blob: 515f956ed00af196ed413edf7cd5af19bdf00840 [file] [log] [blame]
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
3<html>
4<head>
5 <title>LLVM Assembly Language Reference Manual</title>
6 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
7 <meta name="author" content="Chris Lattner">
8 <meta name="description"
9 content="LLVM Assembly Language Reference Manual.">
10 <link rel="stylesheet" href="llvm.css" type="text/css">
11</head>
12
13<body>
14
15<div class="doc_title"> LLVM Language Reference Manual </div>
16<ol>
17 <li><a href="#abstract">Abstract</a></li>
18 <li><a href="#introduction">Introduction</a></li>
19 <li><a href="#identifiers">Identifiers</a></li>
20 <li><a href="#highlevel">High Level Structure</a>
21 <ol>
22 <li><a href="#modulestructure">Module Structure</a></li>
23 <li><a href="#linkage">Linkage Types</a></li>
24 <li><a href="#callingconv">Calling Conventions</a></li>
25 <li><a href="#globalvars">Global Variables</a></li>
26 <li><a href="#functionstructure">Functions</a></li>
27 <li><a href="#aliasstructure">Aliases</a>
28 <li><a href="#paramattrs">Parameter Attributes</a></li>
Gordon Henriksen13fe5e32007-12-10 03:18:06 +000029 <li><a href="#gc">Garbage Collector Names</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +000030 <li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
31 <li><a href="#datalayout">Data Layout</a></li>
32 </ol>
33 </li>
34 <li><a href="#typesystem">Type System</a>
35 <ol>
Chris Lattner488772f2008-01-04 04:32:38 +000036 <li><a href="#t_classifications">Type Classifications</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +000037 <li><a href="#t_primitive">Primitive Types</a>
38 <ol>
Chris Lattner488772f2008-01-04 04:32:38 +000039 <li><a href="#t_floating">Floating Point Types</a></li>
40 <li><a href="#t_void">Void Type</a></li>
41 <li><a href="#t_label">Label Type</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +000042 </ol>
43 </li>
44 <li><a href="#t_derived">Derived Types</a>
45 <ol>
Chris Lattner251ab812007-12-18 06:18:21 +000046 <li><a href="#t_integer">Integer Type</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +000047 <li><a href="#t_array">Array Type</a></li>
48 <li><a href="#t_function">Function Type</a></li>
49 <li><a href="#t_pointer">Pointer Type</a></li>
50 <li><a href="#t_struct">Structure Type</a></li>
51 <li><a href="#t_pstruct">Packed Structure Type</a></li>
52 <li><a href="#t_vector">Vector Type</a></li>
53 <li><a href="#t_opaque">Opaque Type</a></li>
54 </ol>
55 </li>
56 </ol>
57 </li>
58 <li><a href="#constants">Constants</a>
59 <ol>
60 <li><a href="#simpleconstants">Simple Constants</a>
61 <li><a href="#aggregateconstants">Aggregate Constants</a>
62 <li><a href="#globalconstants">Global Variable and Function Addresses</a>
63 <li><a href="#undefvalues">Undefined Values</a>
64 <li><a href="#constantexprs">Constant Expressions</a>
65 </ol>
66 </li>
67 <li><a href="#othervalues">Other Values</a>
68 <ol>
69 <li><a href="#inlineasm">Inline Assembler Expressions</a>
70 </ol>
71 </li>
72 <li><a href="#instref">Instruction Reference</a>
73 <ol>
74 <li><a href="#terminators">Terminator Instructions</a>
75 <ol>
76 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
77 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
78 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
79 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
80 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
81 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
82 </ol>
83 </li>
84 <li><a href="#binaryops">Binary Operations</a>
85 <ol>
86 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
87 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
88 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
89 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
90 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
91 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
92 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
93 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
94 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
95 </ol>
96 </li>
97 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
98 <ol>
99 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
100 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
101 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
102 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
103 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
104 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
105 </ol>
106 </li>
107 <li><a href="#vectorops">Vector Operations</a>
108 <ol>
109 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
110 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
111 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
112 </ol>
113 </li>
114 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
115 <ol>
116 <li><a href="#i_malloc">'<tt>malloc</tt>' Instruction</a></li>
117 <li><a href="#i_free">'<tt>free</tt>' Instruction</a></li>
118 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
119 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
120 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
121 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
122 </ol>
123 </li>
124 <li><a href="#convertops">Conversion Operations</a>
125 <ol>
126 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
127 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
128 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
129 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
130 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
131 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
132 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
133 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
134 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
135 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
136 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
137 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
138 </ol>
139 <li><a href="#otherops">Other Operations</a>
140 <ol>
141 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
142 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
143 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
144 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
145 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
146 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
147 </ol>
148 </li>
149 </ol>
150 </li>
151 <li><a href="#intrinsics">Intrinsic Functions</a>
152 <ol>
153 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
154 <ol>
155 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
156 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
157 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
158 </ol>
159 </li>
160 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
161 <ol>
162 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
163 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
164 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
165 </ol>
166 </li>
167 <li><a href="#int_codegen">Code Generator Intrinsics</a>
168 <ol>
169 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
170 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
171 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
172 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
173 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
174 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
175 <li><a href="#int_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
176 </ol>
177 </li>
178 <li><a href="#int_libc">Standard C Library Intrinsics</a>
179 <ol>
180 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
181 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
182 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
183 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
184 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
Dan Gohman361079c2007-10-15 20:30:11 +0000185 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
186 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
187 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000188 </ol>
189 </li>
190 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
191 <ol>
192 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
193 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
194 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
195 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
196 <li><a href="#int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic </a></li>
197 <li><a href="#int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic </a></li>
198 </ol>
199 </li>
200 <li><a href="#int_debugger">Debugger intrinsics</a></li>
201 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
Duncan Sands7407a9f2007-09-11 14:10:23 +0000202 <li><a href="#int_trampoline">Trampoline Intrinsic</a>
Duncan Sands38947cd2007-07-27 12:58:54 +0000203 <ol>
204 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
Duncan Sands38947cd2007-07-27 12:58:54 +0000205 </ol>
206 </li>
Andrew Lenharth785610d2008-02-16 01:24:58 +0000207 <li><a href="#int_atomics">Atomic intrinsics</a>
208 <ol>
209 <li><a href="#int_memory_barrier"><tt>llvm.memory_barrier</tt></li>
210 </ol>
211 </li>
Reid Spencerb043f672007-07-20 19:59:11 +0000212 <li><a href="#int_general">General intrinsics</a>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000213 <ol>
Reid Spencerb043f672007-07-20 19:59:11 +0000214 <li><a href="#int_var_annotation">
Tanya Lattner51369f32007-09-22 00:01:26 +0000215 <tt>llvm.var.annotation</tt>' Intrinsic</a></li>
Tanya Lattnerb306a9e2007-09-21 22:59:12 +0000216 <li><a href="#int_annotation">
Tanya Lattner51369f32007-09-22 00:01:26 +0000217 <tt>llvm.annotation.*</tt>' Intrinsic</a></li>
Anton Korobeynikove6e764f2008-01-15 22:31:34 +0000218 <li><a href="#int_trap">
219 <tt>llvm.trap</tt>' Intrinsic</a></li>
Tanya Lattnerb306a9e2007-09-21 22:59:12 +0000220 </ol>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000221 </li>
222 </ol>
223 </li>
224</ol>
225
226<div class="doc_author">
227 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
228 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
229</div>
230
231<!-- *********************************************************************** -->
232<div class="doc_section"> <a name="abstract">Abstract </a></div>
233<!-- *********************************************************************** -->
234
235<div class="doc_text">
236<p>This document is a reference manual for the LLVM assembly language.
237LLVM is an SSA based representation that provides type safety,
238low-level operations, flexibility, and the capability of representing
239'all' high-level languages cleanly. It is the common code
240representation used throughout all phases of the LLVM compilation
241strategy.</p>
242</div>
243
244<!-- *********************************************************************** -->
245<div class="doc_section"> <a name="introduction">Introduction</a> </div>
246<!-- *********************************************************************** -->
247
248<div class="doc_text">
249
250<p>The LLVM code representation is designed to be used in three
251different forms: as an in-memory compiler IR, as an on-disk bitcode
252representation (suitable for fast loading by a Just-In-Time compiler),
253and as a human readable assembly language representation. This allows
254LLVM to provide a powerful intermediate representation for efficient
255compiler transformations and analysis, while providing a natural means
256to debug and visualize the transformations. The three different forms
257of LLVM are all equivalent. This document describes the human readable
258representation and notation.</p>
259
260<p>The LLVM representation aims to be light-weight and low-level
261while being expressive, typed, and extensible at the same time. It
262aims to be a "universal IR" of sorts, by being at a low enough level
263that high-level ideas may be cleanly mapped to it (similar to how
264microprocessors are "universal IR's", allowing many source languages to
265be mapped to them). By providing type information, LLVM can be used as
266the target of optimizations: for example, through pointer analysis, it
267can be proven that a C automatic variable is never accessed outside of
268the current function... allowing it to be promoted to a simple SSA
269value instead of a memory location.</p>
270
271</div>
272
273<!-- _______________________________________________________________________ -->
274<div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
275
276<div class="doc_text">
277
278<p>It is important to note that this document describes 'well formed'
279LLVM assembly language. There is a difference between what the parser
280accepts and what is considered 'well formed'. For example, the
281following instruction is syntactically okay, but not well formed:</p>
282
283<div class="doc_code">
284<pre>
285%x = <a href="#i_add">add</a> i32 1, %x
286</pre>
287</div>
288
289<p>...because the definition of <tt>%x</tt> does not dominate all of
290its uses. The LLVM infrastructure provides a verification pass that may
291be used to verify that an LLVM module is well formed. This pass is
292automatically run by the parser after parsing input assembly and by
293the optimizer before it outputs bitcode. The violations pointed out
294by the verifier pass indicate bugs in transformation passes or input to
295the parser.</p>
296</div>
297
Chris Lattnera83fdc02007-10-03 17:34:29 +0000298<!-- Describe the typesetting conventions here. -->
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000299
300<!-- *********************************************************************** -->
301<div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
302<!-- *********************************************************************** -->
303
304<div class="doc_text">
305
Reid Spencerc8245b02007-08-07 14:34:28 +0000306 <p>LLVM identifiers come in two basic types: global and local. Global
307 identifiers (functions, global variables) begin with the @ character. Local
308 identifiers (register names, types) begin with the % character. Additionally,
309 there are three different formats for identifiers, for different purposes:
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000310
311<ol>
Reid Spencerc8245b02007-08-07 14:34:28 +0000312 <li>Named values are represented as a string of characters with their prefix.
313 For example, %foo, @DivisionByZero, %a.really.long.identifier. The actual
314 regular expression used is '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000315 Identifiers which require other characters in their names can be surrounded
Reid Spencerc8245b02007-08-07 14:34:28 +0000316 with quotes. In this way, anything except a <tt>&quot;</tt> character can
317 be used in a named value.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000318
Reid Spencerc8245b02007-08-07 14:34:28 +0000319 <li>Unnamed values are represented as an unsigned numeric value with their
320 prefix. For example, %12, @2, %44.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000321
322 <li>Constants, which are described in a <a href="#constants">section about
323 constants</a>, below.</li>
324</ol>
325
Reid Spencerc8245b02007-08-07 14:34:28 +0000326<p>LLVM requires that values start with a prefix for two reasons: Compilers
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000327don't need to worry about name clashes with reserved words, and the set of
328reserved words may be expanded in the future without penalty. Additionally,
329unnamed identifiers allow a compiler to quickly come up with a temporary
330variable without having to avoid symbol table conflicts.</p>
331
332<p>Reserved words in LLVM are very similar to reserved words in other
333languages. There are keywords for different opcodes
334('<tt><a href="#i_add">add</a></tt>',
335 '<tt><a href="#i_bitcast">bitcast</a></tt>',
336 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
337href="#t_void">void</a></tt>', '<tt><a href="#t_primitive">i32</a></tt>', etc...),
338and others. These reserved words cannot conflict with variable names, because
Reid Spencerc8245b02007-08-07 14:34:28 +0000339none of them start with a prefix character ('%' or '@').</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000340
341<p>Here is an example of LLVM code to multiply the integer variable
342'<tt>%X</tt>' by 8:</p>
343
344<p>The easy way:</p>
345
346<div class="doc_code">
347<pre>
348%result = <a href="#i_mul">mul</a> i32 %X, 8
349</pre>
350</div>
351
352<p>After strength reduction:</p>
353
354<div class="doc_code">
355<pre>
356%result = <a href="#i_shl">shl</a> i32 %X, i8 3
357</pre>
358</div>
359
360<p>And the hard way:</p>
361
362<div class="doc_code">
363<pre>
364<a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
365<a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
366%result = <a href="#i_add">add</a> i32 %1, %1
367</pre>
368</div>
369
370<p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
371important lexical features of LLVM:</p>
372
373<ol>
374
375 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
376 line.</li>
377
378 <li>Unnamed temporaries are created when the result of a computation is not
379 assigned to a named value.</li>
380
381 <li>Unnamed temporaries are numbered sequentially</li>
382
383</ol>
384
385<p>...and it also shows a convention that we follow in this document. When
386demonstrating instructions, we will follow an instruction with a comment that
387defines the type and name of value produced. Comments are shown in italic
388text.</p>
389
390</div>
391
392<!-- *********************************************************************** -->
393<div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
394<!-- *********************************************************************** -->
395
396<!-- ======================================================================= -->
397<div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
398</div>
399
400<div class="doc_text">
401
402<p>LLVM programs are composed of "Module"s, each of which is a
403translation unit of the input programs. Each module consists of
404functions, global variables, and symbol table entries. Modules may be
405combined together with the LLVM linker, which merges function (and
406global variable) definitions, resolves forward declarations, and merges
407symbol table entries. Here is an example of the "hello world" module:</p>
408
409<div class="doc_code">
410<pre><i>; Declare the string constant as a global constant...</i>
411<a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a
412 href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
413
414<i>; External declaration of the puts function</i>
415<a href="#functionstructure">declare</a> i32 @puts(i8 *) <i>; i32(i8 *)* </i>
416
417<i>; Definition of main function</i>
418define i32 @main() { <i>; i32()* </i>
419 <i>; Convert [13x i8 ]* to i8 *...</i>
420 %cast210 = <a
421 href="#i_getelementptr">getelementptr</a> [13 x i8 ]* @.LC0, i64 0, i64 0 <i>; i8 *</i>
422
423 <i>; Call puts function to write out the string to stdout...</i>
424 <a
425 href="#i_call">call</a> i32 @puts(i8 * %cast210) <i>; i32</i>
426 <a
427 href="#i_ret">ret</a> i32 0<br>}<br>
428</pre>
429</div>
430
431<p>This example is made up of a <a href="#globalvars">global variable</a>
432named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
433function, and a <a href="#functionstructure">function definition</a>
434for "<tt>main</tt>".</p>
435
436<p>In general, a module is made up of a list of global values,
437where both functions and global variables are global values. Global values are
438represented by a pointer to a memory location (in this case, a pointer to an
439array of char, and a pointer to a function), and have one of the following <a
440href="#linkage">linkage types</a>.</p>
441
442</div>
443
444<!-- ======================================================================= -->
445<div class="doc_subsection">
446 <a name="linkage">Linkage Types</a>
447</div>
448
449<div class="doc_text">
450
451<p>
452All Global Variables and Functions have one of the following types of linkage:
453</p>
454
455<dl>
456
457 <dt><tt><b><a name="linkage_internal">internal</a></b></tt> </dt>
458
459 <dd>Global values with internal linkage are only directly accessible by
460 objects in the current module. In particular, linking code into a module with
461 an internal global value may cause the internal to be renamed as necessary to
462 avoid collisions. Because the symbol is internal to the module, all
463 references can be updated. This corresponds to the notion of the
464 '<tt>static</tt>' keyword in C.
465 </dd>
466
467 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
468
469 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
470 the same name when linkage occurs. This is typically used to implement
471 inline functions, templates, or other code which must be generated in each
472 translation unit that uses it. Unreferenced <tt>linkonce</tt> globals are
473 allowed to be discarded.
474 </dd>
475
476 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
477
478 <dd>"<tt>weak</tt>" linkage is exactly the same as <tt>linkonce</tt> linkage,
479 except that unreferenced <tt>weak</tt> globals may not be discarded. This is
480 used for globals that may be emitted in multiple translation units, but that
481 are not guaranteed to be emitted into every translation unit that uses them.
482 One example of this are common globals in C, such as "<tt>int X;</tt>" at
483 global scope.
484 </dd>
485
486 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
487
488 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
489 pointer to array type. When two global variables with appending linkage are
490 linked together, the two global arrays are appended together. This is the
491 LLVM, typesafe, equivalent of having the system linker append together
492 "sections" with identical names when .o files are linked.
493 </dd>
494
495 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt>: </dt>
496 <dd>The semantics of this linkage follow the ELF model: the symbol is weak
497 until linked, if not linked, the symbol becomes null instead of being an
498 undefined reference.
499 </dd>
500
501 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
502
503 <dd>If none of the above identifiers are used, the global is externally
504 visible, meaning that it participates in linkage and can be used to resolve
505 external symbol references.
506 </dd>
507</dl>
508
509 <p>
510 The next two types of linkage are targeted for Microsoft Windows platform
511 only. They are designed to support importing (exporting) symbols from (to)
512 DLLs.
513 </p>
514
515 <dl>
516 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt>: </dt>
517
518 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
519 or variable via a global pointer to a pointer that is set up by the DLL
520 exporting the symbol. On Microsoft Windows targets, the pointer name is
521 formed by combining <code>_imp__</code> and the function or variable name.
522 </dd>
523
524 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt>: </dt>
525
526 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
527 pointer to a pointer in a DLL, so that it can be referenced with the
528 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
529 name is formed by combining <code>_imp__</code> and the function or variable
530 name.
531 </dd>
532
533</dl>
534
535<p><a name="linkage_external"></a>For example, since the "<tt>.LC0</tt>"
536variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
537variable and was linked with this one, one of the two would be renamed,
538preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
539external (i.e., lacking any linkage declarations), they are accessible
540outside of the current module.</p>
541<p>It is illegal for a function <i>declaration</i>
542to have any linkage type other than "externally visible", <tt>dllimport</tt>,
543or <tt>extern_weak</tt>.</p>
544<p>Aliases can have only <tt>external</tt>, <tt>internal</tt> and <tt>weak</tt>
545linkages.
546</div>
547
548<!-- ======================================================================= -->
549<div class="doc_subsection">
550 <a name="callingconv">Calling Conventions</a>
551</div>
552
553<div class="doc_text">
554
555<p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
556and <a href="#i_invoke">invokes</a> can all have an optional calling convention
557specified for the call. The calling convention of any pair of dynamic
558caller/callee must match, or the behavior of the program is undefined. The
559following calling conventions are supported by LLVM, and more may be added in
560the future:</p>
561
562<dl>
563 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
564
565 <dd>This calling convention (the default if no other calling convention is
566 specified) matches the target C calling conventions. This calling convention
567 supports varargs function calls and tolerates some mismatch in the declared
568 prototype and implemented declaration of the function (as does normal C).
569 </dd>
570
571 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
572
573 <dd>This calling convention attempts to make calls as fast as possible
574 (e.g. by passing things in registers). This calling convention allows the
575 target to use whatever tricks it wants to produce fast code for the target,
576 without having to conform to an externally specified ABI. Implementations of
577 this convention should allow arbitrary tail call optimization to be supported.
578 This calling convention does not support varargs and requires the prototype of
579 all callees to exactly match the prototype of the function definition.
580 </dd>
581
582 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
583
584 <dd>This calling convention attempts to make code in the caller as efficient
585 as possible under the assumption that the call is not commonly executed. As
586 such, these calls often preserve all registers so that the call does not break
587 any live ranges in the caller side. This calling convention does not support
588 varargs and requires the prototype of all callees to exactly match the
589 prototype of the function definition.
590 </dd>
591
592 <dt><b>"<tt>cc &lt;<em>n</em>&gt;</tt>" - Numbered convention</b>:</dt>
593
594 <dd>Any calling convention may be specified by number, allowing
595 target-specific calling conventions to be used. Target specific calling
596 conventions start at 64.
597 </dd>
598</dl>
599
600<p>More calling conventions can be added/defined on an as-needed basis, to
601support pascal conventions or any other well-known target-independent
602convention.</p>
603
604</div>
605
606<!-- ======================================================================= -->
607<div class="doc_subsection">
608 <a name="visibility">Visibility Styles</a>
609</div>
610
611<div class="doc_text">
612
613<p>
614All Global Variables and Functions have one of the following visibility styles:
615</p>
616
617<dl>
618 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
619
620 <dd>On ELF, default visibility means that the declaration is visible to other
621 modules and, in shared libraries, means that the declared entity may be
622 overridden. On Darwin, default visibility means that the declaration is
623 visible to other modules. Default visibility corresponds to "external
624 linkage" in the language.
625 </dd>
626
627 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
628
629 <dd>Two declarations of an object with hidden visibility refer to the same
630 object if they are in the same shared object. Usually, hidden visibility
631 indicates that the symbol will not be placed into the dynamic symbol table,
632 so no other module (executable or shared library) can reference it
633 directly.
634 </dd>
635
636 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
637
638 <dd>On ELF, protected visibility indicates that the symbol will be placed in
639 the dynamic symbol table, but that references within the defining module will
640 bind to the local symbol. That is, the symbol cannot be overridden by another
641 module.
642 </dd>
643</dl>
644
645</div>
646
647<!-- ======================================================================= -->
648<div class="doc_subsection">
649 <a name="globalvars">Global Variables</a>
650</div>
651
652<div class="doc_text">
653
654<p>Global variables define regions of memory allocated at compilation time
655instead of run-time. Global variables may optionally be initialized, may have
656an explicit section to be placed in, and may have an optional explicit alignment
657specified. A variable may be defined as "thread_local", which means that it
658will not be shared by threads (each thread will have a separated copy of the
659variable). A variable may be defined as a global "constant," which indicates
660that the contents of the variable will <b>never</b> be modified (enabling better
661optimization, allowing the global data to be placed in the read-only section of
662an executable, etc). Note that variables that need runtime initialization
663cannot be marked "constant" as there is a store to the variable.</p>
664
665<p>
666LLVM explicitly allows <em>declarations</em> of global variables to be marked
667constant, even if the final definition of the global is not. This capability
668can be used to enable slightly better optimization of the program, but requires
669the language definition to guarantee that optimizations based on the
670'constantness' are valid for the translation units that do not include the
671definition.
672</p>
673
674<p>As SSA values, global variables define pointer values that are in
675scope (i.e. they dominate) all basic blocks in the program. Global
676variables always define a pointer to their "content" type because they
677describe a region of memory, and all memory objects in LLVM are
678accessed through pointers.</p>
679
Christopher Lambdd0049d2007-12-11 09:31:00 +0000680<p>A global variable may be declared to reside in a target-specifc numbered
681address space. For targets that support them, address spaces may affect how
682optimizations are performed and/or what target instructions are used to access
Christopher Lamb20a39e92007-12-12 08:44:39 +0000683the variable. The default address space is zero. The address space qualifier
684must precede any other attributes.</p>
Christopher Lambdd0049d2007-12-11 09:31:00 +0000685
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000686<p>LLVM allows an explicit section to be specified for globals. If the target
687supports it, it will emit globals to the section specified.</p>
688
689<p>An explicit alignment may be specified for a global. If not present, or if
690the alignment is set to zero, the alignment of the global is set by the target
691to whatever it feels convenient. If an explicit alignment is specified, the
692global is forced to have at least that much alignment. All alignments must be
693a power of 2.</p>
694
Christopher Lambdd0049d2007-12-11 09:31:00 +0000695<p>For example, the following defines a global in a numbered address space with
696an initializer, section, and alignment:</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000697
698<div class="doc_code">
699<pre>
Christopher Lambdd0049d2007-12-11 09:31:00 +0000700@G = constant float 1.0 addrspace(5), section "foo", align 4
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000701</pre>
702</div>
703
704</div>
705
706
707<!-- ======================================================================= -->
708<div class="doc_subsection">
709 <a name="functionstructure">Functions</a>
710</div>
711
712<div class="doc_text">
713
714<p>LLVM function definitions consist of the "<tt>define</tt>" keyord,
715an optional <a href="#linkage">linkage type</a>, an optional
716<a href="#visibility">visibility style</a>, an optional
717<a href="#callingconv">calling convention</a>, a return type, an optional
718<a href="#paramattrs">parameter attribute</a> for the return type, a function
719name, a (possibly empty) argument list (each with optional
720<a href="#paramattrs">parameter attributes</a>), an optional section, an
Gordon Henriksend606f5b2007-12-10 03:30:21 +0000721optional alignment, an optional <a href="#gc">garbage collector name</a>, an
Gordon Henriksen13fe5e32007-12-10 03:18:06 +0000722opening curly brace, a list of basic blocks, and a closing curly brace.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000723
724LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
725optional <a href="#linkage">linkage type</a>, an optional
726<a href="#visibility">visibility style</a>, an optional
727<a href="#callingconv">calling convention</a>, a return type, an optional
728<a href="#paramattrs">parameter attribute</a> for the return type, a function
Gordon Henriksen13fe5e32007-12-10 03:18:06 +0000729name, a possibly empty list of arguments, an optional alignment, and an optional
Gordon Henriksend606f5b2007-12-10 03:30:21 +0000730<a href="#gc">garbage collector name</a>.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000731
732<p>A function definition contains a list of basic blocks, forming the CFG for
733the function. Each basic block may optionally start with a label (giving the
734basic block a symbol table entry), contains a list of instructions, and ends
735with a <a href="#terminators">terminator</a> instruction (such as a branch or
736function return).</p>
737
738<p>The first basic block in a function is special in two ways: it is immediately
739executed on entrance to the function, and it is not allowed to have predecessor
740basic blocks (i.e. there can not be any branches to the entry block of a
741function). Because the block can have no predecessors, it also cannot have any
742<a href="#i_phi">PHI nodes</a>.</p>
743
744<p>LLVM allows an explicit section to be specified for functions. If the target
745supports it, it will emit functions to the section specified.</p>
746
747<p>An explicit alignment may be specified for a function. If not present, or if
748the alignment is set to zero, the alignment of the function is set by the target
749to whatever it feels convenient. If an explicit alignment is specified, the
750function is forced to have at least that much alignment. All alignments must be
751a power of 2.</p>
752
753</div>
754
755
756<!-- ======================================================================= -->
757<div class="doc_subsection">
758 <a name="aliasstructure">Aliases</a>
759</div>
760<div class="doc_text">
761 <p>Aliases act as "second name" for the aliasee value (which can be either
762 function or global variable or bitcast of global value). Aliases may have an
763 optional <a href="#linkage">linkage type</a>, and an
764 optional <a href="#visibility">visibility style</a>.</p>
765
766 <h5>Syntax:</h5>
767
768<div class="doc_code">
769<pre>
770@&lt;Name&gt; = [Linkage] [Visibility] alias &lt;AliaseeTy&gt; @&lt;Aliasee&gt;
771</pre>
772</div>
773
774</div>
775
776
777
778<!-- ======================================================================= -->
779<div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
780<div class="doc_text">
781 <p>The return type and each parameter of a function type may have a set of
782 <i>parameter attributes</i> associated with them. Parameter attributes are
783 used to communicate additional information about the result or parameters of
Duncan Sandsf5588dc2007-11-27 13:23:08 +0000784 a function. Parameter attributes are considered to be part of the function,
785 not of the function type, so functions with different parameter attributes
786 can have the same function type.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000787
788 <p>Parameter attributes are simple keywords that follow the type specified. If
789 multiple parameter attributes are needed, they are space separated. For
790 example:</p>
791
792<div class="doc_code">
793<pre>
Duncan Sandsf5588dc2007-11-27 13:23:08 +0000794declare i32 @printf(i8* noalias , ...) nounwind
795declare i32 @atoi(i8*) nounwind readonly
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000796</pre>
797</div>
798
Duncan Sandsf5588dc2007-11-27 13:23:08 +0000799 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
800 <tt>readonly</tt>) come immediately after the argument list.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000801
802 <p>Currently, only the following parameter attributes are defined:</p>
803 <dl>
Reid Spencerf234bed2007-07-19 23:13:04 +0000804 <dt><tt>zeroext</tt></dt>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000805 <dd>This indicates that the parameter should be zero extended just before
806 a call to this function.</dd>
Chris Lattner275e6be2008-01-11 06:20:47 +0000807
Reid Spencerf234bed2007-07-19 23:13:04 +0000808 <dt><tt>signext</tt></dt>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000809 <dd>This indicates that the parameter should be sign extended just before
810 a call to this function.</dd>
Chris Lattner275e6be2008-01-11 06:20:47 +0000811
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000812 <dt><tt>inreg</tt></dt>
813 <dd>This indicates that the parameter should be placed in register (if
814 possible) during assembling function call. Support for this attribute is
815 target-specific</dd>
Chris Lattner275e6be2008-01-11 06:20:47 +0000816
817 <dt><tt>byval</tt></dt>
Chris Lattner04c86182008-01-15 04:34:22 +0000818 <dd>This indicates that the pointer parameter should really be passed by
819 value to the function. The attribute implies that a hidden copy of the
820 pointee is made between the caller and the callee, so the callee is unable
821 to modify the value in the callee. This attribute is only valid on llvm
822 pointer arguments. It is generally used to pass structs and arrays by
823 value, but is also valid on scalars (even though this is silly).</dd>
Chris Lattner275e6be2008-01-11 06:20:47 +0000824
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000825 <dt><tt>sret</tt></dt>
826 <dd>This indicates that the parameter specifies the address of a structure
827 that is the return value of the function in the source program.</dd>
Chris Lattner275e6be2008-01-11 06:20:47 +0000828
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000829 <dt><tt>noalias</tt></dt>
Owen Andersonc4fc4cd2008-02-18 04:09:01 +0000830 <dd>This indicates that the parameter does not alias any global or any other
831 parameter. The caller is responsible for ensuring that this is the case,
832 usually by placing the value in a stack allocation.</dd>
Chris Lattner275e6be2008-01-11 06:20:47 +0000833
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000834 <dt><tt>noreturn</tt></dt>
835 <dd>This function attribute indicates that the function never returns. This
836 indicates to LLVM that every call to this function should be treated as if
837 an <tt>unreachable</tt> instruction immediately followed the call.</dd>
Chris Lattner275e6be2008-01-11 06:20:47 +0000838
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000839 <dt><tt>nounwind</tt></dt>
840 <dd>This function attribute indicates that the function type does not use
841 the unwind instruction and does not allow stack unwinding to propagate
842 through it.</dd>
Chris Lattner275e6be2008-01-11 06:20:47 +0000843
Duncan Sands4ee46812007-07-27 19:57:41 +0000844 <dt><tt>nest</tt></dt>
845 <dd>This indicates that the parameter can be excised using the
846 <a href="#int_trampoline">trampoline intrinsics</a>.</dd>
Duncan Sands13e13f82007-11-22 20:23:04 +0000847 <dt><tt>readonly</tt></dt>
Duncan Sandsd69c0e62007-11-14 21:14:02 +0000848 <dd>This function attribute indicates that the function has no side-effects
Duncan Sands13e13f82007-11-22 20:23:04 +0000849 except for producing a return value or throwing an exception. The value
850 returned must only depend on the function arguments and/or global variables.
851 It may use values obtained by dereferencing pointers.</dd>
852 <dt><tt>readnone</tt></dt>
853 <dd>A <tt>readnone</tt> function has the same restrictions as a <tt>readonly</tt>
Duncan Sandsd69c0e62007-11-14 21:14:02 +0000854 function, but in addition it is not allowed to dereference any pointer arguments
855 or global variables.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000856 </dl>
857
858</div>
859
860<!-- ======================================================================= -->
861<div class="doc_subsection">
Gordon Henriksen13fe5e32007-12-10 03:18:06 +0000862 <a name="gc">Garbage Collector Names</a>
863</div>
864
865<div class="doc_text">
866<p>Each function may specify a garbage collector name, which is simply a
867string.</p>
868
869<div class="doc_code"><pre
870>define void @f() gc "name" { ...</pre></div>
871
872<p>The compiler declares the supported values of <i>name</i>. Specifying a
873collector which will cause the compiler to alter its output in order to support
874the named garbage collection algorithm.</p>
875</div>
876
877<!-- ======================================================================= -->
878<div class="doc_subsection">
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000879 <a name="moduleasm">Module-Level Inline Assembly</a>
880</div>
881
882<div class="doc_text">
883<p>
884Modules may contain "module-level inline asm" blocks, which corresponds to the
885GCC "file scope inline asm" blocks. These blocks are internally concatenated by
886LLVM and treated as a single unit, but may be separated in the .ll file if
887desired. The syntax is very simple:
888</p>
889
890<div class="doc_code">
891<pre>
892module asm "inline asm code goes here"
893module asm "more can go here"
894</pre>
895</div>
896
897<p>The strings can contain any character by escaping non-printable characters.
898 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
899 for the number.
900</p>
901
902<p>
903 The inline asm code is simply printed to the machine code .s file when
904 assembly code is generated.
905</p>
906</div>
907
908<!-- ======================================================================= -->
909<div class="doc_subsection">
910 <a name="datalayout">Data Layout</a>
911</div>
912
913<div class="doc_text">
914<p>A module may specify a target specific data layout string that specifies how
915data is to be laid out in memory. The syntax for the data layout is simply:</p>
916<pre> target datalayout = "<i>layout specification</i>"</pre>
917<p>The <i>layout specification</i> consists of a list of specifications
918separated by the minus sign character ('-'). Each specification starts with a
919letter and may include other information after the letter to define some
920aspect of the data layout. The specifications accepted are as follows: </p>
921<dl>
922 <dt><tt>E</tt></dt>
923 <dd>Specifies that the target lays out data in big-endian form. That is, the
924 bits with the most significance have the lowest address location.</dd>
925 <dt><tt>e</tt></dt>
926 <dd>Specifies that hte target lays out data in little-endian form. That is,
927 the bits with the least significance have the lowest address location.</dd>
928 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
929 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
930 <i>preferred</i> alignments. All sizes are in bits. Specifying the <i>pref</i>
931 alignment is optional. If omitted, the preceding <tt>:</tt> should be omitted
932 too.</dd>
933 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
934 <dd>This specifies the alignment for an integer type of a given bit
935 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
936 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
937 <dd>This specifies the alignment for a vector type of a given bit
938 <i>size</i>.</dd>
939 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
940 <dd>This specifies the alignment for a floating point type of a given bit
941 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
942 (double).</dd>
943 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
944 <dd>This specifies the alignment for an aggregate type of a given bit
945 <i>size</i>.</dd>
946</dl>
947<p>When constructing the data layout for a given target, LLVM starts with a
948default set of specifications which are then (possibly) overriden by the
949specifications in the <tt>datalayout</tt> keyword. The default specifications
950are given in this list:</p>
951<ul>
952 <li><tt>E</tt> - big endian</li>
953 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
954 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
955 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
956 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
957 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
958 <li><tt>i64:32:64</tt> - i64 has abi alignment of 32-bits but preferred
959 alignment of 64-bits</li>
960 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
961 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
962 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
963 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
964 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
965</ul>
966<p>When llvm is determining the alignment for a given type, it uses the
967following rules:
968<ol>
969 <li>If the type sought is an exact match for one of the specifications, that
970 specification is used.</li>
971 <li>If no match is found, and the type sought is an integer type, then the
972 smallest integer type that is larger than the bitwidth of the sought type is
973 used. If none of the specifications are larger than the bitwidth then the the
974 largest integer type is used. For example, given the default specifications
975 above, the i7 type will use the alignment of i8 (next largest) while both
976 i65 and i256 will use the alignment of i64 (largest specified).</li>
977 <li>If no match is found, and the type sought is a vector type, then the
978 largest vector type that is smaller than the sought vector type will be used
979 as a fall back. This happens because <128 x double> can be implemented in
980 terms of 64 <2 x double>, for example.</li>
981</ol>
982</div>
983
984<!-- *********************************************************************** -->
985<div class="doc_section"> <a name="typesystem">Type System</a> </div>
986<!-- *********************************************************************** -->
987
988<div class="doc_text">
989
990<p>The LLVM type system is one of the most important features of the
991intermediate representation. Being typed enables a number of
992optimizations to be performed on the IR directly, without having to do
993extra analyses on the side before the transformation. A strong type
994system makes it easier to read the generated code and enables novel
995analyses and transformations that are not feasible to perform on normal
996three address code representations.</p>
997
998</div>
999
1000<!-- ======================================================================= -->
Chris Lattner488772f2008-01-04 04:32:38 +00001001<div class="doc_subsection"> <a name="t_classifications">Type
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001002Classifications</a> </div>
1003<div class="doc_text">
Chris Lattner488772f2008-01-04 04:32:38 +00001004<p>The types fall into a few useful
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001005classifications:</p>
1006
1007<table border="1" cellspacing="0" cellpadding="4">
1008 <tbody>
1009 <tr><th>Classification</th><th>Types</th></tr>
1010 <tr>
Chris Lattner488772f2008-01-04 04:32:38 +00001011 <td><a href="#t_integer">integer</a></td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001012 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1013 </tr>
1014 <tr>
Chris Lattner488772f2008-01-04 04:32:38 +00001015 <td><a href="#t_floating">floating point</a></td>
1016 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001017 </tr>
1018 <tr>
1019 <td><a name="t_firstclass">first class</a></td>
Chris Lattner488772f2008-01-04 04:32:38 +00001020 <td><a href="#t_integer">integer</a>,
1021 <a href="#t_floating">floating point</a>,
1022 <a href="#t_pointer">pointer</a>,
1023 <a href="#t_vector">vector</a>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001024 </td>
1025 </tr>
Chris Lattner488772f2008-01-04 04:32:38 +00001026 <tr>
1027 <td><a href="#t_primitive">primitive</a></td>
1028 <td><a href="#t_label">label</a>,
1029 <a href="#t_void">void</a>,
1030 <a href="#t_integer">integer</a>,
1031 <a href="#t_floating">floating point</a>.</td>
1032 </tr>
1033 <tr>
1034 <td><a href="#t_derived">derived</a></td>
1035 <td><a href="#t_integer">integer</a>,
1036 <a href="#t_array">array</a>,
1037 <a href="#t_function">function</a>,
1038 <a href="#t_pointer">pointer</a>,
1039 <a href="#t_struct">structure</a>,
1040 <a href="#t_pstruct">packed structure</a>,
1041 <a href="#t_vector">vector</a>,
1042 <a href="#t_opaque">opaque</a>.
1043 </tr>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001044 </tbody>
1045</table>
1046
1047<p>The <a href="#t_firstclass">first class</a> types are perhaps the
1048most important. Values of these types are the only ones which can be
1049produced by instructions, passed as arguments, or used as operands to
1050instructions. This means that all structures and arrays must be
1051manipulated either by pointer or by component.</p>
1052</div>
1053
1054<!-- ======================================================================= -->
Chris Lattner488772f2008-01-04 04:32:38 +00001055<div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
Chris Lattner86437612008-01-04 04:34:14 +00001056
Chris Lattner488772f2008-01-04 04:32:38 +00001057<div class="doc_text">
1058<p>The primitive types are the fundamental building blocks of the LLVM
1059system.</p>
1060
Chris Lattner86437612008-01-04 04:34:14 +00001061</div>
1062
Chris Lattner488772f2008-01-04 04:32:38 +00001063<!-- _______________________________________________________________________ -->
1064<div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1065
1066<div class="doc_text">
1067 <table>
1068 <tbody>
1069 <tr><th>Type</th><th>Description</th></tr>
1070 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1071 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1072 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1073 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1074 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1075 </tbody>
1076 </table>
1077</div>
1078
1079<!-- _______________________________________________________________________ -->
1080<div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1081
1082<div class="doc_text">
1083<h5>Overview:</h5>
1084<p>The void type does not represent any value and has no size.</p>
1085
1086<h5>Syntax:</h5>
1087
1088<pre>
1089 void
1090</pre>
1091</div>
1092
1093<!-- _______________________________________________________________________ -->
1094<div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1095
1096<div class="doc_text">
1097<h5>Overview:</h5>
1098<p>The label type represents code labels.</p>
1099
1100<h5>Syntax:</h5>
1101
1102<pre>
1103 label
1104</pre>
1105</div>
1106
1107
1108<!-- ======================================================================= -->
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001109<div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1110
1111<div class="doc_text">
1112
1113<p>The real power in LLVM comes from the derived types in the system.
1114This is what allows a programmer to represent arrays, functions,
1115pointers, and other useful types. Note that these derived types may be
1116recursive: For example, it is possible to have a two dimensional array.</p>
1117
1118</div>
1119
1120<!-- _______________________________________________________________________ -->
1121<div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1122
1123<div class="doc_text">
1124
1125<h5>Overview:</h5>
1126<p>The integer type is a very simple derived type that simply specifies an
1127arbitrary bit width for the integer type desired. Any bit width from 1 bit to
11282^23-1 (about 8 million) can be specified.</p>
1129
1130<h5>Syntax:</h5>
1131
1132<pre>
1133 iN
1134</pre>
1135
1136<p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1137value.</p>
1138
1139<h5>Examples:</h5>
1140<table class="layout">
Chris Lattner251ab812007-12-18 06:18:21 +00001141 <tbody>
1142 <tr>
1143 <td><tt>i1</tt></td>
1144 <td>a single-bit integer.</td>
1145 </tr><tr>
1146 <td><tt>i32</tt></td>
1147 <td>a 32-bit integer.</td>
1148 </tr><tr>
1149 <td><tt>i1942652</tt></td>
1150 <td>a really big integer of over 1 million bits.</td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001151 </tr>
Chris Lattner251ab812007-12-18 06:18:21 +00001152 </tbody>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001153</table>
1154</div>
1155
1156<!-- _______________________________________________________________________ -->
1157<div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1158
1159<div class="doc_text">
1160
1161<h5>Overview:</h5>
1162
1163<p>The array type is a very simple derived type that arranges elements
1164sequentially in memory. The array type requires a size (number of
1165elements) and an underlying data type.</p>
1166
1167<h5>Syntax:</h5>
1168
1169<pre>
1170 [&lt;# elements&gt; x &lt;elementtype&gt;]
1171</pre>
1172
1173<p>The number of elements is a constant integer value; elementtype may
1174be any type with a size.</p>
1175
1176<h5>Examples:</h5>
1177<table class="layout">
1178 <tr class="layout">
Chris Lattner7311d222007-12-19 05:04:11 +00001179 <td class="left"><tt>[40 x i32]</tt></td>
1180 <td class="left">Array of 40 32-bit integer values.</td>
1181 </tr>
1182 <tr class="layout">
1183 <td class="left"><tt>[41 x i32]</tt></td>
1184 <td class="left">Array of 41 32-bit integer values.</td>
1185 </tr>
1186 <tr class="layout">
1187 <td class="left"><tt>[4 x i8]</tt></td>
1188 <td class="left">Array of 4 8-bit integer values.</td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001189 </tr>
1190</table>
1191<p>Here are some examples of multidimensional arrays:</p>
1192<table class="layout">
1193 <tr class="layout">
Chris Lattner7311d222007-12-19 05:04:11 +00001194 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1195 <td class="left">3x4 array of 32-bit integer values.</td>
1196 </tr>
1197 <tr class="layout">
1198 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1199 <td class="left">12x10 array of single precision floating point values.</td>
1200 </tr>
1201 <tr class="layout">
1202 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1203 <td class="left">2x3x4 array of 16-bit integer values.</td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001204 </tr>
1205</table>
1206
1207<p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1208length array. Normally, accesses past the end of an array are undefined in
1209LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
1210As a special case, however, zero length arrays are recognized to be variable
1211length. This allows implementation of 'pascal style arrays' with the LLVM
1212type "{ i32, [0 x float]}", for example.</p>
1213
1214</div>
1215
1216<!-- _______________________________________________________________________ -->
1217<div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1218<div class="doc_text">
1219<h5>Overview:</h5>
1220<p>The function type can be thought of as a function signature. It
1221consists of a return type and a list of formal parameter types.
1222Function types are usually used to build virtual function tables
1223(which are structures of pointers to functions), for indirect function
1224calls, and when defining a function.</p>
1225<p>
1226The return type of a function type cannot be an aggregate type.
1227</p>
1228<h5>Syntax:</h5>
1229<pre> &lt;returntype&gt; (&lt;parameter list&gt;)<br></pre>
1230<p>...where '<tt>&lt;parameter list&gt;</tt>' is a comma-separated list of type
1231specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1232which indicates that the function takes a variable number of arguments.
1233Variable argument functions can access their arguments with the <a
1234 href="#int_varargs">variable argument handling intrinsic</a> functions.</p>
1235<h5>Examples:</h5>
1236<table class="layout">
1237 <tr class="layout">
1238 <td class="left"><tt>i32 (i32)</tt></td>
1239 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1240 </td>
1241 </tr><tr class="layout">
Reid Spencerf234bed2007-07-19 23:13:04 +00001242 <td class="left"><tt>float&nbsp;(i16&nbsp;signext,&nbsp;i32&nbsp;*)&nbsp;*
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001243 </tt></td>
1244 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1245 an <tt>i16</tt> that should be sign extended and a
1246 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1247 <tt>float</tt>.
1248 </td>
1249 </tr><tr class="layout">
1250 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1251 <td class="left">A vararg function that takes at least one
1252 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1253 which returns an integer. This is the signature for <tt>printf</tt> in
1254 LLVM.
1255 </td>
1256 </tr>
1257</table>
1258
1259</div>
1260<!-- _______________________________________________________________________ -->
1261<div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1262<div class="doc_text">
1263<h5>Overview:</h5>
1264<p>The structure type is used to represent a collection of data members
1265together in memory. The packing of the field types is defined to match
1266the ABI of the underlying processor. The elements of a structure may
1267be any type that has a size.</p>
1268<p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1269and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1270field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1271instruction.</p>
1272<h5>Syntax:</h5>
1273<pre> { &lt;type list&gt; }<br></pre>
1274<h5>Examples:</h5>
1275<table class="layout">
1276 <tr class="layout">
1277 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1278 <td class="left">A triple of three <tt>i32</tt> values</td>
1279 </tr><tr class="layout">
1280 <td class="left"><tt>{&nbsp;float,&nbsp;i32&nbsp;(i32)&nbsp;*&nbsp;}</tt></td>
1281 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1282 second element is a <a href="#t_pointer">pointer</a> to a
1283 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1284 an <tt>i32</tt>.</td>
1285 </tr>
1286</table>
1287</div>
1288
1289<!-- _______________________________________________________________________ -->
1290<div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1291</div>
1292<div class="doc_text">
1293<h5>Overview:</h5>
1294<p>The packed structure type is used to represent a collection of data members
1295together in memory. There is no padding between fields. Further, the alignment
1296of a packed structure is 1 byte. The elements of a packed structure may
1297be any type that has a size.</p>
1298<p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1299and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1300field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1301instruction.</p>
1302<h5>Syntax:</h5>
1303<pre> &lt; { &lt;type list&gt; } &gt; <br></pre>
1304<h5>Examples:</h5>
1305<table class="layout">
1306 <tr class="layout">
1307 <td class="left"><tt>&lt; { i32, i32, i32 } &gt;</tt></td>
1308 <td class="left">A triple of three <tt>i32</tt> values</td>
1309 </tr><tr class="layout">
Chris Lattner7311d222007-12-19 05:04:11 +00001310 <td class="left"><tt>&lt; { float, i32 (i32)* } &gt;</tt></td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001311 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1312 second element is a <a href="#t_pointer">pointer</a> to a
1313 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1314 an <tt>i32</tt>.</td>
1315 </tr>
1316</table>
1317</div>
1318
1319<!-- _______________________________________________________________________ -->
1320<div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1321<div class="doc_text">
1322<h5>Overview:</h5>
1323<p>As in many languages, the pointer type represents a pointer or
Christopher Lambdd0049d2007-12-11 09:31:00 +00001324reference to another object, which must live in memory. Pointer types may have
1325an optional address space attribute defining the target-specific numbered
1326address space where the pointed-to object resides. The default address space is
1327zero.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001328<h5>Syntax:</h5>
1329<pre> &lt;type&gt; *<br></pre>
1330<h5>Examples:</h5>
1331<table class="layout">
1332 <tr class="layout">
Chris Lattner7311d222007-12-19 05:04:11 +00001333 <td class="left"><tt>[4x i32]*</tt></td>
1334 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1335 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1336 </tr>
1337 <tr class="layout">
1338 <td class="left"><tt>i32 (i32 *) *</tt></td>
1339 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001340 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
Chris Lattner7311d222007-12-19 05:04:11 +00001341 <tt>i32</tt>.</td>
1342 </tr>
1343 <tr class="layout">
1344 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1345 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1346 that resides in address space #5.</td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001347 </tr>
1348</table>
1349</div>
1350
1351<!-- _______________________________________________________________________ -->
1352<div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1353<div class="doc_text">
1354
1355<h5>Overview:</h5>
1356
1357<p>A vector type is a simple derived type that represents a vector
1358of elements. Vector types are used when multiple primitive data
1359are operated in parallel using a single instruction (SIMD).
1360A vector type requires a size (number of
1361elements) and an underlying primitive data type. Vectors must have a power
1362of two length (1, 2, 4, 8, 16 ...). Vector types are
1363considered <a href="#t_firstclass">first class</a>.</p>
1364
1365<h5>Syntax:</h5>
1366
1367<pre>
1368 &lt; &lt;# elements&gt; x &lt;elementtype&gt; &gt;
1369</pre>
1370
1371<p>The number of elements is a constant integer value; elementtype may
1372be any integer or floating point type.</p>
1373
1374<h5>Examples:</h5>
1375
1376<table class="layout">
1377 <tr class="layout">
Chris Lattner7311d222007-12-19 05:04:11 +00001378 <td class="left"><tt>&lt;4 x i32&gt;</tt></td>
1379 <td class="left">Vector of 4 32-bit integer values.</td>
1380 </tr>
1381 <tr class="layout">
1382 <td class="left"><tt>&lt;8 x float&gt;</tt></td>
1383 <td class="left">Vector of 8 32-bit floating-point values.</td>
1384 </tr>
1385 <tr class="layout">
1386 <td class="left"><tt>&lt;2 x i64&gt;</tt></td>
1387 <td class="left">Vector of 2 64-bit integer values.</td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001388 </tr>
1389</table>
1390</div>
1391
1392<!-- _______________________________________________________________________ -->
1393<div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1394<div class="doc_text">
1395
1396<h5>Overview:</h5>
1397
1398<p>Opaque types are used to represent unknown types in the system. This
Gordon Henriksenda0706e2007-10-14 00:34:53 +00001399corresponds (for example) to the C notion of a forward declared structure type.
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001400In LLVM, opaque types can eventually be resolved to any type (not just a
1401structure type).</p>
1402
1403<h5>Syntax:</h5>
1404
1405<pre>
1406 opaque
1407</pre>
1408
1409<h5>Examples:</h5>
1410
1411<table class="layout">
1412 <tr class="layout">
Chris Lattner7311d222007-12-19 05:04:11 +00001413 <td class="left"><tt>opaque</tt></td>
1414 <td class="left">An opaque type.</td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001415 </tr>
1416</table>
1417</div>
1418
1419
1420<!-- *********************************************************************** -->
1421<div class="doc_section"> <a name="constants">Constants</a> </div>
1422<!-- *********************************************************************** -->
1423
1424<div class="doc_text">
1425
1426<p>LLVM has several different basic types of constants. This section describes
1427them all and their syntax.</p>
1428
1429</div>
1430
1431<!-- ======================================================================= -->
1432<div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1433
1434<div class="doc_text">
1435
1436<dl>
1437 <dt><b>Boolean constants</b></dt>
1438
1439 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1440 constants of the <tt><a href="#t_primitive">i1</a></tt> type.
1441 </dd>
1442
1443 <dt><b>Integer constants</b></dt>
1444
1445 <dd>Standard integers (such as '4') are constants of the <a
1446 href="#t_integer">integer</a> type. Negative numbers may be used with
1447 integer types.
1448 </dd>
1449
1450 <dt><b>Floating point constants</b></dt>
1451
1452 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1453 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1454 notation (see below). Floating point constants must have a <a
1455 href="#t_floating">floating point</a> type. </dd>
1456
1457 <dt><b>Null pointer constants</b></dt>
1458
1459 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1460 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1461
1462</dl>
1463
1464<p>The one non-intuitive notation for constants is the optional hexadecimal form
1465of floating point constants. For example, the form '<tt>double
14660x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
14674.5e+15</tt>'. The only time hexadecimal floating point constants are required
1468(and the only time that they are generated by the disassembler) is when a
1469floating point constant must be emitted but it cannot be represented as a
1470decimal floating point number. For example, NaN's, infinities, and other
1471special values are represented in their IEEE hexadecimal format so that
1472assembly and disassembly do not cause any bits to change in the constants.</p>
1473
1474</div>
1475
1476<!-- ======================================================================= -->
1477<div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1478</div>
1479
1480<div class="doc_text">
1481<p>Aggregate constants arise from aggregation of simple constants
1482and smaller aggregate constants.</p>
1483
1484<dl>
1485 <dt><b>Structure constants</b></dt>
1486
1487 <dd>Structure constants are represented with notation similar to structure
1488 type definitions (a comma separated list of elements, surrounded by braces
Chris Lattner6c8de962007-12-25 20:34:52 +00001489 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
1490 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>". Structure constants
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001491 must have <a href="#t_struct">structure type</a>, and the number and
1492 types of elements must match those specified by the type.
1493 </dd>
1494
1495 <dt><b>Array constants</b></dt>
1496
1497 <dd>Array constants are represented with notation similar to array type
1498 definitions (a comma separated list of elements, surrounded by square brackets
1499 (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74 ]</tt>". Array
1500 constants must have <a href="#t_array">array type</a>, and the number and
1501 types of elements must match those specified by the type.
1502 </dd>
1503
1504 <dt><b>Vector constants</b></dt>
1505
1506 <dd>Vector constants are represented with notation similar to vector type
1507 definitions (a comma separated list of elements, surrounded by
1508 less-than/greater-than's (<tt>&lt;&gt;</tt>)). For example: "<tt>&lt; i32 42,
1509 i32 11, i32 74, i32 100 &gt;</tt>". Vector constants must have <a
1510 href="#t_vector">vector type</a>, and the number and types of elements must
1511 match those specified by the type.
1512 </dd>
1513
1514 <dt><b>Zero initialization</b></dt>
1515
1516 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1517 value to zero of <em>any</em> type, including scalar and aggregate types.
1518 This is often used to avoid having to print large zero initializers (e.g. for
1519 large arrays) and is always exactly equivalent to using explicit zero
1520 initializers.
1521 </dd>
1522</dl>
1523
1524</div>
1525
1526<!-- ======================================================================= -->
1527<div class="doc_subsection">
1528 <a name="globalconstants">Global Variable and Function Addresses</a>
1529</div>
1530
1531<div class="doc_text">
1532
1533<p>The addresses of <a href="#globalvars">global variables</a> and <a
1534href="#functionstructure">functions</a> are always implicitly valid (link-time)
1535constants. These constants are explicitly referenced when the <a
1536href="#identifiers">identifier for the global</a> is used and always have <a
1537href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1538file:</p>
1539
1540<div class="doc_code">
1541<pre>
1542@X = global i32 17
1543@Y = global i32 42
1544@Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1545</pre>
1546</div>
1547
1548</div>
1549
1550<!-- ======================================================================= -->
1551<div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1552<div class="doc_text">
1553 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1554 no specific value. Undefined values may be of any type and be used anywhere
1555 a constant is permitted.</p>
1556
1557 <p>Undefined values indicate to the compiler that the program is well defined
1558 no matter what value is used, giving the compiler more freedom to optimize.
1559 </p>
1560</div>
1561
1562<!-- ======================================================================= -->
1563<div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1564</div>
1565
1566<div class="doc_text">
1567
1568<p>Constant expressions are used to allow expressions involving other constants
1569to be used as constants. Constant expressions may be of any <a
1570href="#t_firstclass">first class</a> type and may involve any LLVM operation
1571that does not have side effects (e.g. load and call are not supported). The
1572following is the syntax for constant expressions:</p>
1573
1574<dl>
1575 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1576 <dd>Truncate a constant to another type. The bit size of CST must be larger
1577 than the bit size of TYPE. Both types must be integers.</dd>
1578
1579 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1580 <dd>Zero extend a constant to another type. The bit size of CST must be
1581 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1582
1583 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1584 <dd>Sign extend a constant to another type. The bit size of CST must be
1585 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1586
1587 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1588 <dd>Truncate a floating point constant to another floating point type. The
1589 size of CST must be larger than the size of TYPE. Both types must be
1590 floating point.</dd>
1591
1592 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1593 <dd>Floating point extend a constant to another type. The size of CST must be
1594 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1595
Reid Spencere6adee82007-07-31 14:40:14 +00001596 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001597 <dd>Convert a floating point constant to the corresponding unsigned integer
Nate Begeman78246ca2007-11-17 03:58:34 +00001598 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1599 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1600 of the same number of elements. If the value won't fit in the integer type,
1601 the results are undefined.</dd>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001602
1603 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
1604 <dd>Convert a floating point constant to the corresponding signed integer
Nate Begeman78246ca2007-11-17 03:58:34 +00001605 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1606 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1607 of the same number of elements. If the value won't fit in the integer type,
1608 the results are undefined.</dd>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001609
1610 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
1611 <dd>Convert an unsigned integer constant to the corresponding floating point
Nate Begeman78246ca2007-11-17 03:58:34 +00001612 constant. TYPE must be a scalar or vector floating point type. CST must be of
1613 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1614 of the same number of elements. If the value won't fit in the floating point
1615 type, the results are undefined.</dd>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001616
1617 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
1618 <dd>Convert a signed integer constant to the corresponding floating point
Nate Begeman78246ca2007-11-17 03:58:34 +00001619 constant. TYPE must be a scalar or vector floating point type. CST must be of
1620 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1621 of the same number of elements. If the value won't fit in the floating point
1622 type, the results are undefined.</dd>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001623
1624 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
1625 <dd>Convert a pointer typed constant to the corresponding integer constant
1626 TYPE must be an integer type. CST must be of pointer type. The CST value is
1627 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1628
1629 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
1630 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1631 pointer type. CST must be of integer type. The CST value is zero extended,
1632 truncated, or unchanged to make it fit in a pointer size. This one is
1633 <i>really</i> dangerous!</dd>
1634
1635 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
1636 <dd>Convert a constant, CST, to another TYPE. The size of CST and TYPE must be
1637 identical (same number of bits). The conversion is done as if the CST value
1638 was stored to memory and read back as TYPE. In other words, no bits change
1639 with this operator, just the type. This can be used for conversion of
1640 vector types to any other type, as long as they have the same bit width. For
1641 pointers it is only valid to cast to another pointer type.
1642 </dd>
1643
1644 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1645
1646 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1647 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1648 instruction, the index list may have zero or more indexes, which are required
1649 to make sense for the type of "CSTPTR".</dd>
1650
1651 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1652
1653 <dd>Perform the <a href="#i_select">select operation</a> on
1654 constants.</dd>
1655
1656 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
1657 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
1658
1659 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
1660 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
1661
1662 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1663
1664 <dd>Perform the <a href="#i_extractelement">extractelement
1665 operation</a> on constants.
1666
1667 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1668
1669 <dd>Perform the <a href="#i_insertelement">insertelement
1670 operation</a> on constants.</dd>
1671
1672
1673 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
1674
1675 <dd>Perform the <a href="#i_shufflevector">shufflevector
1676 operation</a> on constants.</dd>
1677
1678 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1679
1680 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1681 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1682 binary</a> operations. The constraints on operands are the same as those for
1683 the corresponding instruction (e.g. no bitwise operations on floating point
1684 values are allowed).</dd>
1685</dl>
1686</div>
1687
1688<!-- *********************************************************************** -->
1689<div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1690<!-- *********************************************************************** -->
1691
1692<!-- ======================================================================= -->
1693<div class="doc_subsection">
1694<a name="inlineasm">Inline Assembler Expressions</a>
1695</div>
1696
1697<div class="doc_text">
1698
1699<p>
1700LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
1701Module-Level Inline Assembly</a>) through the use of a special value. This
1702value represents the inline assembler as a string (containing the instructions
1703to emit), a list of operand constraints (stored as a string), and a flag that
1704indicates whether or not the inline asm expression has side effects. An example
1705inline assembler expression is:
1706</p>
1707
1708<div class="doc_code">
1709<pre>
1710i32 (i32) asm "bswap $0", "=r,r"
1711</pre>
1712</div>
1713
1714<p>
1715Inline assembler expressions may <b>only</b> be used as the callee operand of
1716a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
1717</p>
1718
1719<div class="doc_code">
1720<pre>
1721%X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
1722</pre>
1723</div>
1724
1725<p>
1726Inline asms with side effects not visible in the constraint list must be marked
1727as having side effects. This is done through the use of the
1728'<tt>sideeffect</tt>' keyword, like so:
1729</p>
1730
1731<div class="doc_code">
1732<pre>
1733call void asm sideeffect "eieio", ""()
1734</pre>
1735</div>
1736
1737<p>TODO: The format of the asm and constraints string still need to be
1738documented here. Constraints on what can be done (e.g. duplication, moving, etc
1739need to be documented).
1740</p>
1741
1742</div>
1743
1744<!-- *********************************************************************** -->
1745<div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1746<!-- *********************************************************************** -->
1747
1748<div class="doc_text">
1749
1750<p>The LLVM instruction set consists of several different
1751classifications of instructions: <a href="#terminators">terminator
1752instructions</a>, <a href="#binaryops">binary instructions</a>,
1753<a href="#bitwiseops">bitwise binary instructions</a>, <a
1754 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1755instructions</a>.</p>
1756
1757</div>
1758
1759<!-- ======================================================================= -->
1760<div class="doc_subsection"> <a name="terminators">Terminator
1761Instructions</a> </div>
1762
1763<div class="doc_text">
1764
1765<p>As mentioned <a href="#functionstructure">previously</a>, every
1766basic block in a program ends with a "Terminator" instruction, which
1767indicates which block should be executed after the current block is
1768finished. These terminator instructions typically yield a '<tt>void</tt>'
1769value: they produce control flow, not values (the one exception being
1770the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1771<p>There are six different terminator instructions: the '<a
1772 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1773instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1774the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1775 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1776 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1777
1778</div>
1779
1780<!-- _______________________________________________________________________ -->
1781<div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1782Instruction</a> </div>
1783<div class="doc_text">
1784<h5>Syntax:</h5>
1785<pre> ret &lt;type&gt; &lt;value&gt; <i>; Return a value from a non-void function</i>
1786 ret void <i>; Return from void function</i>
1787</pre>
1788<h5>Overview:</h5>
1789<p>The '<tt>ret</tt>' instruction is used to return control flow (and a
1790value) from a function back to the caller.</p>
1791<p>There are two forms of the '<tt>ret</tt>' instruction: one that
1792returns a value and then causes control flow, and one that just causes
1793control flow to occur.</p>
1794<h5>Arguments:</h5>
1795<p>The '<tt>ret</tt>' instruction may return any '<a
1796 href="#t_firstclass">first class</a>' type. Notice that a function is
1797not <a href="#wellformed">well formed</a> if there exists a '<tt>ret</tt>'
1798instruction inside of the function that returns a value that does not
1799match the return type of the function.</p>
1800<h5>Semantics:</h5>
1801<p>When the '<tt>ret</tt>' instruction is executed, control flow
1802returns back to the calling function's context. If the caller is a "<a
1803 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1804the instruction after the call. If the caller was an "<a
1805 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1806at the beginning of the "normal" destination block. If the instruction
1807returns a value, that value shall set the call or invoke instruction's
1808return value.</p>
1809<h5>Example:</h5>
1810<pre> ret i32 5 <i>; Return an integer value of 5</i>
1811 ret void <i>; Return from a void function</i>
1812</pre>
1813</div>
1814<!-- _______________________________________________________________________ -->
1815<div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1816<div class="doc_text">
1817<h5>Syntax:</h5>
1818<pre> br i1 &lt;cond&gt;, label &lt;iftrue&gt;, label &lt;iffalse&gt;<br> br label &lt;dest&gt; <i>; Unconditional branch</i>
1819</pre>
1820<h5>Overview:</h5>
1821<p>The '<tt>br</tt>' instruction is used to cause control flow to
1822transfer to a different basic block in the current function. There are
1823two forms of this instruction, corresponding to a conditional branch
1824and an unconditional branch.</p>
1825<h5>Arguments:</h5>
1826<p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1827single '<tt>i1</tt>' value and two '<tt>label</tt>' values. The
1828unconditional form of the '<tt>br</tt>' instruction takes a single
1829'<tt>label</tt>' value as a target.</p>
1830<h5>Semantics:</h5>
1831<p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
1832argument is evaluated. If the value is <tt>true</tt>, control flows
1833to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1834control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1835<h5>Example:</h5>
1836<pre>Test:<br> %cond = <a href="#i_icmp">icmp</a> eq, i32 %a, %b<br> br i1 %cond, label %IfEqual, label %IfUnequal<br>IfEqual:<br> <a
1837 href="#i_ret">ret</a> i32 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> i32 0<br></pre>
1838</div>
1839<!-- _______________________________________________________________________ -->
1840<div class="doc_subsubsection">
1841 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
1842</div>
1843
1844<div class="doc_text">
1845<h5>Syntax:</h5>
1846
1847<pre>
1848 switch &lt;intty&gt; &lt;value&gt;, label &lt;defaultdest&gt; [ &lt;intty&gt; &lt;val&gt;, label &lt;dest&gt; ... ]
1849</pre>
1850
1851<h5>Overview:</h5>
1852
1853<p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
1854several different places. It is a generalization of the '<tt>br</tt>'
1855instruction, allowing a branch to occur to one of many possible
1856destinations.</p>
1857
1858
1859<h5>Arguments:</h5>
1860
1861<p>The '<tt>switch</tt>' instruction uses three parameters: an integer
1862comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
1863an array of pairs of comparison value constants and '<tt>label</tt>'s. The
1864table is not allowed to contain duplicate constant entries.</p>
1865
1866<h5>Semantics:</h5>
1867
1868<p>The <tt>switch</tt> instruction specifies a table of values and
1869destinations. When the '<tt>switch</tt>' instruction is executed, this
1870table is searched for the given value. If the value is found, control flow is
1871transfered to the corresponding destination; otherwise, control flow is
1872transfered to the default destination.</p>
1873
1874<h5>Implementation:</h5>
1875
1876<p>Depending on properties of the target machine and the particular
1877<tt>switch</tt> instruction, this instruction may be code generated in different
1878ways. For example, it could be generated as a series of chained conditional
1879branches or with a lookup table.</p>
1880
1881<h5>Example:</h5>
1882
1883<pre>
1884 <i>; Emulate a conditional br instruction</i>
1885 %Val = <a href="#i_zext">zext</a> i1 %value to i32
1886 switch i32 %Val, label %truedest [i32 0, label %falsedest ]
1887
1888 <i>; Emulate an unconditional br instruction</i>
1889 switch i32 0, label %dest [ ]
1890
1891 <i>; Implement a jump table:</i>
1892 switch i32 %val, label %otherwise [ i32 0, label %onzero
1893 i32 1, label %onone
1894 i32 2, label %ontwo ]
1895</pre>
1896</div>
1897
1898<!-- _______________________________________________________________________ -->
1899<div class="doc_subsubsection">
1900 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
1901</div>
1902
1903<div class="doc_text">
1904
1905<h5>Syntax:</h5>
1906
1907<pre>
1908 &lt;result&gt; = invoke [<a href="#callingconv">cconv</a>] &lt;ptr to function ty&gt; %&lt;function ptr val&gt;(&lt;function args&gt;)
1909 to label &lt;normal label&gt; unwind label &lt;exception label&gt;
1910</pre>
1911
1912<h5>Overview:</h5>
1913
1914<p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
1915function, with the possibility of control flow transfer to either the
1916'<tt>normal</tt>' label or the
1917'<tt>exception</tt>' label. If the callee function returns with the
1918"<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
1919"normal" label. If the callee (or any indirect callees) returns with the "<a
1920href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
1921continued at the dynamically nearest "exception" label.</p>
1922
1923<h5>Arguments:</h5>
1924
1925<p>This instruction requires several arguments:</p>
1926
1927<ol>
1928 <li>
1929 The optional "cconv" marker indicates which <a href="#callingconv">calling
1930 convention</a> the call should use. If none is specified, the call defaults
1931 to using C calling conventions.
1932 </li>
1933 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
1934 function value being invoked. In most cases, this is a direct function
1935 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
1936 an arbitrary pointer to function value.
1937 </li>
1938
1939 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
1940 function to be invoked. </li>
1941
1942 <li>'<tt>function args</tt>': argument list whose types match the function
1943 signature argument types. If the function signature indicates the function
1944 accepts a variable number of arguments, the extra arguments can be
1945 specified. </li>
1946
1947 <li>'<tt>normal label</tt>': the label reached when the called function
1948 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
1949
1950 <li>'<tt>exception label</tt>': the label reached when a callee returns with
1951 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
1952
1953</ol>
1954
1955<h5>Semantics:</h5>
1956
1957<p>This instruction is designed to operate as a standard '<tt><a
1958href="#i_call">call</a></tt>' instruction in most regards. The primary
1959difference is that it establishes an association with a label, which is used by
1960the runtime library to unwind the stack.</p>
1961
1962<p>This instruction is used in languages with destructors to ensure that proper
1963cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
1964exception. Additionally, this is important for implementation of
1965'<tt>catch</tt>' clauses in high-level languages that support them.</p>
1966
1967<h5>Example:</h5>
1968<pre>
1969 %retval = invoke i32 %Test(i32 15) to label %Continue
1970 unwind label %TestCleanup <i>; {i32}:retval set</i>
1971 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Test(i32 15) to label %Continue
1972 unwind label %TestCleanup <i>; {i32}:retval set</i>
1973</pre>
1974</div>
1975
1976
1977<!-- _______________________________________________________________________ -->
1978
1979<div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
1980Instruction</a> </div>
1981
1982<div class="doc_text">
1983
1984<h5>Syntax:</h5>
1985<pre>
1986 unwind
1987</pre>
1988
1989<h5>Overview:</h5>
1990
1991<p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
1992at the first callee in the dynamic call stack which used an <a
1993href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
1994primarily used to implement exception handling.</p>
1995
1996<h5>Semantics:</h5>
1997
1998<p>The '<tt>unwind</tt>' intrinsic causes execution of the current function to
1999immediately halt. The dynamic call stack is then searched for the first <a
2000href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
2001execution continues at the "exceptional" destination block specified by the
2002<tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
2003dynamic call chain, undefined behavior results.</p>
2004</div>
2005
2006<!-- _______________________________________________________________________ -->
2007
2008<div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
2009Instruction</a> </div>
2010
2011<div class="doc_text">
2012
2013<h5>Syntax:</h5>
2014<pre>
2015 unreachable
2016</pre>
2017
2018<h5>Overview:</h5>
2019
2020<p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
2021instruction is used to inform the optimizer that a particular portion of the
2022code is not reachable. This can be used to indicate that the code after a
2023no-return function cannot be reached, and other facts.</p>
2024
2025<h5>Semantics:</h5>
2026
2027<p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
2028</div>
2029
2030
2031
2032<!-- ======================================================================= -->
2033<div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
2034<div class="doc_text">
2035<p>Binary operators are used to do most of the computation in a
2036program. They require two operands, execute an operation on them, and
2037produce a single value. The operands might represent
2038multiple data, as is the case with the <a href="#t_vector">vector</a> data type.
2039The result value of a binary operator is not
2040necessarily the same type as its operands.</p>
2041<p>There are several different binary operators:</p>
2042</div>
2043<!-- _______________________________________________________________________ -->
2044<div class="doc_subsubsection"> <a name="i_add">'<tt>add</tt>'
2045Instruction</a> </div>
2046<div class="doc_text">
2047<h5>Syntax:</h5>
2048<pre> &lt;result&gt; = add &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2049</pre>
2050<h5>Overview:</h5>
2051<p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
2052<h5>Arguments:</h5>
2053<p>The two arguments to the '<tt>add</tt>' instruction must be either <a
2054 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a> values.
2055 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2056Both arguments must have identical types.</p>
2057<h5>Semantics:</h5>
2058<p>The value produced is the integer or floating point sum of the two
2059operands.</p>
Chris Lattner9aba1e22008-01-28 00:36:27 +00002060<p>If an integer sum has unsigned overflow, the result returned is the
2061mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2062the result.</p>
2063<p>Because LLVM integers use a two's complement representation, this
2064instruction is appropriate for both signed and unsigned integers.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002065<h5>Example:</h5>
2066<pre> &lt;result&gt; = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
2067</pre>
2068</div>
2069<!-- _______________________________________________________________________ -->
2070<div class="doc_subsubsection"> <a name="i_sub">'<tt>sub</tt>'
2071Instruction</a> </div>
2072<div class="doc_text">
2073<h5>Syntax:</h5>
2074<pre> &lt;result&gt; = sub &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2075</pre>
2076<h5>Overview:</h5>
2077<p>The '<tt>sub</tt>' instruction returns the difference of its two
2078operands.</p>
2079<p>Note that the '<tt>sub</tt>' instruction is used to represent the '<tt>neg</tt>'
2080instruction present in most other intermediate representations.</p>
2081<h5>Arguments:</h5>
2082<p>The two arguments to the '<tt>sub</tt>' instruction must be either <a
2083 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
2084values.
2085This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2086Both arguments must have identical types.</p>
2087<h5>Semantics:</h5>
2088<p>The value produced is the integer or floating point difference of
2089the two operands.</p>
Chris Lattner9aba1e22008-01-28 00:36:27 +00002090<p>If an integer difference has unsigned overflow, the result returned is the
2091mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2092the result.</p>
2093<p>Because LLVM integers use a two's complement representation, this
2094instruction is appropriate for both signed and unsigned integers.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002095<h5>Example:</h5>
2096<pre>
2097 &lt;result&gt; = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
2098 &lt;result&gt; = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
2099</pre>
2100</div>
2101<!-- _______________________________________________________________________ -->
2102<div class="doc_subsubsection"> <a name="i_mul">'<tt>mul</tt>'
2103Instruction</a> </div>
2104<div class="doc_text">
2105<h5>Syntax:</h5>
2106<pre> &lt;result&gt; = mul &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2107</pre>
2108<h5>Overview:</h5>
2109<p>The '<tt>mul</tt>' instruction returns the product of its two
2110operands.</p>
2111<h5>Arguments:</h5>
2112<p>The two arguments to the '<tt>mul</tt>' instruction must be either <a
2113 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
2114values.
2115This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2116Both arguments must have identical types.</p>
2117<h5>Semantics:</h5>
2118<p>The value produced is the integer or floating point product of the
2119two operands.</p>
Chris Lattner9aba1e22008-01-28 00:36:27 +00002120<p>If the result of an integer multiplication has unsigned overflow,
2121the result returned is the mathematical result modulo
21222<sup>n</sup>, where n is the bit width of the result.</p>
2123<p>Because LLVM integers use a two's complement representation, and the
2124result is the same width as the operands, this instruction returns the
2125correct result for both signed and unsigned integers. If a full product
2126(e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands
2127should be sign-extended or zero-extended as appropriate to the
2128width of the full product.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002129<h5>Example:</h5>
2130<pre> &lt;result&gt; = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
2131</pre>
2132</div>
2133<!-- _______________________________________________________________________ -->
2134<div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
2135</a></div>
2136<div class="doc_text">
2137<h5>Syntax:</h5>
2138<pre> &lt;result&gt; = udiv &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2139</pre>
2140<h5>Overview:</h5>
2141<p>The '<tt>udiv</tt>' instruction returns the quotient of its two
2142operands.</p>
2143<h5>Arguments:</h5>
2144<p>The two arguments to the '<tt>udiv</tt>' instruction must be
2145<a href="#t_integer">integer</a> values. Both arguments must have identical
2146types. This instruction can also take <a href="#t_vector">vector</a> versions
2147of the values in which case the elements must be integers.</p>
2148<h5>Semantics:</h5>
Chris Lattner9aba1e22008-01-28 00:36:27 +00002149<p>The value produced is the unsigned integer quotient of the two operands.</p>
2150<p>Note that unsigned integer division and signed integer division are distinct
2151operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
2152<p>Division by zero leads to undefined behavior.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002153<h5>Example:</h5>
2154<pre> &lt;result&gt; = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2155</pre>
2156</div>
2157<!-- _______________________________________________________________________ -->
2158<div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
2159</a> </div>
2160<div class="doc_text">
2161<h5>Syntax:</h5>
2162<pre> &lt;result&gt; = sdiv &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2163</pre>
2164<h5>Overview:</h5>
2165<p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
2166operands.</p>
2167<h5>Arguments:</h5>
2168<p>The two arguments to the '<tt>sdiv</tt>' instruction must be
2169<a href="#t_integer">integer</a> values. Both arguments must have identical
2170types. This instruction can also take <a href="#t_vector">vector</a> versions
2171of the values in which case the elements must be integers.</p>
2172<h5>Semantics:</h5>
Chris Lattner9aba1e22008-01-28 00:36:27 +00002173<p>The value produced is the signed integer quotient of the two operands.</p>
2174<p>Note that signed integer division and unsigned integer division are distinct
2175operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
2176<p>Division by zero leads to undefined behavior. Overflow also leads to
2177undefined behavior; this is a rare case, but can occur, for example,
2178by doing a 32-bit division of -2147483648 by -1.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002179<h5>Example:</h5>
2180<pre> &lt;result&gt; = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2181</pre>
2182</div>
2183<!-- _______________________________________________________________________ -->
2184<div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
2185Instruction</a> </div>
2186<div class="doc_text">
2187<h5>Syntax:</h5>
2188<pre> &lt;result&gt; = fdiv &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2189</pre>
2190<h5>Overview:</h5>
2191<p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
2192operands.</p>
2193<h5>Arguments:</h5>
2194<p>The two arguments to the '<tt>fdiv</tt>' instruction must be
2195<a href="#t_floating">floating point</a> values. Both arguments must have
2196identical types. This instruction can also take <a href="#t_vector">vector</a>
2197versions of floating point values.</p>
2198<h5>Semantics:</h5>
2199<p>The value produced is the floating point quotient of the two operands.</p>
2200<h5>Example:</h5>
2201<pre> &lt;result&gt; = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
2202</pre>
2203</div>
2204<!-- _______________________________________________________________________ -->
2205<div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
2206</div>
2207<div class="doc_text">
2208<h5>Syntax:</h5>
2209<pre> &lt;result&gt; = urem &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2210</pre>
2211<h5>Overview:</h5>
2212<p>The '<tt>urem</tt>' instruction returns the remainder from the
2213unsigned division of its two arguments.</p>
2214<h5>Arguments:</h5>
2215<p>The two arguments to the '<tt>urem</tt>' instruction must be
2216<a href="#t_integer">integer</a> values. Both arguments must have identical
Dan Gohman3e3fd8c2007-11-05 23:35:22 +00002217types. This instruction can also take <a href="#t_vector">vector</a> versions
2218of the values in which case the elements must be integers.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002219<h5>Semantics:</h5>
2220<p>This instruction returns the unsigned integer <i>remainder</i> of a division.
2221This instruction always performs an unsigned division to get the remainder,
2222regardless of whether the arguments are unsigned or not.</p>
Chris Lattner9aba1e22008-01-28 00:36:27 +00002223<p>Note that unsigned integer remainder and signed integer remainder are
2224distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
2225<p>Taking the remainder of a division by zero leads to undefined behavior.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002226<h5>Example:</h5>
2227<pre> &lt;result&gt; = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2228</pre>
2229
2230</div>
2231<!-- _______________________________________________________________________ -->
2232<div class="doc_subsubsection"> <a name="i_srem">'<tt>srem</tt>'
2233Instruction</a> </div>
2234<div class="doc_text">
2235<h5>Syntax:</h5>
2236<pre> &lt;result&gt; = srem &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2237</pre>
2238<h5>Overview:</h5>
2239<p>The '<tt>srem</tt>' instruction returns the remainder from the
Dan Gohman3e3fd8c2007-11-05 23:35:22 +00002240signed division of its two operands. This instruction can also take
2241<a href="#t_vector">vector</a> versions of the values in which case
2242the elements must be integers.</p>
Chris Lattner08497ce2008-01-04 04:33:49 +00002243
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002244<h5>Arguments:</h5>
2245<p>The two arguments to the '<tt>srem</tt>' instruction must be
2246<a href="#t_integer">integer</a> values. Both arguments must have identical
2247types.</p>
2248<h5>Semantics:</h5>
2249<p>This instruction returns the <i>remainder</i> of a division (where the result
2250has the same sign as the dividend, <tt>var1</tt>), not the <i>modulo</i>
2251operator (where the result has the same sign as the divisor, <tt>var2</tt>) of
2252a value. For more information about the difference, see <a
2253 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
2254Math Forum</a>. For a table of how this is implemented in various languages,
2255please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
2256Wikipedia: modulo operation</a>.</p>
Chris Lattner9aba1e22008-01-28 00:36:27 +00002257<p>Note that signed integer remainder and unsigned integer remainder are
2258distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
2259<p>Taking the remainder of a division by zero leads to undefined behavior.
2260Overflow also leads to undefined behavior; this is a rare case, but can occur,
2261for example, by taking the remainder of a 32-bit division of -2147483648 by -1.
2262(The remainder doesn't actually overflow, but this rule lets srem be
2263implemented using instructions that return both the result of the division
2264and the remainder.)</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002265<h5>Example:</h5>
2266<pre> &lt;result&gt; = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2267</pre>
2268
2269</div>
2270<!-- _______________________________________________________________________ -->
2271<div class="doc_subsubsection"> <a name="i_frem">'<tt>frem</tt>'
2272Instruction</a> </div>
2273<div class="doc_text">
2274<h5>Syntax:</h5>
2275<pre> &lt;result&gt; = frem &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2276</pre>
2277<h5>Overview:</h5>
2278<p>The '<tt>frem</tt>' instruction returns the remainder from the
2279division of its two operands.</p>
2280<h5>Arguments:</h5>
2281<p>The two arguments to the '<tt>frem</tt>' instruction must be
2282<a href="#t_floating">floating point</a> values. Both arguments must have
Dan Gohman3e3fd8c2007-11-05 23:35:22 +00002283identical types. This instruction can also take <a href="#t_vector">vector</a>
2284versions of floating point values.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002285<h5>Semantics:</h5>
2286<p>This instruction returns the <i>remainder</i> of a division.</p>
2287<h5>Example:</h5>
2288<pre> &lt;result&gt; = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
2289</pre>
2290</div>
2291
2292<!-- ======================================================================= -->
2293<div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
2294Operations</a> </div>
2295<div class="doc_text">
2296<p>Bitwise binary operators are used to do various forms of
2297bit-twiddling in a program. They are generally very efficient
2298instructions and can commonly be strength reduced from other
2299instructions. They require two operands, execute an operation on them,
2300and produce a single value. The resulting value of the bitwise binary
2301operators is always the same type as its first operand.</p>
2302</div>
2303
2304<!-- _______________________________________________________________________ -->
2305<div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
2306Instruction</a> </div>
2307<div class="doc_text">
2308<h5>Syntax:</h5>
2309<pre> &lt;result&gt; = shl &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2310</pre>
Chris Lattnerd939d9f2007-10-03 21:01:14 +00002311
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002312<h5>Overview:</h5>
Chris Lattnerd939d9f2007-10-03 21:01:14 +00002313
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002314<p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2315the left a specified number of bits.</p>
Chris Lattnerd939d9f2007-10-03 21:01:14 +00002316
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002317<h5>Arguments:</h5>
Chris Lattnerd939d9f2007-10-03 21:01:14 +00002318
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002319<p>Both arguments to the '<tt>shl</tt>' instruction must be the same <a
2320 href="#t_integer">integer</a> type.</p>
Chris Lattnerd939d9f2007-10-03 21:01:14 +00002321
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002322<h5>Semantics:</h5>
Chris Lattnerd939d9f2007-10-03 21:01:14 +00002323
2324<p>The value produced is <tt>var1</tt> * 2<sup><tt>var2</tt></sup>. If
2325<tt>var2</tt> is (statically or dynamically) equal to or larger than the number
2326of bits in <tt>var1</tt>, the result is undefined.</p>
2327
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002328<h5>Example:</h5><pre>
2329 &lt;result&gt; = shl i32 4, %var <i>; yields {i32}: 4 &lt;&lt; %var</i>
2330 &lt;result&gt; = shl i32 4, 2 <i>; yields {i32}: 16</i>
2331 &lt;result&gt; = shl i32 1, 10 <i>; yields {i32}: 1024</i>
Chris Lattnerd939d9f2007-10-03 21:01:14 +00002332 &lt;result&gt; = shl i32 1, 32 <i>; undefined</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002333</pre>
2334</div>
2335<!-- _______________________________________________________________________ -->
2336<div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
2337Instruction</a> </div>
2338<div class="doc_text">
2339<h5>Syntax:</h5>
2340<pre> &lt;result&gt; = lshr &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2341</pre>
2342
2343<h5>Overview:</h5>
2344<p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
2345operand shifted to the right a specified number of bits with zero fill.</p>
2346
2347<h5>Arguments:</h5>
2348<p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
2349<a href="#t_integer">integer</a> type.</p>
2350
2351<h5>Semantics:</h5>
Chris Lattnerd939d9f2007-10-03 21:01:14 +00002352
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002353<p>This instruction always performs a logical shift right operation. The most
2354significant bits of the result will be filled with zero bits after the
Chris Lattnerd939d9f2007-10-03 21:01:14 +00002355shift. If <tt>var2</tt> is (statically or dynamically) equal to or larger than
2356the number of bits in <tt>var1</tt>, the result is undefined.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002357
2358<h5>Example:</h5>
2359<pre>
2360 &lt;result&gt; = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
2361 &lt;result&gt; = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
2362 &lt;result&gt; = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
2363 &lt;result&gt; = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
Chris Lattnerd939d9f2007-10-03 21:01:14 +00002364 &lt;result&gt; = lshr i32 1, 32 <i>; undefined</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002365</pre>
2366</div>
2367
2368<!-- _______________________________________________________________________ -->
2369<div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
2370Instruction</a> </div>
2371<div class="doc_text">
2372
2373<h5>Syntax:</h5>
2374<pre> &lt;result&gt; = ashr &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2375</pre>
2376
2377<h5>Overview:</h5>
2378<p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
2379operand shifted to the right a specified number of bits with sign extension.</p>
2380
2381<h5>Arguments:</h5>
2382<p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
2383<a href="#t_integer">integer</a> type.</p>
2384
2385<h5>Semantics:</h5>
2386<p>This instruction always performs an arithmetic shift right operation,
2387The most significant bits of the result will be filled with the sign bit
Chris Lattnerd939d9f2007-10-03 21:01:14 +00002388of <tt>var1</tt>. If <tt>var2</tt> is (statically or dynamically) equal to or
2389larger than the number of bits in <tt>var1</tt>, the result is undefined.
2390</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002391
2392<h5>Example:</h5>
2393<pre>
2394 &lt;result&gt; = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
2395 &lt;result&gt; = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
2396 &lt;result&gt; = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
2397 &lt;result&gt; = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
Chris Lattnerd939d9f2007-10-03 21:01:14 +00002398 &lt;result&gt; = ashr i32 1, 32 <i>; undefined</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002399</pre>
2400</div>
2401
2402<!-- _______________________________________________________________________ -->
2403<div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
2404Instruction</a> </div>
2405<div class="doc_text">
2406<h5>Syntax:</h5>
2407<pre> &lt;result&gt; = and &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2408</pre>
2409<h5>Overview:</h5>
2410<p>The '<tt>and</tt>' instruction returns the bitwise logical and of
2411its two operands.</p>
2412<h5>Arguments:</h5>
2413<p>The two arguments to the '<tt>and</tt>' instruction must be <a
2414 href="#t_integer">integer</a> values. Both arguments must have
2415identical types.</p>
2416<h5>Semantics:</h5>
2417<p>The truth table used for the '<tt>and</tt>' instruction is:</p>
2418<p> </p>
2419<div style="align: center">
2420<table border="1" cellspacing="0" cellpadding="4">
2421 <tbody>
2422 <tr>
2423 <td>In0</td>
2424 <td>In1</td>
2425 <td>Out</td>
2426 </tr>
2427 <tr>
2428 <td>0</td>
2429 <td>0</td>
2430 <td>0</td>
2431 </tr>
2432 <tr>
2433 <td>0</td>
2434 <td>1</td>
2435 <td>0</td>
2436 </tr>
2437 <tr>
2438 <td>1</td>
2439 <td>0</td>
2440 <td>0</td>
2441 </tr>
2442 <tr>
2443 <td>1</td>
2444 <td>1</td>
2445 <td>1</td>
2446 </tr>
2447 </tbody>
2448</table>
2449</div>
2450<h5>Example:</h5>
2451<pre> &lt;result&gt; = and i32 4, %var <i>; yields {i32}:result = 4 &amp; %var</i>
2452 &lt;result&gt; = and i32 15, 40 <i>; yields {i32}:result = 8</i>
2453 &lt;result&gt; = and i32 4, 8 <i>; yields {i32}:result = 0</i>
2454</pre>
2455</div>
2456<!-- _______________________________________________________________________ -->
2457<div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
2458<div class="doc_text">
2459<h5>Syntax:</h5>
2460<pre> &lt;result&gt; = or &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2461</pre>
2462<h5>Overview:</h5>
2463<p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
2464or of its two operands.</p>
2465<h5>Arguments:</h5>
2466<p>The two arguments to the '<tt>or</tt>' instruction must be <a
2467 href="#t_integer">integer</a> values. Both arguments must have
2468identical types.</p>
2469<h5>Semantics:</h5>
2470<p>The truth table used for the '<tt>or</tt>' instruction is:</p>
2471<p> </p>
2472<div style="align: center">
2473<table border="1" cellspacing="0" cellpadding="4">
2474 <tbody>
2475 <tr>
2476 <td>In0</td>
2477 <td>In1</td>
2478 <td>Out</td>
2479 </tr>
2480 <tr>
2481 <td>0</td>
2482 <td>0</td>
2483 <td>0</td>
2484 </tr>
2485 <tr>
2486 <td>0</td>
2487 <td>1</td>
2488 <td>1</td>
2489 </tr>
2490 <tr>
2491 <td>1</td>
2492 <td>0</td>
2493 <td>1</td>
2494 </tr>
2495 <tr>
2496 <td>1</td>
2497 <td>1</td>
2498 <td>1</td>
2499 </tr>
2500 </tbody>
2501</table>
2502</div>
2503<h5>Example:</h5>
2504<pre> &lt;result&gt; = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
2505 &lt;result&gt; = or i32 15, 40 <i>; yields {i32}:result = 47</i>
2506 &lt;result&gt; = or i32 4, 8 <i>; yields {i32}:result = 12</i>
2507</pre>
2508</div>
2509<!-- _______________________________________________________________________ -->
2510<div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
2511Instruction</a> </div>
2512<div class="doc_text">
2513<h5>Syntax:</h5>
2514<pre> &lt;result&gt; = xor &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2515</pre>
2516<h5>Overview:</h5>
2517<p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
2518or of its two operands. The <tt>xor</tt> is used to implement the
2519"one's complement" operation, which is the "~" operator in C.</p>
2520<h5>Arguments:</h5>
2521<p>The two arguments to the '<tt>xor</tt>' instruction must be <a
2522 href="#t_integer">integer</a> values. Both arguments must have
2523identical types.</p>
2524<h5>Semantics:</h5>
2525<p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
2526<p> </p>
2527<div style="align: center">
2528<table border="1" cellspacing="0" cellpadding="4">
2529 <tbody>
2530 <tr>
2531 <td>In0</td>
2532 <td>In1</td>
2533 <td>Out</td>
2534 </tr>
2535 <tr>
2536 <td>0</td>
2537 <td>0</td>
2538 <td>0</td>
2539 </tr>
2540 <tr>
2541 <td>0</td>
2542 <td>1</td>
2543 <td>1</td>
2544 </tr>
2545 <tr>
2546 <td>1</td>
2547 <td>0</td>
2548 <td>1</td>
2549 </tr>
2550 <tr>
2551 <td>1</td>
2552 <td>1</td>
2553 <td>0</td>
2554 </tr>
2555 </tbody>
2556</table>
2557</div>
2558<p> </p>
2559<h5>Example:</h5>
2560<pre> &lt;result&gt; = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
2561 &lt;result&gt; = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
2562 &lt;result&gt; = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
2563 &lt;result&gt; = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
2564</pre>
2565</div>
2566
2567<!-- ======================================================================= -->
2568<div class="doc_subsection">
2569 <a name="vectorops">Vector Operations</a>
2570</div>
2571
2572<div class="doc_text">
2573
2574<p>LLVM supports several instructions to represent vector operations in a
2575target-independent manner. These instructions cover the element-access and
2576vector-specific operations needed to process vectors effectively. While LLVM
2577does directly support these vector operations, many sophisticated algorithms
2578will want to use target-specific intrinsics to take full advantage of a specific
2579target.</p>
2580
2581</div>
2582
2583<!-- _______________________________________________________________________ -->
2584<div class="doc_subsubsection">
2585 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2586</div>
2587
2588<div class="doc_text">
2589
2590<h5>Syntax:</h5>
2591
2592<pre>
2593 &lt;result&gt; = extractelement &lt;n x &lt;ty&gt;&gt; &lt;val&gt;, i32 &lt;idx&gt; <i>; yields &lt;ty&gt;</i>
2594</pre>
2595
2596<h5>Overview:</h5>
2597
2598<p>
2599The '<tt>extractelement</tt>' instruction extracts a single scalar
2600element from a vector at a specified index.
2601</p>
2602
2603
2604<h5>Arguments:</h5>
2605
2606<p>
2607The first operand of an '<tt>extractelement</tt>' instruction is a
2608value of <a href="#t_vector">vector</a> type. The second operand is
2609an index indicating the position from which to extract the element.
2610The index may be a variable.</p>
2611
2612<h5>Semantics:</h5>
2613
2614<p>
2615The result is a scalar of the same type as the element type of
2616<tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2617<tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2618results are undefined.
2619</p>
2620
2621<h5>Example:</h5>
2622
2623<pre>
2624 %result = extractelement &lt;4 x i32&gt; %vec, i32 0 <i>; yields i32</i>
2625</pre>
2626</div>
2627
2628
2629<!-- _______________________________________________________________________ -->
2630<div class="doc_subsubsection">
2631 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
2632</div>
2633
2634<div class="doc_text">
2635
2636<h5>Syntax:</h5>
2637
2638<pre>
2639 &lt;result&gt; = insertelement &lt;n x &lt;ty&gt;&gt; &lt;val&gt;, &lt;ty&gt; &lt;elt&gt, i32 &lt;idx&gt; <i>; yields &lt;n x &lt;ty&gt;&gt;</i>
2640</pre>
2641
2642<h5>Overview:</h5>
2643
2644<p>
2645The '<tt>insertelement</tt>' instruction inserts a scalar
2646element into a vector at a specified index.
2647</p>
2648
2649
2650<h5>Arguments:</h5>
2651
2652<p>
2653The first operand of an '<tt>insertelement</tt>' instruction is a
2654value of <a href="#t_vector">vector</a> type. The second operand is a
2655scalar value whose type must equal the element type of the first
2656operand. The third operand is an index indicating the position at
2657which to insert the value. The index may be a variable.</p>
2658
2659<h5>Semantics:</h5>
2660
2661<p>
2662The result is a vector of the same type as <tt>val</tt>. Its
2663element values are those of <tt>val</tt> except at position
2664<tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
2665exceeds the length of <tt>val</tt>, the results are undefined.
2666</p>
2667
2668<h5>Example:</h5>
2669
2670<pre>
2671 %result = insertelement &lt;4 x i32&gt; %vec, i32 1, i32 0 <i>; yields &lt;4 x i32&gt;</i>
2672</pre>
2673</div>
2674
2675<!-- _______________________________________________________________________ -->
2676<div class="doc_subsubsection">
2677 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
2678</div>
2679
2680<div class="doc_text">
2681
2682<h5>Syntax:</h5>
2683
2684<pre>
2685 &lt;result&gt; = shufflevector &lt;n x &lt;ty&gt;&gt; &lt;v1&gt;, &lt;n x &lt;ty&gt;&gt; &lt;v2&gt;, &lt;n x i32&gt; &lt;mask&gt; <i>; yields &lt;n x &lt;ty&gt;&gt;</i>
2686</pre>
2687
2688<h5>Overview:</h5>
2689
2690<p>
2691The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
2692from two input vectors, returning a vector of the same type.
2693</p>
2694
2695<h5>Arguments:</h5>
2696
2697<p>
2698The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
2699with types that match each other and types that match the result of the
2700instruction. The third argument is a shuffle mask, which has the same number
2701of elements as the other vector type, but whose element type is always 'i32'.
2702</p>
2703
2704<p>
2705The shuffle mask operand is required to be a constant vector with either
2706constant integer or undef values.
2707</p>
2708
2709<h5>Semantics:</h5>
2710
2711<p>
2712The elements of the two input vectors are numbered from left to right across
2713both of the vectors. The shuffle mask operand specifies, for each element of
2714the result vector, which element of the two input registers the result element
2715gets. The element selector may be undef (meaning "don't care") and the second
2716operand may be undef if performing a shuffle from only one vector.
2717</p>
2718
2719<h5>Example:</h5>
2720
2721<pre>
2722 %result = shufflevector &lt;4 x i32&gt; %v1, &lt;4 x i32&gt; %v2,
2723 &lt;4 x i32&gt; &lt;i32 0, i32 4, i32 1, i32 5&gt; <i>; yields &lt;4 x i32&gt;</i>
2724 %result = shufflevector &lt;4 x i32&gt; %v1, &lt;4 x i32&gt; undef,
2725 &lt;4 x i32&gt; &lt;i32 0, i32 1, i32 2, i32 3&gt; <i>; yields &lt;4 x i32&gt;</i> - Identity shuffle.
2726</pre>
2727</div>
2728
2729
2730<!-- ======================================================================= -->
2731<div class="doc_subsection">
2732 <a name="memoryops">Memory Access and Addressing Operations</a>
2733</div>
2734
2735<div class="doc_text">
2736
2737<p>A key design point of an SSA-based representation is how it
2738represents memory. In LLVM, no memory locations are in SSA form, which
2739makes things very simple. This section describes how to read, write,
2740allocate, and free memory in LLVM.</p>
2741
2742</div>
2743
2744<!-- _______________________________________________________________________ -->
2745<div class="doc_subsubsection">
2746 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
2747</div>
2748
2749<div class="doc_text">
2750
2751<h5>Syntax:</h5>
2752
2753<pre>
2754 &lt;result&gt; = malloc &lt;type&gt;[, i32 &lt;NumElements&gt;][, align &lt;alignment&gt;] <i>; yields {type*}:result</i>
2755</pre>
2756
2757<h5>Overview:</h5>
2758
2759<p>The '<tt>malloc</tt>' instruction allocates memory from the system
Christopher Lambcfe00962007-12-17 01:00:21 +00002760heap and returns a pointer to it. The object is always allocated in the generic
2761address space (address space zero).</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002762
2763<h5>Arguments:</h5>
2764
2765<p>The '<tt>malloc</tt>' instruction allocates
2766<tt>sizeof(&lt;type&gt;)*NumElements</tt>
2767bytes of memory from the operating system and returns a pointer of the
2768appropriate type to the program. If "NumElements" is specified, it is the
Gabor Greif5082cf42008-02-09 22:24:34 +00002769number of elements allocated, otherwise "NumElements" is defaulted to be one.
2770If an alignment is specified, the value result of the allocation is guaranteed to
2771be aligned to at least that boundary. If not specified, or if zero, the target can
2772choose to align the allocation on any convenient boundary.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002773
2774<p>'<tt>type</tt>' must be a sized type.</p>
2775
2776<h5>Semantics:</h5>
2777
2778<p>Memory is allocated using the system "<tt>malloc</tt>" function, and
2779a pointer is returned.</p>
2780
2781<h5>Example:</h5>
2782
2783<pre>
2784 %array = malloc [4 x i8 ] <i>; yields {[%4 x i8]*}:array</i>
2785
2786 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
2787 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
2788 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
2789 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
2790 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
2791</pre>
2792</div>
2793
2794<!-- _______________________________________________________________________ -->
2795<div class="doc_subsubsection">
2796 <a name="i_free">'<tt>free</tt>' Instruction</a>
2797</div>
2798
2799<div class="doc_text">
2800
2801<h5>Syntax:</h5>
2802
2803<pre>
2804 free &lt;type&gt; &lt;value&gt; <i>; yields {void}</i>
2805</pre>
2806
2807<h5>Overview:</h5>
2808
2809<p>The '<tt>free</tt>' instruction returns memory back to the unused
2810memory heap to be reallocated in the future.</p>
2811
2812<h5>Arguments:</h5>
2813
2814<p>'<tt>value</tt>' shall be a pointer value that points to a value
2815that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
2816instruction.</p>
2817
2818<h5>Semantics:</h5>
2819
2820<p>Access to the memory pointed to by the pointer is no longer defined
2821after this instruction executes.</p>
2822
2823<h5>Example:</h5>
2824
2825<pre>
2826 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
2827 free [4 x i8]* %array
2828</pre>
2829</div>
2830
2831<!-- _______________________________________________________________________ -->
2832<div class="doc_subsubsection">
2833 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
2834</div>
2835
2836<div class="doc_text">
2837
2838<h5>Syntax:</h5>
2839
2840<pre>
2841 &lt;result&gt; = alloca &lt;type&gt;[, i32 &lt;NumElements&gt;][, align &lt;alignment&gt;] <i>; yields {type*}:result</i>
2842</pre>
2843
2844<h5>Overview:</h5>
2845
2846<p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
2847currently executing function, to be automatically released when this function
Christopher Lambcfe00962007-12-17 01:00:21 +00002848returns to its caller. The object is always allocated in the generic address
2849space (address space zero).</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002850
2851<h5>Arguments:</h5>
2852
2853<p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(&lt;type&gt;)*NumElements</tt>
2854bytes of memory on the runtime stack, returning a pointer of the
Gabor Greif5082cf42008-02-09 22:24:34 +00002855appropriate type to the program. If "NumElements" is specified, it is the
2856number of elements allocated, otherwise "NumElements" is defaulted to be one.
2857If an alignment is specified, the value result of the allocation is guaranteed
2858to be aligned to at least that boundary. If not specified, or if zero, the target
2859can choose to align the allocation on any convenient boundary.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002860
2861<p>'<tt>type</tt>' may be any sized type.</p>
2862
2863<h5>Semantics:</h5>
2864
2865<p>Memory is allocated; a pointer is returned. '<tt>alloca</tt>'d
2866memory is automatically released when the function returns. The '<tt>alloca</tt>'
2867instruction is commonly used to represent automatic variables that must
2868have an address available. When the function returns (either with the <tt><a
2869 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
2870instructions), the memory is reclaimed.</p>
2871
2872<h5>Example:</h5>
2873
2874<pre>
2875 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
2876 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
2877 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
2878 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
2879</pre>
2880</div>
2881
2882<!-- _______________________________________________________________________ -->
2883<div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
2884Instruction</a> </div>
2885<div class="doc_text">
2886<h5>Syntax:</h5>
2887<pre> &lt;result&gt; = load &lt;ty&gt;* &lt;pointer&gt;[, align &lt;alignment&gt;]<br> &lt;result&gt; = volatile load &lt;ty&gt;* &lt;pointer&gt;[, align &lt;alignment&gt;]<br></pre>
2888<h5>Overview:</h5>
2889<p>The '<tt>load</tt>' instruction is used to read from memory.</p>
2890<h5>Arguments:</h5>
2891<p>The argument to the '<tt>load</tt>' instruction specifies the memory
2892address from which to load. The pointer must point to a <a
2893 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
2894marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
2895the number or order of execution of this <tt>load</tt> with other
2896volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
2897instructions. </p>
Chris Lattner21003c82008-01-06 21:04:43 +00002898<p>
2899The optional "align" argument specifies the alignment of the operation
2900(that is, the alignment of the memory address). A value of 0 or an
2901omitted "align" argument means that the operation has the preferential
2902alignment for the target. It is the responsibility of the code emitter
2903to ensure that the alignment information is correct. Overestimating
2904the alignment results in an undefined behavior. Underestimating the
2905alignment may produce less efficient code. An alignment of 1 is always
2906safe.
2907</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002908<h5>Semantics:</h5>
2909<p>The location of memory pointed to is loaded.</p>
2910<h5>Examples:</h5>
2911<pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
2912 <a
2913 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
2914 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
2915</pre>
2916</div>
2917<!-- _______________________________________________________________________ -->
2918<div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
2919Instruction</a> </div>
2920<div class="doc_text">
2921<h5>Syntax:</h5>
2922<pre> store &lt;ty&gt; &lt;value&gt;, &lt;ty&gt;* &lt;pointer&gt;[, align &lt;alignment&gt;] <i>; yields {void}</i>
2923 volatile store &lt;ty&gt; &lt;value&gt;, &lt;ty&gt;* &lt;pointer&gt;[, align &lt;alignment&gt;] <i>; yields {void}</i>
2924</pre>
2925<h5>Overview:</h5>
2926<p>The '<tt>store</tt>' instruction is used to write to memory.</p>
2927<h5>Arguments:</h5>
2928<p>There are two arguments to the '<tt>store</tt>' instruction: a value
2929to store and an address at which to store it. The type of the '<tt>&lt;pointer&gt;</tt>'
2930operand must be a pointer to the type of the '<tt>&lt;value&gt;</tt>'
2931operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
2932optimizer is not allowed to modify the number or order of execution of
2933this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
2934 href="#i_store">store</a></tt> instructions.</p>
Chris Lattner21003c82008-01-06 21:04:43 +00002935<p>
2936The optional "align" argument specifies the alignment of the operation
2937(that is, the alignment of the memory address). A value of 0 or an
2938omitted "align" argument means that the operation has the preferential
2939alignment for the target. It is the responsibility of the code emitter
2940to ensure that the alignment information is correct. Overestimating
2941the alignment results in an undefined behavior. Underestimating the
2942alignment may produce less efficient code. An alignment of 1 is always
2943safe.
2944</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002945<h5>Semantics:</h5>
2946<p>The contents of memory are updated to contain '<tt>&lt;value&gt;</tt>'
2947at the location specified by the '<tt>&lt;pointer&gt;</tt>' operand.</p>
2948<h5>Example:</h5>
2949<pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
Bill Wendling63ffa142007-10-22 05:10:05 +00002950 store i32 3, i32* %ptr <i>; yields {void}</i>
2951 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002952</pre>
2953</div>
2954
2955<!-- _______________________________________________________________________ -->
2956<div class="doc_subsubsection">
2957 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
2958</div>
2959
2960<div class="doc_text">
2961<h5>Syntax:</h5>
2962<pre>
2963 &lt;result&gt; = getelementptr &lt;ty&gt;* &lt;ptrval&gt;{, &lt;ty&gt; &lt;idx&gt;}*
2964</pre>
2965
2966<h5>Overview:</h5>
2967
2968<p>
2969The '<tt>getelementptr</tt>' instruction is used to get the address of a
2970subelement of an aggregate data structure.</p>
2971
2972<h5>Arguments:</h5>
2973
2974<p>This instruction takes a list of integer operands that indicate what
2975elements of the aggregate object to index to. The actual types of the arguments
2976provided depend on the type of the first pointer argument. The
2977'<tt>getelementptr</tt>' instruction is used to index down through the type
2978levels of a structure or to a specific index in an array. When indexing into a
2979structure, only <tt>i32</tt> integer constants are allowed. When indexing
2980into an array or pointer, only integers of 32 or 64 bits are allowed, and will
2981be sign extended to 64-bit values.</p>
2982
2983<p>For example, let's consider a C code fragment and how it gets
2984compiled to LLVM:</p>
2985
2986<div class="doc_code">
2987<pre>
2988struct RT {
2989 char A;
2990 int B[10][20];
2991 char C;
2992};
2993struct ST {
2994 int X;
2995 double Y;
2996 struct RT Z;
2997};
2998
2999int *foo(struct ST *s) {
3000 return &amp;s[1].Z.B[5][13];
3001}
3002</pre>
3003</div>
3004
3005<p>The LLVM code generated by the GCC frontend is:</p>
3006
3007<div class="doc_code">
3008<pre>
3009%RT = type { i8 , [10 x [20 x i32]], i8 }
3010%ST = type { i32, double, %RT }
3011
3012define i32* %foo(%ST* %s) {
3013entry:
3014 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
3015 ret i32* %reg
3016}
3017</pre>
3018</div>
3019
3020<h5>Semantics:</h5>
3021
3022<p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
3023on the pointer type that is being indexed into. <a href="#t_pointer">Pointer</a>
3024and <a href="#t_array">array</a> types can use a 32-bit or 64-bit
3025<a href="#t_integer">integer</a> type but the value will always be sign extended
3026to 64-bits. <a href="#t_struct">Structure</a> types require <tt>i32</tt>
3027<b>constants</b>.</p>
3028
3029<p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
3030type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
3031}</tt>' type, a structure. The second index indexes into the third element of
3032the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
3033i8 }</tt>' type, another structure. The third index indexes into the second
3034element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
3035array. The two dimensions of the array are subscripted into, yielding an
3036'<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
3037to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
3038
3039<p>Note that it is perfectly legal to index partially through a
3040structure, returning a pointer to an inner element. Because of this,
3041the LLVM code for the given testcase is equivalent to:</p>
3042
3043<pre>
3044 define i32* %foo(%ST* %s) {
3045 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
3046 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
3047 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
3048 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
3049 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
3050 ret i32* %t5
3051 }
3052</pre>
3053
3054<p>Note that it is undefined to access an array out of bounds: array and
3055pointer indexes must always be within the defined bounds of the array type.
3056The one exception for this rules is zero length arrays. These arrays are
3057defined to be accessible as variable length arrays, which requires access
3058beyond the zero'th element.</p>
3059
3060<p>The getelementptr instruction is often confusing. For some more insight
3061into how it works, see <a href="GetElementPtr.html">the getelementptr
3062FAQ</a>.</p>
3063
3064<h5>Example:</h5>
3065
3066<pre>
3067 <i>; yields [12 x i8]*:aptr</i>
3068 %aptr = getelementptr {i32, [12 x i8]}* %sptr, i64 0, i32 1
3069</pre>
3070</div>
3071
3072<!-- ======================================================================= -->
3073<div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
3074</div>
3075<div class="doc_text">
3076<p>The instructions in this category are the conversion instructions (casting)
3077which all take a single operand and a type. They perform various bit conversions
3078on the operand.</p>
3079</div>
3080
3081<!-- _______________________________________________________________________ -->
3082<div class="doc_subsubsection">
3083 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
3084</div>
3085<div class="doc_text">
3086
3087<h5>Syntax:</h5>
3088<pre>
3089 &lt;result&gt; = trunc &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3090</pre>
3091
3092<h5>Overview:</h5>
3093<p>
3094The '<tt>trunc</tt>' instruction truncates its operand to the type <tt>ty2</tt>.
3095</p>
3096
3097<h5>Arguments:</h5>
3098<p>
3099The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
3100be an <a href="#t_integer">integer</a> type, and a type that specifies the size
3101and type of the result, which must be an <a href="#t_integer">integer</a>
3102type. The bit size of <tt>value</tt> must be larger than the bit size of
3103<tt>ty2</tt>. Equal sized types are not allowed.</p>
3104
3105<h5>Semantics:</h5>
3106<p>
3107The '<tt>trunc</tt>' instruction truncates the high order bits in <tt>value</tt>
3108and converts the remaining bits to <tt>ty2</tt>. Since the source size must be
3109larger than the destination size, <tt>trunc</tt> cannot be a <i>no-op cast</i>.
3110It will always truncate bits.</p>
3111
3112<h5>Example:</h5>
3113<pre>
3114 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
3115 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
3116 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
3117</pre>
3118</div>
3119
3120<!-- _______________________________________________________________________ -->
3121<div class="doc_subsubsection">
3122 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
3123</div>
3124<div class="doc_text">
3125
3126<h5>Syntax:</h5>
3127<pre>
3128 &lt;result&gt; = zext &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3129</pre>
3130
3131<h5>Overview:</h5>
3132<p>The '<tt>zext</tt>' instruction zero extends its operand to type
3133<tt>ty2</tt>.</p>
3134
3135
3136<h5>Arguments:</h5>
3137<p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
3138<a href="#t_integer">integer</a> type, and a type to cast it to, which must
3139also be of <a href="#t_integer">integer</a> type. The bit size of the
3140<tt>value</tt> must be smaller than the bit size of the destination type,
3141<tt>ty2</tt>.</p>
3142
3143<h5>Semantics:</h5>
3144<p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
3145bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
3146
3147<p>When zero extending from i1, the result will always be either 0 or 1.</p>
3148
3149<h5>Example:</h5>
3150<pre>
3151 %X = zext i32 257 to i64 <i>; yields i64:257</i>
3152 %Y = zext i1 true to i32 <i>; yields i32:1</i>
3153</pre>
3154</div>
3155
3156<!-- _______________________________________________________________________ -->
3157<div class="doc_subsubsection">
3158 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
3159</div>
3160<div class="doc_text">
3161
3162<h5>Syntax:</h5>
3163<pre>
3164 &lt;result&gt; = sext &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3165</pre>
3166
3167<h5>Overview:</h5>
3168<p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
3169
3170<h5>Arguments:</h5>
3171<p>
3172The '<tt>sext</tt>' instruction takes a value to cast, which must be of
3173<a href="#t_integer">integer</a> type, and a type to cast it to, which must
3174also be of <a href="#t_integer">integer</a> type. The bit size of the
3175<tt>value</tt> must be smaller than the bit size of the destination type,
3176<tt>ty2</tt>.</p>
3177
3178<h5>Semantics:</h5>
3179<p>
3180The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
3181bit (highest order bit) of the <tt>value</tt> until it reaches the bit size of
3182the type <tt>ty2</tt>.</p>
3183
3184<p>When sign extending from i1, the extension always results in -1 or 0.</p>
3185
3186<h5>Example:</h5>
3187<pre>
3188 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
3189 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
3190</pre>
3191</div>
3192
3193<!-- _______________________________________________________________________ -->
3194<div class="doc_subsubsection">
3195 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
3196</div>
3197
3198<div class="doc_text">
3199
3200<h5>Syntax:</h5>
3201
3202<pre>
3203 &lt;result&gt; = fptrunc &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3204</pre>
3205
3206<h5>Overview:</h5>
3207<p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
3208<tt>ty2</tt>.</p>
3209
3210
3211<h5>Arguments:</h5>
3212<p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
3213 point</a> value to cast and a <a href="#t_floating">floating point</a> type to
3214cast it to. The size of <tt>value</tt> must be larger than the size of
3215<tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
3216<i>no-op cast</i>.</p>
3217
3218<h5>Semantics:</h5>
3219<p> The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
3220<a href="#t_floating">floating point</a> type to a smaller
3221<a href="#t_floating">floating point</a> type. If the value cannot fit within
3222the destination type, <tt>ty2</tt>, then the results are undefined.</p>
3223
3224<h5>Example:</h5>
3225<pre>
3226 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
3227 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
3228</pre>
3229</div>
3230
3231<!-- _______________________________________________________________________ -->
3232<div class="doc_subsubsection">
3233 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
3234</div>
3235<div class="doc_text">
3236
3237<h5>Syntax:</h5>
3238<pre>
3239 &lt;result&gt; = fpext &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3240</pre>
3241
3242<h5>Overview:</h5>
3243<p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
3244floating point value.</p>
3245
3246<h5>Arguments:</h5>
3247<p>The '<tt>fpext</tt>' instruction takes a
3248<a href="#t_floating">floating point</a> <tt>value</tt> to cast,
3249and a <a href="#t_floating">floating point</a> type to cast it to. The source
3250type must be smaller than the destination type.</p>
3251
3252<h5>Semantics:</h5>
3253<p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
3254<a href="#t_floating">floating point</a> type to a larger
3255<a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
3256used to make a <i>no-op cast</i> because it always changes bits. Use
3257<tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
3258
3259<h5>Example:</h5>
3260<pre>
3261 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
3262 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
3263</pre>
3264</div>
3265
3266<!-- _______________________________________________________________________ -->
3267<div class="doc_subsubsection">
3268 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
3269</div>
3270<div class="doc_text">
3271
3272<h5>Syntax:</h5>
3273<pre>
Reid Spencere6adee82007-07-31 14:40:14 +00003274 &lt;result&gt; = fptoui &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003275</pre>
3276
3277<h5>Overview:</h5>
Reid Spencere6adee82007-07-31 14:40:14 +00003278<p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003279unsigned integer equivalent of type <tt>ty2</tt>.
3280</p>
3281
3282<h5>Arguments:</h5>
Reid Spencere6adee82007-07-31 14:40:14 +00003283<p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
Nate Begeman78246ca2007-11-17 03:58:34 +00003284scalar or vector <a href="#t_floating">floating point</a> value, and a type
3285to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3286type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3287vector integer type with the same number of elements as <tt>ty</tt></p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003288
3289<h5>Semantics:</h5>
Reid Spencere6adee82007-07-31 14:40:14 +00003290<p> The '<tt>fptoui</tt>' instruction converts its
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003291<a href="#t_floating">floating point</a> operand into the nearest (rounding
3292towards zero) unsigned integer value. If the value cannot fit in <tt>ty2</tt>,
3293the results are undefined.</p>
3294
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003295<h5>Example:</h5>
3296<pre>
Reid Spencere6adee82007-07-31 14:40:14 +00003297 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
Chris Lattner681f1e82007-09-22 03:17:52 +00003298 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
Reid Spencere6adee82007-07-31 14:40:14 +00003299 %X = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003300</pre>
3301</div>
3302
3303<!-- _______________________________________________________________________ -->
3304<div class="doc_subsubsection">
3305 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
3306</div>
3307<div class="doc_text">
3308
3309<h5>Syntax:</h5>
3310<pre>
3311 &lt;result&gt; = fptosi &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3312</pre>
3313
3314<h5>Overview:</h5>
3315<p>The '<tt>fptosi</tt>' instruction converts
3316<a href="#t_floating">floating point</a> <tt>value</tt> to type <tt>ty2</tt>.
3317</p>
3318
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003319<h5>Arguments:</h5>
3320<p> The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
Nate Begeman78246ca2007-11-17 03:58:34 +00003321scalar or vector <a href="#t_floating">floating point</a> value, and a type
3322to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3323type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3324vector integer type with the same number of elements as <tt>ty</tt></p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003325
3326<h5>Semantics:</h5>
3327<p>The '<tt>fptosi</tt>' instruction converts its
3328<a href="#t_floating">floating point</a> operand into the nearest (rounding
3329towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
3330the results are undefined.</p>
3331
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003332<h5>Example:</h5>
3333<pre>
3334 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
Chris Lattner681f1e82007-09-22 03:17:52 +00003335 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003336 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
3337</pre>
3338</div>
3339
3340<!-- _______________________________________________________________________ -->
3341<div class="doc_subsubsection">
3342 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
3343</div>
3344<div class="doc_text">
3345
3346<h5>Syntax:</h5>
3347<pre>
3348 &lt;result&gt; = uitofp &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3349</pre>
3350
3351<h5>Overview:</h5>
3352<p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
3353integer and converts that value to the <tt>ty2</tt> type.</p>
3354
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003355<h5>Arguments:</h5>
Nate Begeman78246ca2007-11-17 03:58:34 +00003356<p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
3357scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3358to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3359type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3360floating point type with the same number of elements as <tt>ty</tt></p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003361
3362<h5>Semantics:</h5>
3363<p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
3364integer quantity and converts it to the corresponding floating point value. If
3365the value cannot fit in the floating point value, the results are undefined.</p>
3366
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003367<h5>Example:</h5>
3368<pre>
3369 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
3370 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
3371</pre>
3372</div>
3373
3374<!-- _______________________________________________________________________ -->
3375<div class="doc_subsubsection">
3376 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
3377</div>
3378<div class="doc_text">
3379
3380<h5>Syntax:</h5>
3381<pre>
3382 &lt;result&gt; = sitofp &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3383</pre>
3384
3385<h5>Overview:</h5>
3386<p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed
3387integer and converts that value to the <tt>ty2</tt> type.</p>
3388
3389<h5>Arguments:</h5>
Nate Begeman78246ca2007-11-17 03:58:34 +00003390<p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
3391scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3392to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3393type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3394floating point type with the same number of elements as <tt>ty</tt></p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003395
3396<h5>Semantics:</h5>
3397<p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed
3398integer quantity and converts it to the corresponding floating point value. If
3399the value cannot fit in the floating point value, the results are undefined.</p>
3400
3401<h5>Example:</h5>
3402<pre>
3403 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
3404 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
3405</pre>
3406</div>
3407
3408<!-- _______________________________________________________________________ -->
3409<div class="doc_subsubsection">
3410 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
3411</div>
3412<div class="doc_text">
3413
3414<h5>Syntax:</h5>
3415<pre>
3416 &lt;result&gt; = ptrtoint &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3417</pre>
3418
3419<h5>Overview:</h5>
3420<p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
3421the integer type <tt>ty2</tt>.</p>
3422
3423<h5>Arguments:</h5>
3424<p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
3425must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
3426<tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.
3427
3428<h5>Semantics:</h5>
3429<p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
3430<tt>ty2</tt> by interpreting the pointer value as an integer and either
3431truncating or zero extending that value to the size of the integer type. If
3432<tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
3433<tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
3434are the same size, then nothing is done (<i>no-op cast</i>) other than a type
3435change.</p>
3436
3437<h5>Example:</h5>
3438<pre>
3439 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
3440 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
3441</pre>
3442</div>
3443
3444<!-- _______________________________________________________________________ -->
3445<div class="doc_subsubsection">
3446 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
3447</div>
3448<div class="doc_text">
3449
3450<h5>Syntax:</h5>
3451<pre>
3452 &lt;result&gt; = inttoptr &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3453</pre>
3454
3455<h5>Overview:</h5>
3456<p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to
3457a pointer type, <tt>ty2</tt>.</p>
3458
3459<h5>Arguments:</h5>
3460<p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
3461value to cast, and a type to cast it to, which must be a
3462<a href="#t_pointer">pointer</a> type.
3463
3464<h5>Semantics:</h5>
3465<p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
3466<tt>ty2</tt> by applying either a zero extension or a truncation depending on
3467the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
3468size of a pointer then a truncation is done. If <tt>value</tt> is smaller than
3469the size of a pointer then a zero extension is done. If they are the same size,
3470nothing is done (<i>no-op cast</i>).</p>
3471
3472<h5>Example:</h5>
3473<pre>
3474 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
3475 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
3476 %Y = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
3477</pre>
3478</div>
3479
3480<!-- _______________________________________________________________________ -->
3481<div class="doc_subsubsection">
3482 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
3483</div>
3484<div class="doc_text">
3485
3486<h5>Syntax:</h5>
3487<pre>
3488 &lt;result&gt; = bitcast &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3489</pre>
3490
3491<h5>Overview:</h5>
3492<p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3493<tt>ty2</tt> without changing any bits.</p>
3494
3495<h5>Arguments:</h5>
3496<p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be
3497a first class value, and a type to cast it to, which must also be a <a
3498 href="#t_firstclass">first class</a> type. The bit sizes of <tt>value</tt>
3499and the destination type, <tt>ty2</tt>, must be identical. If the source
3500type is a pointer, the destination type must also be a pointer.</p>
3501
3502<h5>Semantics:</h5>
3503<p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3504<tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
3505this conversion. The conversion is done as if the <tt>value</tt> had been
3506stored to memory and read back as type <tt>ty2</tt>. Pointer types may only be
3507converted to other pointer types with this instruction. To convert pointers to
3508other types, use the <a href="#i_inttoptr">inttoptr</a> or
3509<a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
3510
3511<h5>Example:</h5>
3512<pre>
3513 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
3514 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
3515 %Z = bitcast <2xint> %V to i64; <i>; yields i64: %V</i>
3516</pre>
3517</div>
3518
3519<!-- ======================================================================= -->
3520<div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
3521<div class="doc_text">
3522<p>The instructions in this category are the "miscellaneous"
3523instructions, which defy better classification.</p>
3524</div>
3525
3526<!-- _______________________________________________________________________ -->
3527<div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
3528</div>
3529<div class="doc_text">
3530<h5>Syntax:</h5>
3531<pre> &lt;result&gt; = icmp &lt;cond&gt; &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {i1}:result</i>
3532</pre>
3533<h5>Overview:</h5>
3534<p>The '<tt>icmp</tt>' instruction returns a boolean value based on comparison
3535of its two integer operands.</p>
3536<h5>Arguments:</h5>
3537<p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
3538the condition code indicating the kind of comparison to perform. It is not
3539a value, just a keyword. The possible condition code are:
3540<ol>
3541 <li><tt>eq</tt>: equal</li>
3542 <li><tt>ne</tt>: not equal </li>
3543 <li><tt>ugt</tt>: unsigned greater than</li>
3544 <li><tt>uge</tt>: unsigned greater or equal</li>
3545 <li><tt>ult</tt>: unsigned less than</li>
3546 <li><tt>ule</tt>: unsigned less or equal</li>
3547 <li><tt>sgt</tt>: signed greater than</li>
3548 <li><tt>sge</tt>: signed greater or equal</li>
3549 <li><tt>slt</tt>: signed less than</li>
3550 <li><tt>sle</tt>: signed less or equal</li>
3551</ol>
3552<p>The remaining two arguments must be <a href="#t_integer">integer</a> or
3553<a href="#t_pointer">pointer</a> typed. They must also be identical types.</p>
3554<h5>Semantics:</h5>
3555<p>The '<tt>icmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3556the condition code given as <tt>cond</tt>. The comparison performed always
3557yields a <a href="#t_primitive">i1</a> result, as follows:
3558<ol>
3559 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
3560 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3561 </li>
3562 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
3563 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3564 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
3565 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3566 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
3567 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3568 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
3569 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3570 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
3571 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3572 <li><tt>sgt</tt>: interprets the operands as signed values and yields
3573 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3574 <li><tt>sge</tt>: interprets the operands as signed values and yields
3575 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3576 <li><tt>slt</tt>: interprets the operands as signed values and yields
3577 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3578 <li><tt>sle</tt>: interprets the operands as signed values and yields
3579 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3580</ol>
3581<p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
3582values are compared as if they were integers.</p>
3583
3584<h5>Example:</h5>
3585<pre> &lt;result&gt; = icmp eq i32 4, 5 <i>; yields: result=false</i>
3586 &lt;result&gt; = icmp ne float* %X, %X <i>; yields: result=false</i>
3587 &lt;result&gt; = icmp ult i16 4, 5 <i>; yields: result=true</i>
3588 &lt;result&gt; = icmp sgt i16 4, 5 <i>; yields: result=false</i>
3589 &lt;result&gt; = icmp ule i16 -4, 5 <i>; yields: result=false</i>
3590 &lt;result&gt; = icmp sge i16 4, 5 <i>; yields: result=false</i>
3591</pre>
3592</div>
3593
3594<!-- _______________________________________________________________________ -->
3595<div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
3596</div>
3597<div class="doc_text">
3598<h5>Syntax:</h5>
3599<pre> &lt;result&gt; = fcmp &lt;cond&gt; &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {i1}:result</i>
3600</pre>
3601<h5>Overview:</h5>
3602<p>The '<tt>fcmp</tt>' instruction returns a boolean value based on comparison
3603of its floating point operands.</p>
3604<h5>Arguments:</h5>
3605<p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
3606the condition code indicating the kind of comparison to perform. It is not
3607a value, just a keyword. The possible condition code are:
3608<ol>
3609 <li><tt>false</tt>: no comparison, always returns false</li>
3610 <li><tt>oeq</tt>: ordered and equal</li>
3611 <li><tt>ogt</tt>: ordered and greater than </li>
3612 <li><tt>oge</tt>: ordered and greater than or equal</li>
3613 <li><tt>olt</tt>: ordered and less than </li>
3614 <li><tt>ole</tt>: ordered and less than or equal</li>
3615 <li><tt>one</tt>: ordered and not equal</li>
3616 <li><tt>ord</tt>: ordered (no nans)</li>
3617 <li><tt>ueq</tt>: unordered or equal</li>
3618 <li><tt>ugt</tt>: unordered or greater than </li>
3619 <li><tt>uge</tt>: unordered or greater than or equal</li>
3620 <li><tt>ult</tt>: unordered or less than </li>
3621 <li><tt>ule</tt>: unordered or less than or equal</li>
3622 <li><tt>une</tt>: unordered or not equal</li>
3623 <li><tt>uno</tt>: unordered (either nans)</li>
3624 <li><tt>true</tt>: no comparison, always returns true</li>
3625</ol>
3626<p><i>Ordered</i> means that neither operand is a QNAN while
3627<i>unordered</i> means that either operand may be a QNAN.</p>
3628<p>The <tt>val1</tt> and <tt>val2</tt> arguments must be
3629<a href="#t_floating">floating point</a> typed. They must have identical
3630types.</p>
3631<h5>Semantics:</h5>
3632<p>The '<tt>fcmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3633the condition code given as <tt>cond</tt>. The comparison performed always
3634yields a <a href="#t_primitive">i1</a> result, as follows:
3635<ol>
3636 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
3637 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3638 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3639 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3640 <tt>var1</tt> is greather than <tt>var2</tt>.</li>
3641 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3642 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3643 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3644 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3645 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3646 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3647 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3648 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3649 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
3650 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
3651 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3652 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
3653 <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3654 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
3655 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3656 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
3657 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3658 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
3659 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3660 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
3661 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3662 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
3663 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
3664</ol>
3665
3666<h5>Example:</h5>
3667<pre> &lt;result&gt; = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
3668 &lt;result&gt; = icmp one float 4.0, 5.0 <i>; yields: result=true</i>
3669 &lt;result&gt; = icmp olt float 4.0, 5.0 <i>; yields: result=true</i>
3670 &lt;result&gt; = icmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
3671</pre>
3672</div>
3673
3674<!-- _______________________________________________________________________ -->
3675<div class="doc_subsubsection"> <a name="i_phi">'<tt>phi</tt>'
3676Instruction</a> </div>
3677<div class="doc_text">
3678<h5>Syntax:</h5>
3679<pre> &lt;result&gt; = phi &lt;ty&gt; [ &lt;val0&gt;, &lt;label0&gt;], ...<br></pre>
3680<h5>Overview:</h5>
3681<p>The '<tt>phi</tt>' instruction is used to implement the &#966; node in
3682the SSA graph representing the function.</p>
3683<h5>Arguments:</h5>
3684<p>The type of the incoming values is specified with the first type
3685field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
3686as arguments, with one pair for each predecessor basic block of the
3687current block. Only values of <a href="#t_firstclass">first class</a>
3688type may be used as the value arguments to the PHI node. Only labels
3689may be used as the label arguments.</p>
3690<p>There must be no non-phi instructions between the start of a basic
3691block and the PHI instructions: i.e. PHI instructions must be first in
3692a basic block.</p>
3693<h5>Semantics:</h5>
3694<p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
3695specified by the pair corresponding to the predecessor basic block that executed
3696just prior to the current block.</p>
3697<h5>Example:</h5>
3698<pre>Loop: ; Infinite loop that counts from 0 on up...<br> %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]<br> %nextindvar = add i32 %indvar, 1<br> br label %Loop<br></pre>
3699</div>
3700
3701<!-- _______________________________________________________________________ -->
3702<div class="doc_subsubsection">
3703 <a name="i_select">'<tt>select</tt>' Instruction</a>
3704</div>
3705
3706<div class="doc_text">
3707
3708<h5>Syntax:</h5>
3709
3710<pre>
3711 &lt;result&gt; = select i1 &lt;cond&gt;, &lt;ty&gt; &lt;val1&gt;, &lt;ty&gt; &lt;val2&gt; <i>; yields ty</i>
3712</pre>
3713
3714<h5>Overview:</h5>
3715
3716<p>
3717The '<tt>select</tt>' instruction is used to choose one value based on a
3718condition, without branching.
3719</p>
3720
3721
3722<h5>Arguments:</h5>
3723
3724<p>
3725The '<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.
3726</p>
3727
3728<h5>Semantics:</h5>
3729
3730<p>
3731If the boolean condition evaluates to true, the instruction returns the first
3732value argument; otherwise, it returns the second value argument.
3733</p>
3734
3735<h5>Example:</h5>
3736
3737<pre>
3738 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
3739</pre>
3740</div>
3741
3742
3743<!-- _______________________________________________________________________ -->
3744<div class="doc_subsubsection">
3745 <a name="i_call">'<tt>call</tt>' Instruction</a>
3746</div>
3747
3748<div class="doc_text">
3749
3750<h5>Syntax:</h5>
3751<pre>
Nick Lewycky93082fc2007-09-08 13:57:50 +00003752 &lt;result&gt; = [tail] call [<a href="#callingconv">cconv</a>] &lt;ty&gt; [&lt;fnty&gt;*] &lt;fnptrval&gt;(&lt;param list&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003753</pre>
3754
3755<h5>Overview:</h5>
3756
3757<p>The '<tt>call</tt>' instruction represents a simple function call.</p>
3758
3759<h5>Arguments:</h5>
3760
3761<p>This instruction requires several arguments:</p>
3762
3763<ol>
3764 <li>
3765 <p>The optional "tail" marker indicates whether the callee function accesses
3766 any allocas or varargs in the caller. If the "tail" marker is present, the
3767 function call is eligible for tail call optimization. Note that calls may
3768 be marked "tail" even if they do not occur before a <a
3769 href="#i_ret"><tt>ret</tt></a> instruction.
3770 </li>
3771 <li>
3772 <p>The optional "cconv" marker indicates which <a href="#callingconv">calling
3773 convention</a> the call should use. If none is specified, the call defaults
3774 to using C calling conventions.
3775 </li>
3776 <li>
Nick Lewycky93082fc2007-09-08 13:57:50 +00003777 <p>'<tt>ty</tt>': the type of the call instruction itself which is also
3778 the type of the return value. Functions that return no value are marked
3779 <tt><a href="#t_void">void</a></tt>.</p>
3780 </li>
3781 <li>
3782 <p>'<tt>fnty</tt>': shall be the signature of the pointer to function
3783 value being invoked. The argument types must match the types implied by
3784 this signature. This type can be omitted if the function is not varargs
3785 and if the function type does not return a pointer to a function.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003786 </li>
3787 <li>
3788 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
3789 be invoked. In most cases, this is a direct function invocation, but
3790 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
3791 to function value.</p>
3792 </li>
3793 <li>
3794 <p>'<tt>function args</tt>': argument list whose types match the
3795 function signature argument types. All arguments must be of
3796 <a href="#t_firstclass">first class</a> type. If the function signature
3797 indicates the function accepts a variable number of arguments, the extra
3798 arguments can be specified.</p>
3799 </li>
3800</ol>
3801
3802<h5>Semantics:</h5>
3803
3804<p>The '<tt>call</tt>' instruction is used to cause control flow to
3805transfer to a specified function, with its incoming arguments bound to
3806the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
3807instruction in the called function, control flow continues with the
3808instruction after the function call, and the return value of the
3809function is bound to the result argument. This is a simpler case of
3810the <a href="#i_invoke">invoke</a> instruction.</p>
3811
3812<h5>Example:</h5>
3813
3814<pre>
Nick Lewycky93082fc2007-09-08 13:57:50 +00003815 %retval = call i32 @test(i32 %argc)
3816 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42);
3817 %X = tail call i32 @foo()
3818 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo()
3819 %Z = call void %foo(i8 97 signext)
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003820</pre>
3821
3822</div>
3823
3824<!-- _______________________________________________________________________ -->
3825<div class="doc_subsubsection">
3826 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
3827</div>
3828
3829<div class="doc_text">
3830
3831<h5>Syntax:</h5>
3832
3833<pre>
3834 &lt;resultval&gt; = va_arg &lt;va_list*&gt; &lt;arglist&gt;, &lt;argty&gt;
3835</pre>
3836
3837<h5>Overview:</h5>
3838
3839<p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
3840the "variable argument" area of a function call. It is used to implement the
3841<tt>va_arg</tt> macro in C.</p>
3842
3843<h5>Arguments:</h5>
3844
3845<p>This instruction takes a <tt>va_list*</tt> value and the type of
3846the argument. It returns a value of the specified argument type and
3847increments the <tt>va_list</tt> to point to the next argument. The
3848actual type of <tt>va_list</tt> is target specific.</p>
3849
3850<h5>Semantics:</h5>
3851
3852<p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
3853type from the specified <tt>va_list</tt> and causes the
3854<tt>va_list</tt> to point to the next argument. For more information,
3855see the variable argument handling <a href="#int_varargs">Intrinsic
3856Functions</a>.</p>
3857
3858<p>It is legal for this instruction to be called in a function which does not
3859take a variable number of arguments, for example, the <tt>vfprintf</tt>
3860function.</p>
3861
3862<p><tt>va_arg</tt> is an LLVM instruction instead of an <a
3863href="#intrinsics">intrinsic function</a> because it takes a type as an
3864argument.</p>
3865
3866<h5>Example:</h5>
3867
3868<p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
3869
3870</div>
3871
3872<!-- *********************************************************************** -->
3873<div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
3874<!-- *********************************************************************** -->
3875
3876<div class="doc_text">
3877
3878<p>LLVM supports the notion of an "intrinsic function". These functions have
3879well known names and semantics and are required to follow certain restrictions.
3880Overall, these intrinsics represent an extension mechanism for the LLVM
3881language that does not require changing all of the transformations in LLVM when
3882adding to the language (or the bitcode reader/writer, the parser, etc...).</p>
3883
3884<p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
3885prefix is reserved in LLVM for intrinsic names; thus, function names may not
3886begin with this prefix. Intrinsic functions must always be external functions:
3887you cannot define the body of intrinsic functions. Intrinsic functions may
3888only be used in call or invoke instructions: it is illegal to take the address
3889of an intrinsic function. Additionally, because intrinsic functions are part
3890of the LLVM language, it is required if any are added that they be documented
3891here.</p>
3892
Chandler Carrutha228e392007-08-04 01:51:18 +00003893<p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
3894a family of functions that perform the same operation but on different data
3895types. Because LLVM can represent over 8 million different integer types,
3896overloading is used commonly to allow an intrinsic function to operate on any
3897integer type. One or more of the argument types or the result type can be
3898overloaded to accept any integer type. Argument types may also be defined as
3899exactly matching a previous argument's type or the result type. This allows an
3900intrinsic function which accepts multiple arguments, but needs all of them to
3901be of the same type, to only be overloaded with respect to a single argument or
3902the result.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003903
Chandler Carrutha228e392007-08-04 01:51:18 +00003904<p>Overloaded intrinsics will have the names of its overloaded argument types
3905encoded into its function name, each preceded by a period. Only those types
3906which are overloaded result in a name suffix. Arguments whose type is matched
3907against another type do not. For example, the <tt>llvm.ctpop</tt> function can
3908take an integer of any width and returns an integer of exactly the same integer
3909width. This leads to a family of functions such as
3910<tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29 %val)</tt>.
3911Only one type, the return type, is overloaded, and only one type suffix is
3912required. Because the argument's type is matched against the return type, it
3913does not require its own name suffix.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003914
3915<p>To learn how to add an intrinsic function, please see the
3916<a href="ExtendingLLVM.html">Extending LLVM Guide</a>.
3917</p>
3918
3919</div>
3920
3921<!-- ======================================================================= -->
3922<div class="doc_subsection">
3923 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
3924</div>
3925
3926<div class="doc_text">
3927
3928<p>Variable argument support is defined in LLVM with the <a
3929 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
3930intrinsic functions. These functions are related to the similarly
3931named macros defined in the <tt>&lt;stdarg.h&gt;</tt> header file.</p>
3932
3933<p>All of these functions operate on arguments that use a
3934target-specific value type "<tt>va_list</tt>". The LLVM assembly
3935language reference manual does not define what this type is, so all
3936transformations should be prepared to handle these functions regardless of
3937the type used.</p>
3938
3939<p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
3940instruction and the variable argument handling intrinsic functions are
3941used.</p>
3942
3943<div class="doc_code">
3944<pre>
3945define i32 @test(i32 %X, ...) {
3946 ; Initialize variable argument processing
3947 %ap = alloca i8*
3948 %ap2 = bitcast i8** %ap to i8*
3949 call void @llvm.va_start(i8* %ap2)
3950
3951 ; Read a single integer argument
3952 %tmp = va_arg i8** %ap, i32
3953
3954 ; Demonstrate usage of llvm.va_copy and llvm.va_end
3955 %aq = alloca i8*
3956 %aq2 = bitcast i8** %aq to i8*
3957 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
3958 call void @llvm.va_end(i8* %aq2)
3959
3960 ; Stop processing of arguments.
3961 call void @llvm.va_end(i8* %ap2)
3962 ret i32 %tmp
3963}
3964
3965declare void @llvm.va_start(i8*)
3966declare void @llvm.va_copy(i8*, i8*)
3967declare void @llvm.va_end(i8*)
3968</pre>
3969</div>
3970
3971</div>
3972
3973<!-- _______________________________________________________________________ -->
3974<div class="doc_subsubsection">
3975 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
3976</div>
3977
3978
3979<div class="doc_text">
3980<h5>Syntax:</h5>
3981<pre> declare void %llvm.va_start(i8* &lt;arglist&gt;)<br></pre>
3982<h5>Overview:</h5>
3983<P>The '<tt>llvm.va_start</tt>' intrinsic initializes
3984<tt>*&lt;arglist&gt;</tt> for subsequent use by <tt><a
3985href="#i_va_arg">va_arg</a></tt>.</p>
3986
3987<h5>Arguments:</h5>
3988
3989<P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
3990
3991<h5>Semantics:</h5>
3992
3993<P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
3994macro available in C. In a target-dependent way, it initializes the
3995<tt>va_list</tt> element to which the argument points, so that the next call to
3996<tt>va_arg</tt> will produce the first variable argument passed to the function.
3997Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
3998last argument of the function as the compiler can figure that out.</p>
3999
4000</div>
4001
4002<!-- _______________________________________________________________________ -->
4003<div class="doc_subsubsection">
4004 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
4005</div>
4006
4007<div class="doc_text">
4008<h5>Syntax:</h5>
4009<pre> declare void @llvm.va_end(i8* &lt;arglist&gt;)<br></pre>
4010<h5>Overview:</h5>
4011
4012<p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*&lt;arglist&gt;</tt>,
4013which has been initialized previously with <tt><a href="#int_va_start">llvm.va_start</a></tt>
4014or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
4015
4016<h5>Arguments:</h5>
4017
4018<p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
4019
4020<h5>Semantics:</h5>
4021
4022<p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
4023macro available in C. In a target-dependent way, it destroys the
4024<tt>va_list</tt> element to which the argument points. Calls to <a
4025href="#int_va_start"><tt>llvm.va_start</tt></a> and <a href="#int_va_copy">
4026<tt>llvm.va_copy</tt></a> must be matched exactly with calls to
4027<tt>llvm.va_end</tt>.</p>
4028
4029</div>
4030
4031<!-- _______________________________________________________________________ -->
4032<div class="doc_subsubsection">
4033 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
4034</div>
4035
4036<div class="doc_text">
4037
4038<h5>Syntax:</h5>
4039
4040<pre>
4041 declare void @llvm.va_copy(i8* &lt;destarglist&gt;, i8* &lt;srcarglist&gt;)
4042</pre>
4043
4044<h5>Overview:</h5>
4045
4046<p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
4047from the source argument list to the destination argument list.</p>
4048
4049<h5>Arguments:</h5>
4050
4051<p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
4052The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
4053
4054
4055<h5>Semantics:</h5>
4056
4057<p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
4058macro available in C. In a target-dependent way, it copies the source
4059<tt>va_list</tt> element into the destination <tt>va_list</tt> element. This
4060intrinsic is necessary because the <tt><a href="#int_va_start">
4061llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
4062example, memory allocation.</p>
4063
4064</div>
4065
4066<!-- ======================================================================= -->
4067<div class="doc_subsection">
4068 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
4069</div>
4070
4071<div class="doc_text">
4072
4073<p>
4074LLVM support for <a href="GarbageCollection.html">Accurate Garbage
4075Collection</a> requires the implementation and generation of these intrinsics.
4076These intrinsics allow identification of <a href="#int_gcroot">GC roots on the
4077stack</a>, as well as garbage collector implementations that require <a
4078href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a> barriers.
4079Front-ends for type-safe garbage collected languages should generate these
4080intrinsics to make use of the LLVM garbage collectors. For more details, see <a
4081href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
4082</p>
Christopher Lambcfe00962007-12-17 01:00:21 +00004083
4084<p>The garbage collection intrinsics only operate on objects in the generic
4085 address space (address space zero).</p>
4086
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004087</div>
4088
4089<!-- _______________________________________________________________________ -->
4090<div class="doc_subsubsection">
4091 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
4092</div>
4093
4094<div class="doc_text">
4095
4096<h5>Syntax:</h5>
4097
4098<pre>
Chris Lattner38bd5dd2007-09-21 17:30:40 +00004099 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004100</pre>
4101
4102<h5>Overview:</h5>
4103
4104<p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
4105the code generator, and allows some metadata to be associated with it.</p>
4106
4107<h5>Arguments:</h5>
4108
4109<p>The first argument specifies the address of a stack object that contains the
4110root pointer. The second pointer (which must be either a constant or a global
4111value address) contains the meta-data to be associated with the root.</p>
4112
4113<h5>Semantics:</h5>
4114
4115<p>At runtime, a call to this intrinsics stores a null pointer into the "ptrloc"
4116location. At compile-time, the code generator generates information to allow
Gordon Henriksen40393542007-12-25 02:31:26 +00004117the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
4118intrinsic may only be used in a function which <a href="#gc">specifies a GC
4119algorithm</a>.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004120
4121</div>
4122
4123
4124<!-- _______________________________________________________________________ -->
4125<div class="doc_subsubsection">
4126 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
4127</div>
4128
4129<div class="doc_text">
4130
4131<h5>Syntax:</h5>
4132
4133<pre>
Chris Lattner38bd5dd2007-09-21 17:30:40 +00004134 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004135</pre>
4136
4137<h5>Overview:</h5>
4138
4139<p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
4140locations, allowing garbage collector implementations that require read
4141barriers.</p>
4142
4143<h5>Arguments:</h5>
4144
4145<p>The second argument is the address to read from, which should be an address
4146allocated from the garbage collector. The first object is a pointer to the
4147start of the referenced object, if needed by the language runtime (otherwise
4148null).</p>
4149
4150<h5>Semantics:</h5>
4151
4152<p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
4153instruction, but may be replaced with substantially more complex code by the
Gordon Henriksen40393542007-12-25 02:31:26 +00004154garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
4155may only be used in a function which <a href="#gc">specifies a GC
4156algorithm</a>.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004157
4158</div>
4159
4160
4161<!-- _______________________________________________________________________ -->
4162<div class="doc_subsubsection">
4163 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
4164</div>
4165
4166<div class="doc_text">
4167
4168<h5>Syntax:</h5>
4169
4170<pre>
Chris Lattner38bd5dd2007-09-21 17:30:40 +00004171 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004172</pre>
4173
4174<h5>Overview:</h5>
4175
4176<p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
4177locations, allowing garbage collector implementations that require write
4178barriers (such as generational or reference counting collectors).</p>
4179
4180<h5>Arguments:</h5>
4181
4182<p>The first argument is the reference to store, the second is the start of the
4183object to store it to, and the third is the address of the field of Obj to
4184store to. If the runtime does not require a pointer to the object, Obj may be
4185null.</p>
4186
4187<h5>Semantics:</h5>
4188
4189<p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
4190instruction, but may be replaced with substantially more complex code by the
Gordon Henriksen40393542007-12-25 02:31:26 +00004191garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
4192may only be used in a function which <a href="#gc">specifies a GC
4193algorithm</a>.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004194
4195</div>
4196
4197
4198
4199<!-- ======================================================================= -->
4200<div class="doc_subsection">
4201 <a name="int_codegen">Code Generator Intrinsics</a>
4202</div>
4203
4204<div class="doc_text">
4205<p>
4206These intrinsics are provided by LLVM to expose special features that may only
4207be implemented with code generator support.
4208</p>
4209
4210</div>
4211
4212<!-- _______________________________________________________________________ -->
4213<div class="doc_subsubsection">
4214 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
4215</div>
4216
4217<div class="doc_text">
4218
4219<h5>Syntax:</h5>
4220<pre>
4221 declare i8 *@llvm.returnaddress(i32 &lt;level&gt;)
4222</pre>
4223
4224<h5>Overview:</h5>
4225
4226<p>
4227The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
4228target-specific value indicating the return address of the current function
4229or one of its callers.
4230</p>
4231
4232<h5>Arguments:</h5>
4233
4234<p>
4235The argument to this intrinsic indicates which function to return the address
4236for. Zero indicates the calling function, one indicates its caller, etc. The
4237argument is <b>required</b> to be a constant integer value.
4238</p>
4239
4240<h5>Semantics:</h5>
4241
4242<p>
4243The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
4244the return address of the specified call frame, or zero if it cannot be
4245identified. The value returned by this intrinsic is likely to be incorrect or 0
4246for arguments other than zero, so it should only be used for debugging purposes.
4247</p>
4248
4249<p>
4250Note that calling this intrinsic does not prevent function inlining or other
4251aggressive transformations, so the value returned may not be that of the obvious
4252source-language caller.
4253</p>
4254</div>
4255
4256
4257<!-- _______________________________________________________________________ -->
4258<div class="doc_subsubsection">
4259 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
4260</div>
4261
4262<div class="doc_text">
4263
4264<h5>Syntax:</h5>
4265<pre>
Chris Lattner38bd5dd2007-09-21 17:30:40 +00004266 declare i8 *@llvm.frameaddress(i32 &lt;level&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004267</pre>
4268
4269<h5>Overview:</h5>
4270
4271<p>
4272The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
4273target-specific frame pointer value for the specified stack frame.
4274</p>
4275
4276<h5>Arguments:</h5>
4277
4278<p>
4279The argument to this intrinsic indicates which function to return the frame
4280pointer for. Zero indicates the calling function, one indicates its caller,
4281etc. The argument is <b>required</b> to be a constant integer value.
4282</p>
4283
4284<h5>Semantics:</h5>
4285
4286<p>
4287The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
4288the frame address of the specified call frame, or zero if it cannot be
4289identified. The value returned by this intrinsic is likely to be incorrect or 0
4290for arguments other than zero, so it should only be used for debugging purposes.
4291</p>
4292
4293<p>
4294Note that calling this intrinsic does not prevent function inlining or other
4295aggressive transformations, so the value returned may not be that of the obvious
4296source-language caller.
4297</p>
4298</div>
4299
4300<!-- _______________________________________________________________________ -->
4301<div class="doc_subsubsection">
4302 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
4303</div>
4304
4305<div class="doc_text">
4306
4307<h5>Syntax:</h5>
4308<pre>
Chris Lattner38bd5dd2007-09-21 17:30:40 +00004309 declare i8 *@llvm.stacksave()
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004310</pre>
4311
4312<h5>Overview:</h5>
4313
4314<p>
4315The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
4316the function stack, for use with <a href="#int_stackrestore">
4317<tt>llvm.stackrestore</tt></a>. This is useful for implementing language
4318features like scoped automatic variable sized arrays in C99.
4319</p>
4320
4321<h5>Semantics:</h5>
4322
4323<p>
4324This intrinsic returns a opaque pointer value that can be passed to <a
4325href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
4326<tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
4327<tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
4328state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
4329practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
4330that were allocated after the <tt>llvm.stacksave</tt> was executed.
4331</p>
4332
4333</div>
4334
4335<!-- _______________________________________________________________________ -->
4336<div class="doc_subsubsection">
4337 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
4338</div>
4339
4340<div class="doc_text">
4341
4342<h5>Syntax:</h5>
4343<pre>
4344 declare void @llvm.stackrestore(i8 * %ptr)
4345</pre>
4346
4347<h5>Overview:</h5>
4348
4349<p>
4350The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
4351the function stack to the state it was in when the corresponding <a
4352href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
4353useful for implementing language features like scoped automatic variable sized
4354arrays in C99.
4355</p>
4356
4357<h5>Semantics:</h5>
4358
4359<p>
4360See the description for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.
4361</p>
4362
4363</div>
4364
4365
4366<!-- _______________________________________________________________________ -->
4367<div class="doc_subsubsection">
4368 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
4369</div>
4370
4371<div class="doc_text">
4372
4373<h5>Syntax:</h5>
4374<pre>
Chris Lattner38bd5dd2007-09-21 17:30:40 +00004375 declare void @llvm.prefetch(i8* &lt;address&gt;, i32 &lt;rw&gt;, i32 &lt;locality&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004376</pre>
4377
4378<h5>Overview:</h5>
4379
4380
4381<p>
4382The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
4383a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
4384no
4385effect on the behavior of the program but can change its performance
4386characteristics.
4387</p>
4388
4389<h5>Arguments:</h5>
4390
4391<p>
4392<tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
4393determining if the fetch should be for a read (0) or write (1), and
4394<tt>locality</tt> is a temporal locality specifier ranging from (0) - no
4395locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
4396<tt>locality</tt> arguments must be constant integers.
4397</p>
4398
4399<h5>Semantics:</h5>
4400
4401<p>
4402This intrinsic does not modify the behavior of the program. In particular,
4403prefetches cannot trap and do not produce a value. On targets that support this
4404intrinsic, the prefetch can provide hints to the processor cache for better
4405performance.
4406</p>
4407
4408</div>
4409
4410<!-- _______________________________________________________________________ -->
4411<div class="doc_subsubsection">
4412 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
4413</div>
4414
4415<div class="doc_text">
4416
4417<h5>Syntax:</h5>
4418<pre>
Chris Lattner38bd5dd2007-09-21 17:30:40 +00004419 declare void @llvm.pcmarker(i32 &lt;id&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004420</pre>
4421
4422<h5>Overview:</h5>
4423
4424
4425<p>
4426The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
4427(PC) in a region of
4428code to simulators and other tools. The method is target specific, but it is
4429expected that the marker will use exported symbols to transmit the PC of the marker.
4430The marker makes no guarantees that it will remain with any specific instruction
4431after optimizations. It is possible that the presence of a marker will inhibit
4432optimizations. The intended use is to be inserted after optimizations to allow
4433correlations of simulation runs.
4434</p>
4435
4436<h5>Arguments:</h5>
4437
4438<p>
4439<tt>id</tt> is a numerical id identifying the marker.
4440</p>
4441
4442<h5>Semantics:</h5>
4443
4444<p>
4445This intrinsic does not modify the behavior of the program. Backends that do not
4446support this intrinisic may ignore it.
4447</p>
4448
4449</div>
4450
4451<!-- _______________________________________________________________________ -->
4452<div class="doc_subsubsection">
4453 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
4454</div>
4455
4456<div class="doc_text">
4457
4458<h5>Syntax:</h5>
4459<pre>
4460 declare i64 @llvm.readcyclecounter( )
4461</pre>
4462
4463<h5>Overview:</h5>
4464
4465
4466<p>
4467The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
4468counter register (or similar low latency, high accuracy clocks) on those targets
4469that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
4470As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
4471should only be used for small timings.
4472</p>
4473
4474<h5>Semantics:</h5>
4475
4476<p>
4477When directly supported, reading the cycle counter should not modify any memory.
4478Implementations are allowed to either return a application specific value or a
4479system wide value. On backends without support, this is lowered to a constant 0.
4480</p>
4481
4482</div>
4483
4484<!-- ======================================================================= -->
4485<div class="doc_subsection">
4486 <a name="int_libc">Standard C Library Intrinsics</a>
4487</div>
4488
4489<div class="doc_text">
4490<p>
4491LLVM provides intrinsics for a few important standard C library functions.
4492These intrinsics allow source-language front-ends to pass information about the
4493alignment of the pointer arguments to the code generator, providing opportunity
4494for more efficient code generation.
4495</p>
4496
4497</div>
4498
4499<!-- _______________________________________________________________________ -->
4500<div class="doc_subsubsection">
4501 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
4502</div>
4503
4504<div class="doc_text">
4505
4506<h5>Syntax:</h5>
4507<pre>
4508 declare void @llvm.memcpy.i32(i8 * &lt;dest&gt;, i8 * &lt;src&gt;,
4509 i32 &lt;len&gt;, i32 &lt;align&gt;)
4510 declare void @llvm.memcpy.i64(i8 * &lt;dest&gt;, i8 * &lt;src&gt;,
4511 i64 &lt;len&gt;, i32 &lt;align&gt;)
4512</pre>
4513
4514<h5>Overview:</h5>
4515
4516<p>
4517The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4518location to the destination location.
4519</p>
4520
4521<p>
4522Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
4523intrinsics do not return a value, and takes an extra alignment argument.
4524</p>
4525
4526<h5>Arguments:</h5>
4527
4528<p>
4529The first argument is a pointer to the destination, the second is a pointer to
4530the source. The third argument is an integer argument
4531specifying the number of bytes to copy, and the fourth argument is the alignment
4532of the source and destination locations.
4533</p>
4534
4535<p>
4536If the call to this intrinisic has an alignment value that is not 0 or 1, then
4537the caller guarantees that both the source and destination pointers are aligned
4538to that boundary.
4539</p>
4540
4541<h5>Semantics:</h5>
4542
4543<p>
4544The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4545location to the destination location, which are not allowed to overlap. It
4546copies "len" bytes of memory over. If the argument is known to be aligned to
4547some boundary, this can be specified as the fourth argument, otherwise it should
4548be set to 0 or 1.
4549</p>
4550</div>
4551
4552
4553<!-- _______________________________________________________________________ -->
4554<div class="doc_subsubsection">
4555 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
4556</div>
4557
4558<div class="doc_text">
4559
4560<h5>Syntax:</h5>
4561<pre>
4562 declare void @llvm.memmove.i32(i8 * &lt;dest&gt;, i8 * &lt;src&gt;,
4563 i32 &lt;len&gt;, i32 &lt;align&gt;)
4564 declare void @llvm.memmove.i64(i8 * &lt;dest&gt;, i8 * &lt;src&gt;,
4565 i64 &lt;len&gt;, i32 &lt;align&gt;)
4566</pre>
4567
4568<h5>Overview:</h5>
4569
4570<p>
4571The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
4572location to the destination location. It is similar to the
Chris Lattnerdba16ea2008-01-06 19:51:52 +00004573'<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to overlap.
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004574</p>
4575
4576<p>
4577Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
4578intrinsics do not return a value, and takes an extra alignment argument.
4579</p>
4580
4581<h5>Arguments:</h5>
4582
4583<p>
4584The first argument is a pointer to the destination, the second is a pointer to
4585the source. The third argument is an integer argument
4586specifying the number of bytes to copy, and the fourth argument is the alignment
4587of the source and destination locations.
4588</p>
4589
4590<p>
4591If the call to this intrinisic has an alignment value that is not 0 or 1, then
4592the caller guarantees that the source and destination pointers are aligned to
4593that boundary.
4594</p>
4595
4596<h5>Semantics:</h5>
4597
4598<p>
4599The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
4600location to the destination location, which may overlap. It
4601copies "len" bytes of memory over. If the argument is known to be aligned to
4602some boundary, this can be specified as the fourth argument, otherwise it should
4603be set to 0 or 1.
4604</p>
4605</div>
4606
4607
4608<!-- _______________________________________________________________________ -->
4609<div class="doc_subsubsection">
4610 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
4611</div>
4612
4613<div class="doc_text">
4614
4615<h5>Syntax:</h5>
4616<pre>
4617 declare void @llvm.memset.i32(i8 * &lt;dest&gt;, i8 &lt;val&gt;,
4618 i32 &lt;len&gt;, i32 &lt;align&gt;)
4619 declare void @llvm.memset.i64(i8 * &lt;dest&gt;, i8 &lt;val&gt;,
4620 i64 &lt;len&gt;, i32 &lt;align&gt;)
4621</pre>
4622
4623<h5>Overview:</h5>
4624
4625<p>
4626The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
4627byte value.
4628</p>
4629
4630<p>
4631Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
4632does not return a value, and takes an extra alignment argument.
4633</p>
4634
4635<h5>Arguments:</h5>
4636
4637<p>
4638The first argument is a pointer to the destination to fill, the second is the
4639byte value to fill it with, the third argument is an integer
4640argument specifying the number of bytes to fill, and the fourth argument is the
4641known alignment of destination location.
4642</p>
4643
4644<p>
4645If the call to this intrinisic has an alignment value that is not 0 or 1, then
4646the caller guarantees that the destination pointer is aligned to that boundary.
4647</p>
4648
4649<h5>Semantics:</h5>
4650
4651<p>
4652The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
4653the
4654destination location. If the argument is known to be aligned to some boundary,
4655this can be specified as the fourth argument, otherwise it should be set to 0 or
46561.
4657</p>
4658</div>
4659
4660
4661<!-- _______________________________________________________________________ -->
4662<div class="doc_subsubsection">
4663 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
4664</div>
4665
4666<div class="doc_text">
4667
4668<h5>Syntax:</h5>
Dale Johannesenf9adbb62007-10-02 17:47:38 +00004669<p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
Dan Gohman361079c2007-10-15 20:30:11 +00004670floating point or vector of floating point type. Not all targets support all
4671types however.
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004672<pre>
Dale Johannesenf9adbb62007-10-02 17:47:38 +00004673 declare float @llvm.sqrt.f32(float %Val)
4674 declare double @llvm.sqrt.f64(double %Val)
4675 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
4676 declare fp128 @llvm.sqrt.f128(fp128 %Val)
4677 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004678</pre>
4679
4680<h5>Overview:</h5>
4681
4682<p>
4683The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
Dan Gohman361079c2007-10-15 20:30:11 +00004684returning the same value as the libm '<tt>sqrt</tt>' functions would. Unlike
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004685<tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
Chris Lattnerf914a002008-01-29 07:00:44 +00004686negative numbers other than -0.0 (which allows for better optimization, because
4687there is no need to worry about errno being set). <tt>llvm.sqrt(-0.0)</tt> is
4688defined to return -0.0 like IEEE sqrt.
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004689</p>
4690
4691<h5>Arguments:</h5>
4692
4693<p>
4694The argument and return value are floating point numbers of the same type.
4695</p>
4696
4697<h5>Semantics:</h5>
4698
4699<p>
4700This function returns the sqrt of the specified operand if it is a nonnegative
4701floating point number.
4702</p>
4703</div>
4704
4705<!-- _______________________________________________________________________ -->
4706<div class="doc_subsubsection">
4707 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
4708</div>
4709
4710<div class="doc_text">
4711
4712<h5>Syntax:</h5>
Dale Johannesenf9adbb62007-10-02 17:47:38 +00004713<p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
Dan Gohman361079c2007-10-15 20:30:11 +00004714floating point or vector of floating point type. Not all targets support all
4715types however.
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004716<pre>
Dale Johannesenf9adbb62007-10-02 17:47:38 +00004717 declare float @llvm.powi.f32(float %Val, i32 %power)
4718 declare double @llvm.powi.f64(double %Val, i32 %power)
4719 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
4720 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
4721 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004722</pre>
4723
4724<h5>Overview:</h5>
4725
4726<p>
4727The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
4728specified (positive or negative) power. The order of evaluation of
Dan Gohman361079c2007-10-15 20:30:11 +00004729multiplications is not defined. When a vector of floating point type is
4730used, the second argument remains a scalar integer value.
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004731</p>
4732
4733<h5>Arguments:</h5>
4734
4735<p>
4736The second argument is an integer power, and the first is a value to raise to
4737that power.
4738</p>
4739
4740<h5>Semantics:</h5>
4741
4742<p>
4743This function returns the first value raised to the second power with an
4744unspecified sequence of rounding operations.</p>
4745</div>
4746
Dan Gohman361079c2007-10-15 20:30:11 +00004747<!-- _______________________________________________________________________ -->
4748<div class="doc_subsubsection">
4749 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
4750</div>
4751
4752<div class="doc_text">
4753
4754<h5>Syntax:</h5>
4755<p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
4756floating point or vector of floating point type. Not all targets support all
4757types however.
4758<pre>
4759 declare float @llvm.sin.f32(float %Val)
4760 declare double @llvm.sin.f64(double %Val)
4761 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
4762 declare fp128 @llvm.sin.f128(fp128 %Val)
4763 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
4764</pre>
4765
4766<h5>Overview:</h5>
4767
4768<p>
4769The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.
4770</p>
4771
4772<h5>Arguments:</h5>
4773
4774<p>
4775The argument and return value are floating point numbers of the same type.
4776</p>
4777
4778<h5>Semantics:</h5>
4779
4780<p>
4781This function returns the sine of the specified operand, returning the
4782same values as the libm <tt>sin</tt> functions would, and handles error
Dan Gohmaneaba92e2007-10-17 18:05:13 +00004783conditions in the same way.</p>
Dan Gohman361079c2007-10-15 20:30:11 +00004784</div>
4785
4786<!-- _______________________________________________________________________ -->
4787<div class="doc_subsubsection">
4788 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
4789</div>
4790
4791<div class="doc_text">
4792
4793<h5>Syntax:</h5>
4794<p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
4795floating point or vector of floating point type. Not all targets support all
4796types however.
4797<pre>
4798 declare float @llvm.cos.f32(float %Val)
4799 declare double @llvm.cos.f64(double %Val)
4800 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
4801 declare fp128 @llvm.cos.f128(fp128 %Val)
4802 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
4803</pre>
4804
4805<h5>Overview:</h5>
4806
4807<p>
4808The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.
4809</p>
4810
4811<h5>Arguments:</h5>
4812
4813<p>
4814The argument and return value are floating point numbers of the same type.
4815</p>
4816
4817<h5>Semantics:</h5>
4818
4819<p>
4820This function returns the cosine of the specified operand, returning the
4821same values as the libm <tt>cos</tt> functions would, and handles error
Dan Gohmaneaba92e2007-10-17 18:05:13 +00004822conditions in the same way.</p>
Dan Gohman361079c2007-10-15 20:30:11 +00004823</div>
4824
4825<!-- _______________________________________________________________________ -->
4826<div class="doc_subsubsection">
4827 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
4828</div>
4829
4830<div class="doc_text">
4831
4832<h5>Syntax:</h5>
4833<p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
4834floating point or vector of floating point type. Not all targets support all
4835types however.
4836<pre>
4837 declare float @llvm.pow.f32(float %Val, float %Power)
4838 declare double @llvm.pow.f64(double %Val, double %Power)
4839 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
4840 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
4841 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
4842</pre>
4843
4844<h5>Overview:</h5>
4845
4846<p>
4847The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
4848specified (positive or negative) power.
4849</p>
4850
4851<h5>Arguments:</h5>
4852
4853<p>
4854The second argument is a floating point power, and the first is a value to
4855raise to that power.
4856</p>
4857
4858<h5>Semantics:</h5>
4859
4860<p>
4861This function returns the first value raised to the second power,
4862returning the
4863same values as the libm <tt>pow</tt> functions would, and handles error
Dan Gohmaneaba92e2007-10-17 18:05:13 +00004864conditions in the same way.</p>
Dan Gohman361079c2007-10-15 20:30:11 +00004865</div>
4866
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004867
4868<!-- ======================================================================= -->
4869<div class="doc_subsection">
4870 <a name="int_manip">Bit Manipulation Intrinsics</a>
4871</div>
4872
4873<div class="doc_text">
4874<p>
4875LLVM provides intrinsics for a few important bit manipulation operations.
4876These allow efficient code generation for some algorithms.
4877</p>
4878
4879</div>
4880
4881<!-- _______________________________________________________________________ -->
4882<div class="doc_subsubsection">
4883 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
4884</div>
4885
4886<div class="doc_text">
4887
4888<h5>Syntax:</h5>
4889<p>This is an overloaded intrinsic function. You can use bswap on any integer
Chandler Carrutha228e392007-08-04 01:51:18 +00004890type that is an even number of bytes (i.e. BitWidth % 16 == 0).
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004891<pre>
Chandler Carrutha228e392007-08-04 01:51:18 +00004892 declare i16 @llvm.bswap.i16(i16 &lt;id&gt;)
4893 declare i32 @llvm.bswap.i32(i32 &lt;id&gt;)
4894 declare i64 @llvm.bswap.i64(i64 &lt;id&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004895</pre>
4896
4897<h5>Overview:</h5>
4898
4899<p>
4900The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
4901values with an even number of bytes (positive multiple of 16 bits). These are
4902useful for performing operations on data that is not in the target's native
4903byte order.
4904</p>
4905
4906<h5>Semantics:</h5>
4907
4908<p>
Chandler Carrutha228e392007-08-04 01:51:18 +00004909The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004910and low byte of the input i16 swapped. Similarly, the <tt>llvm.bswap.i32</tt>
4911intrinsic returns an i32 value that has the four bytes of the input i32
4912swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the returned
Chandler Carrutha228e392007-08-04 01:51:18 +00004913i32 will have its bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i48</tt>,
4914<tt>llvm.bswap.i64</tt> and other intrinsics extend this concept to
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004915additional even-byte lengths (6 bytes, 8 bytes and more, respectively).
4916</p>
4917
4918</div>
4919
4920<!-- _______________________________________________________________________ -->
4921<div class="doc_subsubsection">
4922 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
4923</div>
4924
4925<div class="doc_text">
4926
4927<h5>Syntax:</h5>
4928<p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
4929width. Not all targets support all bit widths however.
4930<pre>
Chandler Carrutha228e392007-08-04 01:51:18 +00004931 declare i8 @llvm.ctpop.i8 (i8 &lt;src&gt;)
4932 declare i16 @llvm.ctpop.i16(i16 &lt;src&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004933 declare i32 @llvm.ctpop.i32(i32 &lt;src&gt;)
Chandler Carrutha228e392007-08-04 01:51:18 +00004934 declare i64 @llvm.ctpop.i64(i64 &lt;src&gt;)
4935 declare i256 @llvm.ctpop.i256(i256 &lt;src&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004936</pre>
4937
4938<h5>Overview:</h5>
4939
4940<p>
4941The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
4942value.
4943</p>
4944
4945<h5>Arguments:</h5>
4946
4947<p>
4948The only argument is the value to be counted. The argument may be of any
4949integer type. The return type must match the argument type.
4950</p>
4951
4952<h5>Semantics:</h5>
4953
4954<p>
4955The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
4956</p>
4957</div>
4958
4959<!-- _______________________________________________________________________ -->
4960<div class="doc_subsubsection">
4961 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
4962</div>
4963
4964<div class="doc_text">
4965
4966<h5>Syntax:</h5>
4967<p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
4968integer bit width. Not all targets support all bit widths however.
4969<pre>
Chandler Carrutha228e392007-08-04 01:51:18 +00004970 declare i8 @llvm.ctlz.i8 (i8 &lt;src&gt;)
4971 declare i16 @llvm.ctlz.i16(i16 &lt;src&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004972 declare i32 @llvm.ctlz.i32(i32 &lt;src&gt;)
Chandler Carrutha228e392007-08-04 01:51:18 +00004973 declare i64 @llvm.ctlz.i64(i64 &lt;src&gt;)
4974 declare i256 @llvm.ctlz.i256(i256 &lt;src&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004975</pre>
4976
4977<h5>Overview:</h5>
4978
4979<p>
4980The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
4981leading zeros in a variable.
4982</p>
4983
4984<h5>Arguments:</h5>
4985
4986<p>
4987The only argument is the value to be counted. The argument may be of any
4988integer type. The return type must match the argument type.
4989</p>
4990
4991<h5>Semantics:</h5>
4992
4993<p>
4994The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
4995in a variable. If the src == 0 then the result is the size in bits of the type
4996of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.
4997</p>
4998</div>
4999
5000
5001
5002<!-- _______________________________________________________________________ -->
5003<div class="doc_subsubsection">
5004 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
5005</div>
5006
5007<div class="doc_text">
5008
5009<h5>Syntax:</h5>
5010<p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
5011integer bit width. Not all targets support all bit widths however.
5012<pre>
Chandler Carrutha228e392007-08-04 01:51:18 +00005013 declare i8 @llvm.cttz.i8 (i8 &lt;src&gt;)
5014 declare i16 @llvm.cttz.i16(i16 &lt;src&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +00005015 declare i32 @llvm.cttz.i32(i32 &lt;src&gt;)
Chandler Carrutha228e392007-08-04 01:51:18 +00005016 declare i64 @llvm.cttz.i64(i64 &lt;src&gt;)
5017 declare i256 @llvm.cttz.i256(i256 &lt;src&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +00005018</pre>
5019
5020<h5>Overview:</h5>
5021
5022<p>
5023The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
5024trailing zeros.
5025</p>
5026
5027<h5>Arguments:</h5>
5028
5029<p>
5030The only argument is the value to be counted. The argument may be of any
5031integer type. The return type must match the argument type.
5032</p>
5033
5034<h5>Semantics:</h5>
5035
5036<p>
5037The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
5038in a variable. If the src == 0 then the result is the size in bits of the type
5039of src. For example, <tt>llvm.cttz(2) = 1</tt>.
5040</p>
5041</div>
5042
5043<!-- _______________________________________________________________________ -->
5044<div class="doc_subsubsection">
5045 <a name="int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic</a>
5046</div>
5047
5048<div class="doc_text">
5049
5050<h5>Syntax:</h5>
5051<p>This is an overloaded intrinsic. You can use <tt>llvm.part.select</tt>
5052on any integer bit width.
5053<pre>
Chandler Carrutha228e392007-08-04 01:51:18 +00005054 declare i17 @llvm.part.select.i17 (i17 %val, i32 %loBit, i32 %hiBit)
5055 declare i29 @llvm.part.select.i29 (i29 %val, i32 %loBit, i32 %hiBit)
Dan Gohmanf17a25c2007-07-18 16:29:46 +00005056</pre>
5057
5058<h5>Overview:</h5>
5059<p>The '<tt>llvm.part.select</tt>' family of intrinsic functions selects a
5060range of bits from an integer value and returns them in the same bit width as
5061the original value.</p>
5062
5063<h5>Arguments:</h5>
5064<p>The first argument, <tt>%val</tt> and the result may be integer types of
5065any bit width but they must have the same bit width. The second and third
5066arguments must be <tt>i32</tt> type since they specify only a bit index.</p>
5067
5068<h5>Semantics:</h5>
5069<p>The operation of the '<tt>llvm.part.select</tt>' intrinsic has two modes
5070of operation: forwards and reverse. If <tt>%loBit</tt> is greater than
5071<tt>%hiBits</tt> then the intrinsic operates in reverse mode. Otherwise it
5072operates in forward mode.</p>
5073<p>In forward mode, this intrinsic is the equivalent of shifting <tt>%val</tt>
5074right by <tt>%loBit</tt> bits and then ANDing it with a mask with
5075only the <tt>%hiBit - %loBit</tt> bits set, as follows:</p>
5076<ol>
5077 <li>The <tt>%val</tt> is shifted right (LSHR) by the number of bits specified
5078 by <tt>%loBits</tt>. This normalizes the value to the low order bits.</li>
5079 <li>The <tt>%loBits</tt> value is subtracted from the <tt>%hiBits</tt> value
5080 to determine the number of bits to retain.</li>
5081 <li>A mask of the retained bits is created by shifting a -1 value.</li>
5082 <li>The mask is ANDed with <tt>%val</tt> to produce the result.
5083</ol>
5084<p>In reverse mode, a similar computation is made except that the bits are
5085returned in the reverse order. So, for example, if <tt>X</tt> has the value
5086<tt>i16 0x0ACF (101011001111)</tt> and we apply
5087<tt>part.select(i16 X, 8, 3)</tt> to it, we get back the value
5088<tt>i16 0x0026 (000000100110)</tt>.</p>
5089</div>
5090
5091<div class="doc_subsubsection">
5092 <a name="int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic</a>
5093</div>
5094
5095<div class="doc_text">
5096
5097<h5>Syntax:</h5>
5098<p>This is an overloaded intrinsic. You can use <tt>llvm.part.set</tt>
5099on any integer bit width.
5100<pre>
Chandler Carrutha228e392007-08-04 01:51:18 +00005101 declare i17 @llvm.part.set.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
5102 declare i29 @llvm.part.set.i29.i9 (i29 %val, i9 %repl, i32 %lo, i32 %hi)
Dan Gohmanf17a25c2007-07-18 16:29:46 +00005103</pre>
5104
5105<h5>Overview:</h5>
5106<p>The '<tt>llvm.part.set</tt>' family of intrinsic functions replaces a range
5107of bits in an integer value with another integer value. It returns the integer
5108with the replaced bits.</p>
5109
5110<h5>Arguments:</h5>
5111<p>The first argument, <tt>%val</tt> and the result may be integer types of
5112any bit width but they must have the same bit width. <tt>%val</tt> is the value
5113whose bits will be replaced. The second argument, <tt>%repl</tt> may be an
5114integer of any bit width. The third and fourth arguments must be <tt>i32</tt>
5115type since they specify only a bit index.</p>
5116
5117<h5>Semantics:</h5>
5118<p>The operation of the '<tt>llvm.part.set</tt>' intrinsic has two modes
5119of operation: forwards and reverse. If <tt>%lo</tt> is greater than
5120<tt>%hi</tt> then the intrinsic operates in reverse mode. Otherwise it
5121operates in forward mode.</p>
5122<p>For both modes, the <tt>%repl</tt> value is prepared for use by either
5123truncating it down to the size of the replacement area or zero extending it
5124up to that size.</p>
5125<p>In forward mode, the bits between <tt>%lo</tt> and <tt>%hi</tt> (inclusive)
5126are replaced with corresponding bits from <tt>%repl</tt>. That is the 0th bit
5127in <tt>%repl</tt> replaces the <tt>%lo</tt>th bit in <tt>%val</tt> and etc. up
5128to the <tt>%hi</tt>th bit.
5129<p>In reverse mode, a similar computation is made except that the bits are
5130reversed. That is, the <tt>0</tt>th bit in <tt>%repl</tt> replaces the
5131<tt>%hi</tt> bit in <tt>%val</tt> and etc. down to the <tt>%lo</tt>th bit.
5132<h5>Examples:</h5>
5133<pre>
5134 llvm.part.set(0xFFFF, 0, 4, 7) -&gt; 0xFF0F
5135 llvm.part.set(0xFFFF, 0, 7, 4) -&gt; 0xFF0F
5136 llvm.part.set(0xFFFF, 1, 7, 4) -&gt; 0xFF8F
5137 llvm.part.set(0xFFFF, F, 8, 3) -&gt; 0xFFE7
5138 llvm.part.set(0xFFFF, 0, 3, 8) -&gt; 0xFE07
5139</pre>
5140</div>
5141
5142<!-- ======================================================================= -->
5143<div class="doc_subsection">
5144 <a name="int_debugger">Debugger Intrinsics</a>
5145</div>
5146
5147<div class="doc_text">
5148<p>
5149The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
5150are described in the <a
5151href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
5152Debugging</a> document.
5153</p>
5154</div>
5155
5156
5157<!-- ======================================================================= -->
5158<div class="doc_subsection">
5159 <a name="int_eh">Exception Handling Intrinsics</a>
5160</div>
5161
5162<div class="doc_text">
5163<p> The LLVM exception handling intrinsics (which all start with
5164<tt>llvm.eh.</tt> prefix), are described in the <a
5165href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
5166Handling</a> document. </p>
5167</div>
5168
5169<!-- ======================================================================= -->
5170<div class="doc_subsection">
Duncan Sands7407a9f2007-09-11 14:10:23 +00005171 <a name="int_trampoline">Trampoline Intrinsic</a>
Duncan Sands38947cd2007-07-27 12:58:54 +00005172</div>
5173
5174<div class="doc_text">
5175<p>
Duncan Sands7407a9f2007-09-11 14:10:23 +00005176 This intrinsic makes it possible to excise one parameter, marked with
Duncan Sands38947cd2007-07-27 12:58:54 +00005177 the <tt>nest</tt> attribute, from a function. The result is a callable
5178 function pointer lacking the nest parameter - the caller does not need
5179 to provide a value for it. Instead, the value to use is stored in
5180 advance in a "trampoline", a block of memory usually allocated
5181 on the stack, which also contains code to splice the nest value into the
5182 argument list. This is used to implement the GCC nested function address
5183 extension.
5184</p>
5185<p>
5186 For example, if the function is
5187 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
Bill Wendlinge262b4c2007-09-22 09:23:55 +00005188 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as follows:</p>
Duncan Sands38947cd2007-07-27 12:58:54 +00005189<pre>
Duncan Sands7407a9f2007-09-11 14:10:23 +00005190 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
5191 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
5192 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
5193 %fp = bitcast i8* %p to i32 (i32, i32)*
Duncan Sands38947cd2007-07-27 12:58:54 +00005194</pre>
Bill Wendlinge262b4c2007-09-22 09:23:55 +00005195 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
5196 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
Duncan Sands38947cd2007-07-27 12:58:54 +00005197</div>
5198
5199<!-- _______________________________________________________________________ -->
5200<div class="doc_subsubsection">
5201 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
5202</div>
5203<div class="doc_text">
5204<h5>Syntax:</h5>
5205<pre>
Duncan Sands7407a9f2007-09-11 14:10:23 +00005206declare i8* @llvm.init.trampoline(i8* &lt;tramp&gt;, i8* &lt;func&gt;, i8* &lt;nval&gt;)
Duncan Sands38947cd2007-07-27 12:58:54 +00005207</pre>
5208<h5>Overview:</h5>
5209<p>
Duncan Sands7407a9f2007-09-11 14:10:23 +00005210 This fills the memory pointed to by <tt>tramp</tt> with code
5211 and returns a function pointer suitable for executing it.
Duncan Sands38947cd2007-07-27 12:58:54 +00005212</p>
5213<h5>Arguments:</h5>
5214<p>
5215 The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
5216 pointers. The <tt>tramp</tt> argument must point to a sufficiently large
5217 and sufficiently aligned block of memory; this memory is written to by the
Duncan Sands35012212007-08-22 23:39:54 +00005218 intrinsic. Note that the size and the alignment are target-specific - LLVM
5219 currently provides no portable way of determining them, so a front-end that
5220 generates this intrinsic needs to have some target-specific knowledge.
5221 The <tt>func</tt> argument must hold a function bitcast to an <tt>i8*</tt>.
Duncan Sands38947cd2007-07-27 12:58:54 +00005222</p>
5223<h5>Semantics:</h5>
5224<p>
5225 The block of memory pointed to by <tt>tramp</tt> is filled with target
Duncan Sands7407a9f2007-09-11 14:10:23 +00005226 dependent code, turning it into a function. A pointer to this function is
5227 returned, but needs to be bitcast to an
Duncan Sands38947cd2007-07-27 12:58:54 +00005228 <a href="#int_trampoline">appropriate function pointer type</a>
Duncan Sands7407a9f2007-09-11 14:10:23 +00005229 before being called. The new function's signature is the same as that of
5230 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
5231 removed. At most one such <tt>nest</tt> argument is allowed, and it must be
5232 of pointer type. Calling the new function is equivalent to calling
5233 <tt>func</tt> with the same argument list, but with <tt>nval</tt> used for the
5234 missing <tt>nest</tt> argument. If, after calling
5235 <tt>llvm.init.trampoline</tt>, the memory pointed to by <tt>tramp</tt> is
5236 modified, then the effect of any later call to the returned function pointer is
5237 undefined.
Duncan Sands38947cd2007-07-27 12:58:54 +00005238</p>
5239</div>
5240
5241<!-- ======================================================================= -->
5242<div class="doc_subsection">
Andrew Lenharth785610d2008-02-16 01:24:58 +00005243 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
5244</div>
5245
5246<div class="doc_text">
5247<p>
5248 These intrinsic functions expand the "universal IR" of LLVM to represent
5249 hardware constructs for atomic operations and memory synchronization. This
5250 provides an interface to the hardware, not an interface to the programmer. It
5251 is aimed at a low enough level to allow any programming models or APIs which
5252 need atomic behaviors to map cleanly onto it. It is also modeled primarily on
5253 hardware behavior. Just as hardware provides a "universal IR" for source
5254 languages, it also provides a starting point for developing a "universal"
5255 atomic operation and synchronization IR.
5256</p>
5257<p>
5258 These do <em>not</em> form an API such as high-level threading libraries,
5259 software transaction memory systems, atomic primitives, and intrinsic
5260 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
5261 application libraries. The hardware interface provided by LLVM should allow
5262 a clean implementation of all of these APIs and parallel programming models.
5263 No one model or paradigm should be selected above others unless the hardware
5264 itself ubiquitously does so.
5265
5266</p>
5267</div>
5268
5269<!-- _______________________________________________________________________ -->
5270<div class="doc_subsubsection">
5271 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
5272</div>
5273<div class="doc_text">
5274<h5>Syntax:</h5>
5275<pre>
5276declare void @llvm.memory.barrier( i1 &lt;ll&gt;, i1 &lt;ls&gt;, i1 &lt;sl&gt;, i1 &lt;ss&gt;,
5277i1 &lt;device&gt; )
5278
5279</pre>
5280<h5>Overview:</h5>
5281<p>
5282 The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
5283 specific pairs of memory access types.
5284</p>
5285<h5>Arguments:</h5>
5286<p>
5287 The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
5288 The first four arguments enables a specific barrier as listed below. The fith
5289 argument specifies that the barrier applies to io or device or uncached memory.
5290
5291</p>
5292 <ul>
5293 <li><tt>ll</tt>: load-load barrier</li>
5294 <li><tt>ls</tt>: load-store barrier</li>
5295 <li><tt>sl</tt>: store-load barrier</li>
5296 <li><tt>ss</tt>: store-store barrier</li>
5297 <li><tt>device</tt>: barrier applies to device and uncached memory also.
5298 </ul>
5299<h5>Semantics:</h5>
5300<p>
5301 This intrinsic causes the system to enforce some ordering constraints upon
5302 the loads and stores of the program. This barrier does not indicate
5303 <em>when</em> any events will occur, it only enforces an <em>order</em> in
5304 which they occur. For any of the specified pairs of load and store operations
5305 (f.ex. load-load, or store-load), all of the first operations preceding the
5306 barrier will complete before any of the second operations succeeding the
5307 barrier begin. Specifically the semantics for each pairing is as follows:
5308</p>
5309 <ul>
5310 <li><tt>ll</tt>: All loads before the barrier must complete before any load
5311 after the barrier begins.</li>
5312
5313 <li><tt>ls</tt>: All loads before the barrier must complete before any
5314 store after the barrier begins.</li>
5315 <li><tt>ss</tt>: All stores before the barrier must complete before any
5316 store after the barrier begins.</li>
5317 <li><tt>sl</tt>: All stores before the barrier must complete before any
5318 load after the barrier begins.</li>
5319 </ul>
5320<p>
5321 These semantics are applied with a logical "and" behavior when more than one
5322 is enabled in a single memory barrier intrinsic.
5323</p>
5324<p>
5325 Backends may implement stronger barriers than those requested when they do not
5326 support as fine grained a barrier as requested. Some architectures do not
5327 need all types of barriers and on such architectures, these become noops.
5328</p>
5329<h5>Example:</h5>
5330<pre>
5331%ptr = malloc i32
5332 store i32 4, %ptr
5333
5334%result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
5335 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
5336 <i>; guarantee the above finishes</i>
5337 store i32 8, %ptr <i>; before this begins</i>
5338</pre>
5339</div>
5340
5341
5342<!-- ======================================================================= -->
5343<div class="doc_subsection">
Dan Gohmanf17a25c2007-07-18 16:29:46 +00005344 <a name="int_general">General Intrinsics</a>
5345</div>
5346
5347<div class="doc_text">
5348<p> This class of intrinsics is designed to be generic and has
5349no specific purpose. </p>
5350</div>
5351
5352<!-- _______________________________________________________________________ -->
5353<div class="doc_subsubsection">
5354 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
5355</div>
5356
5357<div class="doc_text">
5358
5359<h5>Syntax:</h5>
5360<pre>
5361 declare void @llvm.var.annotation(i8* &lt;val&gt;, i8* &lt;str&gt;, i8* &lt;str&gt;, i32 &lt;int&gt; )
5362</pre>
5363
5364<h5>Overview:</h5>
5365
5366<p>
5367The '<tt>llvm.var.annotation</tt>' intrinsic
5368</p>
5369
5370<h5>Arguments:</h5>
5371
5372<p>
5373The first argument is a pointer to a value, the second is a pointer to a
5374global string, the third is a pointer to a global string which is the source
5375file name, and the last argument is the line number.
5376</p>
5377
5378<h5>Semantics:</h5>
5379
5380<p>
Anton Korobeynikove6e764f2008-01-15 22:31:34 +00005381This intrinsic allows annotation of local variables with arbitrary strings.
Dan Gohmanf17a25c2007-07-18 16:29:46 +00005382This can be useful for special purpose optimizations that want to look for these
Anton Korobeynikove6e764f2008-01-15 22:31:34 +00005383annotations. These have no other defined use, they are ignored by code
5384generation and optimization.
5385</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00005386</div>
5387
Tanya Lattnerb306a9e2007-09-21 22:59:12 +00005388<!-- _______________________________________________________________________ -->
5389<div class="doc_subsubsection">
Tanya Lattnerc9869b12007-09-21 23:57:59 +00005390 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
Tanya Lattnerb306a9e2007-09-21 22:59:12 +00005391</div>
5392
5393<div class="doc_text">
5394
5395<h5>Syntax:</h5>
Tanya Lattnere545be72007-09-21 23:56:27 +00005396<p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
5397any integer bit width.
5398</p>
Tanya Lattnerb306a9e2007-09-21 22:59:12 +00005399<pre>
Tanya Lattner09161fe2007-09-22 00:03:01 +00005400 declare i8 @llvm.annotation.i8(i8 &lt;val&gt;, i8* &lt;str&gt;, i8* &lt;str&gt;, i32 &lt;int&gt; )
5401 declare i16 @llvm.annotation.i16(i16 &lt;val&gt;, i8* &lt;str&gt;, i8* &lt;str&gt;, i32 &lt;int&gt; )
5402 declare i32 @llvm.annotation.i32(i32 &lt;val&gt;, i8* &lt;str&gt;, i8* &lt;str&gt;, i32 &lt;int&gt; )
5403 declare i64 @llvm.annotation.i64(i64 &lt;val&gt;, i8* &lt;str&gt;, i8* &lt;str&gt;, i32 &lt;int&gt; )
5404 declare i256 @llvm.annotation.i256(i256 &lt;val&gt;, i8* &lt;str&gt;, i8* &lt;str&gt;, i32 &lt;int&gt; )
Tanya Lattnerb306a9e2007-09-21 22:59:12 +00005405</pre>
5406
5407<h5>Overview:</h5>
Tanya Lattnere545be72007-09-21 23:56:27 +00005408
5409<p>
5410The '<tt>llvm.annotation</tt>' intrinsic.
Tanya Lattnerb306a9e2007-09-21 22:59:12 +00005411</p>
5412
5413<h5>Arguments:</h5>
5414
5415<p>
5416The first argument is an integer value (result of some expression),
5417the second is a pointer to a global string, the third is a pointer to a global
5418string which is the source file name, and the last argument is the line number.
Tanya Lattnere545be72007-09-21 23:56:27 +00005419It returns the value of the first argument.
Tanya Lattnerb306a9e2007-09-21 22:59:12 +00005420</p>
5421
5422<h5>Semantics:</h5>
5423
5424<p>
5425This intrinsic allows annotations to be put on arbitrary expressions
5426with arbitrary strings. This can be useful for special purpose optimizations
5427that want to look for these annotations. These have no other defined use, they
5428are ignored by code generation and optimization.
5429</div>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00005430
Anton Korobeynikove6e764f2008-01-15 22:31:34 +00005431<!-- _______________________________________________________________________ -->
5432<div class="doc_subsubsection">
5433 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
5434</div>
5435
5436<div class="doc_text">
5437
5438<h5>Syntax:</h5>
5439<pre>
5440 declare void @llvm.trap()
5441</pre>
5442
5443<h5>Overview:</h5>
5444
5445<p>
5446The '<tt>llvm.trap</tt>' intrinsic
5447</p>
5448
5449<h5>Arguments:</h5>
5450
5451<p>
5452None
5453</p>
5454
5455<h5>Semantics:</h5>
5456
5457<p>
5458This intrinsics is lowered to the target dependent trap instruction. If the
5459target does not have a trap instruction, this intrinsic will be lowered to the
5460call of the abort() function.
5461</p>
5462</div>
5463
Dan Gohmanf17a25c2007-07-18 16:29:46 +00005464<!-- *********************************************************************** -->
5465<hr>
5466<address>
5467 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
5468 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
5469 <a href="http://validator.w3.org/check/referer"><img
Chris Lattner08497ce2008-01-04 04:33:49 +00005470 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00005471
5472 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
5473 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
5474 Last modified: $Date$
5475</address>
Chris Lattner08497ce2008-01-04 04:33:49 +00005476
Dan Gohmanf17a25c2007-07-18 16:29:46 +00005477</body>
5478</html>