</p>
461 <returntype> (<parameter list>)
464 <p>Where '<tt><parameter list></tt>' is a comma-separated list of type
465 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
466 which indicates that the function takes a variable number of arguments.
467 Variable argument functions can access their arguments with the <a
468 href="#int_varargs">variable argument handling intrinsic</a> functions.</p>
473 <table border="0" cellpadding="0" cellspacing="0">
476 <td><tt>int (int)</tt></td>
477 <td>: function taking an <tt>int</tt>, returning an <tt>int</tt></td>
480 <td><tt>float (int, int *) *</tt></td>
481 <td>: <a href="#t_pointer">Pointer</a> to a function that takes an
482 <tt>int</tt> and a <a href="#t_pointer">pointer</a> to <tt>int</tt>,
483 returning <tt>float</tt>.</td>
486 <td><tt>int (sbyte *, ...)</tt></td>
487 <td>: A vararg function that takes at least one <a
488 href="#t_pointer">pointer</a> to <tt>sbyte</tt> (signed char in C), which
489 returns an integer. This is the signature for <tt>printf</tt> in
497 <!-- _______________________________________________________________________ -->
498 <div class="doc_subsubsection">
499 <a name="t_struct">Structure Type</a>
502 <div class="doc_text">
506 <p>The structure type is used to represent a collection of data members together
507 in memory. The packing of the field types is defined to match the ABI of the
508 underlying processor. The elements of a structure may be any type that has a
511 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt> and
512 '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field with the
513 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
518 { <type list> }
524 <table border="0" cellpadding="0" cellspacing="0">
526 <td><tt>{ int, int, int }</tt></td>
527 <td>: a triple of three <tt>int</tt> values</td>
530 <td><tt>{ float, int (int) * }</tt></td>
531 <td>: A pair, where the first element is a <tt>float</tt> and the second
532 element is a <a href="#t_pointer">pointer</a> to a <a
533 href="t_function">function</a> that takes an <tt>int</tt>, returning an
541 <!-- _______________________________________________________________________ -->
542 <div class="doc_subsubsection">
543 <a name="t_pointer">Pointer Type</a>
546 <div class="doc_text">
550 <p>As in many languages, the pointer type represents a pointer or reference to
551 another object, which must live in memory.</p>
561 <table border="0" cellpadding="0" cellspacing="0">
563 <td><tt>[4x int]*</tt></td>
564 <td>: <a href="#t_pointer">pointer</a> to <a href="#t_array">array</a> of four
565 <tt>int</tt> values</td>
568 <td><tt>int (int *) *</tt></td>
569 <td>: A <a href="#t_pointer">pointer</a> to a <a
570 href="t_function">function</a> that takes an <tt>int</tt>, returning an
578 <!-- _______________________________________________________________________ -->
580 <div class="doc_subsubsection">
581 <a name="t_packed">Packed Type</a>
584 <div class="doc_text">
586 Mention/decide that packed types work with saturation or not. Maybe have a packed+saturated type in addition to just a packed type.<p>
588 Packed types should be 'nonsaturated' because standard data types are not saturated. Maybe have a saturated packed type?<p>
595 <!-- *********************************************************************** -->
596 <div class="doc_section">
597 <a name="highlevel">High Level Structure</a>
599 <!-- *********************************************************************** -->
602 <!-- ======================================================================= -->
603 <div class="doc_subsection">
604 <a name="modulestructure">Module Structure</a>
607 <div class="doc_text">
609 <p>LLVM programs are composed of "Module"s, each of which is a translation unit
610 of the input programs. Each module consists of functions, global variables, and
611 symbol table entries. Modules may be combined together with the LLVM linker,
612 which merges function (and global variable) definitions, resolves forward
613 declarations, and merges symbol table entries. Here is an example of the "hello
617 <i>; Declare the string constant as a global constant...</i>
618 <a href="#identifiers">%.LC0</a> = <a href="#linkage_internal">internal</a> <a href="#globalvars">constant</a> <a href="#t_array">[13 x sbyte]</a> c"hello world\0A\00" <i>; [13 x sbyte]*</i>
620 <i>; External declaration of the puts function</i>
621 <a href="#functionstructure">declare</a> int %puts(sbyte*) <i>; int(sbyte*)* </i>
623 <i>; Definition of main function</i>
624 int %main() { <i>; int()* </i>
625 <i>; Convert [13x sbyte]* to sbyte *...</i>
626 %cast210 = <a href="#i_getelementptr">getelementptr</a> [13 x sbyte]* %.LC0, long 0, long 0 <i>; sbyte*</i>
628 <i>; Call puts function to write out the string to stdout...</i>
629 <a href="#i_call">call</a> int %puts(sbyte* %cast210) <i>; int</i>
630 <a href="#i_ret">ret</a> int 0
634 <p>This example is made up of a <a href="#globalvars">global variable</a> named
635 "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>" function, and a
636 <a href="#functionstructure">function definition</a> for "<tt>main</tt>".</p>
639 In general, a module is made up of a list of global values, where both functions
640 and global variables are global values. Global values are represented by a
641 pointer to a memory location (in this case, a pointer to an array of char, and a
642 pointer to a function), and have one of the following linkage types:<p>
645 <a name="linkage_internal">
646 <dt><tt><b>internal</b></tt>
648 <dd>Global values with internal linkage are only directly accessible by objects
649 in the current module. In particular, linking code into a module with an
650 internal global value may cause the internal to be renamed as necessary to avoid
651 collisions. Because the symbol is internal to the module, all references can be
652 updated. This corresponds to the notion of the '<tt>static</tt>' keyword in C,
653 or the idea of "anonymous namespaces" in C++.<p>
655 <a name="linkage_linkonce">
656 <dt><tt><b>linkonce</b></tt>:
658 <dd>"<tt>linkonce</tt>" linkage is similar to <tt>internal</tt> linkage, with
659 the twist that linking together two modules defining the same <tt>linkonce</tt>
660 globals will cause one of the globals to be discarded. This is typically used
661 to implement inline functions. Unreferenced <tt>linkonce</tt> globals are
662 allowed to be discarded.<p>
664 <a name="linkage_weak">
665 <dt><tt><b>weak</b></tt>:
667 <dd>"<tt>weak</tt>" linkage is exactly the same as <tt>linkonce</tt> linkage,
668 except that unreferenced <tt>weak</tt> globals may not be discarded. This is
669 used to implement constructs in C such as "<tt>int X;</tt>" at global scope.<p>
671 <a name="linkage_appending">
672 <dt><tt><b>appending</b></tt>:
674 <dd>"<tt>appending</tt>" linkage may only applied to global variables of pointer
675 to array type. When two global variables with appending linkage are linked
676 together, the two global arrays are appended together. This is the LLVM,
677 typesafe, equivalent of having the system linker append together "sections" with
678 identical names when .o files are linked.<p>
680 <a name="linkage_external">
681 <dt><tt><b>externally visible</b></tt>:
683 <dd>If none of the above identifiers are used, the global is externally visible,
684 meaning that it participates in linkage and can be used to resolve external
685 symbol references.<p>
689 <p>For example, since the "<tt>.LC0</tt>" variable is defined to be internal, if
690 another module defined a "<tt>.LC0</tt>" variable and was linked with this one,
691 one of the two would be renamed, preventing a collision. Since "<tt>main</tt>"
692 and "<tt>puts</tt>" are external (i.e., lacking any linkage declarations), they
693 are accessible outside of the current module. It is illegal for a function
694 <i>declaration</i> to have any linkage type other than "externally visible".</p>
698 <!-- ======================================================================= -->
699 <div class="doc_subsection">
700 <a name="globalvars">Global Variables</a>
703 <div class="doc_text">
705 <p>Global variables define regions of memory allocated at compilation time
706 instead of run-time. Global variables may optionally be initialized. A
707 variable may be defined as a global "constant", which indicates that the
708 contents of the variable will never be modified (opening options for
709 optimization). Constants must always have an initial value.</p>
711 <p>As SSA values, global variables define pointer values that are in scope
712 (i.e. they dominate) for all basic blocks in the program. Global variables
713 always define a pointer to their "content" type because they describe a region
714 of memory, and all memory objects in LLVM are accessed through pointers.</p>
718 <!-- ======================================================================= -->
719 <div class="doc_subsection">
720 <a name="functionstructure">Functions</a>
723 <div class="doc_text">
725 <p>LLVM functions definitions are composed of a (possibly empty) argument list,
726 an opening curly brace, a list of basic blocks, and a closing curly brace. LLVM
727 function declarations are defined with the "<tt>declare</tt>" keyword, a
728 function name and a function signature.</p>
730 <p>A function definition contains a list of basic blocks, forming the CFG for
731 the function. Each basic block may optionally start with a label (giving the
732 basic block a symbol table entry), contains a list of instructions, and ends
733 with a <a href="#terminators">terminator</a> instruction (such as a branch or
734 function return).</p>
736 <p>The first basic block in program is special in two ways: it is immediately
737 executed on entrance to the function, and it is not allowed to have predecessor
738 basic blocks (i.e. there can not be any branches to the entry block of a
739 function). Because the block can have no predecessors, it also cannot have any
740 <a href="#i_phi">PHI nodes</a>.</p>
744 <!-- *********************************************************************** -->
745 <div class="doc_section">
746 <a name="instref">Instruction Reference</a>
748 <!-- *********************************************************************** -->
750 <div class="doc_text">
752 <p>The LLVM instruction set consists of several different classifications of
753 instructions: <a href="#terminators">terminator instructions</a>, <a
754 href="#binaryops">binary instructions</a>, <a href="#memoryops">memory
755 instructions</a>, and <a href="#otherops">other instructions</a>.</p>
759 <!-- ======================================================================= -->
760 <div class="doc_subsection">
761 <a name="terminators">Terminator Instructions</a>
764 <div class="doc_text">
766 <p>As mentioned <a href="#functionstructure">previously</a>, every basic block
767 in a program ends with a "Terminator" instruction, which indicates which block
768 should be executed after the current block is finished. These terminator
769 instructions typically yield a '<tt>void</tt>' value: they produce control flow,
770 not values (the one exception being the '<a
771 href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
773 <p>There are five different terminator instructions: the '<a
774 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a
775 href="#i_br"><tt>br</tt></a>' instruction, the '<a
776 href="#i_switch"><tt>switch</tt></a>' instruction, the '<a
777 href="#i_invoke"><tt>invoke</tt></a>' instruction, and the '<a
778 href="#i_unwind"><tt>unwind</tt></a>' instruction.</p>
782 <!-- _______________________________________________________________________ -->
783 <div class="doc_subsubsection">
784 <a name="i_ret">'<tt>ret</tt>' Instruction</a>
787 <div class="doc_text">
791 ret <type> <value> <i>; Return a value from a non-void function</i>
792 ret void <i>; Return from void function</i>
797 <p>The '<tt>ret</tt>' instruction is used to return control flow (and a value)
798 from a function, back to the caller.</p>
800 <p>There are two forms of the '<tt>ret</tt>' instructruction: one that returns a
801 value and then causes control flow, and one that just causes control flow to
806 <p>The '<tt>ret</tt>' instruction may return any '<a href="#t_firstclass">first
807 class</a>' type. Notice that a function is not <a href="#wellformed">well
808 formed</a> if there exists a '<tt>ret</tt>' instruction inside of the function
809 that returns a value that does not match the return type of the function.</p>
813 <p>When the '<tt>ret</tt>' instruction is executed, control flow returns back to
814 the calling function's context. If the caller is a "<a
815 href="#i_call"><tt>call</tt></a> instruction, execution continues at the
816 instruction after the call. If the caller was an "<a
817 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues at the
818 beginning "normal" of the destination block. If the instruction returns a
819 value, that value shall set the call or invoke instruction's return value.</p>
823 ret int 5 <i>; Return an integer value of 5</i>
824 ret void <i>; Return from a void function</i>
829 <!-- _______________________________________________________________________ -->
830 <div class="doc_subsubsection">
831 <a name="i_br">'<tt>br</tt>' Instruction</a>
834 <div class="doc_text">
839 br bool <cond>, label <iftrue>, label <iffalse>
840 br label <dest> <i>; Unconditional branch</i>
845 <p>The '<tt>br</tt>' instruction is used to cause control flow to transfer to a
846 different basic block in the current function. There are two forms of this
847 instruction, corresponding to a conditional branch and an unconditional
852 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a single
853 '<tt>bool</tt>' value and two '<tt>label</tt>' values. The unconditional form
854 of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>' value as a
859 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the
860 '<tt>bool</tt>' argument is evaluated. If the value is <tt>true</tt>, control
861 flows to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is
862 <tt>false</tt>, control flows to the '<tt>iffalse</tt>' <tt>label</tt>
869 %cond = <a href="#i_setcc">seteq</a> int %a, %b
870 br bool %cond, label %IfEqual, label %IfUnequal
872 <a href="#i_ret">ret</a> int 1
874 <a href="#i_ret">ret</a> int 0
879 <!-- _______________________________________________________________________ -->
880 <div class="doc_subsubsection">
881 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
884 <div class="doc_text">
889 switch uint <value>, label <defaultdest> [ int <val>, label &dest>, ... ]
894 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
895 several different places. It is a generalization of the '<tt>br</tt>'
896 instruction, allowing a branch to occur to one of many possible
901 <p>The '<tt>switch</tt>' instruction uses three parameters: a '<tt>uint</tt>'
902 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
903 an array of pairs of comparison value constants and '<tt>label</tt>'s.</p>
907 <p>The <tt>switch</tt> instruction specifies a table of values and destinations.
908 When the '<tt>switch</tt>' instruction is executed, this table is searched for
909 the given value. If the value is found, the corresponding destination is
910 branched to, otherwise the default value it transfered to.</p>
912 <h5>Implementation:</h5>
914 <p>Depending on properties of the target machine and the particular
915 <tt>switch</tt> instruction, this instruction may be code generated as a series
916 of chained conditional branches, or with a lookup table.</p>
921 <i>; Emulate a conditional br instruction</i>
922 %Val = <a href="#i_cast">cast</a> bool %value to uint
923 switch uint %Val, label %truedest [int 0, label %falsedest ]
925 <i>; Emulate an unconditional br instruction</i>
926 switch uint 0, label %dest [ ]
928 <i>; Implement a jump table:</i>
929 switch uint %val, label %otherwise [ int 0, label %onzero,
931 int 2, label %ontwo ]
936 <!-- _______________________________________________________________________ -->
937 <div class="doc_subsubsection">
938 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
941 <div class="doc_text">
946 <result> = invoke <ptr to function ty> %<function ptr val>(<function args>)
947 to label <normal label> except label <exception label>
952 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
953 function, with the possibility of control flow transfer to either the
954 '<tt>normal</tt>' <tt>label</tt> label or the '<tt>exception</tt>'
955 <tt>label</tt>. If the callee function returns with the "<tt><a
956 href="#i_ret">ret</a></tt>" instruction, control flow will return to the
957 "normal" label. If the callee (or any indirect callees) returns with the "<a
958 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted, and
959 continued at the dynamically nearest "except" label.</p>
963 <p>This instruction requires several arguments:</p>
967 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
968 function value being invoked. In most cases, this is a direct function
969 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
970 an arbitrary pointer to function value.
972 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
973 function to be invoked.
975 <li>'<tt>function args</tt>': argument list whose types match the function
976 signature argument types. If the function signature indicates the function
977 accepts a variable number of arguments, the extra arguments can be specified.
979 <li>'<tt>normal label</tt>': the label reached when the called function executes
980 a '<tt><a href="#i_ret">ret</a></tt>' instruction.
982 <li>'<tt>exception label</tt>': the label reached when a callee returns with the
983 <a href="#i_unwind"><tt>unwind</tt></a> instruction.
988 <p>This instruction is designed to operate as a standard '<tt><a
989 href="#i_call">call</a></tt>' instruction in most regards. The primary
990 difference is that it establishes an association with a label, which is used by the runtime library to unwind the stack.</p>
992 <p>This instruction is used in languages with destructors to ensure that proper
993 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
994 exception. Additionally, this is important for implementation of
995 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
1000 %retval = invoke int %Test(int 15)
1002 except label %TestCleanup <i>; {int}:retval set</i>
1007 <!-- _______________________________________________________________________ -->
1008 <div class="doc_subsubsection">
1009 <a name="i_unwind">'<tt>unwind</tt>' Instruction</a>
1012 <div class="doc_text">
1022 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
1023 at the first callee in the dynamic call stack which used an <a
1024 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
1025 primarily used to implement exception handling.</p>
1029 <p>The '<tt>unwind</tt>' intrinsic causes execution of the current function to
1030 immediately halt. The dynamic call stack is then searched for the first <a
1031 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
1032 execution continues at the "exceptional" destination block specified by the
1033 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
1034 dynamic call chain, undefined behavior results.</p>
1038 <!-- ======================================================================= -->
1039 <div class="doc_subsection">
1040 <a name="binaryops">Binary Operations</a>
1043 <div class="doc_text">
1045 <p>Binary operators are used to do most of the computation in a program. They
1046 require two operands, execute an operation on them, and produce a single value.
1047 The result value of a binary operator is not necessarily the same type as its
1050 <p>There are several different binary operators:</p>
1054 <!-- _______________________________________________________________________ -->
1055 <div class="doc_subsubsection">
1056 <a name="i_add">'<tt>add</tt>' Instruction</a>
1059 <div class="doc_text">
1064 <result> = add <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1069 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
1073 <p>The two arguments to the '<tt>add</tt>' instruction must be either <a
1074 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1075 values. Both arguments must have identical types.</p>
1079 <p>The value produced is the integer or floating point sum of the two
1085 <result> = add int 4, %var <i>; yields {int}:result = 4 + %var</i>
1090 <!-- _______________________________________________________________________ -->
1091 <div class="doc_subsubsection">
1092 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
1095 <div class="doc_text">
1100 <result> = sub <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1105 <p>The '<tt>sub</tt>' instruction returns the difference of its two
1108 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
1109 '<tt>neg</tt>' instruction present in most other intermediate
1110 representations.</p>
1114 <p>The two arguments to the '<tt>sub</tt>' instruction must be either <a
1115 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1116 values. Both arguments must have identical types.</p>
1120 <p>The value produced is the integer or floating point difference of the two
1126 <result> = sub int 4, %var <i>; yields {int}:result = 4 - %var</i>
1127 <result> = sub int 0, %val <i>; yields {int}:result = -%var</i>
1132 <!-- _______________________________________________________________________ -->
1133 <div class="doc_subsubsection">
1134 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
1137 <div class="doc_text">
1142 <result> = mul <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1147 <p>The '<tt>mul</tt>' instruction returns the product of its two operands.</p>
1151 <p>The two arguments to the '<tt>mul</tt>' instruction must be either <a
1152 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1153 values. Both arguments must have identical types.</p>
1157 <p>The value produced is the integer or floating point product of the two
1160 <p>There is no signed vs unsigned multiplication. The appropriate action is
1161 taken based on the type of the operand.</p>
1166 <result> = mul int 4, %var <i>; yields {int}:result = 4 * %var</i>
1171 <!-- _______________________________________________________________________ -->
1172 <div class="doc_subsubsection">
1173 <a name="i_div">'<tt>div</tt>' Instruction</a>
1176 <div class="doc_text">
1181 <result> = div <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1186 <p>The '<tt>div</tt>' instruction returns the quotient of its two operands.</p>
1190 <p>The two arguments to the '<tt>div</tt>' instruction must be either <a
1191 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1192 values. Both arguments must have identical types.</p>
1196 <p>The value produced is the integer or floating point quotient of the two
1202 <result> = div int 4, %var <i>; yields {int}:result = 4 / %var</i>
1207 <!-- _______________________________________________________________________ -->
1208 <div class="doc_subsubsection">
1209 <a name="i_rem">'<tt>rem</tt>' Instruction</a>
1212 <div class="doc_text">
1217 <result> = rem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1222 <p>The '<tt>rem</tt>' instruction returns the remainder from the division of its
1227 <p>The two arguments to the '<tt>rem</tt>' instruction must be either <a
1228 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1229 values. Both arguments must have identical types.</p>
1233 <p>This returns the <i>remainder</i> of a division (where the result has the
1234 same sign as the divisor), not the <i>modulus</i> (where the result has the same
1235 sign as the dividend) of a value. For more information about the difference,
1236 see: <a href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The Math
1242 <result> = rem int 4, %var <i>; yields {int}:result = 4 % %var</i>
1247 <!-- _______________________________________________________________________ -->
1248 <div class="doc_subsubsection">
1249 <a name="i_setcc">'<tt>set<i>cc</i></tt>' Instructions</a>
1252 <div class="doc_text">
1257 <result> = seteq <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1258 <result> = setne <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1259 <result> = setlt <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1260 <result> = setgt <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1261 <result> = setle <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1262 <result> = setge <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1267 <p>The '<tt>set<i>cc</i></tt>' family of instructions returns a boolean value
1268 based on a comparison of their two operands.</p>
1272 <p>The two arguments to the '<tt>set<i>cc</i></tt>' instructions must be of <a
1273 href="#t_firstclass">first class</a> type (it is not possible to compare
1274 '<tt>label</tt>'s, '<tt>array</tt>'s, '<tt>structure</tt>' or '<tt>void</tt>'
1275 values, etc...). Both arguments must have identical types.</p>
1279 <p>The '<tt>seteq</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>' value
1280 if both operands are equal.<br>
1282 The '<tt>setne</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>' value if
1283 both operands are unequal.<br>
1285 The '<tt>setlt</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>' value if
1286 the first operand is less than the second operand.<br>
1288 The '<tt>setgt</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>' value if
1289 the first operand is greater than the second operand.<br>
1291 The '<tt>setle</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>' value if
1292 the first operand is less than or equal to the second operand.<br>
1294 The '<tt>setge</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>' value if
1295 the first operand is greater than or equal to the second operand.</p>
1300 <result> = seteq int 4, 5 <i>; yields {bool}:result = false</i>
1301 <result> = setne float 4, 5 <i>; yields {bool}:result = true</i>
1302 <result> = setlt uint 4, 5 <i>; yields {bool}:result = true</i>
1303 <result> = setgt sbyte 4, 5 <i>; yields {bool}:result = false</i>
1304 <result> = setle sbyte 4, 5 <i>; yields {bool}:result = true</i>
1305 <result> = setge sbyte 4, 5 <i>; yields {bool}:result = false</i>
1310 <!-- ======================================================================= -->
1311 <div class="doc_subsection">
1312 <a name="bitwiseops">Bitwise Binary Operations</a>
1315 <div class="doc_text">
1317 <p>Bitwise binary operators are used to do various forms of bit-twiddling in a
1318 program. They are generally very efficient instructions, and can commonly be
1319 strength reduced from other instructions. They require two operands, execute an
1320 operation on them, and produce a single value. The resulting value of the
1321 bitwise binary operators is always the same type as its first operand.</p>
1325 <!-- _______________________________________________________________________ -->
1326 <div class="doc_subsubsection">
1327 <a name="i_and">'<tt>and</tt>' Instruction</a>
1330 <div class="doc_text">
1335 <result> = and <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1340 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of its two
1345 <p>The two arguments to the '<tt>and</tt>' instruction must be <a
1346 href="#t_integral">integral</a> values. Both arguments must have identical
1351 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
1355 <table border="1" cellspacing="0" cellpadding="4">
1356 <tr><td>In0</td> <td>In1</td> <td>Out</td></tr>
1357 <tr><td>0</td> <td>0</td> <td>0</td></tr>
1358 <tr><td>0</td> <td>1</td> <td>0</td></tr>
1359 <tr><td>1</td> <td>0</td> <td>0</td></tr>
1360 <tr><td>1</td> <td>1</td> <td>1</td></tr>
1367 <result> = and int 4, %var <i>; yields {int}:result = 4 & %var</i>
1368 <result> = and int 15, 40 <i>; yields {int}:result = 8</i>
1369 <result> = and int 4, 8 <i>; yields {int}:result = 0</i>
1374 <!-- _______________________________________________________________________ -->
1375 <div class="doc_subsubsection">
1376 <a name="i_or">'<tt>or</tt>' Instruction</a>
1379 <div class="doc_text">
1384 <result> = or <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1389 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive or of its
1394 <p>The two arguments to the '<tt>or</tt>' instruction must be <a
1395 href="#t_integral">integral</a> values. Both arguments must have identical
1400 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
1403 <center><table border="1" cellspacing="0" cellpadding="4">
1404 <tr><td>In0</td> <td>In1</td> <td>Out</td></tr>
1405 <tr><td>0</td> <td>0</td> <td>0</td></tr>
1406 <tr><td>0</td> <td>1</td> <td>1</td></tr>
1407 <tr><td>1</td> <td>0</td> <td>1</td></tr>
1408 <tr><td>1</td> <td>1</td> <td>1</td></tr>
1415 <result> = or int 4, %var <i>; yields {int}:result = 4 | %var</i>
1416 <result> = or int 15, 40 <i>; yields {int}:result = 47</i>
1417 <result> = or int 4, 8 <i>; yields {int}:result = 12</i>
1422 <!-- _______________________________________________________________________ -->
1423 <div class="doc_subsubsection">
1424 <a name="i_xor">'<tt>xor</tt>' Instruction</a>
1427 <div class="doc_text">
1432 <result> = xor <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1437 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive or of
1438 its two operands. The <tt>xor</tt> is used to implement the "one's complement"
1439 operation, which is the "~" operator in C.</p>
1443 <p>The two arguments to the '<tt>xor</tt>' instruction must be <a
1444 href="#t_integral">integral</a> values. Both arguments must have identical
1449 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
1452 <center><table border="1" cellspacing="0" cellpadding="4">
1453 <tr><td>In0</td> <td>In1</td> <td>Out</td></tr>
1454 <tr><td>0</td> <td>0</td> <td>0</td></tr>
1455 <tr><td>0</td> <td>1</td> <td>1</td></tr>
1456 <tr><td>1</td> <td>0</td> <td>1</td></tr>
1457 <tr><td>1</td> <td>1</td> <td>0</td></tr>
1464 <result> = xor int 4, %var <i>; yields {int}:result = 4 ^ %var</i>
1465 <result> = xor int 15, 40 <i>; yields {int}:result = 39</i>
1466 <result> = xor int 4, 8 <i>; yields {int}:result = 12</i>
1467 <result> = xor int %V, -1 <i>; yields {int}:result = ~%V</i>
1472 <!-- _______________________________________________________________________ -->
1473 <div class="doc_subsubsection">
1474 <a name="i_shl">'<tt>shl</tt>' Instruction</a>
1477 <div class="doc_text">
1482 <result> = shl <ty> <var1>, ubyte <var2> <i>; yields {ty}:result</i>
1487 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to the left
1488 a specified number of bits.</p>
1492 <p>The first argument to the '<tt>shl</tt>' instruction must be an <a
1493 href="#t_integer">integer</a> type. The second argument must be an
1494 '<tt>ubyte</tt>' type.</p>
1498 <p>The value produced is <tt>var1</tt> * 2<sup><tt>var2</tt></sup>.</p>
1503 <result> = shl int 4, ubyte %var <i>; yields {int}:result = 4 << %var</i>
1504 <result> = shl int 4, ubyte 2 <i>; yields {int}:result = 16</i>
1505 <result> = shl int 1, ubyte 10 <i>; yields {int}:result = 1024</i>
1510 <!-- _______________________________________________________________________ -->
1511 <div class="doc_subsubsection">
1512 <a name="i_shr">'<tt>shr</tt>' Instruction</a>
1515 <div class="doc_text">
1520 <result> = shr <ty> <var1>, ubyte <var2> <i>; yields {ty}:result</i>
1525 <p>The '<tt>shr</tt>' instruction returns the first operand shifted to the right
1526 a specified number of bits.</p>
1530 <p>The first argument to the '<tt>shr</tt>' instruction must be an <a
1531 href="#t_integer">integer</a> type. The second argument must be an
1532 '<tt>ubyte</tt>' type.</p>
1536 <p>If the first argument is a <a href="#t_signed">signed</a> type, the most
1537 significant bit is duplicated in the newly free'd bit positions. If the first
1538 argument is unsigned, zero bits shall fill the empty positions.</p>
1543 <result> = shr int 4, ubyte %var <i>; yields {int}:result = 4 >> %var</i>
1544 <result> = shr uint 4, ubyte 1 <i>; yields {uint}:result = 2</i>
1545 <result> = shr int 4, ubyte 2 <i>; yields {int}:result = 1</i>
1546 <result> = shr sbyte 4, ubyte 3 <i>; yields {sbyte}:result = 0</i>
1547 <result> = shr sbyte -2, ubyte 1 <i>; yields {sbyte}:result = -1</i>
1552 <!-- ======================================================================= -->
1553 <div class="doc_subsection">
1554 <a name="memoryops">Memory Access Operations</div>
1557 <div class="doc_text">
1559 <p>A key design point of an SSA-based representation is how it represents
1560 memory. In LLVM, no memory locations are in SSA form, which makes things very
1561 simple. This section describes how to read, write, allocate and free memory in
1566 <!-- _______________________________________________________________________ -->
1567 <div class="doc_subsubsection">
1568 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
1571 <div class="doc_text">
1576 <result> = malloc <type>, uint <NumElements> <i>; yields {type*}:result</i>
1577 <result> = malloc <type> <i>; yields {type*}:result</i>
1582 <p>The '<tt>malloc</tt>' instruction allocates memory from the system heap and
1583 returns a pointer to it.</p>
1587 <p>The the '<tt>malloc</tt>' instruction allocates
1588 <tt>sizeof(<type>)*NumElements</tt> bytes of memory from the operating
1589 system, and returns a pointer of the appropriate type to the program. The
1590 second form of the instruction is a shorter version of the first instruction
1591 that defaults to allocating one element.</p>
1593 <p>'<tt>type</tt>' must be a sized type.</p>
1597 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and a
1598 pointer is returned.</p>
1603 %array = malloc [4 x ubyte ] <i>; yields {[%4 x ubyte]*}:array</i>
1605 %size = <a href="#i_add">add</a> uint 2, 2 <i>; yields {uint}:size = uint 4</i>
1606 %array1 = malloc ubyte, uint 4 <i>; yields {ubyte*}:array1</i>
1607 %array2 = malloc [12 x ubyte], uint %size <i>; yields {[12 x ubyte]*}:array2</i>
1612 <!-- _______________________________________________________________________ -->
1613 <div class="doc_subsubsection">
1614 <a name="i_free">'<tt>free</tt>' Instruction</a>
1617 <div class="doc_text">
1622 free <type> <value> <i>; yields {void}</i>
1627 <p>The '<tt>free</tt>' instruction returns memory back to the unused memory
1628 heap, to be reallocated in the future.<p>
1632 <p>'<tt>value</tt>' shall be a pointer value that points to a value that was
1633 allocated with the '<tt><a href="#i_malloc">malloc</a></tt>' instruction.</p>
1637 <p>Access to the memory pointed to by the pointer is not longer defined after
1638 this instruction executes.</p>
1642 %array = <a href="#i_malloc">malloc</a> [4 x ubyte] <i>; yields {[4 x ubyte]*}:array</i>
1643 free [4 x ubyte]* %array
1648 <!-- _______________________________________________________________________ -->
1649 <div class="doc_subsubsection">
1650 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
1653 <div class="doc_text">
1658 <result> = alloca <type>, uint <NumElements> <i>; yields {type*}:result</i>
1659 <result> = alloca <type> <i>; yields {type*}:result</i>
1664 <p>The '<tt>alloca</tt>' instruction allocates memory on the current stack frame
1665 of the procedure that is live until the current function returns to its
1670 <p>The the '<tt>alloca</tt>' instruction allocates
1671 <tt>sizeof(<type>)*NumElements</tt> bytes of memory on the runtime stack,
1672 returning a pointer of the appropriate type to the program. The second form of
1673 the instruction is a shorter version of the first that defaults to allocating
1676 <p>'<tt>type</tt>' may be any sized type.</p>
1680 <p>Memory is allocated, a pointer is returned. '<tt>alloca</tt>'d memory is
1681 automatically released when the function returns. The '<tt>alloca</tt>'
1682 instruction is commonly used to represent automatic variables that must have an
1683 address available. When the function returns (either with the <tt><a
1684 href="#i_ret">ret</a></tt> or <tt><a href="#i_invoke">invoke</a></tt>
1685 instructions), the memory is reclaimed.</p>
1690 %ptr = alloca int <i>; yields {int*}:ptr</i>
1691 %ptr = alloca int, uint 4 <i>; yields {int*}:ptr</i>
1696 <!-- _______________________________________________________________________ -->
1697 <div class="doc_subsubsection">
1698 <a name="i_load">'<tt>load</tt>' Instruction</a>
1701 <div class="doc_text">
1706 <result> = load <ty>* <pointer>
1707 <result> = volatile load <ty>* <pointer>
1712 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
1716 <p>The argument to the '<tt>load</tt>' instruction specifies the memory address
1717 to load from. The pointer must point to a <a href="t_firstclass">first
1718 class</a> type. If the <tt>load</tt> is marked as <tt>volatile</tt> then the
1719 optimizer is not allowed to modify the number or order of execution of this
1720 <tt>load</tt> with other volatile <tt>load</tt> and <tt><a
1721 href="#i_store">store</a></tt> instructions. </p>
1725 <p>The location of memory pointed to is loaded.</p>
1730 %ptr = <a href="#i_alloca">alloca</a> int <i>; yields {int*}:ptr</i>
1731 <a href="#i_store">store</a> int 3, int* %ptr <i>; yields {void}</i>
1732 %val = load int* %ptr <i>; yields {int}:val = int 3</i>
1737 <!-- _______________________________________________________________________ -->
1738 <div class="doc_subsubsection">
1739 <a name="i_store">'<tt>store</tt>' Instruction</a>
1745 store <ty> <value>, <ty>* <pointer> <i>; yields {void}</i>
1746 volatile store <ty> <value>, <ty>* <pointer> <i>; yields {void}</i>
1751 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
1755 <p>There are two arguments to the '<tt>store</tt>' instruction: a value to store
1756 and an address to store it into. The type of the '<tt><pointer></tt>'
1757 operand must be a pointer to the type of the '<tt><value></tt>' operand.
1758 If the <tt>store</tt> is marked as <tt>volatile</tt> then the optimizer is not
1759 allowed to modify the number or order of execution of this <tt>store</tt> with
1760 other volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
1765 <p>The contents of memory are updated to contain '<tt><value></tt>' at the
1766 location specified by the '<tt><pointer></tt>' operand.</p>
1771 %ptr = <a href="#i_alloca">alloca</a> int <i>; yields {int*}:ptr</i>
1772 <a href="#i_store">store</a> int 3, int* %ptr <i>; yields {void}</i>
1773 %val = load int* %ptr <i>; yields {int}:val = int 3</i>
1778 <!-- _______________________________________________________________________ -->
1779 <div class="doc_subsubsection">
1780 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
1783 <div class="doc_text">
1788 <result> = getelementptr <ty>* <ptrval>{, long <aidx>|, ubyte <sidx>}*
1793 <p>The '<tt>getelementptr</tt>' instruction is used to get the address of a
1794 subelement of an aggregate data structure.</p>
1798 <p>This instruction takes a list of <tt>long</tt> values and <tt>ubyte</tt>
1799 constants that indicate what form of addressing to perform. The actual types of
1800 the arguments provided depend on the type of the first pointer argument. The
1801 '<tt>getelementptr</tt>' instruction is used to index down through the type
1802 levels of a structure.</p>
1804 <p>For example, let's consider a C code fragment and how it gets compiled to
1819 int *foo(struct ST *s) {
1820 return &s[1].Z.B[5][13];
1824 <p>The LLVM code generated by the GCC frontend is:</p>
1827 %RT = type { sbyte, [10 x [20 x int]], sbyte }
1828 %ST = type { int, double, %RT }
1830 int* "foo"(%ST* %s) {
1831 %reg = getelementptr %ST* %s, long 1, ubyte 2, ubyte 1, long 5, long 13
1838 <p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
1839 on the pointer type that is being index into. <a href="t_pointer">Pointer</a>
1840 and <a href="t_array">array</a> types require '<tt>long</tt>' values, and <a
1841 href="t_struct">structure</a> types require '<tt>ubyte</tt>'
1842 <b>constants</b>.</p>
1844 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
1845 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ int, double, %RT
1846 }</tt>' type, a structure. The second index indexes into the third element of
1847 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ sbyte, [10 x [20 x int]],
1848 sbyte }</tt>' type, another structure. The third index indexes into the second
1849 element of the structure, yielding a '<tt>[10 x [20 x int]]</tt>' type, an
1850 array. The two dimensions of the array are subscripted into, yielding an
1851 '<tt>int</tt>' type. The '<tt>getelementptr</tt>' instruction return a pointer
1852 to this element, thus yielding a '<tt>int*</tt>' type.</p>
1854 <p>Note that it is perfectly legal to index partially through a structure,
1855 returning a pointer to an inner element. Because of this, the LLVM code for the
1856 given testcase is equivalent to:</p>
1859 int* "foo"(%ST* %s) {
1860 %t1 = getelementptr %ST* %s , long 1 <i>; yields %ST*:%t1</i>
1861 %t2 = getelementptr %ST* %t1, long 0, ubyte 2 <i>; yields %RT*:%t2</i>
1862 %t3 = getelementptr %RT* %t2, long 0, ubyte 1 <i>; yields [10 x [20 x int]]*:%t3</i>
1863 %t4 = getelementptr [10 x [20 x int]]* %t3, long 0, long 5 <i>; yields [20 x int]*:%t4</i>
1864 %t5 = getelementptr [20 x int]* %t4, long 0, long 13 <i>; yields int*:%t5</i>
1872 <i>; yields [12 x ubyte]*:aptr</i>
1873 %aptr = getelementptr {int, [12 x ubyte]}* %sptr, long 0, ubyte 1
1878 <!-- ======================================================================= -->
1879 <div class="doc_subsection">
1880 <a name="otherops">Other Operations</a>
1883 <div class="doc_text">
1885 <p>The instructions in this catagory are the "miscellaneous" instructions, which
1886 defy better classification.</p>
1890 <!-- _______________________________________________________________________ -->
1891 <div class="doc_subsubsection">
1892 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
1895 <div class="doc_text">
1900 <result> = phi <ty> [ <val0>, <label0>], ...
1905 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in the SSA
1906 graph representing the function.</p>
1910 <p>The type of the incoming values are specified with the first type field.
1911 After this, the '<tt>phi</tt>' instruction takes a list of pairs as arguments,
1912 with one pair for each predecessor basic block of the current block. Only
1913 values of <a href="#t_firstclass">first class</a> type may be used as the value
1914 arguments to the PHI node. Only labels may be used as the label arguments.</p>
1916 <p>There must be no non-phi instructions between the start of a basic block and
1917 the PHI instructions: i.e. PHI instructions must be first in a basic block.</p>
1921 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
1922 specified by the parameter, depending on which basic block we came from in the
1923 last <a href="#terminators">terminator</a> instruction.</p>
1928 Loop: ; Infinite loop that counts from 0 on up...
1929 %indvar = phi uint [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
1930 %nextindvar = add uint %indvar, 1
1936 <!-- _______________________________________________________________________ -->
1937 <div class="doc_subsubsection">
1938 <a name="i_cast">'<tt>cast .. to</tt>' Instruction</a>
1941 <div class="doc_text">
1946 <result> = cast <ty> <value> to <ty2> <i>; yields ty2</i>
1951 <p>The '<tt>cast</tt>' instruction is used as the primitive means to convert
1952 integers to floating point, change data type sizes, and break type safety (by
1953 casting pointers).</p>
1957 <p>The '<tt>cast</tt>' instruction takes a value to cast, which must be a first
1958 class value, and a type to cast it to, which must also be a <a
1959 href="#t_firstclass">first class</a> type.</p>
1963 <p>This instruction follows the C rules for explicit casts when determining how
1964 the data being cast must change to fit in its new container.</p>
1966 <p>When casting to bool, any value that would be considered true in the context
1967 of a C '<tt>if</tt>' condition is converted to the boolean '<tt>true</tt>'
1968 values, all else are '<tt>false</tt>'.</p>
1970 <p>When extending an integral value from a type of one signness to another (for
1971 example '<tt>sbyte</tt>' to '<tt>ulong</tt>'), the value is sign-extended if the
1972 <b>source</b> value is signed, and zero-extended if the source value is
1973 unsigned. <tt>bool</tt> values are always zero extended into either zero or
1979 %X = cast int 257 to ubyte <i>; yields ubyte:1</i>
1980 %Y = cast int 123 to bool <i>; yields bool:true</i>
1985 <!-- _______________________________________________________________________ -->
1986 <div class="doc_subsubsection">
1987 <a name="i_call">'<tt>call</tt>' Instruction</a>
1990 <div class="doc_text">
1995 <result> = call <ty>* <fnptrval>(<param list>)
2000 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
2004 <p>This instruction requires several arguments:</p>
2008 <li><p>'<tt>ty</tt>': shall be the signature of the pointer to function value
2009 being invoked. The argument types must match the types implied by this
2012 <li><p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function
2013 to be invoked. In most cases, this is a direct function invocation, but
2014 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer to
2015 function values.</p></li>
2017 <li><p>'<tt>function args</tt>': argument list whose types match the function
2018 signature argument types. If the function signature indicates the function
2019 accepts a variable number of arguments, the extra arguments can be
2026 <p>The '<tt>call</tt>' instruction is used to cause control flow to transfer to
2027 a specified function, with its incoming arguments bound to the specified values.
2028 Upon a '<tt><a href="#i_ret">ret</a></tt>' instruction in the called function,
2029 control flow continues with the instruction after the function call, and the
2030 return value of the function is bound to the result argument. This is a simpler
2031 case of the <a href="#i_invoke">invoke</a> instruction.</p>
2036 %retval = call int %test(int %argc)
2037 call int(sbyte*, ...) *%printf(sbyte* %msg, int 12, sbyte 42);
2042 <!-- _______________________________________________________________________ -->
2043 <div class="doc_subsubsection">
2044 <a name="i_vanext">'<tt>vanext</tt>' Instruction</a>
2047 <div class="doc_text">
2052 <resultarglist> = vanext <va_list> <arglist>, <argty>
2057 <p>The '<tt>vanext</tt>' instruction is used to access arguments passed through
2058 the "variable argument" area of a function call. It is used to implement the
2059 <tt>va_arg</tt> macro in C.</p>
2063 <p>This instruction takes a <tt>valist</tt> value and the type of the argument.
2064 It returns another <tt>valist</tt>.</p>
2068 <p>The '<tt>vanext</tt>' instruction advances the specified <tt>valist</tt> past
2069 an argument of the specified type. In conjunction with the <a
2070 href="#i_vaarg"><tt>vaarg</tt></a> instruction, it is used to implement the
2071 <tt>va_arg</tt> macro available in C. For more information, see the variable
2072 argument handling <a href="#int_varargs">Intrinsic Functions</a>.</p>
2074 <p>It is legal for this instruction to be called in a function which does not
2075 take a variable number of arguments, for example, the <tt>vfprintf</tt>
2078 <p><tt>vanext</tt> is an LLVM instruction instead of an <a
2079 href="#intrinsics">intrinsic function</a> because it takes an type as an
2084 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
2088 <!-- _______________________________________________________________________ -->
2089 <div class="doc_subsubsection">
2090 <a name="i_vaarg">'<tt>vaarg</tt>' Instruction</a>
2093 <div class="doc_text">
2098 <resultval> = vaarg <va_list> <arglist>, <argty>
2103 <p>The '<tt>vaarg</tt>' instruction is used to access arguments passed through
2104 the "variable argument" area of a function call. It is used to implement the
2105 <tt>va_arg</tt> macro in C.</p>
2109 <p>This instruction takes a <tt>valist</tt> value and the type of the argument.
2110 It returns a value of the specified argument type.</p>
2114 <p>The '<tt>vaarg</tt>' instruction loads an argument of the specified type from
2115 the specified <tt>va_list</tt>. In conjunction with the <a
2116 href="#i_vanext"><tt>vanext</tt></a> instruction, it is used to implement the
2117 <tt>va_arg</tt> macro available in C. For more information, see the variable
2118 argument handling <a href="#int_varargs">Intrinsic Functions</a>.</p>
2120 <p>It is legal for this instruction to be called in a function which does not
2121 take a variable number of arguments, for example, the <tt>vfprintf</tt>
2124 <p><tt>vaarg</tt> is an LLVM instruction instead of an <a
2125 href="#intrinsics">intrinsic function</a> because it takes an type as an
2130 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
2134 <!-- *********************************************************************** -->
2135 <div class="doc_section">
2136 <a name="intrinsics">Intrinsic Functions</a>
2138 <!-- *********************************************************************** -->
2140 <div class="doc_text">
2142 <p>LLVM supports the notion of an "intrinsic function". These functions have
2143 well known names and semantics, and are required to follow certain restrictions.
2144 Overall, these instructions represent an extension mechanism for the LLVM
2145 language that does not require changing all of the transformations in LLVM to
2146 add to the language (or the bytecode reader/writer, the parser, etc...).</p>
2148 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix, this
2149 prefix is reserved in LLVM for intrinsic names, thus functions may not be named
2150 this. Intrinsic functions must always be external functions: you cannot define
2151 the body of intrinsic functions. Intrinsic functions may only be used in call
2152 or invoke instructions: it is illegal to take the address of an intrinsic
2153 function. Additionally, because intrinsic functions are part of the LLVM
2154 language, it is required that they all be documented here if any are added.</p>
2156 <p>Unless an intrinsic function is target-specific, there must be a lowering
2157 pass to eliminate the intrinsic or all backends must support the intrinsic
2162 <!-- ======================================================================= -->
2163 <div class="doc_subsection">
2164 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
2167 <div class="doc_text">
2169 <p>Variable argument support is defined in LLVM with the <a
2170 href="#i_vanext"><tt>vanext</tt></a> instruction and these three intrinsic
2171 functions. These functions are related to the similarly named macros defined in
2172 the <tt><stdarg.h></tt> header file.</p>
2174 <p>All of these functions operate on arguments that use a target-specific value
2175 type "<tt>va_list</tt>". The LLVM assembly language reference manual does not
2176 define what this type is, so all transformations should be prepared to handle
2177 intrinsics with any type used.</p>
2179 <p>This example shows how the <a href="#i_vanext"><tt>vanext</tt></a>
2180 instruction and the variable argument handling intrinsic functions are used.</p>
2183 int %test(int %X, ...) {
2184 ; Initialize variable argument processing
2185 %ap = call sbyte*()* %<a href="#i_va_start">llvm.va_start</a>()
2187 ; Read a single integer argument
2188 %tmp = vaarg sbyte* %ap, int
2190 ; Advance to the next argument
2191 %ap2 = vanext sbyte* %ap, int
2193 ; Demonstrate usage of llvm.va_copy and llvm.va_end
2194 %aq = call sbyte* (sbyte*)* %<a href="#i_va_copy">llvm.va_copy</a>(sbyte* %ap2)
2195 call void %<a href="#i_va_end">llvm.va_end</a>(sbyte* %aq)
2197 ; Stop processing of arguments.
2198 call void %<a href="#i_va_end">llvm.va_end</a>(sbyte* %ap2)
2205 <!-- _______________________________________________________________________ -->
2206 <div class="doc_subsubsection">
2207 <a name="i_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
2210 <div class="doc_text">
2215 call va_list ()* %llvm.va_start()
2220 <p>The '<tt>llvm.va_start</tt>' intrinsic returns a new <tt><arglist></tt>
2221 for subsequent use by the variable argument intrinsics.</p>
2225 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
2226 macro available in C. In a target-dependent way, it initializes and returns a
2227 <tt>va_list</tt> element, so that the next <tt>vaarg</tt> will produce the first
2228 variable argument passed to the function. Unlike the C <tt>va_start</tt> macro,
2229 this intrinsic does not need to know the last argument of the function, the
2230 compiler can figure that out.</p>
2232 <p>Note that this intrinsic function is only legal to be called from within the
2233 body of a variable argument function.</p>
2237 <!-- _______________________________________________________________________ -->
2238 <div class="doc_subsubsection">
2239 <a name="i_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
2242 <div class="doc_text">
2247 call void (va_list)* %llvm.va_end(va_list <arglist>)
2252 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt><arglist></tt> which
2253 has been initialized previously with <tt><a
2254 href="#i_va_start">llvm.va_start</a></tt> or <tt><a
2255 href="#i_va_copy">llvm.va_copy</a></tt>.</p>
2259 <p>The argument is a <tt>va_list</tt> to destroy.</p>
2263 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
2264 macro available in C. In a target-dependent way, it destroys the
2265 <tt>va_list</tt>. Calls to <a href="#i_va_start"><tt>llvm.va_start</tt></a> and
2266 <a href="#i_va_copy"><tt>llvm.va_copy</tt></a> must be matched exactly with
2267 calls to <tt>llvm.va_end</tt>.</p>
2271 <!-- _______________________________________________________________________ -->
2272 <div class="doc_subsubsection">
2273 <a name="i_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
2276 <div class="doc_text">
2281 call va_list (va_list)* %llvm.va_copy(va_list <destarglist>)
2286 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
2287 from the source argument list to the destination argument list.</p>
2291 <p>The argument is the <tt>va_list</tt> to copy.</p>
2295 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
2296 macro available in C. In a target-dependent way, it copies the source
2297 <tt>va_list</tt> element into the returned list. This intrinsic is necessary
2298 because the <tt><a href="i_va_start">llvm.va_start</a></tt> intrinsic may be
2299 arbitrarily complex and require memory allocation, for example.</p>
2303 <!-- *********************************************************************** -->
2306 <div class="doc_footer">
2307 <address><a href="mailto:sabre@nondot.org">Chris Lattner</a></address>
2308 <a href="http://llvm.cs.uiuc.edu">The LLVM Compiler Infrastructure</a>
2310 Last modified: $Date$