• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1<!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="_static/llvm.css" type="text/css">
11</head>
12
13<body>
14
15<h1>LLVM Language Reference Manual</h1>
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>
24        <ol>
25          <li><a href="#linkage_private">'<tt>private</tt>' Linkage</a></li>
26          <li><a href="#linkage_linker_private">'<tt>linker_private</tt>' Linkage</a></li>
27          <li><a href="#linkage_linker_private_weak">'<tt>linker_private_weak</tt>' Linkage</a></li>
28          <li><a href="#linkage_internal">'<tt>internal</tt>' Linkage</a></li>
29          <li><a href="#linkage_available_externally">'<tt>available_externally</tt>' Linkage</a></li>
30          <li><a href="#linkage_linkonce">'<tt>linkonce</tt>' Linkage</a></li>
31          <li><a href="#linkage_common">'<tt>common</tt>' Linkage</a></li>
32          <li><a href="#linkage_weak">'<tt>weak</tt>' Linkage</a></li>
33          <li><a href="#linkage_appending">'<tt>appending</tt>' Linkage</a></li>
34          <li><a href="#linkage_externweak">'<tt>extern_weak</tt>' Linkage</a></li>
35          <li><a href="#linkage_linkonce_odr">'<tt>linkonce_odr</tt>' Linkage</a></li>
36          <li><a href="#linkage_linkonce_odr_auto_hide">'<tt>linkonce_odr_auto_hide</tt>' Linkage</a></li>
37          <li><a href="#linkage_weak">'<tt>weak_odr</tt>' Linkage</a></li>
38          <li><a href="#linkage_external">'<tt>external</tt>' Linkage</a></li>
39          <li><a href="#linkage_dllimport">'<tt>dllimport</tt>' Linkage</a></li>
40          <li><a href="#linkage_dllexport">'<tt>dllexport</tt>' Linkage</a></li>
41        </ol>
42      </li>
43      <li><a href="#callingconv">Calling Conventions</a></li>
44      <li><a href="#namedtypes">Named Types</a></li>
45      <li><a href="#globalvars">Global Variables</a></li>
46      <li><a href="#functionstructure">Functions</a></li>
47      <li><a href="#aliasstructure">Aliases</a></li>
48      <li><a href="#namedmetadatastructure">Named Metadata</a></li>
49      <li><a href="#paramattrs">Parameter Attributes</a></li>
50      <li><a href="#fnattrs">Function Attributes</a></li>
51      <li><a href="#gc">Garbage Collector Names</a></li>
52      <li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
53      <li><a href="#datalayout">Data Layout</a></li>
54      <li><a href="#pointeraliasing">Pointer Aliasing Rules</a></li>
55      <li><a href="#volatile">Volatile Memory Accesses</a></li>
56      <li><a href="#memmodel">Memory Model for Concurrent Operations</a></li>
57      <li><a href="#ordering">Atomic Memory Ordering Constraints</a></li>
58    </ol>
59  </li>
60  <li><a href="#typesystem">Type System</a>
61    <ol>
62      <li><a href="#t_classifications">Type Classifications</a></li>
63      <li><a href="#t_primitive">Primitive Types</a>
64        <ol>
65          <li><a href="#t_integer">Integer Type</a></li>
66          <li><a href="#t_floating">Floating Point Types</a></li>
67          <li><a href="#t_x86mmx">X86mmx Type</a></li>
68          <li><a href="#t_void">Void Type</a></li>
69          <li><a href="#t_label">Label Type</a></li>
70          <li><a href="#t_metadata">Metadata Type</a></li>
71        </ol>
72      </li>
73      <li><a href="#t_derived">Derived Types</a>
74        <ol>
75          <li><a href="#t_aggregate">Aggregate Types</a>
76            <ol>
77              <li><a href="#t_array">Array Type</a></li>
78              <li><a href="#t_struct">Structure Type</a></li>
79              <li><a href="#t_opaque">Opaque Structure Types</a></li>
80              <li><a href="#t_vector">Vector Type</a></li>
81            </ol>
82          </li>
83          <li><a href="#t_function">Function Type</a></li>
84          <li><a href="#t_pointer">Pointer Type</a></li>
85        </ol>
86      </li>
87    </ol>
88  </li>
89  <li><a href="#constants">Constants</a>
90    <ol>
91      <li><a href="#simpleconstants">Simple Constants</a></li>
92      <li><a href="#complexconstants">Complex Constants</a></li>
93      <li><a href="#globalconstants">Global Variable and Function Addresses</a></li>
94      <li><a href="#undefvalues">Undefined Values</a></li>
95      <li><a href="#poisonvalues">Poison Values</a></li>
96      <li><a href="#blockaddress">Addresses of Basic Blocks</a></li>
97      <li><a href="#constantexprs">Constant Expressions</a></li>
98    </ol>
99  </li>
100  <li><a href="#othervalues">Other Values</a>
101    <ol>
102      <li><a href="#inlineasm">Inline Assembler Expressions</a></li>
103      <li><a href="#metadata">Metadata Nodes and Metadata Strings</a>
104        <ol>
105          <li><a href="#tbaa">'<tt>tbaa</tt>' Metadata</a></li>
106          <li><a href="#fpmath">'<tt>fpmath</tt>' Metadata</a></li>
107          <li><a href="#range">'<tt>range</tt>' Metadata</a></li>
108        </ol>
109      </li>
110    </ol>
111  </li>
112  <li><a href="#module_flags">Module Flags Metadata</a>
113    <ol>
114      <li><a href="#objc_gc_flags">Objective-C Garbage Collection Module Flags Metadata</a></li>
115    </ol>
116  </li>
117  <li><a href="#intrinsic_globals">Intrinsic Global Variables</a>
118    <ol>
119      <li><a href="#intg_used">The '<tt>llvm.used</tt>' Global Variable</a></li>
120      <li><a href="#intg_compiler_used">The '<tt>llvm.compiler.used</tt>'
121          Global Variable</a></li>
122      <li><a href="#intg_global_ctors">The '<tt>llvm.global_ctors</tt>'
123         Global Variable</a></li>
124      <li><a href="#intg_global_dtors">The '<tt>llvm.global_dtors</tt>'
125         Global Variable</a></li>
126    </ol>
127  </li>
128  <li><a href="#instref">Instruction Reference</a>
129    <ol>
130      <li><a href="#terminators">Terminator Instructions</a>
131        <ol>
132          <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
133          <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
134          <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
135          <li><a href="#i_indirectbr">'<tt>indirectbr</tt>' Instruction</a></li>
136          <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
137          <li><a href="#i_resume">'<tt>resume</tt>'  Instruction</a></li>
138          <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
139        </ol>
140      </li>
141      <li><a href="#binaryops">Binary Operations</a>
142        <ol>
143          <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
144          <li><a href="#i_fadd">'<tt>fadd</tt>' Instruction</a></li>
145          <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
146          <li><a href="#i_fsub">'<tt>fsub</tt>' Instruction</a></li>
147          <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
148          <li><a href="#i_fmul">'<tt>fmul</tt>' Instruction</a></li>
149          <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
150          <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
151          <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
152          <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
153          <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
154          <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
155        </ol>
156      </li>
157      <li><a href="#bitwiseops">Bitwise Binary Operations</a>
158        <ol>
159          <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
160          <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
161          <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
162          <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
163          <li><a href="#i_or">'<tt>or</tt>'  Instruction</a></li>
164          <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
165        </ol>
166      </li>
167      <li><a href="#vectorops">Vector Operations</a>
168        <ol>
169          <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
170          <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
171          <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
172        </ol>
173      </li>
174      <li><a href="#aggregateops">Aggregate Operations</a>
175        <ol>
176          <li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li>
177          <li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li>
178        </ol>
179      </li>
180      <li><a href="#memoryops">Memory Access and Addressing Operations</a>
181        <ol>
182          <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
183         <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
184         <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
185         <li><a href="#i_fence">'<tt>fence</tt>' Instruction</a></li>
186         <li><a href="#i_cmpxchg">'<tt>cmpxchg</tt>' Instruction</a></li>
187         <li><a href="#i_atomicrmw">'<tt>atomicrmw</tt>' Instruction</a></li>
188         <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
189        </ol>
190      </li>
191      <li><a href="#convertops">Conversion Operations</a>
192        <ol>
193          <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
194          <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
195          <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
196          <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
197          <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
198          <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
199          <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
200          <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
201          <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
202          <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
203          <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
204          <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
205        </ol>
206      </li>
207      <li><a href="#otherops">Other Operations</a>
208        <ol>
209          <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
210          <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
211          <li><a href="#i_phi">'<tt>phi</tt>'   Instruction</a></li>
212          <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
213          <li><a href="#i_call">'<tt>call</tt>'  Instruction</a></li>
214          <li><a href="#i_va_arg">'<tt>va_arg</tt>'  Instruction</a></li>
215          <li><a href="#i_landingpad">'<tt>landingpad</tt>' Instruction</a></li>
216        </ol>
217      </li>
218    </ol>
219  </li>
220  <li><a href="#intrinsics">Intrinsic Functions</a>
221    <ol>
222      <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
223        <ol>
224          <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
225          <li><a href="#int_va_end">'<tt>llvm.va_end</tt>'   Intrinsic</a></li>
226          <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>'  Intrinsic</a></li>
227        </ol>
228      </li>
229      <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
230        <ol>
231          <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
232          <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
233          <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
234        </ol>
235      </li>
236      <li><a href="#int_codegen">Code Generator Intrinsics</a>
237        <ol>
238          <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
239          <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>'   Intrinsic</a></li>
240          <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
241          <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
242          <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
243          <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
244          <li><a href="#int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
245        </ol>
246      </li>
247      <li><a href="#int_libc">Standard C Library Intrinsics</a>
248        <ol>
249          <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
250          <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
251          <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
252          <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
253          <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
254          <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
255          <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
256          <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
257          <li><a href="#int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a></li>
258          <li><a href="#int_log">'<tt>llvm.log.*</tt>' Intrinsic</a></li>
259          <li><a href="#int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a></li>
260          <li><a href="#int_fabs">'<tt>llvm.fabs.*</tt>' Intrinsic</a></li>
261          <li><a href="#int_floor">'<tt>llvm.floor.*</tt>' Intrinsic</a></li>
262        </ol>
263      </li>
264      <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
265        <ol>
266          <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
267          <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
268          <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
269          <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
270        </ol>
271      </li>
272      <li><a href="#int_overflow">Arithmetic with Overflow Intrinsics</a>
273        <ol>
274          <li><a href="#int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt> Intrinsics</a></li>
275          <li><a href="#int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt> Intrinsics</a></li>
276          <li><a href="#int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt> Intrinsics</a></li>
277          <li><a href="#int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt> Intrinsics</a></li>
278          <li><a href="#int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt> Intrinsics</a></li>
279          <li><a href="#int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt> Intrinsics</a></li>
280        </ol>
281      </li>
282      <li><a href="#spec_arithmetic">Specialised Arithmetic Intrinsics</a>
283        <ol>
284          <li><a href="#fmuladd">'<tt>llvm.fmuladd</tt> Intrinsic</a></li>
285        </ol>
286      </li>
287      <li><a href="#int_fp16">Half Precision Floating Point Intrinsics</a>
288        <ol>
289          <li><a href="#int_convert_to_fp16">'<tt>llvm.convert.to.fp16</tt>' Intrinsic</a></li>
290          <li><a href="#int_convert_from_fp16">'<tt>llvm.convert.from.fp16</tt>' Intrinsic</a></li>
291        </ol>
292      </li>
293      <li><a href="#int_debugger">Debugger intrinsics</a></li>
294      <li><a href="#int_eh">Exception Handling intrinsics</a></li>
295      <li><a href="#int_trampoline">Trampoline Intrinsics</a>
296        <ol>
297          <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
298          <li><a href="#int_at">'<tt>llvm.adjust.trampoline</tt>' Intrinsic</a></li>
299        </ol>
300      </li>
301      <li><a href="#int_memorymarkers">Memory Use Markers</a>
302        <ol>
303          <li><a href="#int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a></li>
304          <li><a href="#int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a></li>
305          <li><a href="#int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a></li>
306          <li><a href="#int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a></li>
307        </ol>
308      </li>
309      <li><a href="#int_general">General intrinsics</a>
310        <ol>
311          <li><a href="#int_var_annotation">
312            '<tt>llvm.var.annotation</tt>' Intrinsic</a></li>
313          <li><a href="#int_annotation">
314            '<tt>llvm.annotation.*</tt>' Intrinsic</a></li>
315          <li><a href="#int_trap">
316            '<tt>llvm.trap</tt>' Intrinsic</a></li>
317          <li><a href="#int_debugtrap">
318            '<tt>llvm.debugtrap</tt>' Intrinsic</a></li>
319          <li><a href="#int_stackprotector">
320            '<tt>llvm.stackprotector</tt>' Intrinsic</a></li>
321          <li><a href="#int_objectsize">
322            '<tt>llvm.objectsize</tt>' Intrinsic</a></li>
323          <li><a href="#int_expect">
324            '<tt>llvm.expect</tt>' Intrinsic</a></li>
325          <li><a href="#int_donothing">
326            '<tt>llvm.donothing</tt>' Intrinsic</a></li>
327        </ol>
328      </li>
329    </ol>
330  </li>
331</ol>
332
333<div class="doc_author">
334  <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
335            and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
336</div>
337
338<!-- *********************************************************************** -->
339<h2><a name="abstract">Abstract</a></h2>
340<!-- *********************************************************************** -->
341
342<div>
343
344<p>This document is a reference manual for the LLVM assembly language. LLVM is
345   a Static Single Assignment (SSA) based representation that provides type
346   safety, low-level operations, flexibility, and the capability of representing
347   'all' high-level languages cleanly.  It is the common code representation
348   used throughout all phases of the LLVM compilation strategy.</p>
349
350</div>
351
352<!-- *********************************************************************** -->
353<h2><a name="introduction">Introduction</a></h2>
354<!-- *********************************************************************** -->
355
356<div>
357
358<p>The LLVM code representation is designed to be used in three different forms:
359   as an in-memory compiler IR, as an on-disk bitcode representation (suitable
360   for fast loading by a Just-In-Time compiler), and as a human readable
361   assembly language representation.  This allows LLVM to provide a powerful
362   intermediate representation for efficient compiler transformations and
363   analysis, while providing a natural means to debug and visualize the
364   transformations.  The three different forms of LLVM are all equivalent.  This
365   document describes the human readable representation and notation.</p>
366
367<p>The LLVM representation aims to be light-weight and low-level while being
368   expressive, typed, and extensible at the same time.  It aims to be a
369   "universal IR" of sorts, by being at a low enough level that high-level ideas
370   may be cleanly mapped to it (similar to how microprocessors are "universal
371   IR's", allowing many source languages to be mapped to them).  By providing
372   type information, LLVM can be used as the target of optimizations: for
373   example, through pointer analysis, it can be proven that a C automatic
374   variable is never accessed outside of the current function, allowing it to
375   be promoted to a simple SSA value instead of a memory location.</p>
376
377<!-- _______________________________________________________________________ -->
378<h4>
379  <a name="wellformed">Well-Formedness</a>
380</h4>
381
382<div>
383
384<p>It is important to note that this document describes 'well formed' LLVM
385   assembly language.  There is a difference between what the parser accepts and
386   what is considered 'well formed'.  For example, the following instruction is
387   syntactically okay, but not well formed:</p>
388
389<pre class="doc_code">
390%x = <a href="#i_add">add</a> i32 1, %x
391</pre>
392
393<p>because the definition of <tt>%x</tt> does not dominate all of its uses. The
394   LLVM infrastructure provides a verification pass that may be used to verify
395   that an LLVM module is well formed.  This pass is automatically run by the
396   parser after parsing input assembly and by the optimizer before it outputs
397   bitcode.  The violations pointed out by the verifier pass indicate bugs in
398   transformation passes or input to the parser.</p>
399
400</div>
401
402</div>
403
404<!-- Describe the typesetting conventions here. -->
405
406<!-- *********************************************************************** -->
407<h2><a name="identifiers">Identifiers</a></h2>
408<!-- *********************************************************************** -->
409
410<div>
411
412<p>LLVM identifiers come in two basic types: global and local. Global
413   identifiers (functions, global variables) begin with the <tt>'@'</tt>
414   character. Local identifiers (register names, types) begin with
415   the <tt>'%'</tt> character. Additionally, there are three different formats
416   for identifiers, for different purposes:</p>
417
418<ol>
419  <li>Named values are represented as a string of characters with their prefix.
420      For example, <tt>%foo</tt>, <tt>@DivisionByZero</tt>,
421      <tt>%a.really.long.identifier</tt>. The actual regular expression used is
422      '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.  Identifiers which require
423      other characters in their names can be surrounded with quotes. Special
424      characters may be escaped using <tt>"\xx"</tt> where <tt>xx</tt> is the
425      ASCII code for the character in hexadecimal.  In this way, any character
426      can be used in a name value, even quotes themselves.</li>
427
428  <li>Unnamed values are represented as an unsigned numeric value with their
429      prefix.  For example, <tt>%12</tt>, <tt>@2</tt>, <tt>%44</tt>.</li>
430
431  <li>Constants, which are described in a <a href="#constants">section about
432      constants</a>, below.</li>
433</ol>
434
435<p>LLVM requires that values start with a prefix for two reasons: Compilers
436   don't need to worry about name clashes with reserved words, and the set of
437   reserved words may be expanded in the future without penalty.  Additionally,
438   unnamed identifiers allow a compiler to quickly come up with a temporary
439   variable without having to avoid symbol table conflicts.</p>
440
441<p>Reserved words in LLVM are very similar to reserved words in other
442   languages. There are keywords for different opcodes
443   ('<tt><a href="#i_add">add</a></tt>',
444   '<tt><a href="#i_bitcast">bitcast</a></tt>',
445   '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names
446   ('<tt><a href="#t_void">void</a></tt>',
447   '<tt><a href="#t_primitive">i32</a></tt>', etc...), and others.  These
448   reserved words cannot conflict with variable names, because none of them
449   start with a prefix character (<tt>'%'</tt> or <tt>'@'</tt>).</p>
450
451<p>Here is an example of LLVM code to multiply the integer variable
452   '<tt>%X</tt>' by 8:</p>
453
454<p>The easy way:</p>
455
456<pre class="doc_code">
457%result = <a href="#i_mul">mul</a> i32 %X, 8
458</pre>
459
460<p>After strength reduction:</p>
461
462<pre class="doc_code">
463%result = <a href="#i_shl">shl</a> i32 %X, i8 3
464</pre>
465
466<p>And the hard way:</p>
467
468<pre class="doc_code">
469%0 = <a href="#i_add">add</a> i32 %X, %X           <i>; yields {i32}:%0</i>
470%1 = <a href="#i_add">add</a> i32 %0, %0           <i>; yields {i32}:%1</i>
471%result = <a href="#i_add">add</a> i32 %1, %1
472</pre>
473
474<p>This last way of multiplying <tt>%X</tt> by 8 illustrates several important
475   lexical features of LLVM:</p>
476
477<ol>
478  <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
479      line.</li>
480
481  <li>Unnamed temporaries are created when the result of a computation is not
482      assigned to a named value.</li>
483
484  <li>Unnamed temporaries are numbered sequentially</li>
485</ol>
486
487<p>It also shows a convention that we follow in this document.  When
488   demonstrating instructions, we will follow an instruction with a comment that
489   defines the type and name of value produced.  Comments are shown in italic
490   text.</p>
491
492</div>
493
494<!-- *********************************************************************** -->
495<h2><a name="highlevel">High Level Structure</a></h2>
496<!-- *********************************************************************** -->
497<div>
498<!-- ======================================================================= -->
499<h3>
500  <a name="modulestructure">Module Structure</a>
501</h3>
502
503<div>
504
505<p>LLVM programs are composed of <tt>Module</tt>s, each of which is a
506   translation unit of the input programs.  Each module consists of functions,
507   global variables, and symbol table entries.  Modules may be combined together
508   with the LLVM linker, which merges function (and global variable)
509   definitions, resolves forward declarations, and merges symbol table
510   entries. Here is an example of the "hello world" module:</p>
511
512<pre class="doc_code">
513<i>; Declare the string constant as a global constant.</i>&nbsp;
514<a href="#identifiers">@.str</a> = <a href="#linkage_private">private</a>&nbsp;<a href="#globalvars">unnamed_addr</a>&nbsp;<a href="#globalvars">constant</a>&nbsp;<a href="#t_array">[13 x i8]</a> c"hello world\0A\00"&nbsp;
515
516<i>; External declaration of the puts function</i>&nbsp;
517<a href="#functionstructure">declare</a> i32 @puts(i8* <a href="#nocapture">nocapture</a>) <a href="#fnattrs">nounwind</a>&nbsp;
518
519<i>; Definition of main function</i>
520define i32 @main() {   <i>; i32()* </i>&nbsp;
521  <i>; Convert [13 x i8]* to i8  *...</i>&nbsp;
522  %cast210 = <a href="#i_getelementptr">getelementptr</a> [13 x i8]* @.str, i64 0, i64 0
523
524  <i>; Call puts function to write out the string to stdout.</i>&nbsp;
525  <a href="#i_call">call</a> i32 @puts(i8* %cast210)
526  <a href="#i_ret">ret</a> i32 0&nbsp;
527}
528
529<i>; Named metadata</i>
530!1 = metadata !{i32 42}
531!foo = !{!1, null}
532</pre>
533
534<p>This example is made up of a <a href="#globalvars">global variable</a> named
535   "<tt>.str</tt>", an external declaration of the "<tt>puts</tt>" function,
536   a <a href="#functionstructure">function definition</a> for
537   "<tt>main</tt>" and <a href="#namedmetadatastructure">named metadata</a>
538   "<tt>foo</tt>".</p>
539
540<p>In general, a module is made up of a list of global values (where both
541   functions and global variables are global values). Global values are
542   represented by a pointer to a memory location (in this case, a pointer to an
543   array of char, and a pointer to a function), and have one of the
544   following <a href="#linkage">linkage types</a>.</p>
545
546</div>
547
548<!-- ======================================================================= -->
549<h3>
550  <a name="linkage">Linkage Types</a>
551</h3>
552
553<div>
554
555<p>All Global Variables and Functions have one of the following types of
556   linkage:</p>
557
558<dl>
559  <dt><tt><b><a name="linkage_private">private</a></b></tt></dt>
560  <dd>Global values with "<tt>private</tt>" linkage are only directly accessible
561      by objects in the current module. In particular, linking code into a
562      module with an private global value may cause the private to be renamed as
563      necessary to avoid collisions.  Because the symbol is private to the
564      module, all references can be updated. This doesn't show up in any symbol
565      table in the object file.</dd>
566
567  <dt><tt><b><a name="linkage_linker_private">linker_private</a></b></tt></dt>
568  <dd>Similar to <tt>private</tt>, but the symbol is passed through the
569      assembler and evaluated by the linker. Unlike normal strong symbols, they
570      are removed by the linker from the final linked image (executable or
571      dynamic library).</dd>
572
573  <dt><tt><b><a name="linkage_linker_private_weak">linker_private_weak</a></b></tt></dt>
574  <dd>Similar to "<tt>linker_private</tt>", but the symbol is weak. Note that
575      <tt>linker_private_weak</tt> symbols are subject to coalescing by the
576      linker. The symbols are removed by the linker from the final linked image
577      (executable or dynamic library).</dd>
578
579  <dt><tt><b><a name="linkage_internal">internal</a></b></tt></dt>
580  <dd>Similar to private, but the value shows as a local symbol
581      (<tt>STB_LOCAL</tt> in the case of ELF) in the object file. This
582      corresponds to the notion of the '<tt>static</tt>' keyword in C.</dd>
583
584  <dt><tt><b><a name="linkage_available_externally">available_externally</a></b></tt></dt>
585  <dd>Globals with "<tt>available_externally</tt>" linkage are never emitted
586      into the object file corresponding to the LLVM module.  They exist to
587      allow inlining and other optimizations to take place given knowledge of
588      the definition of the global, which is known to be somewhere outside the
589      module.  Globals with <tt>available_externally</tt> linkage are allowed to
590      be discarded at will, and are otherwise the same as <tt>linkonce_odr</tt>.
591      This linkage type is only allowed on definitions, not declarations.</dd>
592
593  <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt></dt>
594  <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
595      the same name when linkage occurs.  This can be used to implement
596      some forms of inline functions, templates, or other code which must be
597      generated in each translation unit that uses it, but where the body may
598      be overridden with a more definitive definition later.  Unreferenced
599      <tt>linkonce</tt> globals are allowed to be discarded.  Note that
600      <tt>linkonce</tt> linkage does not actually allow the optimizer to
601      inline the body of this function into callers because it doesn't know if
602      this definition of the function is the definitive definition within the
603      program or whether it will be overridden by a stronger definition.
604      To enable inlining and other optimizations, use "<tt>linkonce_odr</tt>"
605      linkage.</dd>
606
607  <dt><tt><b><a name="linkage_weak">weak</a></b></tt></dt>
608  <dd>"<tt>weak</tt>" linkage has the same merging semantics as
609      <tt>linkonce</tt> linkage, except that unreferenced globals with
610      <tt>weak</tt> linkage may not be discarded.  This is used for globals that
611      are declared "weak" in C source code.</dd>
612
613  <dt><tt><b><a name="linkage_common">common</a></b></tt></dt>
614  <dd>"<tt>common</tt>" linkage is most similar to "<tt>weak</tt>" linkage, but
615      they are used for tentative definitions in C, such as "<tt>int X;</tt>" at
616      global scope.
617      Symbols with "<tt>common</tt>" linkage are merged in the same way as
618      <tt>weak symbols</tt>, and they may not be deleted if unreferenced.
619      <tt>common</tt> symbols may not have an explicit section,
620      must have a zero initializer, and may not be marked '<a
621      href="#globalvars"><tt>constant</tt></a>'.  Functions and aliases may not
622      have common linkage.</dd>
623
624
625  <dt><tt><b><a name="linkage_appending">appending</a></b></tt></dt>
626  <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
627      pointer to array type.  When two global variables with appending linkage
628      are linked together, the two global arrays are appended together.  This is
629      the LLVM, typesafe, equivalent of having the system linker append together
630      "sections" with identical names when .o files are linked.</dd>
631
632  <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt></dt>
633  <dd>The semantics of this linkage follow the ELF object file model: the symbol
634      is weak until linked, if not linked, the symbol becomes null instead of
635      being an undefined reference.</dd>
636
637  <dt><tt><b><a name="linkage_linkonce_odr">linkonce_odr</a></b></tt></dt>
638  <dt><tt><b><a name="linkage_weak_odr">weak_odr</a></b></tt></dt>
639  <dd>Some languages allow differing globals to be merged, such as two functions
640      with different semantics.  Other languages, such as <tt>C++</tt>, ensure
641      that only equivalent globals are ever merged (the "one definition rule"
642      &mdash; "ODR").  Such languages can use the <tt>linkonce_odr</tt>
643      and <tt>weak_odr</tt> linkage types to indicate that the global will only
644      be merged with equivalent globals.  These linkage types are otherwise the
645      same as their non-<tt>odr</tt> versions.</dd>
646
647  <dt><tt><b><a name="linkage_linkonce_odr_auto_hide">linkonce_odr_auto_hide</a></b></tt></dt>
648  <dd>Similar to "<tt>linkonce_odr</tt>", but nothing in the translation unit
649      takes the address of this definition. For instance, functions that had an
650      inline definition, but the compiler decided not to inline it.
651      <tt>linkonce_odr_auto_hide</tt> may have only <tt>default</tt> visibility.
652      The symbols are removed by the linker from the final linked image
653      (executable or dynamic library).</dd>
654
655  <dt><tt><b><a name="linkage_external">external</a></b></tt></dt>
656  <dd>If none of the above identifiers are used, the global is externally
657      visible, meaning that it participates in linkage and can be used to
658      resolve external symbol references.</dd>
659</dl>
660
661<p>The next two types of linkage are targeted for Microsoft Windows platform
662   only. They are designed to support importing (exporting) symbols from (to)
663   DLLs (Dynamic Link Libraries).</p>
664
665<dl>
666  <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt></dt>
667  <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
668      or variable via a global pointer to a pointer that is set up by the DLL
669      exporting the symbol. On Microsoft Windows targets, the pointer name is
670      formed by combining <code>__imp_</code> and the function or variable
671      name.</dd>
672
673  <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt></dt>
674  <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
675      pointer to a pointer in a DLL, so that it can be referenced with the
676      <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
677      name is formed by combining <code>__imp_</code> and the function or
678      variable name.</dd>
679</dl>
680
681<p>For example, since the "<tt>.LC0</tt>" variable is defined to be internal, if
682   another module defined a "<tt>.LC0</tt>" variable and was linked with this
683   one, one of the two would be renamed, preventing a collision.  Since
684   "<tt>main</tt>" and "<tt>puts</tt>" are external (i.e., lacking any linkage
685   declarations), they are accessible outside of the current module.</p>
686
687<p>It is illegal for a function <i>declaration</i> to have any linkage type
688   other than <tt>external</tt>, <tt>dllimport</tt>
689  or <tt>extern_weak</tt>.</p>
690
691<p>Aliases can have only <tt>external</tt>, <tt>internal</tt>, <tt>weak</tt>
692   or <tt>weak_odr</tt> linkages.</p>
693
694</div>
695
696<!-- ======================================================================= -->
697<h3>
698  <a name="callingconv">Calling Conventions</a>
699</h3>
700
701<div>
702
703<p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
704   and <a href="#i_invoke">invokes</a> can all have an optional calling
705   convention specified for the call.  The calling convention of any pair of
706   dynamic caller/callee must match, or the behavior of the program is
707   undefined.  The following calling conventions are supported by LLVM, and more
708   may be added in the future:</p>
709
710<dl>
711  <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
712  <dd>This calling convention (the default if no other calling convention is
713      specified) matches the target C calling conventions.  This calling
714      convention supports varargs function calls and tolerates some mismatch in
715      the declared prototype and implemented declaration of the function (as
716      does normal C).</dd>
717
718  <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
719  <dd>This calling convention attempts to make calls as fast as possible
720      (e.g. by passing things in registers).  This calling convention allows the
721      target to use whatever tricks it wants to produce fast code for the
722      target, without having to conform to an externally specified ABI
723      (Application Binary Interface).
724      <a href="CodeGenerator.html#tailcallopt">Tail calls can only be optimized
725      when this or the GHC convention is used.</a>  This calling convention
726      does not support varargs and requires the prototype of all callees to
727      exactly match the prototype of the function definition.</dd>
728
729  <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
730  <dd>This calling convention attempts to make code in the caller as efficient
731      as possible under the assumption that the call is not commonly executed.
732      As such, these calls often preserve all registers so that the call does
733      not break any live ranges in the caller side.  This calling convention
734      does not support varargs and requires the prototype of all callees to
735      exactly match the prototype of the function definition.</dd>
736
737  <dt><b>"<tt>cc <em>10</em></tt>" - GHC convention</b>:</dt>
738  <dd>This calling convention has been implemented specifically for use by the
739      <a href="http://www.haskell.org/ghc">Glasgow Haskell Compiler (GHC)</a>.
740      It passes everything in registers, going to extremes to achieve this by
741      disabling callee save registers. This calling convention should not be
742      used lightly but only for specific situations such as an alternative to
743      the <em>register pinning</em> performance technique often used when
744      implementing functional programming languages.At the moment only X86
745      supports this convention and it has the following limitations:
746      <ul>
747        <li>On <em>X86-32</em> only supports up to 4 bit type parameters. No
748            floating point types are supported.</li>
749        <li>On <em>X86-64</em> only supports up to 10 bit type parameters and
750            6 floating point parameters.</li>
751      </ul>
752      This calling convention supports
753      <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> but
754      requires both the caller and callee are using it.
755  </dd>
756
757  <dt><b>"<tt>cc &lt;<em>n</em>&gt;</tt>" - Numbered convention</b>:</dt>
758  <dd>Any calling convention may be specified by number, allowing
759      target-specific calling conventions to be used.  Target specific calling
760      conventions start at 64.</dd>
761</dl>
762
763<p>More calling conventions can be added/defined on an as-needed basis, to
764   support Pascal conventions or any other well-known target-independent
765   convention.</p>
766
767</div>
768
769<!-- ======================================================================= -->
770<h3>
771  <a name="visibility">Visibility Styles</a>
772</h3>
773
774<div>
775
776<p>All Global Variables and Functions have one of the following visibility
777   styles:</p>
778
779<dl>
780  <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
781  <dd>On targets that use the ELF object file format, default visibility means
782      that the declaration is visible to other modules and, in shared libraries,
783      means that the declared entity may be overridden. On Darwin, default
784      visibility means that the declaration is visible to other modules. Default
785      visibility corresponds to "external linkage" in the language.</dd>
786
787  <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
788  <dd>Two declarations of an object with hidden visibility refer to the same
789      object if they are in the same shared object. Usually, hidden visibility
790      indicates that the symbol will not be placed into the dynamic symbol
791      table, so no other module (executable or shared library) can reference it
792      directly.</dd>
793
794  <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
795  <dd>On ELF, protected visibility indicates that the symbol will be placed in
796      the dynamic symbol table, but that references within the defining module
797      will bind to the local symbol. That is, the symbol cannot be overridden by
798      another module.</dd>
799</dl>
800
801</div>
802
803<!-- ======================================================================= -->
804<h3>
805  <a name="namedtypes">Named Types</a>
806</h3>
807
808<div>
809
810<p>LLVM IR allows you to specify name aliases for certain types.  This can make
811   it easier to read the IR and make the IR more condensed (particularly when
812   recursive types are involved).  An example of a name specification is:</p>
813
814<pre class="doc_code">
815%mytype = type { %mytype*, i32 }
816</pre>
817
818<p>You may give a name to any <a href="#typesystem">type</a> except
819   "<a href="#t_void">void</a>".  Type name aliases may be used anywhere a type
820   is expected with the syntax "%mytype".</p>
821
822<p>Note that type names are aliases for the structural type that they indicate,
823   and that you can therefore specify multiple names for the same type.  This
824   often leads to confusing behavior when dumping out a .ll file.  Since LLVM IR
825   uses structural typing, the name is not part of the type.  When printing out
826   LLVM IR, the printer will pick <em>one name</em> to render all types of a
827   particular shape.  This means that if you have code where two different
828   source types end up having the same LLVM type, that the dumper will sometimes
829   print the "wrong" or unexpected type.  This is an important design point and
830   isn't going to change.</p>
831
832</div>
833
834<!-- ======================================================================= -->
835<h3>
836  <a name="globalvars">Global Variables</a>
837</h3>
838
839<div>
840
841<p>Global variables define regions of memory allocated at compilation time
842   instead of run-time.  Global variables may optionally be initialized, may
843   have an explicit section to be placed in, and may have an optional explicit
844   alignment specified.</p>
845
846<p>A variable may be defined as <tt>thread_local</tt>, which
847   means that it will not be shared by threads (each thread will have a
848   separated copy of the variable).  Not all targets support thread-local
849   variables.  Optionally, a TLS model may be specified:</p>
850
851<dl>
852  <dt><b><tt>localdynamic</tt></b>:</dt>
853  <dd>For variables that are only used within the current shared library.</dd>
854
855  <dt><b><tt>initialexec</tt></b>:</dt>
856  <dd>For variables in modules that will not be loaded dynamically.</dd>
857
858  <dt><b><tt>localexec</tt></b>:</dt>
859  <dd>For variables defined in the executable and only used within it.</dd>
860</dl>
861
862<p>The models correspond to the ELF TLS models; see
863   <a href="http://people.redhat.com/drepper/tls.pdf">ELF
864   Handling For Thread-Local Storage</a> for more information on under which
865   circumstances the different models may be used.  The target may choose a
866   different TLS model if the specified model is not supported, or if a better
867   choice of model can be made.</p>
868
869<p>A variable may be defined as a global
870   "constant," which indicates that the contents of the variable
871   will <b>never</b> be modified (enabling better optimization, allowing the
872   global data to be placed in the read-only section of an executable, etc).
873   Note that variables that need runtime initialization cannot be marked
874   "constant" as there is a store to the variable.</p>
875
876<p>LLVM explicitly allows <em>declarations</em> of global variables to be marked
877   constant, even if the final definition of the global is not.  This capability
878   can be used to enable slightly better optimization of the program, but
879   requires the language definition to guarantee that optimizations based on the
880   'constantness' are valid for the translation units that do not include the
881   definition.</p>
882
883<p>As SSA values, global variables define pointer values that are in scope
884   (i.e. they dominate) all basic blocks in the program.  Global variables
885   always define a pointer to their "content" type because they describe a
886   region of memory, and all memory objects in LLVM are accessed through
887   pointers.</p>
888
889<p>Global variables can be marked with <tt>unnamed_addr</tt> which indicates
890  that the address is not significant, only the content. Constants marked
891  like this can be merged with other constants if they have the same
892  initializer. Note that a constant with significant address <em>can</em>
893  be merged with a <tt>unnamed_addr</tt> constant, the result being a
894  constant whose address is significant.</p>
895
896<p>A global variable may be declared to reside in a target-specific numbered
897   address space. For targets that support them, address spaces may affect how
898   optimizations are performed and/or what target instructions are used to
899   access the variable. The default address space is zero. The address space
900   qualifier must precede any other attributes.</p>
901
902<p>LLVM allows an explicit section to be specified for globals.  If the target
903   supports it, it will emit globals to the section specified.</p>
904
905<p>An explicit alignment may be specified for a global, which must be a power
906   of 2.  If not present, or if the alignment is set to zero, the alignment of
907   the global is set by the target to whatever it feels convenient.  If an
908   explicit alignment is specified, the global is forced to have exactly that
909   alignment.  Targets and optimizers are not allowed to over-align the global
910   if the global has an assigned section.  In this case, the extra alignment
911   could be observable: for example, code could assume that the globals are
912   densely packed in their section and try to iterate over them as an array,
913   alignment padding would break this iteration.</p>
914
915<p>For example, the following defines a global in a numbered address space with
916   an initializer, section, and alignment:</p>
917
918<pre class="doc_code">
919@G = addrspace(5) constant float 1.0, section "foo", align 4
920</pre>
921
922<p>The following example defines a thread-local global with
923   the <tt>initialexec</tt> TLS model:</p>
924
925<pre class="doc_code">
926@G = thread_local(initialexec) global i32 0, align 4
927</pre>
928
929</div>
930
931
932<!-- ======================================================================= -->
933<h3>
934  <a name="functionstructure">Functions</a>
935</h3>
936
937<div>
938
939<p>LLVM function definitions consist of the "<tt>define</tt>" keyword, an
940   optional <a href="#linkage">linkage type</a>, an optional
941   <a href="#visibility">visibility style</a>, an optional
942   <a href="#callingconv">calling convention</a>,
943   an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
944   <a href="#paramattrs">parameter attribute</a> for the return type, a function
945   name, a (possibly empty) argument list (each with optional
946   <a href="#paramattrs">parameter attributes</a>), optional
947   <a href="#fnattrs">function attributes</a>, an optional section, an optional
948   alignment, an optional <a href="#gc">garbage collector name</a>, an opening
949   curly brace, a list of basic blocks, and a closing curly brace.</p>
950
951<p>LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
952   optional <a href="#linkage">linkage type</a>, an optional
953   <a href="#visibility">visibility style</a>, an optional
954   <a href="#callingconv">calling convention</a>,
955   an optional <tt>unnamed_addr</tt> attribute, a return type, an optional
956   <a href="#paramattrs">parameter attribute</a> for the return type, a function
957   name, a possibly empty list of arguments, an optional alignment, and an
958   optional <a href="#gc">garbage collector name</a>.</p>
959
960<p>A function definition contains a list of basic blocks, forming the CFG
961   (Control Flow Graph) for the function.  Each basic block may optionally start
962   with a label (giving the basic block a symbol table entry), contains a list
963   of instructions, and ends with a <a href="#terminators">terminator</a>
964   instruction (such as a branch or function return).</p>
965
966<p>The first basic block in a function is special in two ways: it is immediately
967   executed on entrance to the function, and it is not allowed to have
968   predecessor basic blocks (i.e. there can not be any branches to the entry
969   block of a function).  Because the block can have no predecessors, it also
970   cannot have any <a href="#i_phi">PHI nodes</a>.</p>
971
972<p>LLVM allows an explicit section to be specified for functions.  If the target
973   supports it, it will emit functions to the section specified.</p>
974
975<p>An explicit alignment may be specified for a function.  If not present, or if
976   the alignment is set to zero, the alignment of the function is set by the
977   target to whatever it feels convenient.  If an explicit alignment is
978   specified, the function is forced to have at least that much alignment.  All
979   alignments must be a power of 2.</p>
980
981<p>If the <tt>unnamed_addr</tt> attribute is given, the address is know to not
982   be significant and two identical functions can be merged.</p>
983
984<h5>Syntax:</h5>
985<pre class="doc_code">
986define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
987       [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
988       &lt;ResultType&gt; @&lt;FunctionName&gt; ([argument list])
989       [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
990       [<a href="#gc">gc</a>] { ... }
991</pre>
992
993</div>
994
995<!-- ======================================================================= -->
996<h3>
997  <a name="aliasstructure">Aliases</a>
998</h3>
999
1000<div>
1001
1002<p>Aliases act as "second name" for the aliasee value (which can be either
1003   function, global variable, another alias or bitcast of global value). Aliases
1004   may have an optional <a href="#linkage">linkage type</a>, and an
1005   optional <a href="#visibility">visibility style</a>.</p>
1006
1007<h5>Syntax:</h5>
1008<pre class="doc_code">
1009@&lt;Name&gt; = alias [Linkage] [Visibility] &lt;AliaseeTy&gt; @&lt;Aliasee&gt;
1010</pre>
1011
1012</div>
1013
1014<!-- ======================================================================= -->
1015<h3>
1016  <a name="namedmetadatastructure">Named Metadata</a>
1017</h3>
1018
1019<div>
1020
1021<p>Named metadata is a collection of metadata. <a href="#metadata">Metadata
1022   nodes</a> (but not metadata strings) are the only valid operands for
1023   a named metadata.</p>
1024
1025<h5>Syntax:</h5>
1026<pre class="doc_code">
1027; Some unnamed metadata nodes, which are referenced by the named metadata.
1028!0 = metadata !{metadata !"zero"}
1029!1 = metadata !{metadata !"one"}
1030!2 = metadata !{metadata !"two"}
1031; A named metadata.
1032!name = !{!0, !1, !2}
1033</pre>
1034
1035</div>
1036
1037<!-- ======================================================================= -->
1038<h3>
1039  <a name="paramattrs">Parameter Attributes</a>
1040</h3>
1041
1042<div>
1043
1044<p>The return type and each parameter of a function type may have a set of
1045   <i>parameter attributes</i> associated with them. Parameter attributes are
1046   used to communicate additional information about the result or parameters of
1047   a function. Parameter attributes are considered to be part of the function,
1048   not of the function type, so functions with different parameter attributes
1049   can have the same function type.</p>
1050
1051<p>Parameter attributes are simple keywords that follow the type specified. If
1052   multiple parameter attributes are needed, they are space separated. For
1053   example:</p>
1054
1055<pre class="doc_code">
1056declare i32 @printf(i8* noalias nocapture, ...)
1057declare i32 @atoi(i8 zeroext)
1058declare signext i8 @returns_signed_char()
1059</pre>
1060
1061<p>Note that any attributes for the function result (<tt>nounwind</tt>,
1062   <tt>readonly</tt>) come immediately after the argument list.</p>
1063
1064<p>Currently, only the following parameter attributes are defined:</p>
1065
1066<dl>
1067  <dt><tt><b>zeroext</b></tt></dt>
1068  <dd>This indicates to the code generator that the parameter or return value
1069      should be zero-extended to the extent required by the target's ABI (which
1070      is usually 32-bits, but is 8-bits for a i1 on x86-64) by the caller (for a
1071      parameter) or the callee (for a return value).</dd>
1072
1073  <dt><tt><b>signext</b></tt></dt>
1074  <dd>This indicates to the code generator that the parameter or return value
1075      should be sign-extended to the extent required by the target's ABI (which
1076      is usually 32-bits) by the caller (for a parameter) or the callee (for a
1077      return value).</dd>
1078
1079  <dt><tt><b>inreg</b></tt></dt>
1080  <dd>This indicates that this parameter or return value should be treated in a
1081      special target-dependent fashion during while emitting code for a function
1082      call or return (usually, by putting it in a register as opposed to memory,
1083      though some targets use it to distinguish between two different kinds of
1084      registers).  Use of this attribute is target-specific.</dd>
1085
1086  <dt><tt><b><a name="byval">byval</a></b></tt></dt>
1087  <dd><p>This indicates that the pointer parameter should really be passed by
1088      value to the function.  The attribute implies that a hidden copy of the
1089      pointee
1090      is made between the caller and the callee, so the callee is unable to
1091      modify the value in the caller.  This attribute is only valid on LLVM
1092      pointer arguments.  It is generally used to pass structs and arrays by
1093      value, but is also valid on pointers to scalars.  The copy is considered
1094      to belong to the caller not the callee (for example,
1095      <tt><a href="#readonly">readonly</a></tt> functions should not write to
1096      <tt>byval</tt> parameters). This is not a valid attribute for return
1097      values.</p>
1098
1099      <p>The byval attribute also supports specifying an alignment with
1100      the align attribute.  It indicates the alignment of the stack slot to
1101      form and the known alignment of the pointer specified to the call site. If
1102      the alignment is not specified, then the code generator makes a
1103      target-specific assumption.</p></dd>
1104
1105  <dt><tt><b><a name="sret">sret</a></b></tt></dt>
1106  <dd>This indicates that the pointer parameter specifies the address of a
1107      structure that is the return value of the function in the source program.
1108      This pointer must be guaranteed by the caller to be valid: loads and
1109      stores to the structure may be assumed by the callee to not to trap.  This
1110      may only be applied to the first parameter. This is not a valid attribute
1111      for return values. </dd>
1112
1113  <dt><tt><b><a name="noalias">noalias</a></b></tt></dt>
1114  <dd>This indicates that pointer values
1115      <a href="#pointeraliasing"><i>based</i></a> on the argument or return
1116      value do not alias pointer values which are not <i>based</i> on it,
1117      ignoring certain "irrelevant" dependencies.
1118      For a call to the parent function, dependencies between memory
1119      references from before or after the call and from those during the call
1120      are "irrelevant" to the <tt>noalias</tt> keyword for the arguments and
1121      return value used in that call.
1122      The caller shares the responsibility with the callee for ensuring that
1123      these requirements are met.
1124      For further details, please see the discussion of the NoAlias response in
1125      <a href="AliasAnalysis.html#MustMayNo">alias analysis</a>.<br>
1126<br>
1127      Note that this definition of <tt>noalias</tt> is intentionally
1128      similar to the definition of <tt>restrict</tt> in C99 for function
1129      arguments, though it is slightly weaker.
1130<br>
1131      For function return values, C99's <tt>restrict</tt> is not meaningful,
1132      while LLVM's <tt>noalias</tt> is.
1133      </dd>
1134
1135  <dt><tt><b><a name="nocapture">nocapture</a></b></tt></dt>
1136  <dd>This indicates that the callee does not make any copies of the pointer
1137      that outlive the callee itself. This is not a valid attribute for return
1138      values.</dd>
1139
1140  <dt><tt><b><a name="nest">nest</a></b></tt></dt>
1141  <dd>This indicates that the pointer parameter can be excised using the
1142      <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
1143      attribute for return values.</dd>
1144</dl>
1145
1146</div>
1147
1148<!-- ======================================================================= -->
1149<h3>
1150  <a name="gc">Garbage Collector Names</a>
1151</h3>
1152
1153<div>
1154
1155<p>Each function may specify a garbage collector name, which is simply a
1156   string:</p>
1157
1158<pre class="doc_code">
1159define void @f() gc "name" { ... }
1160</pre>
1161
1162<p>The compiler declares the supported values of <i>name</i>. Specifying a
1163   collector which will cause the compiler to alter its output in order to
1164   support the named garbage collection algorithm.</p>
1165
1166</div>
1167
1168<!-- ======================================================================= -->
1169<h3>
1170  <a name="fnattrs">Function Attributes</a>
1171</h3>
1172
1173<div>
1174
1175<p>Function attributes are set to communicate additional information about a
1176   function. Function attributes are considered to be part of the function, not
1177   of the function type, so functions with different parameter attributes can
1178   have the same function type.</p>
1179
1180<p>Function attributes are simple keywords that follow the type specified. If
1181   multiple attributes are needed, they are space separated. For example:</p>
1182
1183<pre class="doc_code">
1184define void @f() noinline { ... }
1185define void @f() alwaysinline { ... }
1186define void @f() alwaysinline optsize { ... }
1187define void @f() optsize { ... }
1188</pre>
1189
1190<dl>
1191  <dt><tt><b>address_safety</b></tt></dt>
1192  <dd>This attribute indicates that the address safety analysis
1193  is enabled for this function.  </dd>
1194
1195  <dt><tt><b>alignstack(&lt;<em>n</em>&gt;)</b></tt></dt>
1196  <dd>This attribute indicates that, when emitting the prologue and epilogue,
1197      the backend should forcibly align the stack pointer. Specify the
1198      desired alignment, which must be a power of two, in parentheses.
1199
1200  <dt><tt><b>alwaysinline</b></tt></dt>
1201  <dd>This attribute indicates that the inliner should attempt to inline this
1202      function into callers whenever possible, ignoring any active inlining size
1203      threshold for this caller.</dd>
1204
1205  <dt><tt><b>nonlazybind</b></tt></dt>
1206  <dd>This attribute suppresses lazy symbol binding for the function. This
1207      may make calls to the function faster, at the cost of extra program
1208      startup time if the function is not called during program startup.</dd>
1209
1210  <dt><tt><b>inlinehint</b></tt></dt>
1211  <dd>This attribute indicates that the source code contained a hint that inlining
1212      this function is desirable (such as the "inline" keyword in C/C++).  It
1213      is just a hint; it imposes no requirements on the inliner.</dd>
1214
1215  <dt><tt><b>naked</b></tt></dt>
1216  <dd>This attribute disables prologue / epilogue emission for the function.
1217      This can have very system-specific consequences.</dd>
1218
1219  <dt><tt><b>noimplicitfloat</b></tt></dt>
1220  <dd>This attributes disables implicit floating point instructions.</dd>
1221
1222  <dt><tt><b>noinline</b></tt></dt>
1223  <dd>This attribute indicates that the inliner should never inline this
1224      function in any situation. This attribute may not be used together with
1225      the <tt>alwaysinline</tt> attribute.</dd>
1226
1227  <dt><tt><b>noredzone</b></tt></dt>
1228  <dd>This attribute indicates that the code generator should not use a red
1229      zone, even if the target-specific ABI normally permits it.</dd>
1230
1231  <dt><tt><b>noreturn</b></tt></dt>
1232  <dd>This function attribute indicates that the function never returns
1233      normally.  This produces undefined behavior at runtime if the function
1234      ever does dynamically return.</dd>
1235
1236  <dt><tt><b>nounwind</b></tt></dt>
1237  <dd>This function attribute indicates that the function never returns with an
1238      unwind or exceptional control flow.  If the function does unwind, its
1239      runtime behavior is undefined.</dd>
1240
1241  <dt><tt><b>optsize</b></tt></dt>
1242  <dd>This attribute suggests that optimization passes and code generator passes
1243      make choices that keep the code size of this function low, and otherwise
1244      do optimizations specifically to reduce code size.</dd>
1245
1246  <dt><tt><b>readnone</b></tt></dt>
1247  <dd>This attribute indicates that the function computes its result (or decides
1248      to unwind an exception) based strictly on its arguments, without
1249      dereferencing any pointer arguments or otherwise accessing any mutable
1250      state (e.g. memory, control registers, etc) visible to caller functions.
1251      It does not write through any pointer arguments
1252      (including <tt><a href="#byval">byval</a></tt> arguments) and never
1253      changes any state visible to callers.  This means that it cannot unwind
1254      exceptions by calling the <tt>C++</tt> exception throwing methods.</dd>
1255
1256  <dt><tt><b><a name="readonly">readonly</a></b></tt></dt>
1257  <dd>This attribute indicates that the function does not write through any
1258      pointer arguments (including <tt><a href="#byval">byval</a></tt>
1259      arguments) or otherwise modify any state (e.g. memory, control registers,
1260      etc) visible to caller functions.  It may dereference pointer arguments
1261      and read state that may be set in the caller.  A readonly function always
1262      returns the same value (or unwinds an exception identically) when called
1263      with the same set of arguments and global state.  It cannot unwind an
1264      exception by calling the <tt>C++</tt> exception throwing methods.</dd>
1265
1266  <dt><tt><b><a name="returns_twice">returns_twice</a></b></tt></dt>
1267  <dd>This attribute indicates that this function can return twice. The
1268      C <code>setjmp</code> is an example of such a function.  The compiler
1269      disables some optimizations (like tail calls) in the caller of these
1270      functions.</dd>
1271
1272  <dt><tt><b><a name="ssp">ssp</a></b></tt></dt>
1273  <dd>This attribute indicates that the function should emit a stack smashing
1274      protector. It is in the form of a "canary"&mdash;a random value placed on
1275      the stack before the local variables that's checked upon return from the
1276      function to see if it has been overwritten. A heuristic is used to
1277      determine if a function needs stack protectors or not.<br>
1278<br>
1279      If a function that has an <tt>ssp</tt> attribute is inlined into a
1280      function that doesn't have an <tt>ssp</tt> attribute, then the resulting
1281      function will have an <tt>ssp</tt> attribute.</dd>
1282
1283  <dt><tt><b>sspreq</b></tt></dt>
1284  <dd>This attribute indicates that the function should <em>always</em> emit a
1285      stack smashing protector. This overrides
1286      the <tt><a href="#ssp">ssp</a></tt> function attribute.<br>
1287<br>
1288      If a function that has an <tt>sspreq</tt> attribute is inlined into a
1289      function that doesn't have an <tt>sspreq</tt> attribute or which has
1290      an <tt>ssp</tt> attribute, then the resulting function will have
1291      an <tt>sspreq</tt> attribute.</dd>
1292
1293  <dt><tt><b><a name="uwtable">uwtable</a></b></tt></dt>
1294  <dd>This attribute indicates that the ABI being targeted requires that
1295      an unwind table entry be produce for this function even if we can
1296      show that no exceptions passes by it. This is normally the case for
1297      the ELF x86-64 abi, but it can be disabled for some compilation
1298      units.</dd>
1299</dl>
1300
1301</div>
1302
1303<!-- ======================================================================= -->
1304<h3>
1305  <a name="moduleasm">Module-Level Inline Assembly</a>
1306</h3>
1307
1308<div>
1309
1310<p>Modules may contain "module-level inline asm" blocks, which corresponds to
1311   the GCC "file scope inline asm" blocks.  These blocks are internally
1312   concatenated by LLVM and treated as a single unit, but may be separated in
1313   the <tt>.ll</tt> file if desired.  The syntax is very simple:</p>
1314
1315<pre class="doc_code">
1316module asm "inline asm code goes here"
1317module asm "more can go here"
1318</pre>
1319
1320<p>The strings can contain any character by escaping non-printable characters.
1321   The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1322   for the number.</p>
1323
1324<p>The inline asm code is simply printed to the machine code .s file when
1325   assembly code is generated.</p>
1326
1327</div>
1328
1329<!-- ======================================================================= -->
1330<h3>
1331  <a name="datalayout">Data Layout</a>
1332</h3>
1333
1334<div>
1335
1336<p>A module may specify a target specific data layout string that specifies how
1337   data is to be laid out in memory. The syntax for the data layout is
1338   simply:</p>
1339
1340<pre class="doc_code">
1341target datalayout = "<i>layout specification</i>"
1342</pre>
1343
1344<p>The <i>layout specification</i> consists of a list of specifications
1345   separated by the minus sign character ('-').  Each specification starts with
1346   a letter and may include other information after the letter to define some
1347   aspect of the data layout.  The specifications accepted are as follows:</p>
1348
1349<dl>
1350  <dt><tt>E</tt></dt>
1351  <dd>Specifies that the target lays out data in big-endian form. That is, the
1352      bits with the most significance have the lowest address location.</dd>
1353
1354  <dt><tt>e</tt></dt>
1355  <dd>Specifies that the target lays out data in little-endian form. That is,
1356      the bits with the least significance have the lowest address
1357      location.</dd>
1358
1359  <dt><tt>S<i>size</i></tt></dt>
1360  <dd>Specifies the natural alignment of the stack in bits. Alignment promotion
1361      of stack variables is limited to the natural stack alignment to avoid
1362      dynamic stack realignment. The stack alignment must be a multiple of
1363      8-bits. If omitted, the natural stack alignment defaults to "unspecified",
1364      which does not prevent any alignment promotions.</dd>
1365
1366  <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1367  <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1368      <i>preferred</i> alignments. All sizes are in bits. Specifying
1369      the <i>pref</i> alignment is optional. If omitted, the
1370      preceding <tt>:</tt> should be omitted too.</dd>
1371
1372  <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1373  <dd>This specifies the alignment for an integer type of a given bit
1374      <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1375
1376  <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1377  <dd>This specifies the alignment for a vector type of a given bit
1378      <i>size</i>.</dd>
1379
1380  <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1381  <dd>This specifies the alignment for a floating point type of a given bit
1382      <i>size</i>. Only values of <i>size</i> that are supported by the target
1383      will work.  32 (float) and 64 (double) are supported on all targets;
1384      80 or 128 (different flavors of long double) are also supported on some
1385      targets.
1386
1387  <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1388  <dd>This specifies the alignment for an aggregate type of a given bit
1389      <i>size</i>.</dd>
1390
1391  <dt><tt>s<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1392  <dd>This specifies the alignment for a stack object of a given bit
1393      <i>size</i>.</dd>
1394
1395  <dt><tt>n<i>size1</i>:<i>size2</i>:<i>size3</i>...</tt></dt>
1396  <dd>This specifies a set of native integer widths for the target CPU
1397      in bits.  For example, it might contain "n32" for 32-bit PowerPC,
1398      "n32:64" for PowerPC 64, or "n8:16:32:64" for X86-64.  Elements of
1399      this set are considered to support most general arithmetic
1400      operations efficiently.</dd>
1401</dl>
1402
1403<p>When constructing the data layout for a given target, LLVM starts with a
1404   default set of specifications which are then (possibly) overridden by the
1405   specifications in the <tt>datalayout</tt> keyword. The default specifications
1406   are given in this list:</p>
1407
1408<ul>
1409  <li><tt>E</tt> - big endian</li>
1410  <li><tt>p:64:64:64</tt> - 64-bit pointers with 64-bit alignment</li>
1411  <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1412  <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1413  <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1414  <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1415  <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1416  alignment of 64-bits</li>
1417  <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1418  <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1419  <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1420  <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1421  <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1422  <li><tt>s0:64:64</tt> - stack objects are 64-bit aligned</li>
1423</ul>
1424
1425<p>When LLVM is determining the alignment for a given type, it uses the
1426   following rules:</p>
1427
1428<ol>
1429  <li>If the type sought is an exact match for one of the specifications, that
1430      specification is used.</li>
1431
1432  <li>If no match is found, and the type sought is an integer type, then the
1433      smallest integer type that is larger than the bitwidth of the sought type
1434      is used. If none of the specifications are larger than the bitwidth then
1435      the largest integer type is used. For example, given the default
1436      specifications above, the i7 type will use the alignment of i8 (next
1437      largest) while both i65 and i256 will use the alignment of i64 (largest
1438      specified).</li>
1439
1440  <li>If no match is found, and the type sought is a vector type, then the
1441      largest vector type that is smaller than the sought vector type will be
1442      used as a fall back.  This happens because &lt;128 x double&gt; can be
1443      implemented in terms of 64 &lt;2 x double&gt;, for example.</li>
1444</ol>
1445
1446<p>The function of the data layout string may not be what you expect.  Notably,
1447   this is not a specification from the frontend of what alignment the code
1448   generator should use.</p>
1449
1450<p>Instead, if specified, the target data layout is required to match what the
1451   ultimate <em>code generator</em> expects.  This string is used by the
1452   mid-level optimizers to
1453   improve code, and this only works if it matches what the ultimate code
1454   generator uses.  If you would like to generate IR that does not embed this
1455   target-specific detail into the IR, then you don't have to specify the
1456   string.  This will disable some optimizations that require precise layout
1457   information, but this also prevents those optimizations from introducing
1458   target specificity into the IR.</p>
1459
1460
1461
1462</div>
1463
1464<!-- ======================================================================= -->
1465<h3>
1466  <a name="pointeraliasing">Pointer Aliasing Rules</a>
1467</h3>
1468
1469<div>
1470
1471<p>Any memory access must be done through a pointer value associated
1472with an address range of the memory access, otherwise the behavior
1473is undefined. Pointer values are associated with address ranges
1474according to the following rules:</p>
1475
1476<ul>
1477  <li>A pointer value is associated with the addresses associated with
1478      any value it is <i>based</i> on.
1479  <li>An address of a global variable is associated with the address
1480      range of the variable's storage.</li>
1481  <li>The result value of an allocation instruction is associated with
1482      the address range of the allocated storage.</li>
1483  <li>A null pointer in the default address-space is associated with
1484      no address.</li>
1485  <li>An integer constant other than zero or a pointer value returned
1486      from a function not defined within LLVM may be associated with address
1487      ranges allocated through mechanisms other than those provided by
1488      LLVM. Such ranges shall not overlap with any ranges of addresses
1489      allocated by mechanisms provided by LLVM.</li>
1490</ul>
1491
1492<p>A pointer value is <i>based</i> on another pointer value according
1493   to the following rules:</p>
1494
1495<ul>
1496  <li>A pointer value formed from a
1497      <tt><a href="#i_getelementptr">getelementptr</a></tt> operation
1498      is <i>based</i> on the first operand of the <tt>getelementptr</tt>.</li>
1499  <li>The result value of a
1500      <tt><a href="#i_bitcast">bitcast</a></tt> is <i>based</i> on the operand
1501      of the <tt>bitcast</tt>.</li>
1502  <li>A pointer value formed by an
1503      <tt><a href="#i_inttoptr">inttoptr</a></tt> is <i>based</i> on all
1504      pointer values that contribute (directly or indirectly) to the
1505      computation of the pointer's value.</li>
1506  <li>The "<i>based</i> on" relationship is transitive.</li>
1507</ul>
1508
1509<p>Note that this definition of <i>"based"</i> is intentionally
1510   similar to the definition of <i>"based"</i> in C99, though it is
1511   slightly weaker.</p>
1512
1513<p>LLVM IR does not associate types with memory. The result type of a
1514<tt><a href="#i_load">load</a></tt> merely indicates the size and
1515alignment of the memory from which to load, as well as the
1516interpretation of the value. The first operand type of a
1517<tt><a href="#i_store">store</a></tt> similarly only indicates the size
1518and alignment of the store.</p>
1519
1520<p>Consequently, type-based alias analysis, aka TBAA, aka
1521<tt>-fstrict-aliasing</tt>, is not applicable to general unadorned
1522LLVM IR. <a href="#metadata">Metadata</a> may be used to encode
1523additional information which specialized optimization passes may use
1524to implement type-based alias analysis.</p>
1525
1526</div>
1527
1528<!-- ======================================================================= -->
1529<h3>
1530  <a name="volatile">Volatile Memory Accesses</a>
1531</h3>
1532
1533<div>
1534
1535<p>Certain memory accesses, such as <a href="#i_load"><tt>load</tt></a>s, <a
1536href="#i_store"><tt>store</tt></a>s, and <a
1537href="#int_memcpy"><tt>llvm.memcpy</tt></a>s may be marked <tt>volatile</tt>.
1538The optimizers must not change the number of volatile operations or change their
1539order of execution relative to other volatile operations.  The optimizers
1540<i>may</i> change the order of volatile operations relative to non-volatile
1541operations.  This is not Java's "volatile" and has no cross-thread
1542synchronization behavior.</p>
1543
1544</div>
1545
1546<!-- ======================================================================= -->
1547<h3>
1548  <a name="memmodel">Memory Model for Concurrent Operations</a>
1549</h3>
1550
1551<div>
1552
1553<p>The LLVM IR does not define any way to start parallel threads of execution
1554or to register signal handlers. Nonetheless, there are platform-specific
1555ways to create them, and we define LLVM IR's behavior in their presence. This
1556model is inspired by the C++0x memory model.</p>
1557
1558<p>For a more informal introduction to this model, see the
1559<a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>.
1560
1561<p>We define a <i>happens-before</i> partial order as the least partial order
1562that</p>
1563<ul>
1564  <li>Is a superset of single-thread program order, and</li>
1565  <li>When a <i>synchronizes-with</i> <tt>b</tt>, includes an edge from
1566      <tt>a</tt> to <tt>b</tt>. <i>Synchronizes-with</i> pairs are introduced
1567      by platform-specific techniques, like pthread locks, thread
1568      creation, thread joining, etc., and by atomic instructions.
1569      (See also <a href="#ordering">Atomic Memory Ordering Constraints</a>).
1570      </li>
1571</ul>
1572
1573<p>Note that program order does not introduce <i>happens-before</i> edges
1574between a thread and signals executing inside that thread.</p>
1575
1576<p>Every (defined) read operation (load instructions, memcpy, atomic
1577loads/read-modify-writes, etc.) <var>R</var> reads a series of bytes written by
1578(defined) write operations (store instructions, atomic
1579stores/read-modify-writes, memcpy, etc.). For the purposes of this section,
1580initialized globals are considered to have a write of the initializer which is
1581atomic and happens before any other read or write of the memory in question.
1582For each byte of a read <var>R</var>, <var>R<sub>byte</sub></var> may see
1583any write to the same byte, except:</p>
1584
1585<ul>
1586  <li>If <var>write<sub>1</sub></var> happens before
1587      <var>write<sub>2</sub></var>, and <var>write<sub>2</sub></var> happens
1588      before <var>R<sub>byte</sub></var>, then <var>R<sub>byte</sub></var>
1589      does not see <var>write<sub>1</sub></var>.
1590  <li>If <var>R<sub>byte</sub></var> happens before
1591      <var>write<sub>3</sub></var>, then <var>R<sub>byte</sub></var> does not
1592      see <var>write<sub>3</sub></var>.
1593</ul>
1594
1595<p>Given that definition, <var>R<sub>byte</sub></var> is defined as follows:
1596<ul>
1597  <li>If <var>R</var> is volatile, the result is target-dependent. (Volatile
1598      is supposed to give guarantees which can support
1599      <code>sig_atomic_t</code> in C/C++, and may be used for accesses to
1600      addresses which do not behave like normal memory.  It does not generally
1601      provide cross-thread synchronization.)
1602  <li>Otherwise, if there is no write to the same byte that happens before
1603    <var>R<sub>byte</sub></var>, <var>R<sub>byte</sub></var> returns
1604    <tt>undef</tt> for that byte.
1605  <li>Otherwise, if <var>R<sub>byte</sub></var> may see exactly one write,
1606      <var>R<sub>byte</sub></var> returns the value written by that
1607      write.</li>
1608  <li>Otherwise, if <var>R</var> is atomic, and all the writes
1609      <var>R<sub>byte</sub></var> may see are atomic, it chooses one of the
1610      values written.  See the <a href="#ordering">Atomic Memory Ordering
1611      Constraints</a> section for additional constraints on how the choice
1612      is made.
1613  <li>Otherwise <var>R<sub>byte</sub></var> returns <tt>undef</tt>.</li>
1614</ul>
1615
1616<p><var>R</var> returns the value composed of the series of bytes it read.
1617This implies that some bytes within the value may be <tt>undef</tt>
1618<b>without</b> the entire value being <tt>undef</tt>. Note that this only
1619defines the semantics of the operation; it doesn't mean that targets will
1620emit more than one instruction to read the series of bytes.</p>
1621
1622<p>Note that in cases where none of the atomic intrinsics are used, this model
1623places only one restriction on IR transformations on top of what is required
1624for single-threaded execution: introducing a store to a byte which might not
1625otherwise be stored is not allowed in general.  (Specifically, in the case
1626where another thread might write to and read from an address, introducing a
1627store can change a load that may see exactly one write into a load that may
1628see multiple writes.)</p>
1629
1630<!-- FIXME: This model assumes all targets where concurrency is relevant have
1631a byte-size store which doesn't affect adjacent bytes.  As far as I can tell,
1632none of the backends currently in the tree fall into this category; however,
1633there might be targets which care.  If there are, we want a paragraph
1634like the following:
1635
1636Targets may specify that stores narrower than a certain width are not
1637available; on such a target, for the purposes of this model, treat any
1638non-atomic write with an alignment or width less than the minimum width
1639as if it writes to the relevant surrounding bytes.
1640-->
1641
1642</div>
1643
1644<!-- ======================================================================= -->
1645<h3>
1646      <a name="ordering">Atomic Memory Ordering Constraints</a>
1647</h3>
1648
1649<div>
1650
1651<p>Atomic instructions (<a href="#i_cmpxchg"><code>cmpxchg</code></a>,
1652<a href="#i_atomicrmw"><code>atomicrmw</code></a>,
1653<a href="#i_fence"><code>fence</code></a>,
1654<a href="#i_load"><code>atomic load</code></a>, and
1655<a href="#i_store"><code>atomic store</code></a>) take an ordering parameter
1656that determines which other atomic instructions on the same address they
1657<i>synchronize with</i>.  These semantics are borrowed from Java and C++0x,
1658but are somewhat more colloquial. If these descriptions aren't precise enough,
1659check those specs (see spec references in the
1660<a href="Atomics.html#introduction">atomics guide</a>).
1661<a href="#i_fence"><code>fence</code></a> instructions
1662treat these orderings somewhat differently since they don't take an address.
1663See that instruction's documentation for details.</p>
1664
1665<p>For a simpler introduction to the ordering constraints, see the
1666<a href="Atomics.html">LLVM Atomic Instructions and Concurrency Guide</a>.</p>
1667
1668<dl>
1669<dt><code>unordered</code></dt>
1670<dd>The set of values that can be read is governed by the happens-before
1671partial order. A value cannot be read unless some operation wrote it.
1672This is intended to provide a guarantee strong enough to model Java's
1673non-volatile shared variables.  This ordering cannot be specified for
1674read-modify-write operations; it is not strong enough to make them atomic
1675in any interesting way.</dd>
1676<dt><code>monotonic</code></dt>
1677<dd>In addition to the guarantees of <code>unordered</code>, there is a single
1678total order for modifications by <code>monotonic</code> operations on each
1679address. All modification orders must be compatible with the happens-before
1680order. There is no guarantee that the modification orders can be combined to
1681a global total order for the whole program (and this often will not be
1682possible). The read in an atomic read-modify-write operation
1683(<a href="#i_cmpxchg"><code>cmpxchg</code></a> and
1684<a href="#i_atomicrmw"><code>atomicrmw</code></a>)
1685reads the value in the modification order immediately before the value it
1686writes. If one atomic read happens before another atomic read of the same
1687address, the later read must see the same value or a later value in the
1688address's modification order. This disallows reordering of
1689<code>monotonic</code> (or stronger) operations on the same address. If an
1690address is written <code>monotonic</code>ally by one thread, and other threads
1691<code>monotonic</code>ally read that address repeatedly, the other threads must
1692eventually see the write. This corresponds to the C++0x/C1x
1693<code>memory_order_relaxed</code>.</dd>
1694<dt><code>acquire</code></dt>
1695<dd>In addition to the guarantees of <code>monotonic</code>,
1696a <i>synchronizes-with</i> edge may be formed with a <code>release</code>
1697operation. This is intended to model C++'s <code>memory_order_acquire</code>.</dd>
1698<dt><code>release</code></dt>
1699<dd>In addition to the guarantees of <code>monotonic</code>, if this operation
1700writes a value which is subsequently read by an <code>acquire</code> operation,
1701it <i>synchronizes-with</i> that operation.  (This isn't a complete
1702description; see the C++0x definition of a release sequence.) This corresponds
1703to the C++0x/C1x <code>memory_order_release</code>.</dd>
1704<dt><code>acq_rel</code> (acquire+release)</dt><dd>Acts as both an
1705<code>acquire</code> and <code>release</code> operation on its address.
1706This corresponds to the C++0x/C1x <code>memory_order_acq_rel</code>.</dd>
1707<dt><code>seq_cst</code> (sequentially consistent)</dt><dd>
1708<dd>In addition to the guarantees of <code>acq_rel</code>
1709(<code>acquire</code> for an operation which only reads, <code>release</code>
1710for an operation which only writes), there is a global total order on all
1711sequentially-consistent operations on all addresses, which is consistent with
1712the <i>happens-before</i> partial order and with the modification orders of
1713all the affected addresses. Each sequentially-consistent read sees the last
1714preceding write to the same address in this global order. This corresponds
1715to the C++0x/C1x <code>memory_order_seq_cst</code> and Java volatile.</dd>
1716</dl>
1717
1718<p id="singlethread">If an atomic operation is marked <code>singlethread</code>,
1719it only <i>synchronizes with</i> or participates in modification and seq_cst
1720total orderings with other operations running in the same thread (for example,
1721in signal handlers).</p>
1722
1723</div>
1724
1725</div>
1726
1727<!-- *********************************************************************** -->
1728<h2><a name="typesystem">Type System</a></h2>
1729<!-- *********************************************************************** -->
1730
1731<div>
1732
1733<p>The LLVM type system is one of the most important features of the
1734   intermediate representation.  Being typed enables a number of optimizations
1735   to be performed on the intermediate representation directly, without having
1736   to do extra analyses on the side before the transformation.  A strong type
1737   system makes it easier to read the generated code and enables novel analyses
1738   and transformations that are not feasible to perform on normal three address
1739   code representations.</p>
1740
1741<!-- ======================================================================= -->
1742<h3>
1743  <a name="t_classifications">Type Classifications</a>
1744</h3>
1745
1746<div>
1747
1748<p>The types fall into a few useful classifications:</p>
1749
1750<table border="1" cellspacing="0" cellpadding="4">
1751  <tbody>
1752    <tr><th>Classification</th><th>Types</th></tr>
1753    <tr>
1754      <td><a href="#t_integer">integer</a></td>
1755      <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1756    </tr>
1757    <tr>
1758      <td><a href="#t_floating">floating point</a></td>
1759      <td><tt>half, float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1760    </tr>
1761    <tr>
1762      <td><a name="t_firstclass">first class</a></td>
1763      <td><a href="#t_integer">integer</a>,
1764          <a href="#t_floating">floating point</a>,
1765          <a href="#t_pointer">pointer</a>,
1766          <a href="#t_vector">vector</a>,
1767          <a href="#t_struct">structure</a>,
1768          <a href="#t_array">array</a>,
1769          <a href="#t_label">label</a>,
1770          <a href="#t_metadata">metadata</a>.
1771      </td>
1772    </tr>
1773    <tr>
1774      <td><a href="#t_primitive">primitive</a></td>
1775      <td><a href="#t_label">label</a>,
1776          <a href="#t_void">void</a>,
1777          <a href="#t_integer">integer</a>,
1778          <a href="#t_floating">floating point</a>,
1779          <a href="#t_x86mmx">x86mmx</a>,
1780          <a href="#t_metadata">metadata</a>.</td>
1781    </tr>
1782    <tr>
1783      <td><a href="#t_derived">derived</a></td>
1784      <td><a href="#t_array">array</a>,
1785          <a href="#t_function">function</a>,
1786          <a href="#t_pointer">pointer</a>,
1787          <a href="#t_struct">structure</a>,
1788          <a href="#t_vector">vector</a>,
1789          <a href="#t_opaque">opaque</a>.
1790      </td>
1791    </tr>
1792  </tbody>
1793</table>
1794
1795<p>The <a href="#t_firstclass">first class</a> types are perhaps the most
1796   important.  Values of these types are the only ones which can be produced by
1797   instructions.</p>
1798
1799</div>
1800
1801<!-- ======================================================================= -->
1802<h3>
1803  <a name="t_primitive">Primitive Types</a>
1804</h3>
1805
1806<div>
1807
1808<p>The primitive types are the fundamental building blocks of the LLVM
1809   system.</p>
1810
1811<!-- _______________________________________________________________________ -->
1812<h4>
1813  <a name="t_integer">Integer Type</a>
1814</h4>
1815
1816<div>
1817
1818<h5>Overview:</h5>
1819<p>The integer type is a very simple type that simply specifies an arbitrary
1820   bit width for the integer type desired. Any bit width from 1 bit to
1821   2<sup>23</sup>-1 (about 8 million) can be specified.</p>
1822
1823<h5>Syntax:</h5>
1824<pre>
1825  iN
1826</pre>
1827
1828<p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1829   value.</p>
1830
1831<h5>Examples:</h5>
1832<table class="layout">
1833  <tr class="layout">
1834    <td class="left"><tt>i1</tt></td>
1835    <td class="left">a single-bit integer.</td>
1836  </tr>
1837  <tr class="layout">
1838    <td class="left"><tt>i32</tt></td>
1839    <td class="left">a 32-bit integer.</td>
1840  </tr>
1841  <tr class="layout">
1842    <td class="left"><tt>i1942652</tt></td>
1843    <td class="left">a really big integer of over 1 million bits.</td>
1844  </tr>
1845</table>
1846
1847</div>
1848
1849<!-- _______________________________________________________________________ -->
1850<h4>
1851  <a name="t_floating">Floating Point Types</a>
1852</h4>
1853
1854<div>
1855
1856<table>
1857  <tbody>
1858    <tr><th>Type</th><th>Description</th></tr>
1859    <tr><td><tt>half</tt></td><td>16-bit floating point value</td></tr>
1860    <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1861    <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1862    <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1863    <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1864    <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1865  </tbody>
1866</table>
1867
1868</div>
1869
1870<!-- _______________________________________________________________________ -->
1871<h4>
1872  <a name="t_x86mmx">X86mmx Type</a>
1873</h4>
1874
1875<div>
1876
1877<h5>Overview:</h5>
1878<p>The x86mmx type represents a value held in an MMX register on an x86 machine.  The operations allowed on it are quite limited:  parameters and return values, load and store, and bitcast.  User-specified MMX instructions are represented as intrinsic or asm calls with arguments and/or results of this type.  There are no arrays, vectors or constants of this type.</p>
1879
1880<h5>Syntax:</h5>
1881<pre>
1882  x86mmx
1883</pre>
1884
1885</div>
1886
1887<!-- _______________________________________________________________________ -->
1888<h4>
1889  <a name="t_void">Void Type</a>
1890</h4>
1891
1892<div>
1893
1894<h5>Overview:</h5>
1895<p>The void type does not represent any value and has no size.</p>
1896
1897<h5>Syntax:</h5>
1898<pre>
1899  void
1900</pre>
1901
1902</div>
1903
1904<!-- _______________________________________________________________________ -->
1905<h4>
1906  <a name="t_label">Label Type</a>
1907</h4>
1908
1909<div>
1910
1911<h5>Overview:</h5>
1912<p>The label type represents code labels.</p>
1913
1914<h5>Syntax:</h5>
1915<pre>
1916  label
1917</pre>
1918
1919</div>
1920
1921<!-- _______________________________________________________________________ -->
1922<h4>
1923  <a name="t_metadata">Metadata Type</a>
1924</h4>
1925
1926<div>
1927
1928<h5>Overview:</h5>
1929<p>The metadata type represents embedded metadata. No derived types may be
1930   created from metadata except for <a href="#t_function">function</a>
1931   arguments.
1932
1933<h5>Syntax:</h5>
1934<pre>
1935  metadata
1936</pre>
1937
1938</div>
1939
1940</div>
1941
1942<!-- ======================================================================= -->
1943<h3>
1944  <a name="t_derived">Derived Types</a>
1945</h3>
1946
1947<div>
1948
1949<p>The real power in LLVM comes from the derived types in the system.  This is
1950   what allows a programmer to represent arrays, functions, pointers, and other
1951   useful types.  Each of these types contain one or more element types which
1952   may be a primitive type, or another derived type.  For example, it is
1953   possible to have a two dimensional array, using an array as the element type
1954   of another array.</p>
1955
1956<!-- _______________________________________________________________________ -->
1957<h4>
1958  <a name="t_aggregate">Aggregate Types</a>
1959</h4>
1960
1961<div>
1962
1963<p>Aggregate Types are a subset of derived types that can contain multiple
1964  member types. <a href="#t_array">Arrays</a> and
1965  <a href="#t_struct">structs</a> are aggregate types.
1966  <a href="#t_vector">Vectors</a> are not considered to be aggregate types.</p>
1967
1968</div>
1969
1970<!-- _______________________________________________________________________ -->
1971<h4>
1972  <a name="t_array">Array Type</a>
1973</h4>
1974
1975<div>
1976
1977<h5>Overview:</h5>
1978<p>The array type is a very simple derived type that arranges elements
1979   sequentially in memory.  The array type requires a size (number of elements)
1980   and an underlying data type.</p>
1981
1982<h5>Syntax:</h5>
1983<pre>
1984  [&lt;# elements&gt; x &lt;elementtype&gt;]
1985</pre>
1986
1987<p>The number of elements is a constant integer value; <tt>elementtype</tt> may
1988   be any type with a size.</p>
1989
1990<h5>Examples:</h5>
1991<table class="layout">
1992  <tr class="layout">
1993    <td class="left"><tt>[40 x i32]</tt></td>
1994    <td class="left">Array of 40 32-bit integer values.</td>
1995  </tr>
1996  <tr class="layout">
1997    <td class="left"><tt>[41 x i32]</tt></td>
1998    <td class="left">Array of 41 32-bit integer values.</td>
1999  </tr>
2000  <tr class="layout">
2001    <td class="left"><tt>[4 x i8]</tt></td>
2002    <td class="left">Array of 4 8-bit integer values.</td>
2003  </tr>
2004</table>
2005<p>Here are some examples of multidimensional arrays:</p>
2006<table class="layout">
2007  <tr class="layout">
2008    <td class="left"><tt>[3 x [4 x i32]]</tt></td>
2009    <td class="left">3x4 array of 32-bit integer values.</td>
2010  </tr>
2011  <tr class="layout">
2012    <td class="left"><tt>[12 x [10 x float]]</tt></td>
2013    <td class="left">12x10 array of single precision floating point values.</td>
2014  </tr>
2015  <tr class="layout">
2016    <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
2017    <td class="left">2x3x4 array of 16-bit integer  values.</td>
2018  </tr>
2019</table>
2020
2021<p>There is no restriction on indexing beyond the end of the array implied by
2022   a static type (though there are restrictions on indexing beyond the bounds
2023   of an allocated object in some cases). This means that single-dimension
2024   'variable sized array' addressing can be implemented in LLVM with a zero
2025   length array type. An implementation of 'pascal style arrays' in LLVM could
2026   use the type "<tt>{ i32, [0 x float]}</tt>", for example.</p>
2027
2028</div>
2029
2030<!-- _______________________________________________________________________ -->
2031<h4>
2032  <a name="t_function">Function Type</a>
2033</h4>
2034
2035<div>
2036
2037<h5>Overview:</h5>
2038<p>The function type can be thought of as a function signature.  It consists of
2039   a return type and a list of formal parameter types. The return type of a
2040   function type is a first class type or a void type.</p>
2041
2042<h5>Syntax:</h5>
2043<pre>
2044  &lt;returntype&gt; (&lt;parameter list&gt;)
2045</pre>
2046
2047<p>...where '<tt>&lt;parameter list&gt;</tt>' is a comma-separated list of type
2048   specifiers.  Optionally, the parameter list may include a type <tt>...</tt>,
2049   which indicates that the function takes a variable number of arguments.
2050   Variable argument functions can access their arguments with
2051   the <a href="#int_varargs">variable argument handling intrinsic</a>
2052   functions.  '<tt>&lt;returntype&gt;</tt>' is any type except
2053   <a href="#t_label">label</a>.</p>
2054
2055<h5>Examples:</h5>
2056<table class="layout">
2057  <tr class="layout">
2058    <td class="left"><tt>i32 (i32)</tt></td>
2059    <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
2060    </td>
2061  </tr><tr class="layout">
2062    <td class="left"><tt>float&nbsp;(i16,&nbsp;i32&nbsp;*)&nbsp;*
2063    </tt></td>
2064    <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
2065      an <tt>i16</tt> and a <a href="#t_pointer">pointer</a> to <tt>i32</tt>,
2066      returning <tt>float</tt>.
2067    </td>
2068  </tr><tr class="layout">
2069    <td class="left"><tt>i32 (i8*, ...)</tt></td>
2070    <td class="left">A vararg function that takes at least one
2071      <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
2072      which returns an integer.  This is the signature for <tt>printf</tt> in
2073      LLVM.
2074    </td>
2075  </tr><tr class="layout">
2076    <td class="left"><tt>{i32, i32} (i32)</tt></td>
2077    <td class="left">A function taking an <tt>i32</tt>, returning a
2078        <a href="#t_struct">structure</a> containing two <tt>i32</tt> values
2079    </td>
2080  </tr>
2081</table>
2082
2083</div>
2084
2085<!-- _______________________________________________________________________ -->
2086<h4>
2087  <a name="t_struct">Structure Type</a>
2088</h4>
2089
2090<div>
2091
2092<h5>Overview:</h5>
2093<p>The structure type is used to represent a collection of data members together
2094  in memory.  The elements of a structure may be any type that has a size.</p>
2095
2096<p>Structures in memory are accessed using '<tt><a href="#i_load">load</a></tt>'
2097   and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field
2098   with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
2099   Structures in registers are accessed using the
2100   '<tt><a href="#i_extractvalue">extractvalue</a></tt>' and
2101   '<tt><a href="#i_insertvalue">insertvalue</a></tt>' instructions.</p>
2102
2103<p>Structures may optionally be "packed" structures, which indicate that the
2104  alignment of the struct is one byte, and that there is no padding between
2105  the elements.  In non-packed structs, padding between field types is inserted
2106  as defined by the TargetData string in the module, which is required to match
2107  what the underlying code generator expects.</p>
2108
2109<p>Structures can either be "literal" or "identified".  A literal structure is
2110  defined inline with other types (e.g. <tt>{i32, i32}*</tt>) whereas identified
2111  types are always defined at the top level with a name.  Literal types are
2112  uniqued by their contents and can never be recursive or opaque since there is
2113  no way to write one.  Identified types can be recursive, can be opaqued, and are
2114  never uniqued.
2115</p>
2116
2117<h5>Syntax:</h5>
2118<pre>
2119  %T1 = type { &lt;type list&gt; }     <i>; Identified normal struct type</i>
2120  %T2 = type &lt;{ &lt;type list&gt; }&gt;   <i>; Identified packed struct type</i>
2121</pre>
2122
2123<h5>Examples:</h5>
2124<table class="layout">
2125  <tr class="layout">
2126    <td class="left"><tt>{ i32, i32, i32 }</tt></td>
2127    <td class="left">A triple of three <tt>i32</tt> values</td>
2128  </tr>
2129  <tr class="layout">
2130    <td class="left"><tt>{&nbsp;float,&nbsp;i32&nbsp;(i32)&nbsp;*&nbsp;}</tt></td>
2131    <td class="left">A pair, where the first element is a <tt>float</tt> and the
2132      second element is a <a href="#t_pointer">pointer</a> to a
2133      <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
2134      an <tt>i32</tt>.</td>
2135  </tr>
2136  <tr class="layout">
2137    <td class="left"><tt>&lt;{ i8, i32 }&gt;</tt></td>
2138    <td class="left">A packed struct known to be 5 bytes in size.</td>
2139  </tr>
2140</table>
2141
2142</div>
2143
2144<!-- _______________________________________________________________________ -->
2145<h4>
2146  <a name="t_opaque">Opaque Structure Types</a>
2147</h4>
2148
2149<div>
2150
2151<h5>Overview:</h5>
2152<p>Opaque structure types are used to represent named structure types that do
2153   not have a body specified.  This corresponds (for example) to the C notion of
2154   a forward declared structure.</p>
2155
2156<h5>Syntax:</h5>
2157<pre>
2158  %X = type opaque
2159  %52 = type opaque
2160</pre>
2161
2162<h5>Examples:</h5>
2163<table class="layout">
2164  <tr class="layout">
2165    <td class="left"><tt>opaque</tt></td>
2166    <td class="left">An opaque type.</td>
2167  </tr>
2168</table>
2169
2170</div>
2171
2172
2173
2174<!-- _______________________________________________________________________ -->
2175<h4>
2176  <a name="t_pointer">Pointer Type</a>
2177</h4>
2178
2179<div>
2180
2181<h5>Overview:</h5>
2182<p>The pointer type is used to specify memory locations.
2183   Pointers are commonly used to reference objects in memory.</p>
2184
2185<p>Pointer types may have an optional address space attribute defining the
2186   numbered address space where the pointed-to object resides. The default
2187   address space is number zero. The semantics of non-zero address
2188   spaces are target-specific.</p>
2189
2190<p>Note that LLVM does not permit pointers to void (<tt>void*</tt>) nor does it
2191   permit pointers to labels (<tt>label*</tt>).  Use <tt>i8*</tt> instead.</p>
2192
2193<h5>Syntax:</h5>
2194<pre>
2195  &lt;type&gt; *
2196</pre>
2197
2198<h5>Examples:</h5>
2199<table class="layout">
2200  <tr class="layout">
2201    <td class="left"><tt>[4 x i32]*</tt></td>
2202    <td class="left">A <a href="#t_pointer">pointer</a> to <a
2203                    href="#t_array">array</a> of four <tt>i32</tt> values.</td>
2204  </tr>
2205  <tr class="layout">
2206    <td class="left"><tt>i32 (i32*) *</tt></td>
2207    <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
2208      href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
2209      <tt>i32</tt>.</td>
2210  </tr>
2211  <tr class="layout">
2212    <td class="left"><tt>i32 addrspace(5)*</tt></td>
2213    <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
2214     that resides in address space #5.</td>
2215  </tr>
2216</table>
2217
2218</div>
2219
2220<!-- _______________________________________________________________________ -->
2221<h4>
2222  <a name="t_vector">Vector Type</a>
2223</h4>
2224
2225<div>
2226
2227<h5>Overview:</h5>
2228<p>A vector type is a simple derived type that represents a vector of elements.
2229   Vector types are used when multiple primitive data are operated in parallel
2230   using a single instruction (SIMD).  A vector type requires a size (number of
2231   elements) and an underlying primitive data type.  Vector types are considered
2232   <a href="#t_firstclass">first class</a>.</p>
2233
2234<h5>Syntax:</h5>
2235<pre>
2236  &lt; &lt;# elements&gt; x &lt;elementtype&gt; &gt;
2237</pre>
2238
2239<p>The number of elements is a constant integer value larger than 0; elementtype
2240   may be any integer or floating point type, or a pointer to these types.
2241   Vectors of size zero are not allowed. </p>
2242
2243<h5>Examples:</h5>
2244<table class="layout">
2245  <tr class="layout">
2246    <td class="left"><tt>&lt;4 x i32&gt;</tt></td>
2247    <td class="left">Vector of 4 32-bit integer values.</td>
2248  </tr>
2249  <tr class="layout">
2250    <td class="left"><tt>&lt;8 x float&gt;</tt></td>
2251    <td class="left">Vector of 8 32-bit floating-point values.</td>
2252  </tr>
2253  <tr class="layout">
2254    <td class="left"><tt>&lt;2 x i64&gt;</tt></td>
2255    <td class="left">Vector of 2 64-bit integer values.</td>
2256  </tr>
2257  <tr class="layout">
2258    <td class="left"><tt>&lt;4 x i64*&gt;</tt></td>
2259    <td class="left">Vector of 4 pointers to 64-bit integer values.</td>
2260  </tr>
2261</table>
2262
2263</div>
2264
2265</div>
2266
2267</div>
2268
2269<!-- *********************************************************************** -->
2270<h2><a name="constants">Constants</a></h2>
2271<!-- *********************************************************************** -->
2272
2273<div>
2274
2275<p>LLVM has several different basic types of constants.  This section describes
2276   them all and their syntax.</p>
2277
2278<!-- ======================================================================= -->
2279<h3>
2280  <a name="simpleconstants">Simple Constants</a>
2281</h3>
2282
2283<div>
2284
2285<dl>
2286  <dt><b>Boolean constants</b></dt>
2287  <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
2288      constants of the <tt><a href="#t_integer">i1</a></tt> type.</dd>
2289
2290  <dt><b>Integer constants</b></dt>
2291  <dd>Standard integers (such as '4') are constants of
2292      the <a href="#t_integer">integer</a> type.  Negative numbers may be used
2293      with integer types.</dd>
2294
2295  <dt><b>Floating point constants</b></dt>
2296  <dd>Floating point constants use standard decimal notation (e.g. 123.421),
2297      exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
2298      notation (see below).  The assembler requires the exact decimal value of a
2299      floating-point constant.  For example, the assembler accepts 1.25 but
2300      rejects 1.3 because 1.3 is a repeating decimal in binary.  Floating point
2301      constants must have a <a href="#t_floating">floating point</a> type. </dd>
2302
2303  <dt><b>Null pointer constants</b></dt>
2304  <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
2305      and must be of <a href="#t_pointer">pointer type</a>.</dd>
2306</dl>
2307
2308<p>The one non-intuitive notation for constants is the hexadecimal form of
2309   floating point constants.  For example, the form '<tt>double
2310   0x432ff973cafa8000</tt>' is equivalent to (but harder to read than)
2311   '<tt>double 4.5e+15</tt>'.  The only time hexadecimal floating point
2312   constants are required (and the only time that they are generated by the
2313   disassembler) is when a floating point constant must be emitted but it cannot
2314   be represented as a decimal floating point number in a reasonable number of
2315   digits.  For example, NaN's, infinities, and other special values are
2316   represented in their IEEE hexadecimal format so that assembly and disassembly
2317   do not cause any bits to change in the constants.</p>
2318
2319<p>When using the hexadecimal form, constants of types half, float, and double are
2320   represented using the 16-digit form shown above (which matches the IEEE754
2321   representation for double); half and float values must, however, be exactly
2322   representable as IEE754 half and single precision, respectively.
2323   Hexadecimal format is always used
2324   for long double, and there are three forms of long double.  The 80-bit format
2325   used by x86 is represented as <tt>0xK</tt> followed by 20 hexadecimal digits.
2326   The 128-bit format used by PowerPC (two adjacent doubles) is represented
2327   by <tt>0xM</tt> followed by 32 hexadecimal digits.  The IEEE 128-bit format
2328   is represented by <tt>0xL</tt> followed by 32 hexadecimal digits; no
2329   currently supported target uses this format.  Long doubles will only work if
2330   they match the long double format on your target. The IEEE 16-bit format
2331   (half precision) is represented by <tt>0xH</tt> followed by 4 hexadecimal
2332   digits. All hexadecimal formats are big-endian (sign bit at the left).</p>
2333
2334<p>There are no constants of type x86mmx.</p>
2335</div>
2336
2337<!-- ======================================================================= -->
2338<h3>
2339<a name="aggregateconstants"></a> <!-- old anchor -->
2340<a name="complexconstants">Complex Constants</a>
2341</h3>
2342
2343<div>
2344
2345<p>Complex constants are a (potentially recursive) combination of simple
2346   constants and smaller complex constants.</p>
2347
2348<dl>
2349  <dt><b>Structure constants</b></dt>
2350  <dd>Structure constants are represented with notation similar to structure
2351      type definitions (a comma separated list of elements, surrounded by braces
2352      (<tt>{}</tt>)).  For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
2353      where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>".
2354      Structure constants must have <a href="#t_struct">structure type</a>, and
2355      the number and types of elements must match those specified by the
2356      type.</dd>
2357
2358  <dt><b>Array constants</b></dt>
2359  <dd>Array constants are represented with notation similar to array type
2360     definitions (a comma separated list of elements, surrounded by square
2361     brackets (<tt>[]</tt>)).  For example: "<tt>[ i32 42, i32 11, i32 74
2362     ]</tt>".  Array constants must have <a href="#t_array">array type</a>, and
2363     the number and types of elements must match those specified by the
2364     type.</dd>
2365
2366  <dt><b>Vector constants</b></dt>
2367  <dd>Vector constants are represented with notation similar to vector type
2368      definitions (a comma separated list of elements, surrounded by
2369      less-than/greater-than's (<tt>&lt;&gt;</tt>)).  For example: "<tt>&lt; i32
2370      42, i32 11, i32 74, i32 100 &gt;</tt>".  Vector constants must
2371      have <a href="#t_vector">vector type</a>, and the number and types of
2372      elements must match those specified by the type.</dd>
2373
2374  <dt><b>Zero initialization</b></dt>
2375  <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
2376      value to zero of <em>any</em> type, including scalar and
2377      <a href="#t_aggregate">aggregate</a> types.
2378      This is often used to avoid having to print large zero initializers
2379      (e.g. for large arrays) and is always exactly equivalent to using explicit
2380      zero initializers.</dd>
2381
2382  <dt><b>Metadata node</b></dt>
2383  <dd>A metadata node is a structure-like constant with
2384      <a href="#t_metadata">metadata type</a>.  For example: "<tt>metadata !{
2385      i32 0, metadata !"test" }</tt>".  Unlike other constants that are meant to
2386      be interpreted as part of the instruction stream, metadata is a place to
2387      attach additional information such as debug info.</dd>
2388</dl>
2389
2390</div>
2391
2392<!-- ======================================================================= -->
2393<h3>
2394  <a name="globalconstants">Global Variable and Function Addresses</a>
2395</h3>
2396
2397<div>
2398
2399<p>The addresses of <a href="#globalvars">global variables</a>
2400   and <a href="#functionstructure">functions</a> are always implicitly valid
2401   (link-time) constants.  These constants are explicitly referenced when
2402   the <a href="#identifiers">identifier for the global</a> is used and always
2403   have <a href="#t_pointer">pointer</a> type. For example, the following is a
2404   legal LLVM file:</p>
2405
2406<pre class="doc_code">
2407@X = global i32 17
2408@Y = global i32 42
2409@Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2410</pre>
2411
2412</div>
2413
2414<!-- ======================================================================= -->
2415<h3>
2416  <a name="undefvalues">Undefined Values</a>
2417</h3>
2418
2419<div>
2420
2421<p>The string '<tt>undef</tt>' can be used anywhere a constant is expected, and
2422   indicates that the user of the value may receive an unspecified bit-pattern.
2423   Undefined values may be of any type (other than '<tt>label</tt>'
2424   or '<tt>void</tt>') and be used anywhere a constant is permitted.</p>
2425
2426<p>Undefined values are useful because they indicate to the compiler that the
2427   program is well defined no matter what value is used.  This gives the
2428   compiler more freedom to optimize.  Here are some examples of (potentially
2429   surprising) transformations that are valid (in pseudo IR):</p>
2430
2431
2432<pre class="doc_code">
2433  %A = add %X, undef
2434  %B = sub %X, undef
2435  %C = xor %X, undef
2436Safe:
2437  %A = undef
2438  %B = undef
2439  %C = undef
2440</pre>
2441
2442<p>This is safe because all of the output bits are affected by the undef bits.
2443   Any output bit can have a zero or one depending on the input bits.</p>
2444
2445<pre class="doc_code">
2446  %A = or %X, undef
2447  %B = and %X, undef
2448Safe:
2449  %A = -1
2450  %B = 0
2451Unsafe:
2452  %A = undef
2453  %B = undef
2454</pre>
2455
2456<p>These logical operations have bits that are not always affected by the input.
2457   For example, if <tt>%X</tt> has a zero bit, then the output of the
2458   '<tt>and</tt>' operation will always be a zero for that bit, no matter what
2459   the corresponding bit from the '<tt>undef</tt>' is. As such, it is unsafe to
2460   optimize or assume that the result of the '<tt>and</tt>' is '<tt>undef</tt>'.
2461   However, it is safe to assume that all bits of the '<tt>undef</tt>' could be
2462   0, and optimize the '<tt>and</tt>' to 0. Likewise, it is safe to assume that
2463   all the bits of the '<tt>undef</tt>' operand to the '<tt>or</tt>' could be
2464   set, allowing the '<tt>or</tt>' to be folded to -1.</p>
2465
2466<pre class="doc_code">
2467  %A = select undef, %X, %Y
2468  %B = select undef, 42, %Y
2469  %C = select %X, %Y, undef
2470Safe:
2471  %A = %X     (or %Y)
2472  %B = 42     (or %Y)
2473  %C = %Y
2474Unsafe:
2475  %A = undef
2476  %B = undef
2477  %C = undef
2478</pre>
2479
2480<p>This set of examples shows that undefined '<tt>select</tt>' (and conditional
2481   branch) conditions can go <em>either way</em>, but they have to come from one
2482   of the two operands.  In the <tt>%A</tt> example, if <tt>%X</tt> and
2483   <tt>%Y</tt> were both known to have a clear low bit, then <tt>%A</tt> would
2484   have to have a cleared low bit. However, in the <tt>%C</tt> example, the
2485   optimizer is allowed to assume that the '<tt>undef</tt>' operand could be the
2486   same as <tt>%Y</tt>, allowing the whole '<tt>select</tt>' to be
2487   eliminated.</p>
2488
2489<pre class="doc_code">
2490  %A = xor undef, undef
2491
2492  %B = undef
2493  %C = xor %B, %B
2494
2495  %D = undef
2496  %E = icmp lt %D, 4
2497  %F = icmp gte %D, 4
2498
2499Safe:
2500  %A = undef
2501  %B = undef
2502  %C = undef
2503  %D = undef
2504  %E = undef
2505  %F = undef
2506</pre>
2507
2508<p>This example points out that two '<tt>undef</tt>' operands are not
2509   necessarily the same. This can be surprising to people (and also matches C
2510   semantics) where they assume that "<tt>X^X</tt>" is always zero, even
2511   if <tt>X</tt> is undefined. This isn't true for a number of reasons, but the
2512   short answer is that an '<tt>undef</tt>' "variable" can arbitrarily change
2513   its value over its "live range".  This is true because the variable doesn't
2514   actually <em>have a live range</em>. Instead, the value is logically read
2515   from arbitrary registers that happen to be around when needed, so the value
2516   is not necessarily consistent over time. In fact, <tt>%A</tt> and <tt>%C</tt>
2517   need to have the same semantics or the core LLVM "replace all uses with"
2518   concept would not hold.</p>
2519
2520<pre class="doc_code">
2521  %A = fdiv undef, %X
2522  %B = fdiv %X, undef
2523Safe:
2524  %A = undef
2525b: unreachable
2526</pre>
2527
2528<p>These examples show the crucial difference between an <em>undefined
2529  value</em> and <em>undefined behavior</em>. An undefined value (like
2530  '<tt>undef</tt>') is allowed to have an arbitrary bit-pattern. This means that
2531  the <tt>%A</tt> operation can be constant folded to '<tt>undef</tt>', because
2532  the '<tt>undef</tt>' could be an SNaN, and <tt>fdiv</tt> is not (currently)
2533  defined on SNaN's. However, in the second example, we can make a more
2534  aggressive assumption: because the <tt>undef</tt> is allowed to be an
2535  arbitrary value, we are allowed to assume that it could be zero. Since a
2536  divide by zero has <em>undefined behavior</em>, we are allowed to assume that
2537  the operation does not execute at all. This allows us to delete the divide and
2538  all code after it. Because the undefined operation "can't happen", the
2539  optimizer can assume that it occurs in dead code.</p>
2540
2541<pre class="doc_code">
2542a:  store undef -> %X
2543b:  store %X -> undef
2544Safe:
2545a: &lt;deleted&gt;
2546b: unreachable
2547</pre>
2548
2549<p>These examples reiterate the <tt>fdiv</tt> example: a store <em>of</em> an
2550   undefined value can be assumed to not have any effect; we can assume that the
2551   value is overwritten with bits that happen to match what was already there.
2552   However, a store <em>to</em> an undefined location could clobber arbitrary
2553   memory, therefore, it has undefined behavior.</p>
2554
2555</div>
2556
2557<!-- ======================================================================= -->
2558<h3>
2559  <a name="poisonvalues">Poison Values</a>
2560</h3>
2561
2562<div>
2563
2564<p>Poison values are similar to <a href="#undefvalues">undef values</a>, however
2565   they also represent the fact that an instruction or constant expression which
2566   cannot evoke side effects has nevertheless detected a condition which results
2567   in undefined behavior.</p>
2568
2569<p>There is currently no way of representing a poison value in the IR; they
2570   only exist when produced by operations such as
2571   <a href="#i_add"><tt>add</tt></a> with the <tt>nsw</tt> flag.</p>
2572
2573<p>Poison value behavior is defined in terms of value <i>dependence</i>:</p>
2574
2575<ul>
2576<li>Values other than <a href="#i_phi"><tt>phi</tt></a> nodes depend on
2577    their operands.</li>
2578
2579<li><a href="#i_phi"><tt>Phi</tt></a> nodes depend on the operand corresponding
2580    to their dynamic predecessor basic block.</li>
2581
2582<li>Function arguments depend on the corresponding actual argument values in
2583    the dynamic callers of their functions.</li>
2584
2585<li><a href="#i_call"><tt>Call</tt></a> instructions depend on the
2586    <a href="#i_ret"><tt>ret</tt></a> instructions that dynamically transfer
2587    control back to them.</li>
2588
2589<li><a href="#i_invoke"><tt>Invoke</tt></a> instructions depend on the
2590    <a href="#i_ret"><tt>ret</tt></a>, <a href="#i_resume"><tt>resume</tt></a>,
2591    or exception-throwing call instructions that dynamically transfer control
2592    back to them.</li>
2593
2594<li>Non-volatile loads and stores depend on the most recent stores to all of the
2595    referenced memory addresses, following the order in the IR
2596    (including loads and stores implied by intrinsics such as
2597    <a href="#int_memcpy"><tt>@llvm.memcpy</tt></a>.)</li>
2598
2599<!-- TODO: In the case of multiple threads, this only applies if the store
2600     "happens-before" the load or store. -->
2601
2602<!-- TODO: floating-point exception state -->
2603
2604<li>An instruction with externally visible side effects depends on the most
2605    recent preceding instruction with externally visible side effects, following
2606    the order in the IR. (This includes
2607    <a href="#volatile">volatile operations</a>.)</li>
2608
2609<li>An instruction <i>control-depends</i> on a
2610    <a href="#terminators">terminator instruction</a>
2611    if the terminator instruction has multiple successors and the instruction
2612    is always executed when control transfers to one of the successors, and
2613    may not be executed when control is transferred to another.</li>
2614
2615<li>Additionally, an instruction also <i>control-depends</i> on a terminator
2616    instruction if the set of instructions it otherwise depends on would be
2617    different if the terminator had transferred control to a different
2618    successor.</li>
2619
2620<li>Dependence is transitive.</li>
2621
2622</ul>
2623
2624<p>Poison Values have the same behavior as <a href="#undefvalues">undef values</a>,
2625   with the additional affect that any instruction which has a <i>dependence</i>
2626   on a poison value has undefined behavior.</p>
2627
2628<p>Here are some examples:</p>
2629
2630<pre class="doc_code">
2631entry:
2632  %poison = sub nuw i32 0, 1           ; Results in a poison value.
2633  %still_poison = and i32 %poison, 0   ; 0, but also poison.
2634  %poison_yet_again = getelementptr i32* @h, i32 %still_poison
2635  store i32 0, i32* %poison_yet_again  ; memory at @h[0] is poisoned
2636
2637  store i32 %poison, i32* @g           ; Poison value stored to memory.
2638  %poison2 = load i32* @g              ; Poison value loaded back from memory.
2639
2640  store volatile i32 %poison, i32* @g  ; External observation; undefined behavior.
2641
2642  %narrowaddr = bitcast i32* @g to i16*
2643  %wideaddr = bitcast i32* @g to i64*
2644  %poison3 = load i16* %narrowaddr     ; Returns a poison value.
2645  %poison4 = load i64* %wideaddr       ; Returns a poison value.
2646
2647  %cmp = icmp slt i32 %poison, 0       ; Returns a poison value.
2648  br i1 %cmp, label %true, label %end  ; Branch to either destination.
2649
2650true:
2651  store volatile i32 0, i32* @g        ; This is control-dependent on %cmp, so
2652                                       ; it has undefined behavior.
2653  br label %end
2654
2655end:
2656  %p = phi i32 [ 0, %entry ], [ 1, %true ]
2657                                       ; Both edges into this PHI are
2658                                       ; control-dependent on %cmp, so this
2659                                       ; always results in a poison value.
2660
2661  store volatile i32 0, i32* @g        ; This would depend on the store in %true
2662                                       ; if %cmp is true, or the store in %entry
2663                                       ; otherwise, so this is undefined behavior.
2664
2665  br i1 %cmp, label %second_true, label %second_end
2666                                       ; The same branch again, but this time the
2667                                       ; true block doesn't have side effects.
2668
2669second_true:
2670  ; No side effects!
2671  ret void
2672
2673second_end:
2674  store volatile i32 0, i32* @g        ; This time, the instruction always depends
2675                                       ; on the store in %end. Also, it is
2676                                       ; control-equivalent to %end, so this is
2677                                       ; well-defined (ignoring earlier undefined
2678                                       ; behavior in this example).
2679</pre>
2680
2681</div>
2682
2683<!-- ======================================================================= -->
2684<h3>
2685  <a name="blockaddress">Addresses of Basic Blocks</a>
2686</h3>
2687
2688<div>
2689
2690<p><b><tt>blockaddress(@function, %block)</tt></b></p>
2691
2692<p>The '<tt>blockaddress</tt>' constant computes the address of the specified
2693   basic block in the specified function, and always has an i8* type.  Taking
2694   the address of the entry block is illegal.</p>
2695
2696<p>This value only has defined behavior when used as an operand to the
2697   '<a href="#i_indirectbr"><tt>indirectbr</tt></a>' instruction, or for
2698   comparisons against null. Pointer equality tests between labels addresses
2699   results in undefined behavior &mdash; though, again, comparison against null
2700   is ok, and no label is equal to the null pointer. This may be passed around
2701   as an opaque pointer sized value as long as the bits are not inspected. This
2702   allows <tt>ptrtoint</tt> and arithmetic to be performed on these values so
2703   long as the original value is reconstituted before the <tt>indirectbr</tt>
2704   instruction.</p>
2705
2706<p>Finally, some targets may provide defined semantics when using the value as
2707   the operand to an inline assembly, but that is target specific.</p>
2708
2709</div>
2710
2711
2712<!-- ======================================================================= -->
2713<h3>
2714  <a name="constantexprs">Constant Expressions</a>
2715</h3>
2716
2717<div>
2718
2719<p>Constant expressions are used to allow expressions involving other constants
2720   to be used as constants.  Constant expressions may be of
2721   any <a href="#t_firstclass">first class</a> type and may involve any LLVM
2722   operation that does not have side effects (e.g. load and call are not
2723   supported). The following is the syntax for constant expressions:</p>
2724
2725<dl>
2726  <dt><b><tt>trunc (CST to TYPE)</tt></b></dt>
2727  <dd>Truncate a constant to another type. The bit size of CST must be larger
2728      than the bit size of TYPE. Both types must be integers.</dd>
2729
2730  <dt><b><tt>zext (CST to TYPE)</tt></b></dt>
2731  <dd>Zero extend a constant to another type. The bit size of CST must be
2732      smaller than the bit size of TYPE.  Both types must be integers.</dd>
2733
2734  <dt><b><tt>sext (CST to TYPE)</tt></b></dt>
2735  <dd>Sign extend a constant to another type. The bit size of CST must be
2736      smaller than the bit size of TYPE.  Both types must be integers.</dd>
2737
2738  <dt><b><tt>fptrunc (CST to TYPE)</tt></b></dt>
2739  <dd>Truncate a floating point constant to another floating point type. The
2740      size of CST must be larger than the size of TYPE. Both types must be
2741      floating point.</dd>
2742
2743  <dt><b><tt>fpext (CST to TYPE)</tt></b></dt>
2744  <dd>Floating point extend a constant to another type. The size of CST must be
2745      smaller or equal to the size of TYPE. Both types must be floating
2746      point.</dd>
2747
2748  <dt><b><tt>fptoui (CST to TYPE)</tt></b></dt>
2749  <dd>Convert a floating point constant to the corresponding unsigned integer
2750      constant. TYPE must be a scalar or vector integer type. CST must be of
2751      scalar or vector floating point type. Both CST and TYPE must be scalars,
2752      or vectors of the same number of elements. If the value won't fit in the
2753      integer type, the results are undefined.</dd>
2754
2755  <dt><b><tt>fptosi (CST to TYPE)</tt></b></dt>
2756  <dd>Convert a floating point constant to the corresponding signed integer
2757      constant.  TYPE must be a scalar or vector integer type. CST must be of
2758      scalar or vector floating point type. Both CST and TYPE must be scalars,
2759      or vectors of the same number of elements. If the value won't fit in the
2760      integer type, the results are undefined.</dd>
2761
2762  <dt><b><tt>uitofp (CST to TYPE)</tt></b></dt>
2763  <dd>Convert an unsigned integer constant to the corresponding floating point
2764      constant. TYPE must be a scalar or vector floating point type. CST must be
2765      of scalar or vector integer type. Both CST and TYPE must be scalars, or
2766      vectors of the same number of elements. If the value won't fit in the
2767      floating point type, the results are undefined.</dd>
2768
2769  <dt><b><tt>sitofp (CST to TYPE)</tt></b></dt>
2770  <dd>Convert a signed integer constant to the corresponding floating point
2771      constant. TYPE must be a scalar or vector floating point type. CST must be
2772      of scalar or vector integer type. Both CST and TYPE must be scalars, or
2773      vectors of the same number of elements. If the value won't fit in the
2774      floating point type, the results are undefined.</dd>
2775
2776  <dt><b><tt>ptrtoint (CST to TYPE)</tt></b></dt>
2777  <dd>Convert a pointer typed constant to the corresponding integer constant
2778      <tt>TYPE</tt> must be an integer type. <tt>CST</tt> must be of pointer
2779      type. The <tt>CST</tt> value is zero extended, truncated, or unchanged to
2780      make it fit in <tt>TYPE</tt>.</dd>
2781
2782  <dt><b><tt>inttoptr (CST to TYPE)</tt></b></dt>
2783  <dd>Convert an integer constant to a pointer constant.  TYPE must be a pointer
2784      type.  CST must be of integer type. The CST value is zero extended,
2785      truncated, or unchanged to make it fit in a pointer size. This one is
2786      <i>really</i> dangerous!</dd>
2787
2788  <dt><b><tt>bitcast (CST to TYPE)</tt></b></dt>
2789  <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
2790      are the same as those for the <a href="#i_bitcast">bitcast
2791      instruction</a>.</dd>
2792
2793  <dt><b><tt>getelementptr (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2794  <dt><b><tt>getelementptr inbounds (CSTPTR, IDX0, IDX1, ...)</tt></b></dt>
2795  <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
2796      constants.  As with the <a href="#i_getelementptr">getelementptr</a>
2797      instruction, the index list may have zero or more indexes, which are
2798      required to make sense for the type of "CSTPTR".</dd>
2799
2800  <dt><b><tt>select (COND, VAL1, VAL2)</tt></b></dt>
2801  <dd>Perform the <a href="#i_select">select operation</a> on constants.</dd>
2802
2803  <dt><b><tt>icmp COND (VAL1, VAL2)</tt></b></dt>
2804  <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
2805
2806  <dt><b><tt>fcmp COND (VAL1, VAL2)</tt></b></dt>
2807  <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
2808
2809  <dt><b><tt>extractelement (VAL, IDX)</tt></b></dt>
2810  <dd>Perform the <a href="#i_extractelement">extractelement operation</a> on
2811      constants.</dd>
2812
2813  <dt><b><tt>insertelement (VAL, ELT, IDX)</tt></b></dt>
2814  <dd>Perform the <a href="#i_insertelement">insertelement operation</a> on
2815    constants.</dd>
2816
2817  <dt><b><tt>shufflevector (VEC1, VEC2, IDXMASK)</tt></b></dt>
2818  <dd>Perform the <a href="#i_shufflevector">shufflevector operation</a> on
2819      constants.</dd>
2820
2821  <dt><b><tt>extractvalue (VAL, IDX0, IDX1, ...)</tt></b></dt>
2822  <dd>Perform the <a href="#i_extractvalue">extractvalue operation</a> on
2823    constants. The index list is interpreted in a similar manner as indices in
2824    a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2825    index value must be specified.</dd>
2826
2827  <dt><b><tt>insertvalue (VAL, ELT, IDX0, IDX1, ...)</tt></b></dt>
2828  <dd>Perform the <a href="#i_insertvalue">insertvalue operation</a> on
2829    constants. The index list is interpreted in a similar manner as indices in
2830    a '<a href="#i_getelementptr">getelementptr</a>' operation. At least one
2831    index value must be specified.</dd>
2832
2833  <dt><b><tt>OPCODE (LHS, RHS)</tt></b></dt>
2834  <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2835      be any of the <a href="#binaryops">binary</a>
2836      or <a href="#bitwiseops">bitwise binary</a> operations.  The constraints
2837      on operands are the same as those for the corresponding instruction
2838      (e.g. no bitwise operations on floating point values are allowed).</dd>
2839</dl>
2840
2841</div>
2842
2843</div>
2844
2845<!-- *********************************************************************** -->
2846<h2><a name="othervalues">Other Values</a></h2>
2847<!-- *********************************************************************** -->
2848<div>
2849<!-- ======================================================================= -->
2850<h3>
2851<a name="inlineasm">Inline Assembler Expressions</a>
2852</h3>
2853
2854<div>
2855
2856<p>LLVM supports inline assembler expressions (as opposed
2857   to <a href="#moduleasm">Module-Level Inline Assembly</a>) through the use of
2858   a special value.  This value represents the inline assembler as a string
2859   (containing the instructions to emit), a list of operand constraints (stored
2860   as a string), a flag that indicates whether or not the inline asm
2861   expression has side effects, and a flag indicating whether the function
2862   containing the asm needs to align its stack conservatively.  An example
2863   inline assembler expression is:</p>
2864
2865<pre class="doc_code">
2866i32 (i32) asm "bswap $0", "=r,r"
2867</pre>
2868
2869<p>Inline assembler expressions may <b>only</b> be used as the callee operand of
2870   a <a href="#i_call"><tt>call</tt></a> or an
2871   <a href="#i_invoke"><tt>invoke</tt></a> instruction.
2872   Thus, typically we have:</p>
2873
2874<pre class="doc_code">
2875%X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
2876</pre>
2877
2878<p>Inline asms with side effects not visible in the constraint list must be
2879   marked as having side effects.  This is done through the use of the
2880   '<tt>sideeffect</tt>' keyword, like so:</p>
2881
2882<pre class="doc_code">
2883call void asm sideeffect "eieio", ""()
2884</pre>
2885
2886<p>In some cases inline asms will contain code that will not work unless the
2887   stack is aligned in some way, such as calls or SSE instructions on x86,
2888   yet will not contain code that does that alignment within the asm.
2889   The compiler should make conservative assumptions about what the asm might
2890   contain and should generate its usual stack alignment code in the prologue
2891   if the '<tt>alignstack</tt>' keyword is present:</p>
2892
2893<pre class="doc_code">
2894call void asm alignstack "eieio", ""()
2895</pre>
2896
2897<p>Inline asms also support using non-standard assembly dialects.  The assumed
2898   dialect is ATT.  When the '<tt>inteldialect</tt>' keyword is present, the
2899   inline asm is using the Intel dialect.  Currently, ATT and Intel are the
2900   only supported dialects.  An example is:</p>
2901
2902<pre class="doc_code">
2903call void asm inteldialect "eieio", ""()
2904</pre>
2905
2906<p>If multiple keywords appear the '<tt>sideeffect</tt>' keyword must come
2907   first, the '<tt>alignstack</tt>' keyword second and the
2908   '<tt>inteldialect</tt>' keyword last.</p>
2909
2910<!--
2911<p>TODO: The format of the asm and constraints string still need to be
2912   documented here.  Constraints on what can be done (e.g. duplication, moving,
2913   etc need to be documented).  This is probably best done by reference to
2914   another document that covers inline asm from a holistic perspective.</p>
2915  -->
2916
2917<!-- _______________________________________________________________________ -->
2918<h4>
2919  <a name="inlineasm_md">Inline Asm Metadata</a>
2920</h4>
2921
2922<div>
2923
2924<p>The call instructions that wrap inline asm nodes may have a
2925   "<tt>!srcloc</tt>" MDNode attached to it that contains a list of constant
2926   integers.  If present, the code generator will use the integer as the
2927   location cookie value when report errors through the <tt>LLVMContext</tt>
2928   error reporting mechanisms.  This allows a front-end to correlate backend
2929   errors that occur with inline asm back to the source code that produced it.
2930   For example:</p>
2931
2932<pre class="doc_code">
2933call void asm sideeffect "something bad", ""()<b>, !srcloc !42</b>
2934...
2935!42 = !{ i32 1234567 }
2936</pre>
2937
2938<p>It is up to the front-end to make sense of the magic numbers it places in the
2939   IR. If the MDNode contains multiple constants, the code generator will use
2940   the one that corresponds to the line of the asm that the error occurs on.</p>
2941
2942</div>
2943
2944</div>
2945
2946<!-- ======================================================================= -->
2947<h3>
2948  <a name="metadata">Metadata Nodes and Metadata Strings</a>
2949</h3>
2950
2951<div>
2952
2953<p>LLVM IR allows metadata to be attached to instructions in the program that
2954   can convey extra information about the code to the optimizers and code
2955   generator.  One example application of metadata is source-level debug
2956   information.  There are two metadata primitives: strings and nodes. All
2957   metadata has the <tt>metadata</tt> type and is identified in syntax by a
2958   preceding exclamation point ('<tt>!</tt>').</p>
2959
2960<p>A metadata string is a string surrounded by double quotes.  It can contain
2961   any character by escaping non-printable characters with "<tt>\xx</tt>" where
2962   "<tt>xx</tt>" is the two digit hex code.  For example:
2963   "<tt>!"test\00"</tt>".</p>
2964
2965<p>Metadata nodes are represented with notation similar to structure constants
2966   (a comma separated list of elements, surrounded by braces and preceded by an
2967   exclamation point). Metadata nodes can have any values as their operand. For
2968   example:</p>
2969
2970<div class="doc_code">
2971<pre>
2972!{ metadata !"test\00", i32 10}
2973</pre>
2974</div>
2975
2976<p>A <a href="#namedmetadatastructure">named metadata</a> is a collection of
2977   metadata nodes, which can be looked up in the module symbol table. For
2978   example:</p>
2979
2980<div class="doc_code">
2981<pre>
2982!foo =  metadata !{!4, !3}
2983</pre>
2984</div>
2985
2986<p>Metadata can be used as function arguments. Here <tt>llvm.dbg.value</tt>
2987   function is using two metadata arguments:</p>
2988
2989<div class="doc_code">
2990<pre>
2991call void @llvm.dbg.value(metadata !24, i64 0, metadata !25)
2992</pre>
2993</div>
2994
2995<p>Metadata can be attached with an instruction. Here metadata <tt>!21</tt> is
2996   attached to the <tt>add</tt> instruction using the <tt>!dbg</tt>
2997   identifier:</p>
2998
2999<div class="doc_code">
3000<pre>
3001%indvar.next = add i64 %indvar, 1, !dbg !21
3002</pre>
3003</div>
3004
3005<p>More information about specific metadata nodes recognized by the optimizers
3006   and code generator is found below.</p>
3007
3008<!-- _______________________________________________________________________ -->
3009<h4>
3010  <a name="tbaa">'<tt>tbaa</tt>' Metadata</a>
3011</h4>
3012
3013<div>
3014
3015<p>In LLVM IR, memory does not have types, so LLVM's own type system is not
3016   suitable for doing TBAA. Instead, metadata is added to the IR to describe
3017   a type system of a higher level language. This can be used to implement
3018   typical C/C++ TBAA, but it can also be used to implement custom alias
3019   analysis behavior for other languages.</p>
3020
3021<p>The current metadata format is very simple. TBAA metadata nodes have up to
3022   three fields, e.g.:</p>
3023
3024<div class="doc_code">
3025<pre>
3026!0 = metadata !{ metadata !"an example type tree" }
3027!1 = metadata !{ metadata !"int", metadata !0 }
3028!2 = metadata !{ metadata !"float", metadata !0 }
3029!3 = metadata !{ metadata !"const float", metadata !2, i64 1 }
3030</pre>
3031</div>
3032
3033<p>The first field is an identity field. It can be any value, usually
3034   a metadata string, which uniquely identifies the type. The most important
3035   name in the tree is the name of the root node. Two trees with
3036   different root node names are entirely disjoint, even if they
3037   have leaves with common names.</p>
3038
3039<p>The second field identifies the type's parent node in the tree, or
3040   is null or omitted for a root node. A type is considered to alias
3041   all of its descendants and all of its ancestors in the tree. Also,
3042   a type is considered to alias all types in other trees, so that
3043   bitcode produced from multiple front-ends is handled conservatively.</p>
3044
3045<p>If the third field is present, it's an integer which if equal to 1
3046   indicates that the type is "constant" (meaning
3047   <tt>pointsToConstantMemory</tt> should return true; see
3048   <a href="AliasAnalysis.html#OtherItfs">other useful
3049   <tt>AliasAnalysis</tt> methods</a>).</p>
3050
3051</div>
3052
3053<!-- _______________________________________________________________________ -->
3054<h4>
3055  <a name="fpmath">'<tt>fpmath</tt>' Metadata</a>
3056</h4>
3057
3058<div>
3059
3060<p><tt>fpmath</tt> metadata may be attached to any instruction of floating point
3061  type.  It can be used to express the maximum acceptable error in the result of
3062  that instruction, in ULPs, thus potentially allowing the compiler to use a
3063  more efficient but less accurate method of computing it.  ULP is defined as
3064  follows:</p>
3065
3066<blockquote>
3067
3068<p>If <tt>x</tt> is a real number that lies between two finite consecutive
3069   floating-point numbers <tt>a</tt> and <tt>b</tt>, without being equal to one
3070   of them, then <tt>ulp(x) = |b - a|</tt>, otherwise <tt>ulp(x)</tt> is the
3071   distance between the two non-equal finite floating-point numbers nearest
3072   <tt>x</tt>. Moreover, <tt>ulp(NaN)</tt> is <tt>NaN</tt>.</p>
3073
3074</blockquote>
3075
3076<p>The metadata node shall consist of a single positive floating point number
3077   representing the maximum relative error, for example:</p>
3078
3079<div class="doc_code">
3080<pre>
3081!0 = metadata !{ float 2.5 } ; maximum acceptable inaccuracy is 2.5 ULPs
3082</pre>
3083</div>
3084
3085</div>
3086
3087<!-- _______________________________________________________________________ -->
3088<h4>
3089  <a name="range">'<tt>range</tt>' Metadata</a>
3090</h4>
3091
3092<div>
3093<p><tt>range</tt> metadata may be attached only to loads of integer types. It
3094   expresses the possible ranges the loaded value is in. The ranges are
3095   represented with a flattened list of integers. The loaded value is known to
3096   be in the union of the ranges defined by each consecutive pair. Each pair
3097   has the following properties:</p>
3098<ul>
3099   <li>The type must match the type loaded by the instruction.</li>
3100   <li>The pair <tt>a,b</tt> represents the range <tt>[a,b)</tt>.</li>
3101   <li>Both <tt>a</tt> and <tt>b</tt> are constants.</li>
3102   <li>The range is allowed to wrap.</li>
3103   <li>The range should not represent the full or empty set. That is,
3104       <tt>a!=b</tt>. </li>
3105</ul>
3106<p> In addition, the pairs must be in signed order of the lower bound and
3107  they must be non-contiguous.</p>
3108
3109<p>Examples:</p>
3110<div class="doc_code">
3111<pre>
3112  %a = load i8* %x, align 1, !range !0 ; Can only be 0 or 1
3113  %b = load i8* %y, align 1, !range !1 ; Can only be 255 (-1), 0 or 1
3114  %c = load i8* %z, align 1, !range !2 ; Can only be 0, 1, 3, 4 or 5
3115  %d = load i8* %z, align 1, !range !3 ; Can only be -2, -1, 3, 4 or 5
3116...
3117!0 = metadata !{ i8 0, i8 2 }
3118!1 = metadata !{ i8 255, i8 2 }
3119!2 = metadata !{ i8 0, i8 2, i8 3, i8 6 }
3120!3 = metadata !{ i8 -2, i8 0, i8 3, i8 6 }
3121</pre>
3122</div>
3123</div>
3124</div>
3125
3126</div>
3127
3128<!-- *********************************************************************** -->
3129<h2>
3130  <a name="module_flags">Module Flags Metadata</a>
3131</h2>
3132<!-- *********************************************************************** -->
3133
3134<div>
3135
3136<p>Information about the module as a whole is difficult to convey to LLVM's
3137   subsystems. The LLVM IR isn't sufficient to transmit this
3138   information. The <tt>llvm.module.flags</tt> named metadata exists in order to
3139   facilitate this. These flags are in the form of key / value pairs &mdash;
3140   much like a dictionary &mdash; making it easy for any subsystem who cares
3141   about a flag to look it up.</p>
3142
3143<p>The <tt>llvm.module.flags</tt> metadata contains a list of metadata
3144   triplets. Each triplet has the following form:</p>
3145
3146<ul>
3147  <li>The first element is a <i>behavior</i> flag, which specifies the behavior
3148      when two (or more) modules are merged together, and it encounters two (or
3149      more) metadata with the same ID. The supported behaviors are described
3150      below.</li>
3151
3152  <li>The second element is a metadata string that is a unique ID for the
3153      metadata. How each ID is interpreted is documented below.</li>
3154
3155  <li>The third element is the value of the flag.</li>
3156</ul>
3157
3158<p>When two (or more) modules are merged together, the resulting
3159   <tt>llvm.module.flags</tt> metadata is the union of the
3160   modules' <tt>llvm.module.flags</tt> metadata. The only exception being a flag
3161   with the <i>Override</i> behavior, which may override another flag's value
3162   (see below).</p>
3163
3164<p>The following behaviors are supported:</p>
3165
3166<table border="1" cellspacing="0" cellpadding="4">
3167  <tbody>
3168    <tr>
3169      <th>Value</th>
3170      <th>Behavior</th>
3171    </tr>
3172    <tr>
3173      <td>1</td>
3174      <td align="left">
3175        <dl>
3176          <dt><b>Error</b></dt>
3177          <dd>Emits an error if two values disagree. It is an error to have an ID
3178              with both an Error and a Warning behavior.</dd>
3179        </dl>
3180      </td>
3181    </tr>
3182    <tr>
3183      <td>2</td>
3184      <td align="left">
3185        <dl>
3186          <dt><b>Warning</b></dt>
3187          <dd>Emits a warning if two values disagree.</dd>
3188        </dl>
3189      </td>
3190    </tr>
3191    <tr>
3192      <td>3</td>
3193      <td align="left">
3194        <dl>
3195          <dt><b>Require</b></dt>
3196          <dd>Emits an error when the specified value is not present or doesn't
3197              have the specified value. It is an error for two (or more)
3198              <tt>llvm.module.flags</tt> with the same ID to have the Require
3199              behavior but different values. There may be multiple Require flags
3200              per ID.</dd>
3201        </dl>
3202      </td>
3203    </tr>
3204    <tr>
3205      <td>4</td>
3206      <td align="left">
3207        <dl>
3208          <dt><b>Override</b></dt>
3209          <dd>Uses the specified value if the two values disagree. It is an
3210              error for two (or more) <tt>llvm.module.flags</tt> with the same
3211              ID to have the Override behavior but different values.</dd>
3212        </dl>
3213      </td>
3214    </tr>
3215  </tbody>
3216</table>
3217
3218<p>An example of module flags:</p>
3219
3220<pre class="doc_code">
3221!0 = metadata !{ i32 1, metadata !"foo", i32 1 }
3222!1 = metadata !{ i32 4, metadata !"bar", i32 37 }
3223!2 = metadata !{ i32 2, metadata !"qux", i32 42 }
3224!3 = metadata !{ i32 3, metadata !"qux",
3225  metadata !{
3226    metadata !"foo", i32 1
3227  }
3228}
3229!llvm.module.flags = !{ !0, !1, !2, !3 }
3230</pre>
3231
3232<ul>
3233  <li><p>Metadata <tt>!0</tt> has the ID <tt>!"foo"</tt> and the value '1'. The
3234         behavior if two or more <tt>!"foo"</tt> flags are seen is to emit an
3235         error if their values are not equal.</p></li>
3236
3237  <li><p>Metadata <tt>!1</tt> has the ID <tt>!"bar"</tt> and the value '37'. The
3238         behavior if two or more <tt>!"bar"</tt> flags are seen is to use the
3239         value '37' if their values are not equal.</p></li>
3240
3241  <li><p>Metadata <tt>!2</tt> has the ID <tt>!"qux"</tt> and the value '42'. The
3242         behavior if two or more <tt>!"qux"</tt> flags are seen is to emit a
3243         warning if their values are not equal.</p></li>
3244
3245  <li><p>Metadata <tt>!3</tt> has the ID <tt>!"qux"</tt> and the value:</p>
3246
3247<pre class="doc_code">
3248metadata !{ metadata !"foo", i32 1 }
3249</pre>
3250
3251      <p>The behavior is to emit an error if the <tt>llvm.module.flags</tt> does
3252         not contain a flag with the ID <tt>!"foo"</tt> that has the value
3253         '1'. If two or more <tt>!"qux"</tt> flags exist, then they must have
3254         the same value or an error will be issued.</p></li>
3255</ul>
3256
3257
3258<!-- ======================================================================= -->
3259<h3>
3260<a name="objc_gc_flags">Objective-C Garbage Collection Module Flags Metadata</a>
3261</h3>
3262
3263<div>
3264
3265<p>On the Mach-O platform, Objective-C stores metadata about garbage collection
3266   in a special section called "image info". The metadata consists of a version
3267   number and a bitmask specifying what types of garbage collection are
3268   supported (if any) by the file. If two or more modules are linked together
3269   their garbage collection metadata needs to be merged rather than appended
3270   together.</p>
3271
3272<p>The Objective-C garbage collection module flags metadata consists of the
3273   following key-value pairs:</p>
3274
3275<table border="1" cellspacing="0" cellpadding="4">
3276  <col width="30%">
3277  <tbody>
3278    <tr>
3279      <th>Key</th>
3280      <th>Value</th>
3281    </tr>
3282    <tr>
3283      <td><tt>Objective-C&nbsp;Version</tt></td>
3284      <td align="left"><b>[Required]</b> &mdash; The Objective-C ABI
3285         version. Valid values are 1 and 2.</td>
3286    </tr>
3287    <tr>
3288      <td><tt>Objective-C&nbsp;Image&nbsp;Info&nbsp;Version</tt></td>
3289      <td align="left"><b>[Required]</b> &mdash; The version of the image info
3290         section. Currently always 0.</td>
3291    </tr>
3292    <tr>
3293      <td><tt>Objective-C&nbsp;Image&nbsp;Info&nbsp;Section</tt></td>
3294      <td align="left"><b>[Required]</b> &mdash; The section to place the
3295         metadata. Valid values are <tt>"__OBJC, __image_info, regular"</tt> for
3296         Objective-C ABI version 1, and <tt>"__DATA,__objc_imageinfo, regular,
3297         no_dead_strip"</tt> for Objective-C ABI version 2.</td>
3298    </tr>
3299    <tr>
3300      <td><tt>Objective-C&nbsp;Garbage&nbsp;Collection</tt></td>
3301      <td align="left"><b>[Required]</b> &mdash; Specifies whether garbage
3302          collection is supported or not. Valid values are 0, for no garbage
3303          collection, and 2, for garbage collection supported.</td>
3304    </tr>
3305    <tr>
3306      <td><tt>Objective-C&nbsp;GC&nbsp;Only</tt></td>
3307      <td align="left"><b>[Optional]</b> &mdash; Specifies that only garbage
3308         collection is supported. If present, its value must be 6. This flag
3309         requires that the <tt>Objective-C Garbage Collection</tt> flag have the
3310         value 2.</td>
3311    </tr>
3312  </tbody>
3313</table>
3314
3315<p>Some important flag interactions:</p>
3316
3317<ul>
3318  <li>If a module with <tt>Objective-C Garbage Collection</tt> set to 0 is
3319      merged with a module with <tt>Objective-C Garbage Collection</tt> set to
3320      2, then the resulting module has the <tt>Objective-C Garbage
3321      Collection</tt> flag set to 0.</li>
3322
3323  <li>A module with <tt>Objective-C Garbage Collection</tt> set to 0 cannot be
3324      merged with a module with <tt>Objective-C GC Only</tt> set to 6.</li>
3325</ul>
3326
3327</div>
3328
3329</div>
3330
3331<!-- *********************************************************************** -->
3332<h2>
3333  <a name="intrinsic_globals">Intrinsic Global Variables</a>
3334</h2>
3335<!-- *********************************************************************** -->
3336<div>
3337<p>LLVM has a number of "magic" global variables that contain data that affect
3338code generation or other IR semantics.  These are documented here.  All globals
3339of this sort should have a section specified as "<tt>llvm.metadata</tt>".  This
3340section and all globals that start with "<tt>llvm.</tt>" are reserved for use
3341by LLVM.</p>
3342
3343<!-- ======================================================================= -->
3344<h3>
3345<a name="intg_used">The '<tt>llvm.used</tt>' Global Variable</a>
3346</h3>
3347
3348<div>
3349
3350<p>The <tt>@llvm.used</tt> global is an array with i8* element type which has <a
3351href="#linkage_appending">appending linkage</a>.  This array contains a list of
3352pointers to global variables and functions which may optionally have a pointer
3353cast formed of bitcast or getelementptr.  For example, a legal use of it is:</p>
3354
3355<div class="doc_code">
3356<pre>
3357@X = global i8 4
3358@Y = global i32 123
3359
3360@llvm.used = appending global [2 x i8*] [
3361   i8* @X,
3362   i8* bitcast (i32* @Y to i8*)
3363], section "llvm.metadata"
3364</pre>
3365</div>
3366
3367<p>If a global variable appears in the <tt>@llvm.used</tt> list, then the
3368   compiler, assembler, and linker are required to treat the symbol as if there
3369   is a reference to the global that it cannot see.  For example, if a variable
3370   has internal linkage and no references other than that from
3371   the <tt>@llvm.used</tt> list, it cannot be deleted.  This is commonly used to
3372   represent references from inline asms and other things the compiler cannot
3373   "see", and corresponds to "<tt>attribute((used))</tt>" in GNU C.</p>
3374
3375<p>On some targets, the code generator must emit a directive to the assembler or
3376   object file to prevent the assembler and linker from molesting the
3377   symbol.</p>
3378
3379</div>
3380
3381<!-- ======================================================================= -->
3382<h3>
3383  <a name="intg_compiler_used">
3384    The '<tt>llvm.compiler.used</tt>' Global Variable
3385  </a>
3386</h3>
3387
3388<div>
3389
3390<p>The <tt>@llvm.compiler.used</tt> directive is the same as the
3391   <tt>@llvm.used</tt> directive, except that it only prevents the compiler from
3392   touching the symbol.  On targets that support it, this allows an intelligent
3393   linker to optimize references to the symbol without being impeded as it would
3394   be by <tt>@llvm.used</tt>.</p>
3395
3396<p>This is a rare construct that should only be used in rare circumstances, and
3397   should not be exposed to source languages.</p>
3398
3399</div>
3400
3401<!-- ======================================================================= -->
3402<h3>
3403<a name="intg_global_ctors">The '<tt>llvm.global_ctors</tt>' Global Variable</a>
3404</h3>
3405
3406<div>
3407
3408<div class="doc_code">
3409<pre>
3410%0 = type { i32, void ()* }
3411@llvm.global_ctors = appending global [1 x %0] [%0 { i32 65535, void ()* @ctor }]
3412</pre>
3413</div>
3414
3415<p>The <tt>@llvm.global_ctors</tt> array contains a list of constructor
3416   functions and associated priorities.  The functions referenced by this array
3417   will be called in ascending order of priority (i.e. lowest first) when the
3418   module is loaded.  The order of functions with the same priority is not
3419   defined.</p>
3420
3421</div>
3422
3423<!-- ======================================================================= -->
3424<h3>
3425<a name="intg_global_dtors">The '<tt>llvm.global_dtors</tt>' Global Variable</a>
3426</h3>
3427
3428<div>
3429
3430<div class="doc_code">
3431<pre>
3432%0 = type { i32, void ()* }
3433@llvm.global_dtors = appending global [1 x %0] [%0 { i32 65535, void ()* @dtor }]
3434</pre>
3435</div>
3436
3437<p>The <tt>@llvm.global_dtors</tt> array contains a list of destructor functions
3438   and associated priorities.  The functions referenced by this array will be
3439   called in descending order of priority (i.e. highest first) when the module
3440   is loaded.  The order of functions with the same priority is not defined.</p>
3441
3442</div>
3443
3444</div>
3445
3446<!-- *********************************************************************** -->
3447<h2><a name="instref">Instruction Reference</a></h2>
3448<!-- *********************************************************************** -->
3449
3450<div>
3451
3452<p>The LLVM instruction set consists of several different classifications of
3453   instructions: <a href="#terminators">terminator
3454   instructions</a>, <a href="#binaryops">binary instructions</a>,
3455   <a href="#bitwiseops">bitwise binary instructions</a>,
3456   <a href="#memoryops">memory instructions</a>, and
3457   <a href="#otherops">other instructions</a>.</p>
3458
3459<!-- ======================================================================= -->
3460<h3>
3461  <a name="terminators">Terminator Instructions</a>
3462</h3>
3463
3464<div>
3465
3466<p>As mentioned <a href="#functionstructure">previously</a>, every basic block
3467   in a program ends with a "Terminator" instruction, which indicates which
3468   block should be executed after the current block is finished. These
3469   terminator instructions typically yield a '<tt>void</tt>' value: they produce
3470   control flow, not values (the one exception being the
3471   '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
3472
3473<p>The terminator instructions are:
3474   '<a href="#i_ret"><tt>ret</tt></a>',
3475   '<a href="#i_br"><tt>br</tt></a>',
3476   '<a href="#i_switch"><tt>switch</tt></a>',
3477   '<a href="#i_indirectbr"><tt>indirectbr</tt></a>',
3478   '<a href="#i_invoke"><tt>invoke</tt></a>',
3479   '<a href="#i_resume"><tt>resume</tt></a>', and
3480   '<a href="#i_unreachable"><tt>unreachable</tt></a>'.</p>
3481
3482<!-- _______________________________________________________________________ -->
3483<h4>
3484  <a name="i_ret">'<tt>ret</tt>' Instruction</a>
3485</h4>
3486
3487<div>
3488
3489<h5>Syntax:</h5>
3490<pre>
3491  ret &lt;type&gt; &lt;value&gt;       <i>; Return a value from a non-void function</i>
3492  ret void                 <i>; Return from void function</i>
3493</pre>
3494
3495<h5>Overview:</h5>
3496<p>The '<tt>ret</tt>' instruction is used to return control flow (and optionally
3497   a value) from a function back to the caller.</p>
3498
3499<p>There are two forms of the '<tt>ret</tt>' instruction: one that returns a
3500   value and then causes control flow, and one that just causes control flow to
3501   occur.</p>
3502
3503<h5>Arguments:</h5>
3504<p>The '<tt>ret</tt>' instruction optionally accepts a single argument, the
3505   return value. The type of the return value must be a
3506   '<a href="#t_firstclass">first class</a>' type.</p>
3507
3508<p>A function is not <a href="#wellformed">well formed</a> if it it has a
3509   non-void return type and contains a '<tt>ret</tt>' instruction with no return
3510   value or a return value with a type that does not match its type, or if it
3511   has a void return type and contains a '<tt>ret</tt>' instruction with a
3512   return value.</p>
3513
3514<h5>Semantics:</h5>
3515<p>When the '<tt>ret</tt>' instruction is executed, control flow returns back to
3516   the calling function's context.  If the caller is a
3517   "<a href="#i_call"><tt>call</tt></a>" instruction, execution continues at the
3518   instruction after the call.  If the caller was an
3519   "<a href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues at
3520   the beginning of the "normal" destination block.  If the instruction returns
3521   a value, that value shall set the call or invoke instruction's return
3522   value.</p>
3523
3524<h5>Example:</h5>
3525<pre>
3526  ret i32 5                       <i>; Return an integer value of 5</i>
3527  ret void                        <i>; Return from a void function</i>
3528  ret { i32, i8 } { i32 4, i8 2 } <i>; Return a struct of values 4 and 2</i>
3529</pre>
3530
3531</div>
3532<!-- _______________________________________________________________________ -->
3533<h4>
3534  <a name="i_br">'<tt>br</tt>' Instruction</a>
3535</h4>
3536
3537<div>
3538
3539<h5>Syntax:</h5>
3540<pre>
3541  br i1 &lt;cond&gt;, label &lt;iftrue&gt;, label &lt;iffalse&gt;
3542  br label &lt;dest&gt;          <i>; Unconditional branch</i>
3543</pre>
3544
3545<h5>Overview:</h5>
3546<p>The '<tt>br</tt>' instruction is used to cause control flow to transfer to a
3547   different basic block in the current function.  There are two forms of this
3548   instruction, corresponding to a conditional branch and an unconditional
3549   branch.</p>
3550
3551<h5>Arguments:</h5>
3552<p>The conditional branch form of the '<tt>br</tt>' instruction takes a single
3553   '<tt>i1</tt>' value and two '<tt>label</tt>' values.  The unconditional form
3554   of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>' value as a
3555   target.</p>
3556
3557<h5>Semantics:</h5>
3558<p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
3559   argument is evaluated.  If the value is <tt>true</tt>, control flows to the
3560   '<tt>iftrue</tt>' <tt>label</tt> argument.  If "cond" is <tt>false</tt>,
3561   control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
3562
3563<h5>Example:</h5>
3564<pre>
3565Test:
3566  %cond = <a href="#i_icmp">icmp</a> eq i32 %a, %b
3567  br i1 %cond, label %IfEqual, label %IfUnequal
3568IfEqual:
3569  <a href="#i_ret">ret</a> i32 1
3570IfUnequal:
3571  <a href="#i_ret">ret</a> i32 0
3572</pre>
3573
3574</div>
3575
3576<!-- _______________________________________________________________________ -->
3577<h4>
3578   <a name="i_switch">'<tt>switch</tt>' Instruction</a>
3579</h4>
3580
3581<div>
3582
3583<h5>Syntax:</h5>
3584<pre>
3585  switch &lt;intty&gt; &lt;value&gt;, label &lt;defaultdest&gt; [ &lt;intty&gt; &lt;val&gt;, label &lt;dest&gt; ... ]
3586</pre>
3587
3588<h5>Overview:</h5>
3589<p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
3590   several different places.  It is a generalization of the '<tt>br</tt>'
3591   instruction, allowing a branch to occur to one of many possible
3592   destinations.</p>
3593
3594<h5>Arguments:</h5>
3595<p>The '<tt>switch</tt>' instruction uses three parameters: an integer
3596   comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination,
3597   and an array of pairs of comparison value constants and '<tt>label</tt>'s.
3598   The table is not allowed to contain duplicate constant entries.</p>
3599
3600<h5>Semantics:</h5>
3601<p>The <tt>switch</tt> instruction specifies a table of values and
3602   destinations. When the '<tt>switch</tt>' instruction is executed, this table
3603   is searched for the given value.  If the value is found, control flow is
3604   transferred to the corresponding destination; otherwise, control flow is
3605   transferred to the default destination.</p>
3606
3607<h5>Implementation:</h5>
3608<p>Depending on properties of the target machine and the particular
3609   <tt>switch</tt> instruction, this instruction may be code generated in
3610   different ways.  For example, it could be generated as a series of chained
3611   conditional branches or with a lookup table.</p>
3612
3613<h5>Example:</h5>
3614<pre>
3615 <i>; Emulate a conditional br instruction</i>
3616 %Val = <a href="#i_zext">zext</a> i1 %value to i32
3617 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
3618
3619 <i>; Emulate an unconditional br instruction</i>
3620 switch i32 0, label %dest [ ]
3621
3622 <i>; Implement a jump table:</i>
3623 switch i32 %val, label %otherwise [ i32 0, label %onzero
3624                                     i32 1, label %onone
3625                                     i32 2, label %ontwo ]
3626</pre>
3627
3628</div>
3629
3630
3631<!-- _______________________________________________________________________ -->
3632<h4>
3633   <a name="i_indirectbr">'<tt>indirectbr</tt>' Instruction</a>
3634</h4>
3635
3636<div>
3637
3638<h5>Syntax:</h5>
3639<pre>
3640  indirectbr &lt;somety&gt;* &lt;address&gt;, [ label &lt;dest1&gt;, label &lt;dest2&gt;, ... ]
3641</pre>
3642
3643<h5>Overview:</h5>
3644
3645<p>The '<tt>indirectbr</tt>' instruction implements an indirect branch to a label
3646   within the current function, whose address is specified by
3647   "<tt>address</tt>".  Address must be derived from a <a
3648   href="#blockaddress">blockaddress</a> constant.</p>
3649
3650<h5>Arguments:</h5>
3651
3652<p>The '<tt>address</tt>' argument is the address of the label to jump to.  The
3653   rest of the arguments indicate the full set of possible destinations that the
3654   address may point to.  Blocks are allowed to occur multiple times in the
3655   destination list, though this isn't particularly useful.</p>
3656
3657<p>This destination list is required so that dataflow analysis has an accurate
3658   understanding of the CFG.</p>
3659
3660<h5>Semantics:</h5>
3661
3662<p>Control transfers to the block specified in the address argument.  All
3663   possible destination blocks must be listed in the label list, otherwise this
3664   instruction has undefined behavior.  This implies that jumps to labels
3665   defined in other functions have undefined behavior as well.</p>
3666
3667<h5>Implementation:</h5>
3668
3669<p>This is typically implemented with a jump through a register.</p>
3670
3671<h5>Example:</h5>
3672<pre>
3673 indirectbr i8* %Addr, [ label %bb1, label %bb2, label %bb3 ]
3674</pre>
3675
3676</div>
3677
3678
3679<!-- _______________________________________________________________________ -->
3680<h4>
3681  <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
3682</h4>
3683
3684<div>
3685
3686<h5>Syntax:</h5>
3687<pre>
3688  &lt;result&gt; = invoke [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>] &lt;ptr to function ty&gt; &lt;function ptr val&gt;(&lt;function args&gt;) [<a href="#fnattrs">fn attrs</a>]
3689                to label &lt;normal label&gt; unwind label &lt;exception label&gt;
3690</pre>
3691
3692<h5>Overview:</h5>
3693<p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
3694   function, with the possibility of control flow transfer to either the
3695   '<tt>normal</tt>' label or the '<tt>exception</tt>' label.  If the callee
3696   function returns with the "<tt><a href="#i_ret">ret</a></tt>" instruction,
3697   control flow will return to the "normal" label.  If the callee (or any
3698   indirect callees) returns via the "<a href="#i_resume"><tt>resume</tt></a>"
3699   instruction or other exception handling mechanism, control is interrupted and
3700   continued at the dynamically nearest "exception" label.</p>
3701
3702<p>The '<tt>exception</tt>' label is a
3703   <i><a href="ExceptionHandling.html#overview">landing pad</a></i> for the
3704   exception. As such, '<tt>exception</tt>' label is required to have the
3705   "<a href="#i_landingpad"><tt>landingpad</tt></a>" instruction, which contains
3706   the information about the behavior of the program after unwinding
3707   happens, as its first non-PHI instruction. The restrictions on the
3708   "<tt>landingpad</tt>" instruction's tightly couples it to the
3709   "<tt>invoke</tt>" instruction, so that the important information contained
3710   within the "<tt>landingpad</tt>" instruction can't be lost through normal
3711   code motion.</p>
3712
3713<h5>Arguments:</h5>
3714<p>This instruction requires several arguments:</p>
3715
3716<ol>
3717  <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
3718      convention</a> the call should use.  If none is specified, the call
3719      defaults to using C calling conventions.</li>
3720
3721  <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
3722      return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
3723      '<tt>inreg</tt>' attributes are valid here.</li>
3724
3725  <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
3726      function value being invoked.  In most cases, this is a direct function
3727      invocation, but indirect <tt>invoke</tt>s are just as possible, branching
3728      off an arbitrary pointer to function value.</li>
3729
3730  <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
3731      function to be invoked. </li>
3732
3733  <li>'<tt>function args</tt>': argument list whose types match the function
3734      signature argument types and parameter attributes. All arguments must be
3735      of <a href="#t_firstclass">first class</a> type. If the function
3736      signature indicates the function accepts a variable number of arguments,
3737      the extra arguments can be specified.</li>
3738
3739  <li>'<tt>normal label</tt>': the label reached when the called function
3740      executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
3741
3742  <li>'<tt>exception label</tt>': the label reached when a callee returns via
3743      the <a href="#i_resume"><tt>resume</tt></a> instruction or other exception
3744      handling mechanism.</li>
3745
3746  <li>The optional <a href="#fnattrs">function attributes</a> list. Only
3747      '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
3748      '<tt>readnone</tt>' attributes are valid here.</li>
3749</ol>
3750
3751<h5>Semantics:</h5>
3752<p>This instruction is designed to operate as a standard
3753   '<tt><a href="#i_call">call</a></tt>' instruction in most regards.  The
3754   primary difference is that it establishes an association with a label, which
3755   is used by the runtime library to unwind the stack.</p>
3756
3757<p>This instruction is used in languages with destructors to ensure that proper
3758   cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
3759   exception.  Additionally, this is important for implementation of
3760   '<tt>catch</tt>' clauses in high-level languages that support them.</p>
3761
3762<p>For the purposes of the SSA form, the definition of the value returned by the
3763   '<tt>invoke</tt>' instruction is deemed to occur on the edge from the current
3764   block to the "normal" label. If the callee unwinds then no return value is
3765   available.</p>
3766
3767<h5>Example:</h5>
3768<pre>
3769  %retval = invoke i32 @Test(i32 15) to label %Continue
3770              unwind label %TestCleanup              <i>; {i32}:retval set</i>
3771  %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
3772              unwind label %TestCleanup              <i>; {i32}:retval set</i>
3773</pre>
3774
3775</div>
3776
3777 <!-- _______________________________________________________________________ -->
3778
3779<h4>
3780  <a name="i_resume">'<tt>resume</tt>' Instruction</a>
3781</h4>
3782
3783<div>
3784
3785<h5>Syntax:</h5>
3786<pre>
3787  resume &lt;type&gt; &lt;value&gt;
3788</pre>
3789
3790<h5>Overview:</h5>
3791<p>The '<tt>resume</tt>' instruction is a terminator instruction that has no
3792   successors.</p>
3793
3794<h5>Arguments:</h5>
3795<p>The '<tt>resume</tt>' instruction requires one argument, which must have the
3796   same type as the result of any '<tt>landingpad</tt>' instruction in the same
3797   function.</p>
3798
3799<h5>Semantics:</h5>
3800<p>The '<tt>resume</tt>' instruction resumes propagation of an existing
3801   (in-flight) exception whose unwinding was interrupted with
3802   a <a href="#i_landingpad"><tt>landingpad</tt></a> instruction.</p>
3803
3804<h5>Example:</h5>
3805<pre>
3806  resume { i8*, i32 } %exn
3807</pre>
3808
3809</div>
3810
3811<!-- _______________________________________________________________________ -->
3812
3813<h4>
3814  <a name="i_unreachable">'<tt>unreachable</tt>' Instruction</a>
3815</h4>
3816
3817<div>
3818
3819<h5>Syntax:</h5>
3820<pre>
3821  unreachable
3822</pre>
3823
3824<h5>Overview:</h5>
3825<p>The '<tt>unreachable</tt>' instruction has no defined semantics.  This
3826   instruction is used to inform the optimizer that a particular portion of the
3827   code is not reachable.  This can be used to indicate that the code after a
3828   no-return function cannot be reached, and other facts.</p>
3829
3830<h5>Semantics:</h5>
3831<p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
3832
3833</div>
3834
3835</div>
3836
3837<!-- ======================================================================= -->
3838<h3>
3839  <a name="binaryops">Binary Operations</a>
3840</h3>
3841
3842<div>
3843
3844<p>Binary operators are used to do most of the computation in a program.  They
3845   require two operands of the same type, execute an operation on them, and
3846   produce a single value.  The operands might represent multiple data, as is
3847   the case with the <a href="#t_vector">vector</a> data type.  The result value
3848   has the same type as its operands.</p>
3849
3850<p>There are several different binary operators:</p>
3851
3852<!-- _______________________________________________________________________ -->
3853<h4>
3854  <a name="i_add">'<tt>add</tt>' Instruction</a>
3855</h4>
3856
3857<div>
3858
3859<h5>Syntax:</h5>
3860<pre>
3861  &lt;result&gt; = add &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt;          <i>; yields {ty}:result</i>
3862  &lt;result&gt; = add nuw &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt;      <i>; yields {ty}:result</i>
3863  &lt;result&gt; = add nsw &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt;      <i>; yields {ty}:result</i>
3864  &lt;result&gt; = add nuw nsw &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt;  <i>; yields {ty}:result</i>
3865</pre>
3866
3867<h5>Overview:</h5>
3868<p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
3869
3870<h5>Arguments:</h5>
3871<p>The two arguments to the '<tt>add</tt>' instruction must
3872   be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3873   integer values. Both arguments must have identical types.</p>
3874
3875<h5>Semantics:</h5>
3876<p>The value produced is the integer sum of the two operands.</p>
3877
3878<p>If the sum has unsigned overflow, the result returned is the mathematical
3879   result modulo 2<sup>n</sup>, where n is the bit width of the result.</p>
3880
3881<p>Because LLVM integers use a two's complement representation, this instruction
3882   is appropriate for both signed and unsigned integers.</p>
3883
3884<p><tt>nuw</tt> and <tt>nsw</tt> stand for &quot;No Unsigned Wrap&quot;
3885   and &quot;No Signed Wrap&quot;, respectively. If the <tt>nuw</tt> and/or
3886   <tt>nsw</tt> keywords are present, the result value of the <tt>add</tt>
3887   is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow,
3888   respectively, occurs.</p>
3889
3890<h5>Example:</h5>
3891<pre>
3892  &lt;result&gt; = add i32 4, %var          <i>; yields {i32}:result = 4 + %var</i>
3893</pre>
3894
3895</div>
3896
3897<!-- _______________________________________________________________________ -->
3898<h4>
3899  <a name="i_fadd">'<tt>fadd</tt>' Instruction</a>
3900</h4>
3901
3902<div>
3903
3904<h5>Syntax:</h5>
3905<pre>
3906  &lt;result&gt; = fadd &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt;   <i>; yields {ty}:result</i>
3907</pre>
3908
3909<h5>Overview:</h5>
3910<p>The '<tt>fadd</tt>' instruction returns the sum of its two operands.</p>
3911
3912<h5>Arguments:</h5>
3913<p>The two arguments to the '<tt>fadd</tt>' instruction must be
3914   <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3915   floating point values. Both arguments must have identical types.</p>
3916
3917<h5>Semantics:</h5>
3918<p>The value produced is the floating point sum of the two operands.</p>
3919
3920<h5>Example:</h5>
3921<pre>
3922  &lt;result&gt; = fadd float 4.0, %var          <i>; yields {float}:result = 4.0 + %var</i>
3923</pre>
3924
3925</div>
3926
3927<!-- _______________________________________________________________________ -->
3928<h4>
3929   <a name="i_sub">'<tt>sub</tt>' Instruction</a>
3930</h4>
3931
3932<div>
3933
3934<h5>Syntax:</h5>
3935<pre>
3936  &lt;result&gt; = sub &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt;          <i>; yields {ty}:result</i>
3937  &lt;result&gt; = sub nuw &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt;      <i>; yields {ty}:result</i>
3938  &lt;result&gt; = sub nsw &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt;      <i>; yields {ty}:result</i>
3939  &lt;result&gt; = sub nuw nsw &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt;  <i>; yields {ty}:result</i>
3940</pre>
3941
3942<h5>Overview:</h5>
3943<p>The '<tt>sub</tt>' instruction returns the difference of its two
3944   operands.</p>
3945
3946<p>Note that the '<tt>sub</tt>' instruction is used to represent the
3947   '<tt>neg</tt>' instruction present in most other intermediate
3948   representations.</p>
3949
3950<h5>Arguments:</h5>
3951<p>The two arguments to the '<tt>sub</tt>' instruction must
3952   be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3953   integer values.  Both arguments must have identical types.</p>
3954
3955<h5>Semantics:</h5>
3956<p>The value produced is the integer difference of the two operands.</p>
3957
3958<p>If the difference has unsigned overflow, the result returned is the
3959   mathematical result modulo 2<sup>n</sup>, where n is the bit width of the
3960   result.</p>
3961
3962<p>Because LLVM integers use a two's complement representation, this instruction
3963   is appropriate for both signed and unsigned integers.</p>
3964
3965<p><tt>nuw</tt> and <tt>nsw</tt> stand for &quot;No Unsigned Wrap&quot;
3966   and &quot;No Signed Wrap&quot;, respectively. If the <tt>nuw</tt> and/or
3967   <tt>nsw</tt> keywords are present, the result value of the <tt>sub</tt>
3968   is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow,
3969   respectively, occurs.</p>
3970
3971<h5>Example:</h5>
3972<pre>
3973  &lt;result&gt; = sub i32 4, %var          <i>; yields {i32}:result = 4 - %var</i>
3974  &lt;result&gt; = sub i32 0, %val          <i>; yields {i32}:result = -%var</i>
3975</pre>
3976
3977</div>
3978
3979<!-- _______________________________________________________________________ -->
3980<h4>
3981   <a name="i_fsub">'<tt>fsub</tt>' Instruction</a>
3982</h4>
3983
3984<div>
3985
3986<h5>Syntax:</h5>
3987<pre>
3988  &lt;result&gt; = fsub &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt;   <i>; yields {ty}:result</i>
3989</pre>
3990
3991<h5>Overview:</h5>
3992<p>The '<tt>fsub</tt>' instruction returns the difference of its two
3993   operands.</p>
3994
3995<p>Note that the '<tt>fsub</tt>' instruction is used to represent the
3996   '<tt>fneg</tt>' instruction present in most other intermediate
3997   representations.</p>
3998
3999<h5>Arguments:</h5>
4000<p>The two arguments to the '<tt>fsub</tt>' instruction must be
4001   <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4002   floating point values.  Both arguments must have identical types.</p>
4003
4004<h5>Semantics:</h5>
4005<p>The value produced is the floating point difference of the two operands.</p>
4006
4007<h5>Example:</h5>
4008<pre>
4009  &lt;result&gt; = fsub float 4.0, %var           <i>; yields {float}:result = 4.0 - %var</i>
4010  &lt;result&gt; = fsub float -0.0, %val          <i>; yields {float}:result = -%var</i>
4011</pre>
4012
4013</div>
4014
4015<!-- _______________________________________________________________________ -->
4016<h4>
4017  <a name="i_mul">'<tt>mul</tt>' Instruction</a>
4018</h4>
4019
4020<div>
4021
4022<h5>Syntax:</h5>
4023<pre>
4024  &lt;result&gt; = mul &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt;          <i>; yields {ty}:result</i>
4025  &lt;result&gt; = mul nuw &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt;      <i>; yields {ty}:result</i>
4026  &lt;result&gt; = mul nsw &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt;      <i>; yields {ty}:result</i>
4027  &lt;result&gt; = mul nuw nsw &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt;  <i>; yields {ty}:result</i>
4028</pre>
4029
4030<h5>Overview:</h5>
4031<p>The '<tt>mul</tt>' instruction returns the product of its two operands.</p>
4032
4033<h5>Arguments:</h5>
4034<p>The two arguments to the '<tt>mul</tt>' instruction must
4035   be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
4036   integer values.  Both arguments must have identical types.</p>
4037
4038<h5>Semantics:</h5>
4039<p>The value produced is the integer product of the two operands.</p>
4040
4041<p>If the result of the multiplication has unsigned overflow, the result
4042   returned is the mathematical result modulo 2<sup>n</sup>, where n is the bit
4043   width of the result.</p>
4044
4045<p>Because LLVM integers use a two's complement representation, and the result
4046   is the same width as the operands, this instruction returns the correct
4047   result for both signed and unsigned integers.  If a full product
4048   (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands should
4049   be sign-extended or zero-extended as appropriate to the width of the full
4050   product.</p>
4051
4052<p><tt>nuw</tt> and <tt>nsw</tt> stand for &quot;No Unsigned Wrap&quot;
4053   and &quot;No Signed Wrap&quot;, respectively. If the <tt>nuw</tt> and/or
4054   <tt>nsw</tt> keywords are present, the result value of the <tt>mul</tt>
4055   is a <a href="#poisonvalues">poison value</a> if unsigned and/or signed overflow,
4056   respectively, occurs.</p>
4057
4058<h5>Example:</h5>
4059<pre>
4060  &lt;result&gt; = mul i32 4, %var          <i>; yields {i32}:result = 4 * %var</i>
4061</pre>
4062
4063</div>
4064
4065<!-- _______________________________________________________________________ -->
4066<h4>
4067  <a name="i_fmul">'<tt>fmul</tt>' Instruction</a>
4068</h4>
4069
4070<div>
4071
4072<h5>Syntax:</h5>
4073<pre>
4074  &lt;result&gt; = fmul &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt;   <i>; yields {ty}:result</i>
4075</pre>
4076
4077<h5>Overview:</h5>
4078<p>The '<tt>fmul</tt>' instruction returns the product of its two operands.</p>
4079
4080<h5>Arguments:</h5>
4081<p>The two arguments to the '<tt>fmul</tt>' instruction must be
4082   <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4083   floating point values.  Both arguments must have identical types.</p>
4084
4085<h5>Semantics:</h5>
4086<p>The value produced is the floating point product of the two operands.</p>
4087
4088<h5>Example:</h5>
4089<pre>
4090  &lt;result&gt; = fmul float 4.0, %var          <i>; yields {float}:result = 4.0 * %var</i>
4091</pre>
4092
4093</div>
4094
4095<!-- _______________________________________________________________________ -->
4096<h4>
4097  <a name="i_udiv">'<tt>udiv</tt>' Instruction</a>
4098</h4>
4099
4100<div>
4101
4102<h5>Syntax:</h5>
4103<pre>
4104  &lt;result&gt; = udiv &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt;         <i>; yields {ty}:result</i>
4105  &lt;result&gt; = udiv exact &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt;   <i>; yields {ty}:result</i>
4106</pre>
4107
4108<h5>Overview:</h5>
4109<p>The '<tt>udiv</tt>' instruction returns the quotient of its two operands.</p>
4110
4111<h5>Arguments:</h5>
4112<p>The two arguments to the '<tt>udiv</tt>' instruction must be
4113   <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4114   values.  Both arguments must have identical types.</p>
4115
4116<h5>Semantics:</h5>
4117<p>The value produced is the unsigned integer quotient of the two operands.</p>
4118
4119<p>Note that unsigned integer division and signed integer division are distinct
4120   operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
4121
4122<p>Division by zero leads to undefined behavior.</p>
4123
4124<p>If the <tt>exact</tt> keyword is present, the result value of the
4125   <tt>udiv</tt> is a <a href="#poisonvalues">poison value</a> if %op1 is not a
4126  multiple of %op2 (as such, "((a udiv exact b) mul b) == a").</p>
4127
4128
4129<h5>Example:</h5>
4130<pre>
4131  &lt;result&gt; = udiv i32 4, %var          <i>; yields {i32}:result = 4 / %var</i>
4132</pre>
4133
4134</div>
4135
4136<!-- _______________________________________________________________________ -->
4137<h4>
4138  <a name="i_sdiv">'<tt>sdiv</tt>' Instruction</a>
4139</h4>
4140
4141<div>
4142
4143<h5>Syntax:</h5>
4144<pre>
4145  &lt;result&gt; = sdiv &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt;         <i>; yields {ty}:result</i>
4146  &lt;result&gt; = sdiv exact &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt;   <i>; yields {ty}:result</i>
4147</pre>
4148
4149<h5>Overview:</h5>
4150<p>The '<tt>sdiv</tt>' instruction returns the quotient of its two operands.</p>
4151
4152<h5>Arguments:</h5>
4153<p>The two arguments to the '<tt>sdiv</tt>' instruction must be
4154   <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4155   values.  Both arguments must have identical types.</p>
4156
4157<h5>Semantics:</h5>
4158<p>The value produced is the signed integer quotient of the two operands rounded
4159   towards zero.</p>
4160
4161<p>Note that signed integer division and unsigned integer division are distinct
4162   operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
4163
4164<p>Division by zero leads to undefined behavior. Overflow also leads to
4165   undefined behavior; this is a rare case, but can occur, for example, by doing
4166   a 32-bit division of -2147483648 by -1.</p>
4167
4168<p>If the <tt>exact</tt> keyword is present, the result value of the
4169   <tt>sdiv</tt> is a <a href="#poisonvalues">poison value</a> if the result would
4170   be rounded.</p>
4171
4172<h5>Example:</h5>
4173<pre>
4174  &lt;result&gt; = sdiv i32 4, %var          <i>; yields {i32}:result = 4 / %var</i>
4175</pre>
4176
4177</div>
4178
4179<!-- _______________________________________________________________________ -->
4180<h4>
4181  <a name="i_fdiv">'<tt>fdiv</tt>' Instruction</a>
4182</h4>
4183
4184<div>
4185
4186<h5>Syntax:</h5>
4187<pre>
4188  &lt;result&gt; = fdiv &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt;   <i>; yields {ty}:result</i>
4189</pre>
4190
4191<h5>Overview:</h5>
4192<p>The '<tt>fdiv</tt>' instruction returns the quotient of its two operands.</p>
4193
4194<h5>Arguments:</h5>
4195<p>The two arguments to the '<tt>fdiv</tt>' instruction must be
4196   <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4197   floating point values.  Both arguments must have identical types.</p>
4198
4199<h5>Semantics:</h5>
4200<p>The value produced is the floating point quotient of the two operands.</p>
4201
4202<h5>Example:</h5>
4203<pre>
4204  &lt;result&gt; = fdiv float 4.0, %var          <i>; yields {float}:result = 4.0 / %var</i>
4205</pre>
4206
4207</div>
4208
4209<!-- _______________________________________________________________________ -->
4210<h4>
4211  <a name="i_urem">'<tt>urem</tt>' Instruction</a>
4212</h4>
4213
4214<div>
4215
4216<h5>Syntax:</h5>
4217<pre>
4218  &lt;result&gt; = urem &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt;   <i>; yields {ty}:result</i>
4219</pre>
4220
4221<h5>Overview:</h5>
4222<p>The '<tt>urem</tt>' instruction returns the remainder from the unsigned
4223   division of its two arguments.</p>
4224
4225<h5>Arguments:</h5>
4226<p>The two arguments to the '<tt>urem</tt>' instruction must be
4227   <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4228   values.  Both arguments must have identical types.</p>
4229
4230<h5>Semantics:</h5>
4231<p>This instruction returns the unsigned integer <i>remainder</i> of a division.
4232   This instruction always performs an unsigned division to get the
4233   remainder.</p>
4234
4235<p>Note that unsigned integer remainder and signed integer remainder are
4236   distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
4237
4238<p>Taking the remainder of a division by zero leads to undefined behavior.</p>
4239
4240<h5>Example:</h5>
4241<pre>
4242  &lt;result&gt; = urem i32 4, %var          <i>; yields {i32}:result = 4 % %var</i>
4243</pre>
4244
4245</div>
4246
4247<!-- _______________________________________________________________________ -->
4248<h4>
4249  <a name="i_srem">'<tt>srem</tt>' Instruction</a>
4250</h4>
4251
4252<div>
4253
4254<h5>Syntax:</h5>
4255<pre>
4256  &lt;result&gt; = srem &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt;   <i>; yields {ty}:result</i>
4257</pre>
4258
4259<h5>Overview:</h5>
4260<p>The '<tt>srem</tt>' instruction returns the remainder from the signed
4261   division of its two operands. This instruction can also take
4262   <a href="#t_vector">vector</a> versions of the values in which case the
4263   elements must be integers.</p>
4264
4265<h5>Arguments:</h5>
4266<p>The two arguments to the '<tt>srem</tt>' instruction must be
4267   <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4268   values.  Both arguments must have identical types.</p>
4269
4270<h5>Semantics:</h5>
4271<p>This instruction returns the <i>remainder</i> of a division (where the result
4272   is either zero or has the same sign as the dividend, <tt>op1</tt>), not the
4273   <i>modulo</i> operator (where the result is either zero or has the same sign
4274   as the divisor, <tt>op2</tt>) of a value.
4275   For more information about the difference,
4276   see <a href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
4277   Math Forum</a>. For a table of how this is implemented in various languages,
4278   please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
4279   Wikipedia: modulo operation</a>.</p>
4280
4281<p>Note that signed integer remainder and unsigned integer remainder are
4282   distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
4283
4284<p>Taking the remainder of a division by zero leads to undefined behavior.
4285   Overflow also leads to undefined behavior; this is a rare case, but can
4286   occur, for example, by taking the remainder of a 32-bit division of
4287   -2147483648 by -1.  (The remainder doesn't actually overflow, but this rule
4288   lets srem be implemented using instructions that return both the result of
4289   the division and the remainder.)</p>
4290
4291<h5>Example:</h5>
4292<pre>
4293  &lt;result&gt; = srem i32 4, %var          <i>; yields {i32}:result = 4 % %var</i>
4294</pre>
4295
4296</div>
4297
4298<!-- _______________________________________________________________________ -->
4299<h4>
4300  <a name="i_frem">'<tt>frem</tt>' Instruction</a>
4301</h4>
4302
4303<div>
4304
4305<h5>Syntax:</h5>
4306<pre>
4307  &lt;result&gt; = frem &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt;   <i>; yields {ty}:result</i>
4308</pre>
4309
4310<h5>Overview:</h5>
4311<p>The '<tt>frem</tt>' instruction returns the remainder from the division of
4312   its two operands.</p>
4313
4314<h5>Arguments:</h5>
4315<p>The two arguments to the '<tt>frem</tt>' instruction must be
4316   <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
4317   floating point values.  Both arguments must have identical types.</p>
4318
4319<h5>Semantics:</h5>
4320<p>This instruction returns the <i>remainder</i> of a division.  The remainder
4321   has the same sign as the dividend.</p>
4322
4323<h5>Example:</h5>
4324<pre>
4325  &lt;result&gt; = frem float 4.0, %var          <i>; yields {float}:result = 4.0 % %var</i>
4326</pre>
4327
4328</div>
4329
4330</div>
4331
4332<!-- ======================================================================= -->
4333<h3>
4334  <a name="bitwiseops">Bitwise Binary Operations</a>
4335</h3>
4336
4337<div>
4338
4339<p>Bitwise binary operators are used to do various forms of bit-twiddling in a
4340   program.  They are generally very efficient instructions and can commonly be
4341   strength reduced from other instructions.  They require two operands of the
4342   same type, execute an operation on them, and produce a single value.  The
4343   resulting value is the same type as its operands.</p>
4344
4345<!-- _______________________________________________________________________ -->
4346<h4>
4347  <a name="i_shl">'<tt>shl</tt>' Instruction</a>
4348</h4>
4349
4350<div>
4351
4352<h5>Syntax:</h5>
4353<pre>
4354  &lt;result&gt; = shl &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt;           <i>; yields {ty}:result</i>
4355  &lt;result&gt; = shl nuw &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt;       <i>; yields {ty}:result</i>
4356  &lt;result&gt; = shl nsw &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt;       <i>; yields {ty}:result</i>
4357  &lt;result&gt; = shl nuw nsw &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt;   <i>; yields {ty}:result</i>
4358</pre>
4359
4360<h5>Overview:</h5>
4361<p>The '<tt>shl</tt>' instruction returns the first operand shifted to the left
4362   a specified number of bits.</p>
4363
4364<h5>Arguments:</h5>
4365<p>Both arguments to the '<tt>shl</tt>' instruction must be the
4366    same <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
4367    integer type.  '<tt>op2</tt>' is treated as an unsigned value.</p>
4368
4369<h5>Semantics:</h5>
4370<p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod
4371   2<sup>n</sup>, where <tt>n</tt> is the width of the result.  If <tt>op2</tt>
4372   is (statically or dynamically) negative or equal to or larger than the number
4373   of bits in <tt>op1</tt>, the result is undefined.  If the arguments are
4374   vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
4375   shift amount in <tt>op2</tt>.</p>
4376
4377<p>If the <tt>nuw</tt> keyword is present, then the shift produces a
4378   <a href="#poisonvalues">poison value</a> if it shifts out any non-zero bits.  If
4379   the <tt>nsw</tt> keyword is present, then the shift produces a
4380   <a href="#poisonvalues">poison value</a> if it shifts out any bits that disagree
4381   with the resultant sign bit.  As such, NUW/NSW have the same semantics as
4382   they would if the shift were expressed as a mul instruction with the same
4383   nsw/nuw bits in (mul %op1, (shl 1, %op2)).</p>
4384
4385<h5>Example:</h5>
4386<pre>
4387  &lt;result&gt; = shl i32 4, %var   <i>; yields {i32}: 4 &lt;&lt; %var</i>
4388  &lt;result&gt; = shl i32 4, 2      <i>; yields {i32}: 16</i>
4389  &lt;result&gt; = shl i32 1, 10     <i>; yields {i32}: 1024</i>
4390  &lt;result&gt; = shl i32 1, 32     <i>; undefined</i>
4391  &lt;result&gt; = shl &lt;2 x i32&gt; &lt; i32 1, i32 1&gt;, &lt; i32 1, i32 2&gt;   <i>; yields: result=&lt;2 x i32&gt; &lt; i32 2, i32 4&gt;</i>
4392</pre>
4393
4394</div>
4395
4396<!-- _______________________________________________________________________ -->
4397<h4>
4398  <a name="i_lshr">'<tt>lshr</tt>' Instruction</a>
4399</h4>
4400
4401<div>
4402
4403<h5>Syntax:</h5>
4404<pre>
4405  &lt;result&gt; = lshr &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt;         <i>; yields {ty}:result</i>
4406  &lt;result&gt; = lshr exact &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt;   <i>; yields {ty}:result</i>
4407</pre>
4408
4409<h5>Overview:</h5>
4410<p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
4411   operand shifted to the right a specified number of bits with zero fill.</p>
4412
4413<h5>Arguments:</h5>
4414<p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
4415   <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4416   type. '<tt>op2</tt>' is treated as an unsigned value.</p>
4417
4418<h5>Semantics:</h5>
4419<p>This instruction always performs a logical shift right operation. The most
4420   significant bits of the result will be filled with zero bits after the shift.
4421   If <tt>op2</tt> is (statically or dynamically) equal to or larger than the
4422   number of bits in <tt>op1</tt>, the result is undefined. If the arguments are
4423   vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
4424   shift amount in <tt>op2</tt>.</p>
4425
4426<p>If the <tt>exact</tt> keyword is present, the result value of the
4427   <tt>lshr</tt> is a <a href="#poisonvalues">poison value</a> if any of the bits
4428   shifted out are non-zero.</p>
4429
4430
4431<h5>Example:</h5>
4432<pre>
4433  &lt;result&gt; = lshr i32 4, 1   <i>; yields {i32}:result = 2</i>
4434  &lt;result&gt; = lshr i32 4, 2   <i>; yields {i32}:result = 1</i>
4435  &lt;result&gt; = lshr i8  4, 3   <i>; yields {i8}:result = 0</i>
4436  &lt;result&gt; = lshr i8 -2, 1   <i>; yields {i8}:result = 0x7FFFFFFF </i>
4437  &lt;result&gt; = lshr i32 1, 32  <i>; undefined</i>
4438  &lt;result&gt; = lshr &lt;2 x i32&gt; &lt; i32 -2, i32 4&gt;, &lt; i32 1, i32 2&gt;   <i>; yields: result=&lt;2 x i32&gt; &lt; i32 0x7FFFFFFF, i32 1&gt;</i>
4439</pre>
4440
4441</div>
4442
4443<!-- _______________________________________________________________________ -->
4444<h4>
4445  <a name="i_ashr">'<tt>ashr</tt>' Instruction</a>
4446</h4>
4447
4448<div>
4449
4450<h5>Syntax:</h5>
4451<pre>
4452  &lt;result&gt; = ashr &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt;         <i>; yields {ty}:result</i>
4453  &lt;result&gt; = ashr exact &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt;   <i>; yields {ty}:result</i>
4454</pre>
4455
4456<h5>Overview:</h5>
4457<p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
4458   operand shifted to the right a specified number of bits with sign
4459   extension.</p>
4460
4461<h5>Arguments:</h5>
4462<p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
4463   <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4464   type.  '<tt>op2</tt>' is treated as an unsigned value.</p>
4465
4466<h5>Semantics:</h5>
4467<p>This instruction always performs an arithmetic shift right operation, The
4468   most significant bits of the result will be filled with the sign bit
4469   of <tt>op1</tt>.  If <tt>op2</tt> is (statically or dynamically) equal to or
4470   larger than the number of bits in <tt>op1</tt>, the result is undefined. If
4471   the arguments are vectors, each vector element of <tt>op1</tt> is shifted by
4472   the corresponding shift amount in <tt>op2</tt>.</p>
4473
4474<p>If the <tt>exact</tt> keyword is present, the result value of the
4475   <tt>ashr</tt> is a <a href="#poisonvalues">poison value</a> if any of the bits
4476   shifted out are non-zero.</p>
4477
4478<h5>Example:</h5>
4479<pre>
4480  &lt;result&gt; = ashr i32 4, 1   <i>; yields {i32}:result = 2</i>
4481  &lt;result&gt; = ashr i32 4, 2   <i>; yields {i32}:result = 1</i>
4482  &lt;result&gt; = ashr i8  4, 3   <i>; yields {i8}:result = 0</i>
4483  &lt;result&gt; = ashr i8 -2, 1   <i>; yields {i8}:result = -1</i>
4484  &lt;result&gt; = ashr i32 1, 32  <i>; undefined</i>
4485  &lt;result&gt; = ashr &lt;2 x i32&gt; &lt; i32 -2, i32 4&gt;, &lt; i32 1, i32 3&gt;   <i>; yields: result=&lt;2 x i32&gt; &lt; i32 -1, i32 0&gt;</i>
4486</pre>
4487
4488</div>
4489
4490<!-- _______________________________________________________________________ -->
4491<h4>
4492  <a name="i_and">'<tt>and</tt>' Instruction</a>
4493</h4>
4494
4495<div>
4496
4497<h5>Syntax:</h5>
4498<pre>
4499  &lt;result&gt; = and &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt;   <i>; yields {ty}:result</i>
4500</pre>
4501
4502<h5>Overview:</h5>
4503<p>The '<tt>and</tt>' instruction returns the bitwise logical and of its two
4504   operands.</p>
4505
4506<h5>Arguments:</h5>
4507<p>The two arguments to the '<tt>and</tt>' instruction must be
4508   <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4509   values.  Both arguments must have identical types.</p>
4510
4511<h5>Semantics:</h5>
4512<p>The truth table used for the '<tt>and</tt>' instruction is:</p>
4513
4514<table border="1" cellspacing="0" cellpadding="4">
4515  <tbody>
4516    <tr>
4517      <th>In0</th>
4518      <th>In1</th>
4519      <th>Out</th>
4520    </tr>
4521    <tr>
4522      <td>0</td>
4523      <td>0</td>
4524      <td>0</td>
4525    </tr>
4526    <tr>
4527      <td>0</td>
4528      <td>1</td>
4529      <td>0</td>
4530    </tr>
4531    <tr>
4532      <td>1</td>
4533      <td>0</td>
4534      <td>0</td>
4535    </tr>
4536    <tr>
4537      <td>1</td>
4538      <td>1</td>
4539      <td>1</td>
4540    </tr>
4541  </tbody>
4542</table>
4543
4544<h5>Example:</h5>
4545<pre>
4546  &lt;result&gt; = and i32 4, %var         <i>; yields {i32}:result = 4 &amp; %var</i>
4547  &lt;result&gt; = and i32 15, 40          <i>; yields {i32}:result = 8</i>
4548  &lt;result&gt; = and i32 4, 8            <i>; yields {i32}:result = 0</i>
4549</pre>
4550</div>
4551<!-- _______________________________________________________________________ -->
4552<h4>
4553  <a name="i_or">'<tt>or</tt>' Instruction</a>
4554</h4>
4555
4556<div>
4557
4558<h5>Syntax:</h5>
4559<pre>
4560  &lt;result&gt; = or &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt;   <i>; yields {ty}:result</i>
4561</pre>
4562
4563<h5>Overview:</h5>
4564<p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive or of its
4565   two operands.</p>
4566
4567<h5>Arguments:</h5>
4568<p>The two arguments to the '<tt>or</tt>' instruction must be
4569   <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4570   values.  Both arguments must have identical types.</p>
4571
4572<h5>Semantics:</h5>
4573<p>The truth table used for the '<tt>or</tt>' instruction is:</p>
4574
4575<table border="1" cellspacing="0" cellpadding="4">
4576  <tbody>
4577    <tr>
4578      <th>In0</th>
4579      <th>In1</th>
4580      <th>Out</th>
4581    </tr>
4582    <tr>
4583      <td>0</td>
4584      <td>0</td>
4585      <td>0</td>
4586    </tr>
4587    <tr>
4588      <td>0</td>
4589      <td>1</td>
4590      <td>1</td>
4591    </tr>
4592    <tr>
4593      <td>1</td>
4594      <td>0</td>
4595      <td>1</td>
4596    </tr>
4597    <tr>
4598      <td>1</td>
4599      <td>1</td>
4600      <td>1</td>
4601    </tr>
4602  </tbody>
4603</table>
4604
4605<h5>Example:</h5>
4606<pre>
4607  &lt;result&gt; = or i32 4, %var         <i>; yields {i32}:result = 4 | %var</i>
4608  &lt;result&gt; = or i32 15, 40          <i>; yields {i32}:result = 47</i>
4609  &lt;result&gt; = or i32 4, 8            <i>; yields {i32}:result = 12</i>
4610</pre>
4611
4612</div>
4613
4614<!-- _______________________________________________________________________ -->
4615<h4>
4616  <a name="i_xor">'<tt>xor</tt>' Instruction</a>
4617</h4>
4618
4619<div>
4620
4621<h5>Syntax:</h5>
4622<pre>
4623  &lt;result&gt; = xor &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt;   <i>; yields {ty}:result</i>
4624</pre>
4625
4626<h5>Overview:</h5>
4627<p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive or of
4628   its two operands.  The <tt>xor</tt> is used to implement the "one's
4629   complement" operation, which is the "~" operator in C.</p>
4630
4631<h5>Arguments:</h5>
4632<p>The two arguments to the '<tt>xor</tt>' instruction must be
4633   <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
4634   values.  Both arguments must have identical types.</p>
4635
4636<h5>Semantics:</h5>
4637<p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
4638
4639<table border="1" cellspacing="0" cellpadding="4">
4640  <tbody>
4641    <tr>
4642      <th>In0</th>
4643      <th>In1</th>
4644      <th>Out</th>
4645    </tr>
4646    <tr>
4647      <td>0</td>
4648      <td>0</td>
4649      <td>0</td>
4650    </tr>
4651    <tr>
4652      <td>0</td>
4653      <td>1</td>
4654      <td>1</td>
4655    </tr>
4656    <tr>
4657      <td>1</td>
4658      <td>0</td>
4659      <td>1</td>
4660    </tr>
4661    <tr>
4662      <td>1</td>
4663      <td>1</td>
4664      <td>0</td>
4665    </tr>
4666  </tbody>
4667</table>
4668
4669<h5>Example:</h5>
4670<pre>
4671  &lt;result&gt; = xor i32 4, %var         <i>; yields {i32}:result = 4 ^ %var</i>
4672  &lt;result&gt; = xor i32 15, 40          <i>; yields {i32}:result = 39</i>
4673  &lt;result&gt; = xor i32 4, 8            <i>; yields {i32}:result = 12</i>
4674  &lt;result&gt; = xor i32 %V, -1          <i>; yields {i32}:result = ~%V</i>
4675</pre>
4676
4677</div>
4678
4679</div>
4680
4681<!-- ======================================================================= -->
4682<h3>
4683  <a name="vectorops">Vector Operations</a>
4684</h3>
4685
4686<div>
4687
4688<p>LLVM supports several instructions to represent vector operations in a
4689   target-independent manner.  These instructions cover the element-access and
4690   vector-specific operations needed to process vectors effectively.  While LLVM
4691   does directly support these vector operations, many sophisticated algorithms
4692   will want to use target-specific intrinsics to take full advantage of a
4693   specific target.</p>
4694
4695<!-- _______________________________________________________________________ -->
4696<h4>
4697   <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
4698</h4>
4699
4700<div>
4701
4702<h5>Syntax:</h5>
4703<pre>
4704  &lt;result&gt; = extractelement &lt;n x &lt;ty&gt;&gt; &lt;val&gt;, i32 &lt;idx&gt;    <i>; yields &lt;ty&gt;</i>
4705</pre>
4706
4707<h5>Overview:</h5>
4708<p>The '<tt>extractelement</tt>' instruction extracts a single scalar element
4709   from a vector at a specified index.</p>
4710
4711
4712<h5>Arguments:</h5>
4713<p>The first operand of an '<tt>extractelement</tt>' instruction is a value
4714   of <a href="#t_vector">vector</a> type.  The second operand is an index
4715   indicating the position from which to extract the element.  The index may be
4716   a variable.</p>
4717
4718<h5>Semantics:</h5>
4719<p>The result is a scalar of the same type as the element type of
4720   <tt>val</tt>.  Its value is the value at position <tt>idx</tt> of
4721   <tt>val</tt>.  If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4722   results are undefined.</p>
4723
4724<h5>Example:</h5>
4725<pre>
4726  &lt;result&gt; = extractelement &lt;4 x i32&gt; %vec, i32 0    <i>; yields i32</i>
4727</pre>
4728
4729</div>
4730
4731<!-- _______________________________________________________________________ -->
4732<h4>
4733   <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
4734</h4>
4735
4736<div>
4737
4738<h5>Syntax:</h5>
4739<pre>
4740  &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>
4741</pre>
4742
4743<h5>Overview:</h5>
4744<p>The '<tt>insertelement</tt>' instruction inserts a scalar element into a
4745   vector at a specified index.</p>
4746
4747<h5>Arguments:</h5>
4748<p>The first operand of an '<tt>insertelement</tt>' instruction is a value
4749   of <a href="#t_vector">vector</a> type.  The second operand is a scalar value
4750   whose type must equal the element type of the first operand.  The third
4751   operand is an index indicating the position at which to insert the value.
4752   The index may be a variable.</p>
4753
4754<h5>Semantics:</h5>
4755<p>The result is a vector of the same type as <tt>val</tt>.  Its element values
4756   are those of <tt>val</tt> except at position <tt>idx</tt>, where it gets the
4757   value <tt>elt</tt>.  If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
4758   results are undefined.</p>
4759
4760<h5>Example:</h5>
4761<pre>
4762  &lt;result&gt; = insertelement &lt;4 x i32&gt; %vec, i32 1, i32 0    <i>; yields &lt;4 x i32&gt;</i>
4763</pre>
4764
4765</div>
4766
4767<!-- _______________________________________________________________________ -->
4768<h4>
4769   <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
4770</h4>
4771
4772<div>
4773
4774<h5>Syntax:</h5>
4775<pre>
4776  &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;m x i32&gt; &lt;mask&gt;    <i>; yields &lt;m x &lt;ty&gt;&gt;</i>
4777</pre>
4778
4779<h5>Overview:</h5>
4780<p>The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
4781   from two input vectors, returning a vector with the same element type as the
4782   input and length that is the same as the shuffle mask.</p>
4783
4784<h5>Arguments:</h5>
4785<p>The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
4786   with the same type.  The third argument is a shuffle mask whose
4787   element type is always 'i32'.  The result of the instruction is a vector
4788   whose length is the same as the shuffle mask and whose element type is the
4789   same as the element type of the first two operands.</p>
4790
4791<p>The shuffle mask operand is required to be a constant vector with either
4792   constant integer or undef values.</p>
4793
4794<h5>Semantics:</h5>
4795<p>The elements of the two input vectors are numbered from left to right across
4796   both of the vectors.  The shuffle mask operand specifies, for each element of
4797   the result vector, which element of the two input vectors the result element
4798   gets.  The element selector may be undef (meaning "don't care") and the
4799   second operand may be undef if performing a shuffle from only one vector.</p>
4800
4801<h5>Example:</h5>
4802<pre>
4803  &lt;result&gt; = shufflevector &lt;4 x i32&gt; %v1, &lt;4 x i32&gt; %v2,
4804                          &lt;4 x i32&gt; &lt;i32 0, i32 4, i32 1, i32 5&gt;  <i>; yields &lt;4 x i32&gt;</i>
4805  &lt;result&gt; = shufflevector &lt;4 x i32&gt; %v1, &lt;4 x i32&gt; undef,
4806                          &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.
4807  &lt;result&gt; = shufflevector &lt;8 x i32&gt; %v1, &lt;8 x i32&gt; undef,
4808                          &lt;4 x i32&gt; &lt;i32 0, i32 1, i32 2, i32 3&gt;  <i>; yields &lt;4 x i32&gt;</i>
4809  &lt;result&gt; = shufflevector &lt;4 x i32&gt; %v1, &lt;4 x i32&gt; %v2,
4810                          &lt;8 x i32&gt; &lt;i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 &gt;  <i>; yields &lt;8 x i32&gt;</i>
4811</pre>
4812
4813</div>
4814
4815</div>
4816
4817<!-- ======================================================================= -->
4818<h3>
4819  <a name="aggregateops">Aggregate Operations</a>
4820</h3>
4821
4822<div>
4823
4824<p>LLVM supports several instructions for working with
4825  <a href="#t_aggregate">aggregate</a> values.</p>
4826
4827<!-- _______________________________________________________________________ -->
4828<h4>
4829   <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
4830</h4>
4831
4832<div>
4833
4834<h5>Syntax:</h5>
4835<pre>
4836  &lt;result&gt; = extractvalue &lt;aggregate type&gt; &lt;val&gt;, &lt;idx&gt;{, &lt;idx&gt;}*
4837</pre>
4838
4839<h5>Overview:</h5>
4840<p>The '<tt>extractvalue</tt>' instruction extracts the value of a member field
4841   from an <a href="#t_aggregate">aggregate</a> value.</p>
4842
4843<h5>Arguments:</h5>
4844<p>The first operand of an '<tt>extractvalue</tt>' instruction is a value
4845   of <a href="#t_struct">struct</a> or
4846   <a href="#t_array">array</a> type.  The operands are constant indices to
4847   specify which value to extract in a similar manner as indices in a
4848   '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
4849   <p>The major differences to <tt>getelementptr</tt> indexing are:</p>
4850     <ul>
4851       <li>Since the value being indexed is not a pointer, the first index is
4852           omitted and assumed to be zero.</li>
4853       <li>At least one index must be specified.</li>
4854       <li>Not only struct indices but also array indices must be in
4855           bounds.</li>
4856     </ul>
4857
4858<h5>Semantics:</h5>
4859<p>The result is the value at the position in the aggregate specified by the
4860   index operands.</p>
4861
4862<h5>Example:</h5>
4863<pre>
4864  &lt;result&gt; = extractvalue {i32, float} %agg, 0    <i>; yields i32</i>
4865</pre>
4866
4867</div>
4868
4869<!-- _______________________________________________________________________ -->
4870<h4>
4871   <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
4872</h4>
4873
4874<div>
4875
4876<h5>Syntax:</h5>
4877<pre>
4878  &lt;result&gt; = insertvalue &lt;aggregate type&gt; &lt;val&gt;, &lt;ty&gt; &lt;elt&gt;, &lt;idx&gt;{, &lt;idx&gt;}*    <i>; yields &lt;aggregate type&gt;</i>
4879</pre>
4880
4881<h5>Overview:</h5>
4882<p>The '<tt>insertvalue</tt>' instruction inserts a value into a member field
4883   in an <a href="#t_aggregate">aggregate</a> value.</p>
4884
4885<h5>Arguments:</h5>
4886<p>The first operand of an '<tt>insertvalue</tt>' instruction is a value
4887   of <a href="#t_struct">struct</a> or
4888   <a href="#t_array">array</a> type.  The second operand is a first-class
4889   value to insert.  The following operands are constant indices indicating
4890   the position at which to insert the value in a similar manner as indices in a
4891   '<tt><a href="#i_extractvalue">extractvalue</a></tt>' instruction.  The
4892   value to insert must have the same type as the value identified by the
4893   indices.</p>
4894
4895<h5>Semantics:</h5>
4896<p>The result is an aggregate of the same type as <tt>val</tt>.  Its value is
4897   that of <tt>val</tt> except that the value at the position specified by the
4898   indices is that of <tt>elt</tt>.</p>
4899
4900<h5>Example:</h5>
4901<pre>
4902  %agg1 = insertvalue {i32, float} undef, i32 1, 0              <i>; yields {i32 1, float undef}</i>
4903  %agg2 = insertvalue {i32, float} %agg1, float %val, 1         <i>; yields {i32 1, float %val}</i>
4904  %agg3 = insertvalue {i32, {float}} %agg1, float %val, 1, 0    <i>; yields {i32 1, float %val}</i>
4905</pre>
4906
4907</div>
4908
4909</div>
4910
4911<!-- ======================================================================= -->
4912<h3>
4913  <a name="memoryops">Memory Access and Addressing Operations</a>
4914</h3>
4915
4916<div>
4917
4918<p>A key design point of an SSA-based representation is how it represents
4919   memory.  In LLVM, no memory locations are in SSA form, which makes things
4920   very simple.  This section describes how to read, write, and allocate
4921   memory in LLVM.</p>
4922
4923<!-- _______________________________________________________________________ -->
4924<h4>
4925  <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
4926</h4>
4927
4928<div>
4929
4930<h5>Syntax:</h5>
4931<pre>
4932  &lt;result&gt; = alloca &lt;type&gt;[, &lt;ty&gt; &lt;NumElements&gt;][, align &lt;alignment&gt;]     <i>; yields {type*}:result</i>
4933</pre>
4934
4935<h5>Overview:</h5>
4936<p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
4937   currently executing function, to be automatically released when this function
4938   returns to its caller. The object is always allocated in the generic address
4939   space (address space zero).</p>
4940
4941<h5>Arguments:</h5>
4942<p>The '<tt>alloca</tt>' instruction
4943   allocates <tt>sizeof(&lt;type&gt;)*NumElements</tt> bytes of memory on the
4944   runtime stack, returning a pointer of the appropriate type to the program.
4945   If "NumElements" is specified, it is the number of elements allocated,
4946   otherwise "NumElements" is defaulted to be one.  If a constant alignment is
4947   specified, the value result of the allocation is guaranteed to be aligned to
4948   at least that boundary.  If not specified, or if zero, the target can choose
4949   to align the allocation on any convenient boundary compatible with the
4950   type.</p>
4951
4952<p>'<tt>type</tt>' may be any sized type.</p>
4953
4954<h5>Semantics:</h5>
4955<p>Memory is allocated; a pointer is returned.  The operation is undefined if
4956   there is insufficient stack space for the allocation.  '<tt>alloca</tt>'d
4957   memory is automatically released when the function returns.  The
4958   '<tt>alloca</tt>' instruction is commonly used to represent automatic
4959   variables that must have an address available.  When the function returns
4960   (either with the <tt><a href="#i_ret">ret</a></tt>
4961   or <tt><a href="#i_resume">resume</a></tt> instructions), the memory is
4962   reclaimed.  Allocating zero bytes is legal, but the result is undefined.
4963   The order in which memory is allocated (ie., which way the stack grows) is
4964   not specified.</p>
4965
4966<p>
4967
4968<h5>Example:</h5>
4969<pre>
4970  %ptr = alloca i32                             <i>; yields {i32*}:ptr</i>
4971  %ptr = alloca i32, i32 4                      <i>; yields {i32*}:ptr</i>
4972  %ptr = alloca i32, i32 4, align 1024          <i>; yields {i32*}:ptr</i>
4973  %ptr = alloca i32, align 1024                 <i>; yields {i32*}:ptr</i>
4974</pre>
4975
4976</div>
4977
4978<!-- _______________________________________________________________________ -->
4979<h4>
4980  <a name="i_load">'<tt>load</tt>' Instruction</a>
4981</h4>
4982
4983<div>
4984
4985<h5>Syntax:</h5>
4986<pre>
4987  &lt;result&gt; = load [volatile] &lt;ty&gt;* &lt;pointer&gt;[, align &lt;alignment&gt;][, !nontemporal !&lt;index&gt;][, !invariant.load !&lt;index&gt;]
4988  &lt;result&gt; = load atomic [volatile] &lt;ty&gt;* &lt;pointer&gt; [singlethread] &lt;ordering&gt;, align &lt;alignment&gt;
4989  !&lt;index&gt; = !{ i32 1 }
4990</pre>
4991
4992<h5>Overview:</h5>
4993<p>The '<tt>load</tt>' instruction is used to read from memory.</p>
4994
4995<h5>Arguments:</h5>
4996<p>The argument to the '<tt>load</tt>' instruction specifies the memory address
4997   from which to load.  The pointer must point to
4998   a <a href="#t_firstclass">first class</a> type.  If the <tt>load</tt> is
4999   marked as <tt>volatile</tt>, then the optimizer is not allowed to modify the
5000   number or order of execution of this <tt>load</tt> with other <a
5001   href="#volatile">volatile operations</a>.</p>
5002
5003<p>If the <code>load</code> is marked as <code>atomic</code>, it takes an extra
5004   <a href="#ordering">ordering</a> and optional <code>singlethread</code>
5005   argument.  The <code>release</code> and <code>acq_rel</code> orderings are
5006   not valid on <code>load</code> instructions.  Atomic loads produce <a
5007   href="#memorymodel">defined</a> results when they may see multiple atomic
5008   stores.  The type of the pointee must be an integer type whose bit width
5009   is a power of two greater than or equal to eight and less than or equal
5010   to a target-specific size limit. <code>align</code> must be explicitly
5011   specified on atomic loads, and the load has undefined behavior if the
5012   alignment is not set to a value which is at least the size in bytes of
5013   the pointee. <code>!nontemporal</code> does not have any defined semantics
5014   for atomic loads.</p>
5015
5016<p>The optional constant <tt>align</tt> argument specifies the alignment of the
5017   operation (that is, the alignment of the memory address). A value of 0 or an
5018   omitted <tt>align</tt> argument means that the operation has the preferential
5019   alignment for the target. It is the responsibility of the code emitter to
5020   ensure that the alignment information is correct. Overestimating the
5021   alignment results in undefined behavior. Underestimating the alignment may
5022   produce less efficient code. An alignment of 1 is always safe.</p>
5023
5024<p>The optional <tt>!nontemporal</tt> metadata must reference a single
5025   metatadata name &lt;index&gt; corresponding to a metadata node with
5026   one <tt>i32</tt> entry of value 1.  The existence of
5027   the <tt>!nontemporal</tt> metatadata on the instruction tells the optimizer
5028   and code generator that this load is not expected to be reused in the cache.
5029   The code generator may select special instructions to save cache bandwidth,
5030   such as the <tt>MOVNT</tt> instruction on x86.</p>
5031
5032<p>The optional <tt>!invariant.load</tt> metadata must reference a single
5033   metatadata name &lt;index&gt; corresponding to a metadata node with no
5034   entries.  The existence of the <tt>!invariant.load</tt> metatadata on the
5035   instruction tells the optimizer and code generator that this load address
5036   points to memory which does not change value during program execution.
5037   The optimizer may then move this load around, for example, by hoisting it
5038   out of loops using loop invariant code motion.</p>
5039
5040<h5>Semantics:</h5>
5041<p>The location of memory pointed to is loaded.  If the value being loaded is of
5042   scalar type then the number of bytes read does not exceed the minimum number
5043   of bytes needed to hold all bits of the type.  For example, loading an
5044   <tt>i24</tt> reads at most three bytes.  When loading a value of a type like
5045   <tt>i20</tt> with a size that is not an integral number of bytes, the result
5046   is undefined if the value was not originally written using a store of the
5047   same type.</p>
5048
5049<h5>Examples:</h5>
5050<pre>
5051  %ptr = <a href="#i_alloca">alloca</a> i32                               <i>; yields {i32*}:ptr</i>
5052  <a href="#i_store">store</a> i32 3, i32* %ptr                          <i>; yields {void}</i>
5053  %val = load i32* %ptr                           <i>; yields {i32}:val = i32 3</i>
5054</pre>
5055
5056</div>
5057
5058<!-- _______________________________________________________________________ -->
5059<h4>
5060  <a name="i_store">'<tt>store</tt>' Instruction</a>
5061</h4>
5062
5063<div>
5064
5065<h5>Syntax:</h5>
5066<pre>
5067  store [volatile] &lt;ty&gt; &lt;value&gt;, &lt;ty&gt;* &lt;pointer&gt;[, align &lt;alignment&gt;][, !nontemporal !&lt;index&gt;]        <i>; yields {void}</i>
5068  store atomic [volatile] &lt;ty&gt; &lt;value&gt;, &lt;ty&gt;* &lt;pointer&gt; [singlethread] &lt;ordering&gt;, align &lt;alignment&gt;  <i>; yields {void}</i>
5069</pre>
5070
5071<h5>Overview:</h5>
5072<p>The '<tt>store</tt>' instruction is used to write to memory.</p>
5073
5074<h5>Arguments:</h5>
5075<p>There are two arguments to the '<tt>store</tt>' instruction: a value to store
5076   and an address at which to store it.  The type of the
5077   '<tt>&lt;pointer&gt;</tt>' operand must be a pointer to
5078   the <a href="#t_firstclass">first class</a> type of the
5079   '<tt>&lt;value&gt;</tt>' operand. If the <tt>store</tt> is marked as
5080   <tt>volatile</tt>, then the optimizer is not allowed to modify the number or
5081   order of execution of this <tt>store</tt> with other <a
5082   href="#volatile">volatile operations</a>.</p>
5083
5084<p>If the <code>store</code> is marked as <code>atomic</code>, it takes an extra
5085   <a href="#ordering">ordering</a> and optional <code>singlethread</code>
5086   argument.  The <code>acquire</code> and <code>acq_rel</code> orderings aren't
5087   valid on <code>store</code> instructions.  Atomic loads produce <a
5088   href="#memorymodel">defined</a> results when they may see multiple atomic
5089   stores. The type of the pointee must be an integer type whose bit width
5090   is a power of two greater than or equal to eight and less than or equal
5091   to a target-specific size limit. <code>align</code> must be explicitly
5092   specified on atomic stores, and the store has undefined behavior if the
5093   alignment is not set to a value which is at least the size in bytes of
5094   the pointee. <code>!nontemporal</code> does not have any defined semantics
5095   for atomic stores.</p>
5096
5097<p>The optional constant "align" argument specifies the alignment of the
5098   operation (that is, the alignment of the memory address). A value of 0 or an
5099   omitted "align" argument means that the operation has the preferential
5100   alignment for the target. It is the responsibility of the code emitter to
5101   ensure that the alignment information is correct. Overestimating the
5102   alignment results in an undefined behavior. Underestimating the alignment may
5103   produce less efficient code. An alignment of 1 is always safe.</p>
5104
5105<p>The optional !nontemporal metadata must reference a single metatadata
5106   name &lt;index&gt; corresponding to a metadata node with one i32 entry of
5107   value 1.  The existence of the !nontemporal metatadata on the
5108   instruction tells the optimizer and code generator that this load is
5109   not expected to be reused in the cache.  The code generator may
5110   select special instructions to save cache bandwidth, such as the
5111   MOVNT instruction on x86.</p>
5112
5113
5114<h5>Semantics:</h5>
5115<p>The contents of memory are updated to contain '<tt>&lt;value&gt;</tt>' at the
5116   location specified by the '<tt>&lt;pointer&gt;</tt>' operand.  If
5117   '<tt>&lt;value&gt;</tt>' is of scalar type then the number of bytes written
5118   does not exceed the minimum number of bytes needed to hold all bits of the
5119   type.  For example, storing an <tt>i24</tt> writes at most three bytes.  When
5120   writing a value of a type like <tt>i20</tt> with a size that is not an
5121   integral number of bytes, it is unspecified what happens to the extra bits
5122   that do not belong to the type, but they will typically be overwritten.</p>
5123
5124<h5>Example:</h5>
5125<pre>
5126  %ptr = <a href="#i_alloca">alloca</a> i32                               <i>; yields {i32*}:ptr</i>
5127  store i32 3, i32* %ptr                          <i>; yields {void}</i>
5128  %val = <a href="#i_load">load</a> i32* %ptr                           <i>; yields {i32}:val = i32 3</i>
5129</pre>
5130
5131</div>
5132
5133<!-- _______________________________________________________________________ -->
5134<h4>
5135<a name="i_fence">'<tt>fence</tt>' Instruction</a>
5136</h4>
5137
5138<div>
5139
5140<h5>Syntax:</h5>
5141<pre>
5142  fence [singlethread] &lt;ordering&gt;                   <i>; yields {void}</i>
5143</pre>
5144
5145<h5>Overview:</h5>
5146<p>The '<tt>fence</tt>' instruction is used to introduce happens-before edges
5147between operations.</p>
5148
5149<h5>Arguments:</h5> <p>'<code>fence</code>' instructions take an <a
5150href="#ordering">ordering</a> argument which defines what
5151<i>synchronizes-with</i> edges they add.  They can only be given
5152<code>acquire</code>, <code>release</code>, <code>acq_rel</code>, and
5153<code>seq_cst</code> orderings.</p>
5154
5155<h5>Semantics:</h5>
5156<p>A fence <var>A</var> which has (at least) <code>release</code> ordering
5157semantics <i>synchronizes with</i> a fence <var>B</var> with (at least)
5158<code>acquire</code> ordering semantics if and only if there exist atomic
5159operations <var>X</var> and <var>Y</var>, both operating on some atomic object
5160<var>M</var>, such that <var>A</var> is sequenced before <var>X</var>,
5161<var>X</var> modifies <var>M</var> (either directly or through some side effect
5162of a sequence headed by <var>X</var>), <var>Y</var> is sequenced before
5163<var>B</var>, and <var>Y</var> observes <var>M</var>. This provides a
5164<i>happens-before</i> dependency between <var>A</var> and <var>B</var>. Rather
5165than an explicit <code>fence</code>, one (but not both) of the atomic operations
5166<var>X</var> or <var>Y</var> might provide a <code>release</code> or
5167<code>acquire</code> (resp.) ordering constraint and still
5168<i>synchronize-with</i> the explicit <code>fence</code> and establish the
5169<i>happens-before</i> edge.</p>
5170
5171<p>A <code>fence</code> which has <code>seq_cst</code> ordering, in addition to
5172having both <code>acquire</code> and <code>release</code> semantics specified
5173above, participates in the global program order of other <code>seq_cst</code>
5174operations and/or fences.</p>
5175
5176<p>The optional "<a href="#singlethread"><code>singlethread</code></a>" argument
5177specifies that the fence only synchronizes with other fences in the same
5178thread.  (This is useful for interacting with signal handlers.)</p>
5179
5180<h5>Example:</h5>
5181<pre>
5182  fence acquire                          <i>; yields {void}</i>
5183  fence singlethread seq_cst             <i>; yields {void}</i>
5184</pre>
5185
5186</div>
5187
5188<!-- _______________________________________________________________________ -->
5189<h4>
5190<a name="i_cmpxchg">'<tt>cmpxchg</tt>' Instruction</a>
5191</h4>
5192
5193<div>
5194
5195<h5>Syntax:</h5>
5196<pre>
5197  cmpxchg [volatile] &lt;ty&gt;* &lt;pointer&gt;, &lt;ty&gt; &lt;cmp&gt;, &lt;ty&gt; &lt;new&gt; [singlethread] &lt;ordering&gt;  <i>; yields {ty}</i>
5198</pre>
5199
5200<h5>Overview:</h5>
5201<p>The '<tt>cmpxchg</tt>' instruction is used to atomically modify memory.
5202It loads a value in memory and compares it to a given value. If they are
5203equal, it stores a new value into the memory.</p>
5204
5205<h5>Arguments:</h5>
5206<p>There are three arguments to the '<code>cmpxchg</code>' instruction: an
5207address to operate on, a value to compare to the value currently be at that
5208address, and a new value to place at that address if the compared values are
5209equal.  The type of '<var>&lt;cmp&gt;</var>' must be an integer type whose
5210bit width is a power of two greater than or equal to eight and less than
5211or equal to a target-specific size limit. '<var>&lt;cmp&gt;</var>' and
5212'<var>&lt;new&gt;</var>' must have the same type, and the type of
5213'<var>&lt;pointer&gt;</var>' must be a pointer to that type. If the
5214<code>cmpxchg</code> is marked as <code>volatile</code>, then the
5215optimizer is not allowed to modify the number or order of execution
5216of this <code>cmpxchg</code> with other <a href="#volatile">volatile
5217operations</a>.</p>
5218
5219<!-- FIXME: Extend allowed types. -->
5220
5221<p>The <a href="#ordering"><var>ordering</var></a> argument specifies how this
5222<code>cmpxchg</code> synchronizes with other atomic operations.</p>
5223
5224<p>The optional "<code>singlethread</code>" argument declares that the
5225<code>cmpxchg</code> is only atomic with respect to code (usually signal
5226handlers) running in the same thread as the <code>cmpxchg</code>.  Otherwise the
5227cmpxchg is atomic with respect to all other code in the system.</p>
5228
5229<p>The pointer passed into cmpxchg must have alignment greater than or equal to
5230the size in memory of the operand.
5231
5232<h5>Semantics:</h5>
5233<p>The contents of memory at the location specified by the
5234'<tt>&lt;pointer&gt;</tt>' operand is read and compared to
5235'<tt>&lt;cmp&gt;</tt>'; if the read value is the equal,
5236'<tt>&lt;new&gt;</tt>' is written.  The original value at the location
5237is returned.
5238
5239<p>A successful <code>cmpxchg</code> is a read-modify-write instruction for the
5240purpose of identifying <a href="#release_sequence">release sequences</a>.  A
5241failed <code>cmpxchg</code> is equivalent to an atomic load with an ordering
5242parameter determined by dropping any <code>release</code> part of the
5243<code>cmpxchg</code>'s ordering.</p>
5244
5245<!--
5246FIXME: Is compare_exchange_weak() necessary?  (Consider after we've done
5247optimization work on ARM.)
5248
5249FIXME: Is a weaker ordering constraint on failure helpful in practice?
5250-->
5251
5252<h5>Example:</h5>
5253<pre>
5254entry:
5255  %orig = atomic <a href="#i_load">load</a> i32* %ptr unordered                   <i>; yields {i32}</i>
5256  <a href="#i_br">br</a> label %loop
5257
5258loop:
5259  %cmp = <a href="#i_phi">phi</a> i32 [ %orig, %entry ], [%old, %loop]
5260  %squared = <a href="#i_mul">mul</a> i32 %cmp, %cmp
5261  %old = cmpxchg i32* %ptr, i32 %cmp, i32 %squared          <i>; yields {i32}</i>
5262  %success = <a href="#i_icmp">icmp</a> eq i32 %cmp, %old
5263  <a href="#i_br">br</a> i1 %success, label %done, label %loop
5264
5265done:
5266  ...
5267</pre>
5268
5269</div>
5270
5271<!-- _______________________________________________________________________ -->
5272<h4>
5273<a name="i_atomicrmw">'<tt>atomicrmw</tt>' Instruction</a>
5274</h4>
5275
5276<div>
5277
5278<h5>Syntax:</h5>
5279<pre>
5280  atomicrmw [volatile] &lt;operation&gt; &lt;ty&gt;* &lt;pointer&gt;, &lt;ty&gt; &lt;value&gt; [singlethread] &lt;ordering&gt;                   <i>; yields {ty}</i>
5281</pre>
5282
5283<h5>Overview:</h5>
5284<p>The '<tt>atomicrmw</tt>' instruction is used to atomically modify memory.</p>
5285
5286<h5>Arguments:</h5>
5287<p>There are three arguments to the '<code>atomicrmw</code>' instruction: an
5288operation to apply, an address whose value to modify, an argument to the
5289operation.  The operation must be one of the following keywords:</p>
5290<ul>
5291  <li>xchg</li>
5292  <li>add</li>
5293  <li>sub</li>
5294  <li>and</li>
5295  <li>nand</li>
5296  <li>or</li>
5297  <li>xor</li>
5298  <li>max</li>
5299  <li>min</li>
5300  <li>umax</li>
5301  <li>umin</li>
5302</ul>
5303
5304<p>The type of '<var>&lt;value&gt;</var>' must be an integer type whose
5305bit width is a power of two greater than or equal to eight and less than
5306or equal to a target-specific size limit.  The type of the
5307'<code>&lt;pointer&gt;</code>' operand must be a pointer to that type.
5308If the <code>atomicrmw</code> is marked as <code>volatile</code>, then the
5309optimizer is not allowed to modify the number or order of execution of this
5310<code>atomicrmw</code> with other <a href="#volatile">volatile
5311  operations</a>.</p>
5312
5313<!-- FIXME: Extend allowed types. -->
5314
5315<h5>Semantics:</h5>
5316<p>The contents of memory at the location specified by the
5317'<tt>&lt;pointer&gt;</tt>' operand are atomically read, modified, and written
5318back.  The original value at the location is returned.  The modification is
5319specified by the <var>operation</var> argument:</p>
5320
5321<ul>
5322  <li>xchg: <code>*ptr = val</code></li>
5323  <li>add: <code>*ptr = *ptr + val</code></li>
5324  <li>sub: <code>*ptr = *ptr - val</code></li>
5325  <li>and: <code>*ptr = *ptr &amp; val</code></li>
5326  <li>nand: <code>*ptr = ~(*ptr &amp; val)</code></li>
5327  <li>or: <code>*ptr = *ptr | val</code></li>
5328  <li>xor: <code>*ptr = *ptr ^ val</code></li>
5329  <li>max: <code>*ptr = *ptr &gt; val ? *ptr : val</code> (using a signed comparison)</li>
5330  <li>min: <code>*ptr = *ptr &lt; val ? *ptr : val</code> (using a signed comparison)</li>
5331  <li>umax: <code>*ptr = *ptr &gt; val ? *ptr : val</code> (using an unsigned comparison)</li>
5332  <li>umin: <code>*ptr = *ptr &lt; val ? *ptr : val</code> (using an unsigned comparison)</li>
5333</ul>
5334
5335<h5>Example:</h5>
5336<pre>
5337  %old = atomicrmw add i32* %ptr, i32 1 acquire                        <i>; yields {i32}</i>
5338</pre>
5339
5340</div>
5341
5342<!-- _______________________________________________________________________ -->
5343<h4>
5344   <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
5345</h4>
5346
5347<div>
5348
5349<h5>Syntax:</h5>
5350<pre>
5351  &lt;result&gt; = getelementptr &lt;pty&gt;* &lt;ptrval&gt;{, &lt;ty&gt; &lt;idx&gt;}*
5352  &lt;result&gt; = getelementptr inbounds &lt;pty&gt;* &lt;ptrval&gt;{, &lt;ty&gt; &lt;idx&gt;}*
5353  &lt;result&gt; = getelementptr &lt;ptr vector&gt; ptrval, &lt;vector index type&gt; idx
5354</pre>
5355
5356<h5>Overview:</h5>
5357<p>The '<tt>getelementptr</tt>' instruction is used to get the address of a
5358   subelement of an <a href="#t_aggregate">aggregate</a> data structure.
5359   It performs address calculation only and does not access memory.</p>
5360
5361<h5>Arguments:</h5>
5362<p>The first argument is always a pointer or a vector of pointers,
5363   and forms the basis of the
5364   calculation. The remaining arguments are indices that indicate which of the
5365   elements of the aggregate object are indexed. The interpretation of each
5366   index is dependent on the type being indexed into. The first index always
5367   indexes the pointer value given as the first argument, the second index
5368   indexes a value of the type pointed to (not necessarily the value directly
5369   pointed to, since the first index can be non-zero), etc. The first type
5370   indexed into must be a pointer value, subsequent types can be arrays,
5371   vectors, and structs. Note that subsequent types being indexed into
5372   can never be pointers, since that would require loading the pointer before
5373   continuing calculation.</p>
5374
5375<p>The type of each index argument depends on the type it is indexing into.
5376   When indexing into a (optionally packed) structure, only <tt>i32</tt>
5377   integer <b>constants</b> are allowed.  When indexing into an array, pointer
5378   or vector, integers of any width are allowed, and they are not required to be
5379   constant.  These integers are treated as signed values where relevant.</p>
5380
5381<p>For example, let's consider a C code fragment and how it gets compiled to
5382   LLVM:</p>
5383
5384<pre class="doc_code">
5385struct RT {
5386  char A;
5387  int B[10][20];
5388  char C;
5389};
5390struct ST {
5391  int X;
5392  double Y;
5393  struct RT Z;
5394};
5395
5396int *foo(struct ST *s) {
5397  return &amp;s[1].Z.B[5][13];
5398}
5399</pre>
5400
5401<p>The LLVM code generated by Clang is:</p>
5402
5403<pre class="doc_code">
5404%struct.RT = <a href="#namedtypes">type</a> { i8, [10 x [20 x i32]], i8 }
5405%struct.ST = <a href="#namedtypes">type</a> { i32, double, %struct.RT }
5406
5407define i32* @foo(%struct.ST* %s) nounwind uwtable readnone optsize ssp {
5408entry:
5409  %arrayidx = getelementptr inbounds %struct.ST* %s, i64 1, i32 2, i32 1, i64 5, i64 13
5410  ret i32* %arrayidx
5411}
5412</pre>
5413
5414<h5>Semantics:</h5>
5415<p>In the example above, the first index is indexing into the
5416   '<tt>%struct.ST*</tt>' type, which is a pointer, yielding a
5417   '<tt>%struct.ST</tt>' = '<tt>{ i32, double, %struct.RT }</tt>' type, a
5418   structure. The second index indexes into the third element of the structure,
5419   yielding a '<tt>%struct.RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]], i8 }</tt>'
5420   type, another structure. The third index indexes into the second element of
5421   the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an array. The
5422   two dimensions of the array are subscripted into, yielding an '<tt>i32</tt>'
5423   type. The '<tt>getelementptr</tt>' instruction returns a pointer to this
5424   element, thus computing a value of '<tt>i32*</tt>' type.</p>
5425
5426<p>Note that it is perfectly legal to index partially through a structure,
5427   returning a pointer to an inner element.  Because of this, the LLVM code for
5428   the given testcase is equivalent to:</p>
5429
5430<pre class="doc_code">
5431define i32* @foo(%struct.ST* %s) {
5432  %t1 = getelementptr %struct.ST* %s, i32 1                 <i>; yields %struct.ST*:%t1</i>
5433  %t2 = getelementptr %struct.ST* %t1, i32 0, i32 2         <i>; yields %struct.RT*:%t2</i>
5434  %t3 = getelementptr %struct.RT* %t2, i32 0, i32 1         <i>; yields [10 x [20 x i32]]*:%t3</i>
5435  %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5  <i>; yields [20 x i32]*:%t4</i>
5436  %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13        <i>; yields i32*:%t5</i>
5437  ret i32* %t5
5438}
5439</pre>
5440
5441<p>If the <tt>inbounds</tt> keyword is present, the result value of the
5442   <tt>getelementptr</tt> is a <a href="#poisonvalues">poison value</a> if the
5443   base pointer is not an <i>in bounds</i> address of an allocated object,
5444   or if any of the addresses that would be formed by successive addition of
5445   the offsets implied by the indices to the base address with infinitely
5446   precise signed arithmetic are not an <i>in bounds</i> address of that
5447   allocated object. The <i>in bounds</i> addresses for an allocated object
5448   are all the addresses that point into the object, plus the address one
5449   byte past the end.
5450   In cases where the base is a vector of pointers the <tt>inbounds</tt> keyword
5451   applies to each of the computations element-wise. </p>
5452
5453<p>If the <tt>inbounds</tt> keyword is not present, the offsets are added to
5454   the base address with silently-wrapping two's complement arithmetic. If the
5455   offsets have a different width from the pointer, they are sign-extended or
5456   truncated to the width of the pointer. The result value of the
5457   <tt>getelementptr</tt> may be outside the object pointed to by the base
5458   pointer. The result value may not necessarily be used to access memory
5459   though, even if it happens to point into allocated storage. See the
5460   <a href="#pointeraliasing">Pointer Aliasing Rules</a> section for more
5461   information.</p>
5462
5463<p>The getelementptr instruction is often confusing.  For some more insight into
5464   how it works, see <a href="GetElementPtr.html">the getelementptr FAQ</a>.</p>
5465
5466<h5>Example:</h5>
5467<pre>
5468    <i>; yields [12 x i8]*:aptr</i>
5469    %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
5470    <i>; yields i8*:vptr</i>
5471    %vptr = getelementptr {i32, &lt;2 x i8&gt;}* %svptr, i64 0, i32 1, i32 1
5472    <i>; yields i8*:eptr</i>
5473    %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
5474    <i>; yields i32*:iptr</i>
5475    %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
5476</pre>
5477
5478<p>In cases where the pointer argument is a vector of pointers, only a
5479   single index may be used, and the number of vector elements has to be
5480   the same.  For example: </p>
5481<pre class="doc_code">
5482 %A = getelementptr <4 x i8*> %ptrs, <4 x i64> %offsets,
5483</pre>
5484
5485</div>
5486
5487</div>
5488
5489<!-- ======================================================================= -->
5490<h3>
5491  <a name="convertops">Conversion Operations</a>
5492</h3>
5493
5494<div>
5495
5496<p>The instructions in this category are the conversion instructions (casting)
5497   which all take a single operand and a type. They perform various bit
5498   conversions on the operand.</p>
5499
5500<!-- _______________________________________________________________________ -->
5501<h4>
5502   <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
5503</h4>
5504
5505<div>
5506
5507<h5>Syntax:</h5>
5508<pre>
5509  &lt;result&gt; = trunc &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt;             <i>; yields ty2</i>
5510</pre>
5511
5512<h5>Overview:</h5>
5513<p>The '<tt>trunc</tt>' instruction truncates its operand to the
5514   type <tt>ty2</tt>.</p>
5515
5516<h5>Arguments:</h5>
5517<p>The '<tt>trunc</tt>' instruction takes a value to trunc, and a type to trunc it to.
5518   Both types must be of <a href="#t_integer">integer</a> types, or vectors
5519   of the same number of integers.
5520   The bit size of the <tt>value</tt> must be larger than
5521   the bit size of the destination type, <tt>ty2</tt>.
5522   Equal sized types are not allowed.</p>
5523
5524<h5>Semantics:</h5>
5525<p>The '<tt>trunc</tt>' instruction truncates the high order bits
5526   in <tt>value</tt> and converts the remaining bits to <tt>ty2</tt>. Since the
5527   source size must be larger than the destination size, <tt>trunc</tt> cannot
5528   be a <i>no-op cast</i>.  It will always truncate bits.</p>
5529
5530<h5>Example:</h5>
5531<pre>
5532  %X = trunc i32 257 to i8                        <i>; yields i8:1</i>
5533  %Y = trunc i32 123 to i1                        <i>; yields i1:true</i>
5534  %Z = trunc i32 122 to i1                        <i>; yields i1:false</i>
5535  %W = trunc &lt;2 x i16&gt; &lt;i16 8, i16 7&gt; to &lt;2 x i8&gt; <i>; yields &lt;i8 8, i8 7&gt;</i>
5536</pre>
5537
5538</div>
5539
5540<!-- _______________________________________________________________________ -->
5541<h4>
5542   <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
5543</h4>
5544
5545<div>
5546
5547<h5>Syntax:</h5>
5548<pre>
5549  &lt;result&gt; = zext &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt;             <i>; yields ty2</i>
5550</pre>
5551
5552<h5>Overview:</h5>
5553<p>The '<tt>zext</tt>' instruction zero extends its operand to type
5554   <tt>ty2</tt>.</p>
5555
5556
5557<h5>Arguments:</h5>
5558<p>The '<tt>zext</tt>' instruction takes a value to cast, and a type to cast it to.
5559   Both types must be of <a href="#t_integer">integer</a> types, or vectors
5560   of the same number of integers.
5561   The bit size of the <tt>value</tt> must be smaller than
5562   the bit size of the destination type,
5563   <tt>ty2</tt>.</p>
5564
5565<h5>Semantics:</h5>
5566<p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
5567   bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
5568
5569<p>When zero extending from i1, the result will always be either 0 or 1.</p>
5570
5571<h5>Example:</h5>
5572<pre>
5573  %X = zext i32 257 to i64              <i>; yields i64:257</i>
5574  %Y = zext i1 true to i32              <i>; yields i32:1</i>
5575  %Z = zext &lt;2 x i16&gt; &lt;i16 8, i16 7&gt; to &lt;2 x i32&gt; <i>; yields &lt;i32 8, i32 7&gt;</i>
5576</pre>
5577
5578</div>
5579
5580<!-- _______________________________________________________________________ -->
5581<h4>
5582   <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
5583</h4>
5584
5585<div>
5586
5587<h5>Syntax:</h5>
5588<pre>
5589  &lt;result&gt; = sext &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt;             <i>; yields ty2</i>
5590</pre>
5591
5592<h5>Overview:</h5>
5593<p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
5594
5595<h5>Arguments:</h5>
5596<p>The '<tt>sext</tt>' instruction takes a value to cast, and a type to cast it to.
5597   Both types must be of <a href="#t_integer">integer</a> types, or vectors
5598   of the same number of integers.
5599   The bit size of the <tt>value</tt> must be smaller than
5600   the bit size of the destination type,
5601   <tt>ty2</tt>.</p>
5602
5603<h5>Semantics:</h5>
5604<p>The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
5605   bit (highest order bit) of the <tt>value</tt> until it reaches the bit size
5606   of the type <tt>ty2</tt>.</p>
5607
5608<p>When sign extending from i1, the extension always results in -1 or 0.</p>
5609
5610<h5>Example:</h5>
5611<pre>
5612  %X = sext i8  -1 to i16              <i>; yields i16   :65535</i>
5613  %Y = sext i1 true to i32             <i>; yields i32:-1</i>
5614  %Z = sext &lt;2 x i16&gt; &lt;i16 8, i16 7&gt; to &lt;2 x i32&gt; <i>; yields &lt;i32 8, i32 7&gt;</i>
5615</pre>
5616
5617</div>
5618
5619<!-- _______________________________________________________________________ -->
5620<h4>
5621   <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
5622</h4>
5623
5624<div>
5625
5626<h5>Syntax:</h5>
5627<pre>
5628  &lt;result&gt; = fptrunc &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt;             <i>; yields ty2</i>
5629</pre>
5630
5631<h5>Overview:</h5>
5632<p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
5633   <tt>ty2</tt>.</p>
5634
5635<h5>Arguments:</h5>
5636<p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
5637   point</a> value to cast and a <a href="#t_floating">floating point</a> type
5638   to cast it to. The size of <tt>value</tt> must be larger than the size of
5639   <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
5640   <i>no-op cast</i>.</p>
5641
5642<h5>Semantics:</h5>
5643<p>The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
5644   <a href="#t_floating">floating point</a> type to a smaller
5645   <a href="#t_floating">floating point</a> type.  If the value cannot fit
5646   within the destination type, <tt>ty2</tt>, then the results are
5647   undefined.</p>
5648
5649<h5>Example:</h5>
5650<pre>
5651  %X = fptrunc double 123.0 to float         <i>; yields float:123.0</i>
5652  %Y = fptrunc double 1.0E+300 to float      <i>; yields undefined</i>
5653</pre>
5654
5655</div>
5656
5657<!-- _______________________________________________________________________ -->
5658<h4>
5659   <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
5660</h4>
5661
5662<div>
5663
5664<h5>Syntax:</h5>
5665<pre>
5666  &lt;result&gt; = fpext &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt;             <i>; yields ty2</i>
5667</pre>
5668
5669<h5>Overview:</h5>
5670<p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
5671   floating point value.</p>
5672
5673<h5>Arguments:</h5>
5674<p>The '<tt>fpext</tt>' instruction takes a
5675   <a href="#t_floating">floating point</a> <tt>value</tt> to cast, and
5676   a <a href="#t_floating">floating point</a> type to cast it to. The source
5677   type must be smaller than the destination type.</p>
5678
5679<h5>Semantics:</h5>
5680<p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
5681   <a href="#t_floating">floating point</a> type to a larger
5682   <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
5683   used to make a <i>no-op cast</i> because it always changes bits. Use
5684   <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
5685
5686<h5>Example:</h5>
5687<pre>
5688  %X = fpext float 3.125 to double         <i>; yields double:3.125000e+00</i>
5689  %Y = fpext double %X to fp128            <i>; yields fp128:0xL00000000000000004000900000000000</i>
5690</pre>
5691
5692</div>
5693
5694<!-- _______________________________________________________________________ -->
5695<h4>
5696   <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
5697</h4>
5698
5699<div>
5700
5701<h5>Syntax:</h5>
5702<pre>
5703  &lt;result&gt; = fptoui &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt;             <i>; yields ty2</i>
5704</pre>
5705
5706<h5>Overview:</h5>
5707<p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
5708   unsigned integer equivalent of type <tt>ty2</tt>.</p>
5709
5710<h5>Arguments:</h5>
5711<p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
5712   scalar or vector <a href="#t_floating">floating point</a> value, and a type
5713   to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5714   type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5715   vector integer type with the same number of elements as <tt>ty</tt></p>
5716
5717<h5>Semantics:</h5>
5718<p>The '<tt>fptoui</tt>' instruction converts its
5719   <a href="#t_floating">floating point</a> operand into the nearest (rounding
5720   towards zero) unsigned integer value. If the value cannot fit
5721   in <tt>ty2</tt>, the results are undefined.</p>
5722
5723<h5>Example:</h5>
5724<pre>
5725  %X = fptoui double 123.0 to i32      <i>; yields i32:123</i>
5726  %Y = fptoui float 1.0E+300 to i1     <i>; yields undefined:1</i>
5727  %Z = fptoui float 1.04E+17 to i8     <i>; yields undefined:1</i>
5728</pre>
5729
5730</div>
5731
5732<!-- _______________________________________________________________________ -->
5733<h4>
5734   <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
5735</h4>
5736
5737<div>
5738
5739<h5>Syntax:</h5>
5740<pre>
5741  &lt;result&gt; = fptosi &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt;             <i>; yields ty2</i>
5742</pre>
5743
5744<h5>Overview:</h5>
5745<p>The '<tt>fptosi</tt>' instruction converts
5746   <a href="#t_floating">floating point</a> <tt>value</tt> to
5747   type <tt>ty2</tt>.</p>
5748
5749<h5>Arguments:</h5>
5750<p>The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
5751   scalar or vector <a href="#t_floating">floating point</a> value, and a type
5752   to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
5753   type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
5754   vector integer type with the same number of elements as <tt>ty</tt></p>
5755
5756<h5>Semantics:</h5>
5757<p>The '<tt>fptosi</tt>' instruction converts its
5758   <a href="#t_floating">floating point</a> operand into the nearest (rounding
5759   towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
5760   the results are undefined.</p>
5761
5762<h5>Example:</h5>
5763<pre>
5764  %X = fptosi double -123.0 to i32      <i>; yields i32:-123</i>
5765  %Y = fptosi float 1.0E-247 to i1      <i>; yields undefined:1</i>
5766  %Z = fptosi float 1.04E+17 to i8      <i>; yields undefined:1</i>
5767</pre>
5768
5769</div>
5770
5771<!-- _______________________________________________________________________ -->
5772<h4>
5773   <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
5774</h4>
5775
5776<div>
5777
5778<h5>Syntax:</h5>
5779<pre>
5780  &lt;result&gt; = uitofp &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt;             <i>; yields ty2</i>
5781</pre>
5782
5783<h5>Overview:</h5>
5784<p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
5785   integer and converts that value to the <tt>ty2</tt> type.</p>
5786
5787<h5>Arguments:</h5>
5788<p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
5789   scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
5790   it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
5791   type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
5792   floating point type with the same number of elements as <tt>ty</tt></p>
5793
5794<h5>Semantics:</h5>
5795<p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
5796   integer quantity and converts it to the corresponding floating point
5797   value. If the value cannot fit in the floating point value, the results are
5798   undefined.</p>
5799
5800<h5>Example:</h5>
5801<pre>
5802  %X = uitofp i32 257 to float         <i>; yields float:257.0</i>
5803  %Y = uitofp i8 -1 to double          <i>; yields double:255.0</i>
5804</pre>
5805
5806</div>
5807
5808<!-- _______________________________________________________________________ -->
5809<h4>
5810   <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
5811</h4>
5812
5813<div>
5814
5815<h5>Syntax:</h5>
5816<pre>
5817  &lt;result&gt; = sitofp &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt;             <i>; yields ty2</i>
5818</pre>
5819
5820<h5>Overview:</h5>
5821<p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed integer
5822   and converts that value to the <tt>ty2</tt> type.</p>
5823
5824<h5>Arguments:</h5>
5825<p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
5826   scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
5827   it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
5828   type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
5829   floating point type with the same number of elements as <tt>ty</tt></p>
5830
5831<h5>Semantics:</h5>
5832<p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed integer
5833   quantity and converts it to the corresponding floating point value. If the
5834   value cannot fit in the floating point value, the results are undefined.</p>
5835
5836<h5>Example:</h5>
5837<pre>
5838  %X = sitofp i32 257 to float         <i>; yields float:257.0</i>
5839  %Y = sitofp i8 -1 to double          <i>; yields double:-1.0</i>
5840</pre>
5841
5842</div>
5843
5844<!-- _______________________________________________________________________ -->
5845<h4>
5846   <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
5847</h4>
5848
5849<div>
5850
5851<h5>Syntax:</h5>
5852<pre>
5853  &lt;result&gt; = ptrtoint &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt;             <i>; yields ty2</i>
5854</pre>
5855
5856<h5>Overview:</h5>
5857<p>The '<tt>ptrtoint</tt>' instruction converts the pointer or a vector of
5858   pointers <tt>value</tt> to
5859   the integer (or vector of integers) type <tt>ty2</tt>.</p>
5860
5861<h5>Arguments:</h5>
5862<p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
5863   must be a a value of type <a href="#t_pointer">pointer</a> or a vector of
5864    pointers, and a type to cast it to
5865   <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> or a vector
5866   of integers type.</p>
5867
5868<h5>Semantics:</h5>
5869<p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
5870   <tt>ty2</tt> by interpreting the pointer value as an integer and either
5871   truncating or zero extending that value to the size of the integer type. If
5872   <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
5873   <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
5874   are the same size, then nothing is done (<i>no-op cast</i>) other than a type
5875   change.</p>
5876
5877<h5>Example:</h5>
5878<pre>
5879  %X = ptrtoint i32* %P to i8                         <i>; yields truncation on 32-bit architecture</i>
5880  %Y = ptrtoint i32* %P to i64                        <i>; yields zero extension on 32-bit architecture</i>
5881  %Z = ptrtoint &lt;4 x i32*&gt; %P to &lt;4 x i64&gt;<i>; yields vector zero extension for a vector of addresses on 32-bit architecture</i>
5882</pre>
5883
5884</div>
5885
5886<!-- _______________________________________________________________________ -->
5887<h4>
5888   <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
5889</h4>
5890
5891<div>
5892
5893<h5>Syntax:</h5>
5894<pre>
5895  &lt;result&gt; = inttoptr &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt;             <i>; yields ty2</i>
5896</pre>
5897
5898<h5>Overview:</h5>
5899<p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to a
5900   pointer type, <tt>ty2</tt>.</p>
5901
5902<h5>Arguments:</h5>
5903<p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
5904   value to cast, and a type to cast it to, which must be a
5905   <a href="#t_pointer">pointer</a> type.</p>
5906
5907<h5>Semantics:</h5>
5908<p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
5909   <tt>ty2</tt> by applying either a zero extension or a truncation depending on
5910   the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
5911   size of a pointer then a truncation is done. If <tt>value</tt> is smaller
5912   than the size of a pointer then a zero extension is done. If they are the
5913   same size, nothing is done (<i>no-op cast</i>).</p>
5914
5915<h5>Example:</h5>
5916<pre>
5917  %X = inttoptr i32 255 to i32*          <i>; yields zero extension on 64-bit architecture</i>
5918  %Y = inttoptr i32 255 to i32*          <i>; yields no-op on 32-bit architecture</i>
5919  %Z = inttoptr i64 0 to i32*            <i>; yields truncation on 32-bit architecture</i>
5920  %Z = inttoptr &lt;4 x i32&gt; %G to &lt;4 x i8*&gt;<i>; yields truncation of vector G to four pointers</i>
5921</pre>
5922
5923</div>
5924
5925<!-- _______________________________________________________________________ -->
5926<h4>
5927   <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
5928</h4>
5929
5930<div>
5931
5932<h5>Syntax:</h5>
5933<pre>
5934  &lt;result&gt; = bitcast &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt;             <i>; yields ty2</i>
5935</pre>
5936
5937<h5>Overview:</h5>
5938<p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
5939   <tt>ty2</tt> without changing any bits.</p>
5940
5941<h5>Arguments:</h5>
5942<p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be a
5943   non-aggregate first class value, and a type to cast it to, which must also be
5944   a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes
5945   of <tt>value</tt> and the destination type, <tt>ty2</tt>, must be
5946   identical. If the source type is a pointer, the destination type must also be
5947   a pointer.  This instruction supports bitwise conversion of vectors to
5948   integers and to vectors of other types (as long as they have the same
5949   size).</p>
5950
5951<h5>Semantics:</h5>
5952<p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
5953   <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
5954   this conversion.  The conversion is done as if the <tt>value</tt> had been
5955   stored to memory and read back as type <tt>ty2</tt>.
5956   Pointer (or vector of pointers) types may only be converted to other pointer
5957   (or vector of pointers) types with this instruction. To convert
5958   pointers to other types, use the <a href="#i_inttoptr">inttoptr</a> or
5959   <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
5960
5961<h5>Example:</h5>
5962<pre>
5963  %X = bitcast i8 255 to i8              <i>; yields i8 :-1</i>
5964  %Y = bitcast i32* %x to sint*          <i>; yields sint*:%x</i>
5965  %Z = bitcast &lt;2 x int&gt; %V to i64;        <i>; yields i64: %V</i>
5966  %Z = bitcast &lt;2 x i32*&gt; %V to &lt;2 x i64*&gt; <i>; yields &lt;2 x i64*&gt;</i>
5967</pre>
5968
5969</div>
5970
5971</div>
5972
5973<!-- ======================================================================= -->
5974<h3>
5975  <a name="otherops">Other Operations</a>
5976</h3>
5977
5978<div>
5979
5980<p>The instructions in this category are the "miscellaneous" instructions, which
5981   defy better classification.</p>
5982
5983<!-- _______________________________________________________________________ -->
5984<h4>
5985  <a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
5986</h4>
5987
5988<div>
5989
5990<h5>Syntax:</h5>
5991<pre>
5992  &lt;result&gt; = icmp &lt;cond&gt; &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt;   <i>; yields {i1} or {&lt;N x i1&gt;}:result</i>
5993</pre>
5994
5995<h5>Overview:</h5>
5996<p>The '<tt>icmp</tt>' instruction returns a boolean value or a vector of
5997   boolean values based on comparison of its two integer, integer vector,
5998   pointer, or pointer vector operands.</p>
5999
6000<h5>Arguments:</h5>
6001<p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
6002   the condition code indicating the kind of comparison to perform. It is not a
6003   value, just a keyword. The possible condition code are:</p>
6004
6005<ol>
6006  <li><tt>eq</tt>: equal</li>
6007  <li><tt>ne</tt>: not equal </li>
6008  <li><tt>ugt</tt>: unsigned greater than</li>
6009  <li><tt>uge</tt>: unsigned greater or equal</li>
6010  <li><tt>ult</tt>: unsigned less than</li>
6011  <li><tt>ule</tt>: unsigned less or equal</li>
6012  <li><tt>sgt</tt>: signed greater than</li>
6013  <li><tt>sge</tt>: signed greater or equal</li>
6014  <li><tt>slt</tt>: signed less than</li>
6015  <li><tt>sle</tt>: signed less or equal</li>
6016</ol>
6017
6018<p>The remaining two arguments must be <a href="#t_integer">integer</a> or
6019   <a href="#t_pointer">pointer</a> or integer <a href="#t_vector">vector</a>
6020   typed.  They must also be identical types.</p>
6021
6022<h5>Semantics:</h5>
6023<p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to the
6024   condition code given as <tt>cond</tt>. The comparison performed always yields
6025   either an <a href="#t_integer"><tt>i1</tt></a> or vector of <tt>i1</tt>
6026   result, as follows:</p>
6027
6028<ol>
6029  <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
6030      <tt>false</tt> otherwise. No sign interpretation is necessary or
6031      performed.</li>
6032
6033  <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
6034      <tt>false</tt> otherwise. No sign interpretation is necessary or
6035      performed.</li>
6036
6037  <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
6038      <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
6039
6040  <li><tt>uge</tt>: interprets the operands as unsigned values and yields
6041      <tt>true</tt> if <tt>op1</tt> is greater than or equal
6042      to <tt>op2</tt>.</li>
6043
6044  <li><tt>ult</tt>: interprets the operands as unsigned values and yields
6045      <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
6046
6047  <li><tt>ule</tt>: interprets the operands as unsigned values and yields
6048      <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6049
6050  <li><tt>sgt</tt>: interprets the operands as signed values and yields
6051      <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
6052
6053  <li><tt>sge</tt>: interprets the operands as signed values and yields
6054      <tt>true</tt> if <tt>op1</tt> is greater than or equal
6055      to <tt>op2</tt>.</li>
6056
6057  <li><tt>slt</tt>: interprets the operands as signed values and yields
6058      <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
6059
6060  <li><tt>sle</tt>: interprets the operands as signed values and yields
6061      <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6062</ol>
6063
6064<p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
6065   values are compared as if they were integers.</p>
6066
6067<p>If the operands are integer vectors, then they are compared element by
6068   element. The result is an <tt>i1</tt> vector with the same number of elements
6069   as the values being compared.  Otherwise, the result is an <tt>i1</tt>.</p>
6070
6071<h5>Example:</h5>
6072<pre>
6073  &lt;result&gt; = icmp eq i32 4, 5          <i>; yields: result=false</i>
6074  &lt;result&gt; = icmp ne float* %X, %X     <i>; yields: result=false</i>
6075  &lt;result&gt; = icmp ult i16  4, 5        <i>; yields: result=true</i>
6076  &lt;result&gt; = icmp sgt i16  4, 5        <i>; yields: result=false</i>
6077  &lt;result&gt; = icmp ule i16 -4, 5        <i>; yields: result=false</i>
6078  &lt;result&gt; = icmp sge i16  4, 5        <i>; yields: result=false</i>
6079</pre>
6080
6081<p>Note that the code generator does not yet support vector types with
6082   the <tt>icmp</tt> instruction.</p>
6083
6084</div>
6085
6086<!-- _______________________________________________________________________ -->
6087<h4>
6088  <a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
6089</h4>
6090
6091<div>
6092
6093<h5>Syntax:</h5>
6094<pre>
6095  &lt;result&gt; = fcmp &lt;cond&gt; &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt;     <i>; yields {i1} or {&lt;N x i1&gt;}:result</i>
6096</pre>
6097
6098<h5>Overview:</h5>
6099<p>The '<tt>fcmp</tt>' instruction returns a boolean value or vector of boolean
6100   values based on comparison of its operands.</p>
6101
6102<p>If the operands are floating point scalars, then the result type is a boolean
6103(<a href="#t_integer"><tt>i1</tt></a>).</p>
6104
6105<p>If the operands are floating point vectors, then the result type is a vector
6106   of boolean with the same number of elements as the operands being
6107   compared.</p>
6108
6109<h5>Arguments:</h5>
6110<p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
6111   the condition code indicating the kind of comparison to perform. It is not a
6112   value, just a keyword. The possible condition code are:</p>
6113
6114<ol>
6115  <li><tt>false</tt>: no comparison, always returns false</li>
6116  <li><tt>oeq</tt>: ordered and equal</li>
6117  <li><tt>ogt</tt>: ordered and greater than </li>
6118  <li><tt>oge</tt>: ordered and greater than or equal</li>
6119  <li><tt>olt</tt>: ordered and less than </li>
6120  <li><tt>ole</tt>: ordered and less than or equal</li>
6121  <li><tt>one</tt>: ordered and not equal</li>
6122  <li><tt>ord</tt>: ordered (no nans)</li>
6123  <li><tt>ueq</tt>: unordered or equal</li>
6124  <li><tt>ugt</tt>: unordered or greater than </li>
6125  <li><tt>uge</tt>: unordered or greater than or equal</li>
6126  <li><tt>ult</tt>: unordered or less than </li>
6127  <li><tt>ule</tt>: unordered or less than or equal</li>
6128  <li><tt>une</tt>: unordered or not equal</li>
6129  <li><tt>uno</tt>: unordered (either nans)</li>
6130  <li><tt>true</tt>: no comparison, always returns true</li>
6131</ol>
6132
6133<p><i>Ordered</i> means that neither operand is a QNAN while
6134   <i>unordered</i> means that either operand may be a QNAN.</p>
6135
6136<p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be either
6137   a <a href="#t_floating">floating point</a> type or
6138   a <a href="#t_vector">vector</a> of floating point type.  They must have
6139   identical types.</p>
6140
6141<h5>Semantics:</h5>
6142<p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
6143   according to the condition code given as <tt>cond</tt>.  If the operands are
6144   vectors, then the vectors are compared element by element.  Each comparison
6145   performed always yields an <a href="#t_integer">i1</a> result, as
6146   follows:</p>
6147
6148<ol>
6149  <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
6150
6151  <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6152      <tt>op1</tt> is equal to <tt>op2</tt>.</li>
6153
6154  <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6155      <tt>op1</tt> is greater than <tt>op2</tt>.</li>
6156
6157  <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6158      <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
6159
6160  <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6161      <tt>op1</tt> is less than <tt>op2</tt>.</li>
6162
6163  <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6164      <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6165
6166  <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
6167      <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
6168
6169  <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
6170
6171  <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
6172      <tt>op1</tt> is equal to <tt>op2</tt>.</li>
6173
6174  <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
6175      <tt>op1</tt> is greater than <tt>op2</tt>.</li>
6176
6177  <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
6178      <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
6179
6180  <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
6181      <tt>op1</tt> is less than <tt>op2</tt>.</li>
6182
6183  <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
6184      <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
6185
6186  <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
6187      <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
6188
6189  <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
6190
6191  <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
6192</ol>
6193
6194<h5>Example:</h5>
6195<pre>
6196  &lt;result&gt; = fcmp oeq float 4.0, 5.0    <i>; yields: result=false</i>
6197  &lt;result&gt; = fcmp one float 4.0, 5.0    <i>; yields: result=true</i>
6198  &lt;result&gt; = fcmp olt float 4.0, 5.0    <i>; yields: result=true</i>
6199  &lt;result&gt; = fcmp ueq double 1.0, 2.0   <i>; yields: result=false</i>
6200</pre>
6201
6202<p>Note that the code generator does not yet support vector types with
6203   the <tt>fcmp</tt> instruction.</p>
6204
6205</div>
6206
6207<!-- _______________________________________________________________________ -->
6208<h4>
6209  <a name="i_phi">'<tt>phi</tt>' Instruction</a>
6210</h4>
6211
6212<div>
6213
6214<h5>Syntax:</h5>
6215<pre>
6216  &lt;result&gt; = phi &lt;ty&gt; [ &lt;val0&gt;, &lt;label0&gt;], ...
6217</pre>
6218
6219<h5>Overview:</h5>
6220<p>The '<tt>phi</tt>' instruction is used to implement the &#966; node in the
6221   SSA graph representing the function.</p>
6222
6223<h5>Arguments:</h5>
6224<p>The type of the incoming values is specified with the first type field. After
6225   this, the '<tt>phi</tt>' instruction takes a list of pairs as arguments, with
6226   one pair for each predecessor basic block of the current block.  Only values
6227   of <a href="#t_firstclass">first class</a> type may be used as the value
6228   arguments to the PHI node.  Only labels may be used as the label
6229   arguments.</p>
6230
6231<p>There must be no non-phi instructions between the start of a basic block and
6232   the PHI instructions: i.e. PHI instructions must be first in a basic
6233   block.</p>
6234
6235<p>For the purposes of the SSA form, the use of each incoming value is deemed to
6236   occur on the edge from the corresponding predecessor block to the current
6237   block (but after any definition of an '<tt>invoke</tt>' instruction's return
6238   value on the same edge).</p>
6239
6240<h5>Semantics:</h5>
6241<p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
6242   specified by the pair corresponding to the predecessor basic block that
6243   executed just prior to the current block.</p>
6244
6245<h5>Example:</h5>
6246<pre>
6247Loop:       ; Infinite loop that counts from 0 on up...
6248  %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
6249  %nextindvar = add i32 %indvar, 1
6250  br label %Loop
6251</pre>
6252
6253</div>
6254
6255<!-- _______________________________________________________________________ -->
6256<h4>
6257   <a name="i_select">'<tt>select</tt>' Instruction</a>
6258</h4>
6259
6260<div>
6261
6262<h5>Syntax:</h5>
6263<pre>
6264  &lt;result&gt; = select <i>selty</i> &lt;cond&gt;, &lt;ty&gt; &lt;val1&gt;, &lt;ty&gt; &lt;val2&gt;             <i>; yields ty</i>
6265
6266  <i>selty</i> is either i1 or {&lt;N x i1&gt;}
6267</pre>
6268
6269<h5>Overview:</h5>
6270<p>The '<tt>select</tt>' instruction is used to choose one value based on a
6271   condition, without branching.</p>
6272
6273
6274<h5>Arguments:</h5>
6275<p>The '<tt>select</tt>' instruction requires an 'i1' value or a vector of 'i1'
6276   values indicating the condition, and two values of the
6277   same <a href="#t_firstclass">first class</a> type.  If the val1/val2 are
6278   vectors and the condition is a scalar, then entire vectors are selected, not
6279   individual elements.</p>
6280
6281<h5>Semantics:</h5>
6282<p>If the condition is an i1 and it evaluates to 1, the instruction returns the
6283   first value argument; otherwise, it returns the second value argument.</p>
6284
6285<p>If the condition is a vector of i1, then the value arguments must be vectors
6286   of the same size, and the selection is done element by element.</p>
6287
6288<h5>Example:</h5>
6289<pre>
6290  %X = select i1 true, i8 17, i8 42          <i>; yields i8:17</i>
6291</pre>
6292
6293</div>
6294
6295<!-- _______________________________________________________________________ -->
6296<h4>
6297  <a name="i_call">'<tt>call</tt>' Instruction</a>
6298</h4>
6299
6300<div>
6301
6302<h5>Syntax:</h5>
6303<pre>
6304  &lt;result&gt; = [tail] call [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>] &lt;ty&gt; [&lt;fnty&gt;*] &lt;fnptrval&gt;(&lt;function args&gt;) [<a href="#fnattrs">fn attrs</a>]
6305</pre>
6306
6307<h5>Overview:</h5>
6308<p>The '<tt>call</tt>' instruction represents a simple function call.</p>
6309
6310<h5>Arguments:</h5>
6311<p>This instruction requires several arguments:</p>
6312
6313<ol>
6314  <li>The optional "tail" marker indicates that the callee function does not
6315      access any allocas or varargs in the caller.  Note that calls may be
6316      marked "tail" even if they do not occur before
6317      a <a href="#i_ret"><tt>ret</tt></a> instruction.  If the "tail" marker is
6318      present, the function call is eligible for tail call optimization,
6319      but <a href="CodeGenerator.html#tailcallopt">might not in fact be
6320      optimized into a jump</a>.  The code generator may optimize calls marked
6321      "tail" with either 1) automatic <a href="CodeGenerator.html#sibcallopt">
6322      sibling call optimization</a> when the caller and callee have
6323      matching signatures, or 2) forced tail call optimization when the
6324      following extra requirements are met:
6325      <ul>
6326        <li>Caller and callee both have the calling
6327            convention <tt>fastcc</tt>.</li>
6328        <li>The call is in tail position (ret immediately follows call and ret
6329            uses value of call or is void).</li>
6330        <li>Option <tt>-tailcallopt</tt> is enabled,
6331            or <code>llvm::GuaranteedTailCallOpt</code> is <code>true</code>.</li>
6332        <li><a href="CodeGenerator.html#tailcallopt">Platform specific
6333            constraints are met.</a></li>
6334      </ul>
6335  </li>
6336
6337  <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
6338      convention</a> the call should use.  If none is specified, the call
6339      defaults to using C calling conventions.  The calling convention of the
6340      call must match the calling convention of the target function, or else the
6341      behavior is undefined.</li>
6342
6343  <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
6344      return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
6345      '<tt>inreg</tt>' attributes are valid here.</li>
6346
6347  <li>'<tt>ty</tt>': the type of the call instruction itself which is also the
6348      type of the return value.  Functions that return no value are marked
6349      <tt><a href="#t_void">void</a></tt>.</li>
6350
6351  <li>'<tt>fnty</tt>': shall be the signature of the pointer to function value
6352      being invoked.  The argument types must match the types implied by this
6353      signature.  This type can be omitted if the function is not varargs and if
6354      the function type does not return a pointer to a function.</li>
6355
6356  <li>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
6357      be invoked. In most cases, this is a direct function invocation, but
6358      indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
6359      to function value.</li>
6360
6361  <li>'<tt>function args</tt>': argument list whose types match the function
6362      signature argument types and parameter attributes. All arguments must be
6363      of <a href="#t_firstclass">first class</a> type. If the function
6364      signature indicates the function accepts a variable number of arguments,
6365      the extra arguments can be specified.</li>
6366
6367  <li>The optional <a href="#fnattrs">function attributes</a> list. Only
6368      '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
6369      '<tt>readnone</tt>' attributes are valid here.</li>
6370</ol>
6371
6372<h5>Semantics:</h5>
6373<p>The '<tt>call</tt>' instruction is used to cause control flow to transfer to
6374   a specified function, with its incoming arguments bound to the specified
6375   values. Upon a '<tt><a href="#i_ret">ret</a></tt>' instruction in the called
6376   function, control flow continues with the instruction after the function
6377   call, and the return value of the function is bound to the result
6378   argument.</p>
6379
6380<h5>Example:</h5>
6381<pre>
6382  %retval = call i32 @test(i32 %argc)
6383  call i32 (i8*, ...)* @printf(i8* %msg, i32 12, i8 42)        <i>; yields i32</i>
6384  %X = tail call i32 @foo()                                    <i>; yields i32</i>
6385  %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo()  <i>; yields i32</i>
6386  call void %foo(i8 97 signext)
6387
6388  %struct.A = type { i32, i8 }
6389  %r = call %struct.A @foo()                        <i>; yields { 32, i8 }</i>
6390  %gr = extractvalue %struct.A %r, 0                <i>; yields i32</i>
6391  %gr1 = extractvalue %struct.A %r, 1               <i>; yields i8</i>
6392  %Z = call void @foo() noreturn                    <i>; indicates that %foo never returns normally</i>
6393  %ZZ = call zeroext i32 @bar()                     <i>; Return value is %zero extended</i>
6394</pre>
6395
6396<p>llvm treats calls to some functions with names and arguments that match the
6397standard C99 library as being the C99 library functions, and may perform
6398optimizations or generate code for them under that assumption.  This is
6399something we'd like to change in the future to provide better support for
6400freestanding environments and non-C-based languages.</p>
6401
6402</div>
6403
6404<!-- _______________________________________________________________________ -->
6405<h4>
6406  <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
6407</h4>
6408
6409<div>
6410
6411<h5>Syntax:</h5>
6412<pre>
6413  &lt;resultval&gt; = va_arg &lt;va_list*&gt; &lt;arglist&gt;, &lt;argty&gt;
6414</pre>
6415
6416<h5>Overview:</h5>
6417<p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
6418   the "variable argument" area of a function call.  It is used to implement the
6419   <tt>va_arg</tt> macro in C.</p>
6420
6421<h5>Arguments:</h5>
6422<p>This instruction takes a <tt>va_list*</tt> value and the type of the
6423   argument. It returns a value of the specified argument type and increments
6424   the <tt>va_list</tt> to point to the next argument.  The actual type
6425   of <tt>va_list</tt> is target specific.</p>
6426
6427<h5>Semantics:</h5>
6428<p>The '<tt>va_arg</tt>' instruction loads an argument of the specified type
6429   from the specified <tt>va_list</tt> and causes the <tt>va_list</tt> to point
6430   to the next argument.  For more information, see the variable argument
6431   handling <a href="#int_varargs">Intrinsic Functions</a>.</p>
6432
6433<p>It is legal for this instruction to be called in a function which does not
6434   take a variable number of arguments, for example, the <tt>vfprintf</tt>
6435   function.</p>
6436
6437<p><tt>va_arg</tt> is an LLVM instruction instead of
6438   an <a href="#intrinsics">intrinsic function</a> because it takes a type as an
6439   argument.</p>
6440
6441<h5>Example:</h5>
6442<p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
6443
6444<p>Note that the code generator does not yet fully support va_arg on many
6445   targets. Also, it does not currently support va_arg with aggregate types on
6446   any target.</p>
6447
6448</div>
6449
6450<!-- _______________________________________________________________________ -->
6451<h4>
6452  <a name="i_landingpad">'<tt>landingpad</tt>' Instruction</a>
6453</h4>
6454
6455<div>
6456
6457<h5>Syntax:</h5>
6458<pre>
6459  &lt;resultval&gt; = landingpad &lt;resultty&gt; personality &lt;type&gt; &lt;pers_fn&gt; &lt;clause&gt;+
6460  &lt;resultval&gt; = landingpad &lt;resultty&gt; personality &lt;type&gt; &lt;pers_fn&gt; cleanup &lt;clause&gt;*
6461
6462  &lt;clause&gt; := catch &lt;type&gt; &lt;value&gt;
6463  &lt;clause&gt; := filter &lt;array constant type&gt; &lt;array constant&gt;
6464</pre>
6465
6466<h5>Overview:</h5>
6467<p>The '<tt>landingpad</tt>' instruction is used by
6468   <a href="ExceptionHandling.html#overview">LLVM's exception handling
6469   system</a> to specify that a basic block is a landing pad &mdash; one where
6470   the exception lands, and corresponds to the code found in the
6471   <i><tt>catch</tt></i> portion of a <i><tt>try/catch</tt></i> sequence. It
6472   defines values supplied by the personality function (<tt>pers_fn</tt>) upon
6473   re-entry to the function. The <tt>resultval</tt> has the
6474   type <tt>resultty</tt>.</p>
6475
6476<h5>Arguments:</h5>
6477<p>This instruction takes a <tt>pers_fn</tt> value. This is the personality
6478   function associated with the unwinding mechanism. The optional
6479   <tt>cleanup</tt> flag indicates that the landing pad block is a cleanup.</p>
6480
6481<p>A <tt>clause</tt> begins with the clause type &mdash; <tt>catch</tt>
6482   or <tt>filter</tt> &mdash; and contains the global variable representing the
6483   "type" that may be caught or filtered respectively. Unlike the
6484   <tt>catch</tt> clause, the <tt>filter</tt> clause takes an array constant as
6485   its argument. Use "<tt>[0 x i8**] undef</tt>" for a filter which cannot
6486   throw. The '<tt>landingpad</tt>' instruction must contain <em>at least</em>
6487   one <tt>clause</tt> or the <tt>cleanup</tt> flag.</p>
6488
6489<h5>Semantics:</h5>
6490<p>The '<tt>landingpad</tt>' instruction defines the values which are set by the
6491   personality function (<tt>pers_fn</tt>) upon re-entry to the function, and
6492   therefore the "result type" of the <tt>landingpad</tt> instruction. As with
6493   calling conventions, how the personality function results are represented in
6494   LLVM IR is target specific.</p>
6495
6496<p>The clauses are applied in order from top to bottom. If two
6497   <tt>landingpad</tt> instructions are merged together through inlining, the
6498   clauses from the calling function are appended to the list of clauses.
6499   When the call stack is being unwound due to an exception being thrown, the
6500   exception is compared against each <tt>clause</tt> in turn.  If it doesn't
6501   match any of the clauses, and the <tt>cleanup</tt> flag is not set, then
6502   unwinding continues further up the call stack.</p>
6503
6504<p>The <tt>landingpad</tt> instruction has several restrictions:</p>
6505
6506<ul>
6507  <li>A landing pad block is a basic block which is the unwind destination of an
6508      '<tt>invoke</tt>' instruction.</li>
6509  <li>A landing pad block must have a '<tt>landingpad</tt>' instruction as its
6510      first non-PHI instruction.</li>
6511  <li>There can be only one '<tt>landingpad</tt>' instruction within the landing
6512      pad block.</li>
6513  <li>A basic block that is not a landing pad block may not include a
6514      '<tt>landingpad</tt>' instruction.</li>
6515  <li>All '<tt>landingpad</tt>' instructions in a function must have the same
6516      personality function.</li>
6517</ul>
6518
6519<h5>Example:</h5>
6520<pre>
6521  ;; A landing pad which can catch an integer.
6522  %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6523           catch i8** @_ZTIi
6524  ;; A landing pad that is a cleanup.
6525  %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6526           cleanup
6527  ;; A landing pad which can catch an integer and can only throw a double.
6528  %res = landingpad { i8*, i32 } personality i32 (...)* @__gxx_personality_v0
6529           catch i8** @_ZTIi
6530           filter [1 x i8**] [@_ZTId]
6531</pre>
6532
6533</div>
6534
6535</div>
6536
6537</div>
6538
6539<!-- *********************************************************************** -->
6540<h2><a name="intrinsics">Intrinsic Functions</a></h2>
6541<!-- *********************************************************************** -->
6542
6543<div>
6544
6545<p>LLVM supports the notion of an "intrinsic function".  These functions have
6546   well known names and semantics and are required to follow certain
6547   restrictions.  Overall, these intrinsics represent an extension mechanism for
6548   the LLVM language that does not require changing all of the transformations
6549   in LLVM when adding to the language (or the bitcode reader/writer, the
6550   parser, etc...).</p>
6551
6552<p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
6553   prefix is reserved in LLVM for intrinsic names; thus, function names may not
6554   begin with this prefix.  Intrinsic functions must always be external
6555   functions: you cannot define the body of intrinsic functions.  Intrinsic
6556   functions may only be used in call or invoke instructions: it is illegal to
6557   take the address of an intrinsic function.  Additionally, because intrinsic
6558   functions are part of the LLVM language, it is required if any are added that
6559   they be documented here.</p>
6560
6561<p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents a
6562   family of functions that perform the same operation but on different data
6563   types. Because LLVM can represent over 8 million different integer types,
6564   overloading is used commonly to allow an intrinsic function to operate on any
6565   integer type. One or more of the argument types or the result type can be
6566   overloaded to accept any integer type. Argument types may also be defined as
6567   exactly matching a previous argument's type or the result type. This allows
6568   an intrinsic function which accepts multiple arguments, but needs all of them
6569   to be of the same type, to only be overloaded with respect to a single
6570   argument or the result.</p>
6571
6572<p>Overloaded intrinsics will have the names of its overloaded argument types
6573   encoded into its function name, each preceded by a period. Only those types
6574   which are overloaded result in a name suffix. Arguments whose type is matched
6575   against another type do not. For example, the <tt>llvm.ctpop</tt> function
6576   can take an integer of any width and returns an integer of exactly the same
6577   integer width. This leads to a family of functions such as
6578   <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29
6579   %val)</tt>.  Only one type, the return type, is overloaded, and only one type
6580   suffix is required. Because the argument's type is matched against the return
6581   type, it does not require its own name suffix.</p>
6582
6583<p>To learn how to add an intrinsic function, please see the
6584   <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.</p>
6585
6586<!-- ======================================================================= -->
6587<h3>
6588  <a name="int_varargs">Variable Argument Handling Intrinsics</a>
6589</h3>
6590
6591<div>
6592
6593<p>Variable argument support is defined in LLVM with
6594   the <a href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
6595   intrinsic functions.  These functions are related to the similarly named
6596   macros defined in the <tt>&lt;stdarg.h&gt;</tt> header file.</p>
6597
6598<p>All of these functions operate on arguments that use a target-specific value
6599   type "<tt>va_list</tt>".  The LLVM assembly language reference manual does
6600   not define what this type is, so all transformations should be prepared to
6601   handle these functions regardless of the type used.</p>
6602
6603<p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
6604   instruction and the variable argument handling intrinsic functions are
6605   used.</p>
6606
6607<pre class="doc_code">
6608define i32 @test(i32 %X, ...) {
6609  ; Initialize variable argument processing
6610  %ap = alloca i8*
6611  %ap2 = bitcast i8** %ap to i8*
6612  call void @llvm.va_start(i8* %ap2)
6613
6614  ; Read a single integer argument
6615  %tmp = va_arg i8** %ap, i32
6616
6617  ; Demonstrate usage of llvm.va_copy and llvm.va_end
6618  %aq = alloca i8*
6619  %aq2 = bitcast i8** %aq to i8*
6620  call void @llvm.va_copy(i8* %aq2, i8* %ap2)
6621  call void @llvm.va_end(i8* %aq2)
6622
6623  ; Stop processing of arguments.
6624  call void @llvm.va_end(i8* %ap2)
6625  ret i32 %tmp
6626}
6627
6628declare void @llvm.va_start(i8*)
6629declare void @llvm.va_copy(i8*, i8*)
6630declare void @llvm.va_end(i8*)
6631</pre>
6632
6633<!-- _______________________________________________________________________ -->
6634<h4>
6635  <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
6636</h4>
6637
6638
6639<div>
6640
6641<h5>Syntax:</h5>
6642<pre>
6643  declare void %llvm.va_start(i8* &lt;arglist&gt;)
6644</pre>
6645
6646<h5>Overview:</h5>
6647<p>The '<tt>llvm.va_start</tt>' intrinsic initializes <tt>*&lt;arglist&gt;</tt>
6648   for subsequent use by <tt><a href="#i_va_arg">va_arg</a></tt>.</p>
6649
6650<h5>Arguments:</h5>
6651<p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
6652
6653<h5>Semantics:</h5>
6654<p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
6655   macro available in C.  In a target-dependent way, it initializes
6656   the <tt>va_list</tt> element to which the argument points, so that the next
6657   call to <tt>va_arg</tt> will produce the first variable argument passed to
6658   the function.  Unlike the C <tt>va_start</tt> macro, this intrinsic does not
6659   need to know the last argument of the function as the compiler can figure
6660   that out.</p>
6661
6662</div>
6663
6664<!-- _______________________________________________________________________ -->
6665<h4>
6666 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
6667</h4>
6668
6669<div>
6670
6671<h5>Syntax:</h5>
6672<pre>
6673  declare void @llvm.va_end(i8* &lt;arglist&gt;)
6674</pre>
6675
6676<h5>Overview:</h5>
6677<p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*&lt;arglist&gt;</tt>,
6678   which has been initialized previously
6679   with <tt><a href="#int_va_start">llvm.va_start</a></tt>
6680   or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
6681
6682<h5>Arguments:</h5>
6683<p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
6684
6685<h5>Semantics:</h5>
6686<p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
6687   macro available in C.  In a target-dependent way, it destroys
6688   the <tt>va_list</tt> element to which the argument points.  Calls
6689   to <a href="#int_va_start"><tt>llvm.va_start</tt></a>
6690   and <a href="#int_va_copy"> <tt>llvm.va_copy</tt></a> must be matched exactly
6691   with calls to <tt>llvm.va_end</tt>.</p>
6692
6693</div>
6694
6695<!-- _______________________________________________________________________ -->
6696<h4>
6697  <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
6698</h4>
6699
6700<div>
6701
6702<h5>Syntax:</h5>
6703<pre>
6704  declare void @llvm.va_copy(i8* &lt;destarglist&gt;, i8* &lt;srcarglist&gt;)
6705</pre>
6706
6707<h5>Overview:</h5>
6708<p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
6709   from the source argument list to the destination argument list.</p>
6710
6711<h5>Arguments:</h5>
6712<p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
6713   The second argument is a pointer to a <tt>va_list</tt> element to copy
6714   from.</p>
6715
6716<h5>Semantics:</h5>
6717<p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
6718   macro available in C.  In a target-dependent way, it copies the
6719   source <tt>va_list</tt> element into the destination <tt>va_list</tt>
6720   element.  This intrinsic is necessary because
6721   the <tt><a href="#int_va_start"> llvm.va_start</a></tt> intrinsic may be
6722   arbitrarily complex and require, for example, memory allocation.</p>
6723
6724</div>
6725
6726</div>
6727
6728<!-- ======================================================================= -->
6729<h3>
6730  <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
6731</h3>
6732
6733<div>
6734
6735<p>LLVM support for <a href="GarbageCollection.html">Accurate Garbage
6736Collection</a> (GC) requires the implementation and generation of these
6737intrinsics. These intrinsics allow identification of <a href="#int_gcroot">GC
6738roots on the stack</a>, as well as garbage collector implementations that
6739require <a href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a>
6740barriers.  Front-ends for type-safe garbage collected languages should generate
6741these intrinsics to make use of the LLVM garbage collectors.  For more details,
6742see <a href="GarbageCollection.html">Accurate Garbage Collection with
6743LLVM</a>.</p>
6744
6745<p>The garbage collection intrinsics only operate on objects in the generic
6746   address space (address space zero).</p>
6747
6748<!-- _______________________________________________________________________ -->
6749<h4>
6750  <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
6751</h4>
6752
6753<div>
6754
6755<h5>Syntax:</h5>
6756<pre>
6757  declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
6758</pre>
6759
6760<h5>Overview:</h5>
6761<p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
6762   the code generator, and allows some metadata to be associated with it.</p>
6763
6764<h5>Arguments:</h5>
6765<p>The first argument specifies the address of a stack object that contains the
6766   root pointer.  The second pointer (which must be either a constant or a
6767   global value address) contains the meta-data to be associated with the
6768   root.</p>
6769
6770<h5>Semantics:</h5>
6771<p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
6772   location.  At compile-time, the code generator generates information to allow
6773   the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
6774   intrinsic may only be used in a function which <a href="#gc">specifies a GC
6775   algorithm</a>.</p>
6776
6777</div>
6778
6779<!-- _______________________________________________________________________ -->
6780<h4>
6781  <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
6782</h4>
6783
6784<div>
6785
6786<h5>Syntax:</h5>
6787<pre>
6788  declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
6789</pre>
6790
6791<h5>Overview:</h5>
6792<p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
6793   locations, allowing garbage collector implementations that require read
6794   barriers.</p>
6795
6796<h5>Arguments:</h5>
6797<p>The second argument is the address to read from, which should be an address
6798   allocated from the garbage collector.  The first object is a pointer to the
6799   start of the referenced object, if needed by the language runtime (otherwise
6800   null).</p>
6801
6802<h5>Semantics:</h5>
6803<p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
6804   instruction, but may be replaced with substantially more complex code by the
6805   garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
6806   may only be used in a function which <a href="#gc">specifies a GC
6807   algorithm</a>.</p>
6808
6809</div>
6810
6811<!-- _______________________________________________________________________ -->
6812<h4>
6813  <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
6814</h4>
6815
6816<div>
6817
6818<h5>Syntax:</h5>
6819<pre>
6820  declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
6821</pre>
6822
6823<h5>Overview:</h5>
6824<p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
6825   locations, allowing garbage collector implementations that require write
6826   barriers (such as generational or reference counting collectors).</p>
6827
6828<h5>Arguments:</h5>
6829<p>The first argument is the reference to store, the second is the start of the
6830   object to store it to, and the third is the address of the field of Obj to
6831   store to.  If the runtime does not require a pointer to the object, Obj may
6832   be null.</p>
6833
6834<h5>Semantics:</h5>
6835<p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
6836   instruction, but may be replaced with substantially more complex code by the
6837   garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
6838   may only be used in a function which <a href="#gc">specifies a GC
6839   algorithm</a>.</p>
6840
6841</div>
6842
6843</div>
6844
6845<!-- ======================================================================= -->
6846<h3>
6847  <a name="int_codegen">Code Generator Intrinsics</a>
6848</h3>
6849
6850<div>
6851
6852<p>These intrinsics are provided by LLVM to expose special features that may
6853   only be implemented with code generator support.</p>
6854
6855<!-- _______________________________________________________________________ -->
6856<h4>
6857  <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
6858</h4>
6859
6860<div>
6861
6862<h5>Syntax:</h5>
6863<pre>
6864  declare i8  *@llvm.returnaddress(i32 &lt;level&gt;)
6865</pre>
6866
6867<h5>Overview:</h5>
6868<p>The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
6869   target-specific value indicating the return address of the current function
6870   or one of its callers.</p>
6871
6872<h5>Arguments:</h5>
6873<p>The argument to this intrinsic indicates which function to return the address
6874   for.  Zero indicates the calling function, one indicates its caller, etc.
6875   The argument is <b>required</b> to be a constant integer value.</p>
6876
6877<h5>Semantics:</h5>
6878<p>The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer
6879   indicating the return address of the specified call frame, or zero if it
6880   cannot be identified.  The value returned by this intrinsic is likely to be
6881   incorrect or 0 for arguments other than zero, so it should only be used for
6882   debugging purposes.</p>
6883
6884<p>Note that calling this intrinsic does not prevent function inlining or other
6885   aggressive transformations, so the value returned may not be that of the
6886   obvious source-language caller.</p>
6887
6888</div>
6889
6890<!-- _______________________________________________________________________ -->
6891<h4>
6892  <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
6893</h4>
6894
6895<div>
6896
6897<h5>Syntax:</h5>
6898<pre>
6899  declare i8* @llvm.frameaddress(i32 &lt;level&gt;)
6900</pre>
6901
6902<h5>Overview:</h5>
6903<p>The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
6904   target-specific frame pointer value for the specified stack frame.</p>
6905
6906<h5>Arguments:</h5>
6907<p>The argument to this intrinsic indicates which function to return the frame
6908   pointer for.  Zero indicates the calling function, one indicates its caller,
6909   etc.  The argument is <b>required</b> to be a constant integer value.</p>
6910
6911<h5>Semantics:</h5>
6912<p>The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer
6913   indicating the frame address of the specified call frame, or zero if it
6914   cannot be identified.  The value returned by this intrinsic is likely to be
6915   incorrect or 0 for arguments other than zero, so it should only be used for
6916   debugging purposes.</p>
6917
6918<p>Note that calling this intrinsic does not prevent function inlining or other
6919   aggressive transformations, so the value returned may not be that of the
6920   obvious source-language caller.</p>
6921
6922</div>
6923
6924<!-- _______________________________________________________________________ -->
6925<h4>
6926  <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
6927</h4>
6928
6929<div>
6930
6931<h5>Syntax:</h5>
6932<pre>
6933  declare i8* @llvm.stacksave()
6934</pre>
6935
6936<h5>Overview:</h5>
6937<p>The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state
6938   of the function stack, for use
6939   with <a href="#int_stackrestore"> <tt>llvm.stackrestore</tt></a>.  This is
6940   useful for implementing language features like scoped automatic variable
6941   sized arrays in C99.</p>
6942
6943<h5>Semantics:</h5>
6944<p>This intrinsic returns a opaque pointer value that can be passed
6945   to <a href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>.  When
6946   an <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved
6947   from <tt>llvm.stacksave</tt>, it effectively restores the state of the stack
6948   to the state it was in when the <tt>llvm.stacksave</tt> intrinsic executed.
6949   In practice, this pops any <a href="#i_alloca">alloca</a> blocks from the
6950   stack that were allocated after the <tt>llvm.stacksave</tt> was executed.</p>
6951
6952</div>
6953
6954<!-- _______________________________________________________________________ -->
6955<h4>
6956  <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
6957</h4>
6958
6959<div>
6960
6961<h5>Syntax:</h5>
6962<pre>
6963  declare void @llvm.stackrestore(i8* %ptr)
6964</pre>
6965
6966<h5>Overview:</h5>
6967<p>The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
6968   the function stack to the state it was in when the
6969   corresponding <a href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic
6970   executed.  This is useful for implementing language features like scoped
6971   automatic variable sized arrays in C99.</p>
6972
6973<h5>Semantics:</h5>
6974<p>See the description
6975   for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.</p>
6976
6977</div>
6978
6979<!-- _______________________________________________________________________ -->
6980<h4>
6981  <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
6982</h4>
6983
6984<div>
6985
6986<h5>Syntax:</h5>
6987<pre>
6988  declare void @llvm.prefetch(i8* &lt;address&gt;, i32 &lt;rw&gt;, i32 &lt;locality&gt;, i32 &lt;cache type&gt;)
6989</pre>
6990
6991<h5>Overview:</h5>
6992<p>The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to
6993   insert a prefetch instruction if supported; otherwise, it is a noop.
6994   Prefetches have no effect on the behavior of the program but can change its
6995   performance characteristics.</p>
6996
6997<h5>Arguments:</h5>
6998<p><tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the
6999   specifier determining if the fetch should be for a read (0) or write (1),
7000   and <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
7001   locality, to (3) - extremely local keep in cache. The <tt>cache type</tt>
7002   specifies whether the prefetch is performed on the data (1) or instruction (0)
7003   cache. The <tt>rw</tt>, <tt>locality</tt> and <tt>cache type</tt> arguments
7004   must be constant integers.</p>
7005
7006<h5>Semantics:</h5>
7007<p>This intrinsic does not modify the behavior of the program.  In particular,
7008   prefetches cannot trap and do not produce a value.  On targets that support
7009   this intrinsic, the prefetch can provide hints to the processor cache for
7010   better performance.</p>
7011
7012</div>
7013
7014<!-- _______________________________________________________________________ -->
7015<h4>
7016  <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
7017</h4>
7018
7019<div>
7020
7021<h5>Syntax:</h5>
7022<pre>
7023  declare void @llvm.pcmarker(i32 &lt;id&gt;)
7024</pre>
7025
7026<h5>Overview:</h5>
7027<p>The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program
7028   Counter (PC) in a region of code to simulators and other tools.  The method
7029   is target specific, but it is expected that the marker will use exported
7030   symbols to transmit the PC of the marker.  The marker makes no guarantees
7031   that it will remain with any specific instruction after optimizations.  It is
7032   possible that the presence of a marker will inhibit optimizations.  The
7033   intended use is to be inserted after optimizations to allow correlations of
7034   simulation runs.</p>
7035
7036<h5>Arguments:</h5>
7037<p><tt>id</tt> is a numerical id identifying the marker.</p>
7038
7039<h5>Semantics:</h5>
7040<p>This intrinsic does not modify the behavior of the program.  Backends that do
7041   not support this intrinsic may ignore it.</p>
7042
7043</div>
7044
7045<!-- _______________________________________________________________________ -->
7046<h4>
7047  <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
7048</h4>
7049
7050<div>
7051
7052<h5>Syntax:</h5>
7053<pre>
7054  declare i64 @llvm.readcyclecounter()
7055</pre>
7056
7057<h5>Overview:</h5>
7058<p>The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
7059   counter register (or similar low latency, high accuracy clocks) on those
7060   targets that support it.  On X86, it should map to RDTSC.  On Alpha, it
7061   should map to RPCC.  As the backing counters overflow quickly (on the order
7062   of 9 seconds on alpha), this should only be used for small timings.</p>
7063
7064<h5>Semantics:</h5>
7065<p>When directly supported, reading the cycle counter should not modify any
7066   memory.  Implementations are allowed to either return a application specific
7067   value or a system wide value.  On backends without support, this is lowered
7068   to a constant 0.</p>
7069
7070</div>
7071
7072</div>
7073
7074<!-- ======================================================================= -->
7075<h3>
7076  <a name="int_libc">Standard C Library Intrinsics</a>
7077</h3>
7078
7079<div>
7080
7081<p>LLVM provides intrinsics for a few important standard C library functions.
7082   These intrinsics allow source-language front-ends to pass information about
7083   the alignment of the pointer arguments to the code generator, providing
7084   opportunity for more efficient code generation.</p>
7085
7086<!-- _______________________________________________________________________ -->
7087<h4>
7088  <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
7089</h4>
7090
7091<div>
7092
7093<h5>Syntax:</h5>
7094<p>This is an overloaded intrinsic. You can use <tt>llvm.memcpy</tt> on any
7095   integer bit width and for different address spaces. Not all targets support
7096   all bit widths however.</p>
7097
7098<pre>
7099  declare void @llvm.memcpy.p0i8.p0i8.i32(i8* &lt;dest&gt;, i8* &lt;src&gt;,
7100                                          i32 &lt;len&gt;, i32 &lt;align&gt;, i1 &lt;isvolatile&gt;)
7101  declare void @llvm.memcpy.p0i8.p0i8.i64(i8* &lt;dest&gt;, i8* &lt;src&gt;,
7102                                          i64 &lt;len&gt;, i32 &lt;align&gt;, i1 &lt;isvolatile&gt;)
7103</pre>
7104
7105<h5>Overview:</h5>
7106<p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
7107   source location to the destination location.</p>
7108
7109<p>Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
7110   intrinsics do not return a value, takes extra alignment/isvolatile arguments
7111   and the pointers can be in specified address spaces.</p>
7112
7113<h5>Arguments:</h5>
7114
7115<p>The first argument is a pointer to the destination, the second is a pointer
7116   to the source.  The third argument is an integer argument specifying the
7117   number of bytes to copy, the fourth argument is the alignment of the
7118   source and destination locations, and the fifth is a boolean indicating a
7119   volatile access.</p>
7120
7121<p>If the call to this intrinsic has an alignment value that is not 0 or 1,
7122   then the caller guarantees that both the source and destination pointers are
7123   aligned to that boundary.</p>
7124
7125<p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
7126   <tt>llvm.memcpy</tt> call is a <a href="#volatile">volatile operation</a>.
7127   The detailed access behavior is not very cleanly specified and it is unwise
7128   to depend on it.</p>
7129
7130<h5>Semantics:</h5>
7131
7132<p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
7133   source location to the destination location, which are not allowed to
7134   overlap.  It copies "len" bytes of memory over.  If the argument is known to
7135   be aligned to some boundary, this can be specified as the fourth argument,
7136   otherwise it should be set to 0 or 1.</p>
7137
7138</div>
7139
7140<!-- _______________________________________________________________________ -->
7141<h4>
7142  <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
7143</h4>
7144
7145<div>
7146
7147<h5>Syntax:</h5>
7148<p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
7149   width and for different address space. Not all targets support all bit
7150   widths however.</p>
7151
7152<pre>
7153  declare void @llvm.memmove.p0i8.p0i8.i32(i8* &lt;dest&gt;, i8* &lt;src&gt;,
7154                                           i32 &lt;len&gt;, i32 &lt;align&gt;, i1 &lt;isvolatile&gt;)
7155  declare void @llvm.memmove.p0i8.p0i8.i64(i8* &lt;dest&gt;, i8* &lt;src&gt;,
7156                                           i64 &lt;len&gt;, i32 &lt;align&gt;, i1 &lt;isvolatile&gt;)
7157</pre>
7158
7159<h5>Overview:</h5>
7160<p>The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the
7161   source location to the destination location. It is similar to the
7162   '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to
7163   overlap.</p>
7164
7165<p>Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
7166   intrinsics do not return a value, takes extra alignment/isvolatile arguments
7167   and the pointers can be in specified address spaces.</p>
7168
7169<h5>Arguments:</h5>
7170
7171<p>The first argument is a pointer to the destination, the second is a pointer
7172   to the source.  The third argument is an integer argument specifying the
7173   number of bytes to copy, the fourth argument is the alignment of the
7174   source and destination locations, and the fifth is a boolean indicating a
7175   volatile access.</p>
7176
7177<p>If the call to this intrinsic has an alignment value that is not 0 or 1,
7178   then the caller guarantees that the source and destination pointers are
7179   aligned to that boundary.</p>
7180
7181<p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
7182   <tt>llvm.memmove</tt> call is a <a href="#volatile">volatile operation</a>.
7183   The detailed access behavior is not very cleanly specified and it is unwise
7184   to depend on it.</p>
7185
7186<h5>Semantics:</h5>
7187
7188<p>The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the
7189   source location to the destination location, which may overlap.  It copies
7190   "len" bytes of memory over.  If the argument is known to be aligned to some
7191   boundary, this can be specified as the fourth argument, otherwise it should
7192   be set to 0 or 1.</p>
7193
7194</div>
7195
7196<!-- _______________________________________________________________________ -->
7197<h4>
7198  <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
7199</h4>
7200
7201<div>
7202
7203<h5>Syntax:</h5>
7204<p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
7205   width and for different address spaces. However, not all targets support all
7206   bit widths.</p>
7207
7208<pre>
7209  declare void @llvm.memset.p0i8.i32(i8* &lt;dest&gt;, i8 &lt;val&gt;,
7210                                     i32 &lt;len&gt;, i32 &lt;align&gt;, i1 &lt;isvolatile&gt;)
7211  declare void @llvm.memset.p0i8.i64(i8* &lt;dest&gt;, i8 &lt;val&gt;,
7212                                     i64 &lt;len&gt;, i32 &lt;align&gt;, i1 &lt;isvolatile&gt;)
7213</pre>
7214
7215<h5>Overview:</h5>
7216<p>The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a
7217   particular byte value.</p>
7218
7219<p>Note that, unlike the standard libc function, the <tt>llvm.memset</tt>
7220   intrinsic does not return a value and takes extra alignment/volatile
7221   arguments.  Also, the destination can be in an arbitrary address space.</p>
7222
7223<h5>Arguments:</h5>
7224<p>The first argument is a pointer to the destination to fill, the second is the
7225   byte value with which to fill it, the third argument is an integer argument
7226   specifying the number of bytes to fill, and the fourth argument is the known
7227   alignment of the destination location.</p>
7228
7229<p>If the call to this intrinsic has an alignment value that is not 0 or 1,
7230   then the caller guarantees that the destination pointer is aligned to that
7231   boundary.</p>
7232
7233<p>If the <tt>isvolatile</tt> parameter is <tt>true</tt>, the
7234   <tt>llvm.memset</tt> call is a <a href="#volatile">volatile operation</a>.
7235   The detailed access behavior is not very cleanly specified and it is unwise
7236   to depend on it.</p>
7237
7238<h5>Semantics:</h5>
7239<p>The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting
7240   at the destination location.  If the argument is known to be aligned to some
7241   boundary, this can be specified as the fourth argument, otherwise it should
7242   be set to 0 or 1.</p>
7243
7244</div>
7245
7246<!-- _______________________________________________________________________ -->
7247<h4>
7248  <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
7249</h4>
7250
7251<div>
7252
7253<h5>Syntax:</h5>
7254<p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
7255   floating point or vector of floating point type. Not all targets support all
7256   types however.</p>
7257
7258<pre>
7259  declare float     @llvm.sqrt.f32(float %Val)
7260  declare double    @llvm.sqrt.f64(double %Val)
7261  declare x86_fp80  @llvm.sqrt.f80(x86_fp80 %Val)
7262  declare fp128     @llvm.sqrt.f128(fp128 %Val)
7263  declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
7264</pre>
7265
7266<h5>Overview:</h5>
7267<p>The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
7268   returning the same value as the libm '<tt>sqrt</tt>' functions would.
7269   Unlike <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined
7270   behavior for negative numbers other than -0.0 (which allows for better
7271   optimization, because there is no need to worry about errno being
7272   set).  <tt>llvm.sqrt(-0.0)</tt> is defined to return -0.0 like IEEE sqrt.</p>
7273
7274<h5>Arguments:</h5>
7275<p>The argument and return value are floating point numbers of the same
7276   type.</p>
7277
7278<h5>Semantics:</h5>
7279<p>This function returns the sqrt of the specified operand if it is a
7280   nonnegative floating point number.</p>
7281
7282</div>
7283
7284<!-- _______________________________________________________________________ -->
7285<h4>
7286  <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
7287</h4>
7288
7289<div>
7290
7291<h5>Syntax:</h5>
7292<p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
7293   floating point or vector of floating point type. Not all targets support all
7294   types however.</p>
7295
7296<pre>
7297  declare float     @llvm.powi.f32(float  %Val, i32 %power)
7298  declare double    @llvm.powi.f64(double %Val, i32 %power)
7299  declare x86_fp80  @llvm.powi.f80(x86_fp80  %Val, i32 %power)
7300  declare fp128     @llvm.powi.f128(fp128 %Val, i32 %power)
7301  declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128  %Val, i32 %power)
7302</pre>
7303
7304<h5>Overview:</h5>
7305<p>The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
7306   specified (positive or negative) power.  The order of evaluation of
7307   multiplications is not defined.  When a vector of floating point type is
7308   used, the second argument remains a scalar integer value.</p>
7309
7310<h5>Arguments:</h5>
7311<p>The second argument is an integer power, and the first is a value to raise to
7312   that power.</p>
7313
7314<h5>Semantics:</h5>
7315<p>This function returns the first value raised to the second power with an
7316   unspecified sequence of rounding operations.</p>
7317
7318</div>
7319
7320<!-- _______________________________________________________________________ -->
7321<h4>
7322  <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
7323</h4>
7324
7325<div>
7326
7327<h5>Syntax:</h5>
7328<p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
7329   floating point or vector of floating point type. Not all targets support all
7330   types however.</p>
7331
7332<pre>
7333  declare float     @llvm.sin.f32(float  %Val)
7334  declare double    @llvm.sin.f64(double %Val)
7335  declare x86_fp80  @llvm.sin.f80(x86_fp80  %Val)
7336  declare fp128     @llvm.sin.f128(fp128 %Val)
7337  declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128  %Val)
7338</pre>
7339
7340<h5>Overview:</h5>
7341<p>The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.</p>
7342
7343<h5>Arguments:</h5>
7344<p>The argument and return value are floating point numbers of the same
7345   type.</p>
7346
7347<h5>Semantics:</h5>
7348<p>This function returns the sine of the specified operand, returning the same
7349   values as the libm <tt>sin</tt> functions would, and handles error conditions
7350   in the same way.</p>
7351
7352</div>
7353
7354<!-- _______________________________________________________________________ -->
7355<h4>
7356  <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
7357</h4>
7358
7359<div>
7360
7361<h5>Syntax:</h5>
7362<p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
7363   floating point or vector of floating point type. Not all targets support all
7364   types however.</p>
7365
7366<pre>
7367  declare float     @llvm.cos.f32(float  %Val)
7368  declare double    @llvm.cos.f64(double %Val)
7369  declare x86_fp80  @llvm.cos.f80(x86_fp80  %Val)
7370  declare fp128     @llvm.cos.f128(fp128 %Val)
7371  declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128  %Val)
7372</pre>
7373
7374<h5>Overview:</h5>
7375<p>The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.</p>
7376
7377<h5>Arguments:</h5>
7378<p>The argument and return value are floating point numbers of the same
7379   type.</p>
7380
7381<h5>Semantics:</h5>
7382<p>This function returns the cosine of the specified operand, returning the same
7383   values as the libm <tt>cos</tt> functions would, and handles error conditions
7384   in the same way.</p>
7385
7386</div>
7387
7388<!-- _______________________________________________________________________ -->
7389<h4>
7390  <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
7391</h4>
7392
7393<div>
7394
7395<h5>Syntax:</h5>
7396<p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
7397   floating point or vector of floating point type. Not all targets support all
7398   types however.</p>
7399
7400<pre>
7401  declare float     @llvm.pow.f32(float  %Val, float %Power)
7402  declare double    @llvm.pow.f64(double %Val, double %Power)
7403  declare x86_fp80  @llvm.pow.f80(x86_fp80  %Val, x86_fp80 %Power)
7404  declare fp128     @llvm.pow.f128(fp128 %Val, fp128 %Power)
7405  declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128  %Val, ppc_fp128 Power)
7406</pre>
7407
7408<h5>Overview:</h5>
7409<p>The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
7410   specified (positive or negative) power.</p>
7411
7412<h5>Arguments:</h5>
7413<p>The second argument is a floating point power, and the first is a value to
7414   raise to that power.</p>
7415
7416<h5>Semantics:</h5>
7417<p>This function returns the first value raised to the second power, returning
7418   the same values as the libm <tt>pow</tt> functions would, and handles error
7419   conditions in the same way.</p>
7420
7421</div>
7422
7423<!-- _______________________________________________________________________ -->
7424<h4>
7425  <a name="int_exp">'<tt>llvm.exp.*</tt>' Intrinsic</a>
7426</h4>
7427
7428<div>
7429
7430<h5>Syntax:</h5>
7431<p>This is an overloaded intrinsic. You can use <tt>llvm.exp</tt> on any
7432   floating point or vector of floating point type. Not all targets support all
7433   types however.</p>
7434
7435<pre>
7436  declare float     @llvm.exp.f32(float  %Val)
7437  declare double    @llvm.exp.f64(double %Val)
7438  declare x86_fp80  @llvm.exp.f80(x86_fp80  %Val)
7439  declare fp128     @llvm.exp.f128(fp128 %Val)
7440  declare ppc_fp128 @llvm.exp.ppcf128(ppc_fp128  %Val)
7441</pre>
7442
7443<h5>Overview:</h5>
7444<p>The '<tt>llvm.exp.*</tt>' intrinsics perform the exp function.</p>
7445
7446<h5>Arguments:</h5>
7447<p>The argument and return value are floating point numbers of the same
7448   type.</p>
7449
7450<h5>Semantics:</h5>
7451<p>This function returns the same values as the libm <tt>exp</tt> functions
7452   would, and handles error conditions in the same way.</p>
7453
7454</div>
7455
7456<!-- _______________________________________________________________________ -->
7457<h4>
7458  <a name="int_log">'<tt>llvm.log.*</tt>' Intrinsic</a>
7459</h4>
7460
7461<div>
7462
7463<h5>Syntax:</h5>
7464<p>This is an overloaded intrinsic. You can use <tt>llvm.log</tt> on any
7465   floating point or vector of floating point type. Not all targets support all
7466   types however.</p>
7467
7468<pre>
7469  declare float     @llvm.log.f32(float  %Val)
7470  declare double    @llvm.log.f64(double %Val)
7471  declare x86_fp80  @llvm.log.f80(x86_fp80  %Val)
7472  declare fp128     @llvm.log.f128(fp128 %Val)
7473  declare ppc_fp128 @llvm.log.ppcf128(ppc_fp128  %Val)
7474</pre>
7475
7476<h5>Overview:</h5>
7477<p>The '<tt>llvm.log.*</tt>' intrinsics perform the log function.</p>
7478
7479<h5>Arguments:</h5>
7480<p>The argument and return value are floating point numbers of the same
7481   type.</p>
7482
7483<h5>Semantics:</h5>
7484<p>This function returns the same values as the libm <tt>log</tt> functions
7485   would, and handles error conditions in the same way.</p>
7486
7487</div>
7488
7489<!-- _______________________________________________________________________ -->
7490<h4>
7491  <a name="int_fma">'<tt>llvm.fma.*</tt>' Intrinsic</a>
7492</h4>
7493
7494<div>
7495
7496<h5>Syntax:</h5>
7497<p>This is an overloaded intrinsic. You can use <tt>llvm.fma</tt> on any
7498   floating point or vector of floating point type. Not all targets support all
7499   types however.</p>
7500
7501<pre>
7502  declare float     @llvm.fma.f32(float  %a, float  %b, float  %c)
7503  declare double    @llvm.fma.f64(double %a, double %b, double %c)
7504  declare x86_fp80  @llvm.fma.f80(x86_fp80 %a, x86_fp80 %b, x86_fp80 %c)
7505  declare fp128     @llvm.fma.f128(fp128 %a, fp128 %b, fp128 %c)
7506  declare ppc_fp128 @llvm.fma.ppcf128(ppc_fp128 %a, ppc_fp128 %b, ppc_fp128 %c)
7507</pre>
7508
7509<h5>Overview:</h5>
7510<p>The '<tt>llvm.fma.*</tt>' intrinsics perform the fused multiply-add
7511   operation.</p>
7512
7513<h5>Arguments:</h5>
7514<p>The argument and return value are floating point numbers of the same
7515   type.</p>
7516
7517<h5>Semantics:</h5>
7518<p>This function returns the same values as the libm <tt>fma</tt> functions
7519   would.</p>
7520
7521</div>
7522
7523<!-- _______________________________________________________________________ -->
7524<h4>
7525  <a name="int_fabs">'<tt>llvm.fabs.*</tt>' Intrinsic</a>
7526</h4>
7527
7528<div>
7529
7530<h5>Syntax:</h5>
7531<p>This is an overloaded intrinsic. You can use <tt>llvm.fabs</tt> on any
7532   floating point or vector of floating point type. Not all targets support all
7533   types however.</p>
7534
7535<pre>
7536  declare float     @llvm.fabs.f32(float  %Val)
7537  declare double    @llvm.fabs.f64(double %Val)
7538  declare x86_fp80  @llvm.fabs.f80(x86_fp80  %Val)
7539  declare fp128     @llvm.fabs.f128(fp128 %Val)
7540  declare ppc_fp128 @llvm.fabs.ppcf128(ppc_fp128  %Val)
7541</pre>
7542
7543<h5>Overview:</h5>
7544<p>The '<tt>llvm.fabs.*</tt>' intrinsics return the absolute value of
7545   the operand.</p>
7546
7547<h5>Arguments:</h5>
7548<p>The argument and return value are floating point numbers of the same
7549   type.</p>
7550
7551<h5>Semantics:</h5>
7552<p>This function returns the same values as the libm <tt>fabs</tt> functions
7553   would, and handles error conditions in the same way.</p>
7554
7555</div>
7556
7557<!-- _______________________________________________________________________ -->
7558<h4>
7559  <a name="int_floor">'<tt>llvm.floor.*</tt>' Intrinsic</a>
7560</h4>
7561
7562<div>
7563
7564<h5>Syntax:</h5>
7565<p>This is an overloaded intrinsic. You can use <tt>llvm.floor</tt> on any
7566   floating point or vector of floating point type. Not all targets support all
7567   types however.</p>
7568
7569<pre>
7570  declare float     @llvm.floor.f32(float  %Val)
7571  declare double    @llvm.floor.f64(double %Val)
7572  declare x86_fp80  @llvm.floor.f80(x86_fp80  %Val)
7573  declare fp128     @llvm.floor.f128(fp128 %Val)
7574  declare ppc_fp128 @llvm.floor.ppcf128(ppc_fp128  %Val)
7575</pre>
7576
7577<h5>Overview:</h5>
7578<p>The '<tt>llvm.floor.*</tt>' intrinsics return the floor of
7579   the operand.</p>
7580
7581<h5>Arguments:</h5>
7582<p>The argument and return value are floating point numbers of the same
7583   type.</p>
7584
7585<h5>Semantics:</h5>
7586<p>This function returns the same values as the libm <tt>floor</tt> functions
7587   would, and handles error conditions in the same way.</p>
7588
7589</div>
7590
7591</div>
7592
7593<!-- ======================================================================= -->
7594<h3>
7595  <a name="int_manip">Bit Manipulation Intrinsics</a>
7596</h3>
7597
7598<div>
7599
7600<p>LLVM provides intrinsics for a few important bit manipulation operations.
7601   These allow efficient code generation for some algorithms.</p>
7602
7603<!-- _______________________________________________________________________ -->
7604<h4>
7605  <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
7606</h4>
7607
7608<div>
7609
7610<h5>Syntax:</h5>
7611<p>This is an overloaded intrinsic function. You can use bswap on any integer
7612   type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
7613
7614<pre>
7615  declare i16 @llvm.bswap.i16(i16 &lt;id&gt;)
7616  declare i32 @llvm.bswap.i32(i32 &lt;id&gt;)
7617  declare i64 @llvm.bswap.i64(i64 &lt;id&gt;)
7618</pre>
7619
7620<h5>Overview:</h5>
7621<p>The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
7622   values with an even number of bytes (positive multiple of 16 bits).  These
7623   are useful for performing operations on data that is not in the target's
7624   native byte order.</p>
7625
7626<h5>Semantics:</h5>
7627<p>The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
7628   and low byte of the input i16 swapped.  Similarly,
7629   the <tt>llvm.bswap.i32</tt> intrinsic returns an i32 value that has the four
7630   bytes of the input i32 swapped, so that if the input bytes are numbered 0, 1,
7631   2, 3 then the returned i32 will have its bytes in 3, 2, 1, 0 order.
7632   The <tt>llvm.bswap.i48</tt>, <tt>llvm.bswap.i64</tt> and other intrinsics
7633   extend this concept to additional even-byte lengths (6 bytes, 8 bytes and
7634   more, respectively).</p>
7635
7636</div>
7637
7638<!-- _______________________________________________________________________ -->
7639<h4>
7640  <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
7641</h4>
7642
7643<div>
7644
7645<h5>Syntax:</h5>
7646<p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
7647   width, or on any vector with integer elements. Not all targets support all
7648  bit widths or vector types, however.</p>
7649
7650<pre>
7651  declare i8 @llvm.ctpop.i8(i8  &lt;src&gt;)
7652  declare i16 @llvm.ctpop.i16(i16 &lt;src&gt;)
7653  declare i32 @llvm.ctpop.i32(i32 &lt;src&gt;)
7654  declare i64 @llvm.ctpop.i64(i64 &lt;src&gt;)
7655  declare i256 @llvm.ctpop.i256(i256 &lt;src&gt;)
7656  declare &lt;2 x i32&gt; @llvm.ctpop.v2i32(&lt;2 x i32&gt; &lt;src&gt;)
7657</pre>
7658
7659<h5>Overview:</h5>
7660<p>The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set
7661   in a value.</p>
7662
7663<h5>Arguments:</h5>
7664<p>The only argument is the value to be counted.  The argument may be of any
7665   integer type, or a vector with integer elements.
7666   The return type must match the argument type.</p>
7667
7668<h5>Semantics:</h5>
7669<p>The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable, or within each
7670   element of a vector.</p>
7671
7672</div>
7673
7674<!-- _______________________________________________________________________ -->
7675<h4>
7676  <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
7677</h4>
7678
7679<div>
7680
7681<h5>Syntax:</h5>
7682<p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
7683   integer bit width, or any vector whose elements are integers. Not all
7684   targets support all bit widths or vector types, however.</p>
7685
7686<pre>
7687  declare i8   @llvm.ctlz.i8  (i8   &lt;src&gt;, i1 &lt;is_zero_undef&gt;)
7688  declare i16  @llvm.ctlz.i16 (i16  &lt;src&gt;, i1 &lt;is_zero_undef&gt;)
7689  declare i32  @llvm.ctlz.i32 (i32  &lt;src&gt;, i1 &lt;is_zero_undef&gt;)
7690  declare i64  @llvm.ctlz.i64 (i64  &lt;src&gt;, i1 &lt;is_zero_undef&gt;)
7691  declare i256 @llvm.ctlz.i256(i256 &lt;src&gt;, i1 &lt;is_zero_undef&gt;)
7692  declase &lt;2 x i32&gt; @llvm.ctlz.v2i32(&lt;2 x i32&gt; &lt;src&gt;, i1 &lt;is_zero_undef&gt;)
7693</pre>
7694
7695<h5>Overview:</h5>
7696<p>The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
7697   leading zeros in a variable.</p>
7698
7699<h5>Arguments:</h5>
7700<p>The first argument is the value to be counted. This argument may be of any
7701   integer type, or a vectory with integer element type. The return type
7702   must match the first argument type.</p>
7703
7704<p>The second argument must be a constant and is a flag to indicate whether the
7705   intrinsic should ensure that a zero as the first argument produces a defined
7706   result. Historically some architectures did not provide a defined result for
7707   zero values as efficiently, and many algorithms are now predicated on
7708   avoiding zero-value inputs.</p>
7709
7710<h5>Semantics:</h5>
7711<p>The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant)
7712   zeros in a variable, or within each element of the vector.
7713   If <tt>src == 0</tt> then the result is the size in bits of the type of
7714   <tt>src</tt> if <tt>is_zero_undef == 0</tt> and <tt>undef</tt> otherwise.
7715   For example, <tt>llvm.ctlz(i32 2) = 30</tt>.</p>
7716
7717</div>
7718
7719<!-- _______________________________________________________________________ -->
7720<h4>
7721  <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
7722</h4>
7723
7724<div>
7725
7726<h5>Syntax:</h5>
7727<p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
7728   integer bit width, or any vector of integer elements. Not all targets
7729   support all bit widths or vector types, however.</p>
7730
7731<pre>
7732  declare i8   @llvm.cttz.i8  (i8   &lt;src&gt;, i1 &lt;is_zero_undef&gt;)
7733  declare i16  @llvm.cttz.i16 (i16  &lt;src&gt;, i1 &lt;is_zero_undef&gt;)
7734  declare i32  @llvm.cttz.i32 (i32  &lt;src&gt;, i1 &lt;is_zero_undef&gt;)
7735  declare i64  @llvm.cttz.i64 (i64  &lt;src&gt;, i1 &lt;is_zero_undef&gt;)
7736  declare i256 @llvm.cttz.i256(i256 &lt;src&gt;, i1 &lt;is_zero_undef&gt;)
7737  declase &lt;2 x i32&gt; @llvm.cttz.v2i32(&lt;2 x i32&gt; &lt;src&gt;, i1 &lt;is_zero_undef&gt;)
7738</pre>
7739
7740<h5>Overview:</h5>
7741<p>The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
7742   trailing zeros.</p>
7743
7744<h5>Arguments:</h5>
7745<p>The first argument is the value to be counted. This argument may be of any
7746   integer type, or a vectory with integer element type. The return type
7747   must match the first argument type.</p>
7748
7749<p>The second argument must be a constant and is a flag to indicate whether the
7750   intrinsic should ensure that a zero as the first argument produces a defined
7751   result. Historically some architectures did not provide a defined result for
7752   zero values as efficiently, and many algorithms are now predicated on
7753   avoiding zero-value inputs.</p>
7754
7755<h5>Semantics:</h5>
7756<p>The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant)
7757   zeros in a variable, or within each element of a vector.
7758   If <tt>src == 0</tt> then the result is the size in bits of the type of
7759   <tt>src</tt> if <tt>is_zero_undef == 0</tt> and <tt>undef</tt> otherwise.
7760   For example, <tt>llvm.cttz(2) = 1</tt>.</p>
7761
7762</div>
7763
7764</div>
7765
7766<!-- ======================================================================= -->
7767<h3>
7768  <a name="int_overflow">Arithmetic with Overflow Intrinsics</a>
7769</h3>
7770
7771<div>
7772
7773<p>LLVM provides intrinsics for some arithmetic with overflow operations.</p>
7774
7775<!-- _______________________________________________________________________ -->
7776<h4>
7777  <a name="int_sadd_overflow">
7778    '<tt>llvm.sadd.with.overflow.*</tt>' Intrinsics
7779  </a>
7780</h4>
7781
7782<div>
7783
7784<h5>Syntax:</h5>
7785<p>This is an overloaded intrinsic. You can use <tt>llvm.sadd.with.overflow</tt>
7786   on any integer bit width.</p>
7787
7788<pre>
7789  declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
7790  declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7791  declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
7792</pre>
7793
7794<h5>Overview:</h5>
7795<p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
7796   a signed addition of the two arguments, and indicate whether an overflow
7797   occurred during the signed summation.</p>
7798
7799<h5>Arguments:</h5>
7800<p>The arguments (%a and %b) and the first element of the result structure may
7801   be of integer types of any bit width, but they must have the same bit
7802   width. The second element of the result structure must be of
7803   type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7804   undergo signed addition.</p>
7805
7806<h5>Semantics:</h5>
7807<p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
7808   a signed addition of the two variables. They return a structure &mdash; the
7809   first element of which is the signed summation, and the second element of
7810   which is a bit specifying if the signed summation resulted in an
7811   overflow.</p>
7812
7813<h5>Examples:</h5>
7814<pre>
7815  %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
7816  %sum = extractvalue {i32, i1} %res, 0
7817  %obit = extractvalue {i32, i1} %res, 1
7818  br i1 %obit, label %overflow, label %normal
7819</pre>
7820
7821</div>
7822
7823<!-- _______________________________________________________________________ -->
7824<h4>
7825  <a name="int_uadd_overflow">
7826    '<tt>llvm.uadd.with.overflow.*</tt>' Intrinsics
7827  </a>
7828</h4>
7829
7830<div>
7831
7832<h5>Syntax:</h5>
7833<p>This is an overloaded intrinsic. You can use <tt>llvm.uadd.with.overflow</tt>
7834   on any integer bit width.</p>
7835
7836<pre>
7837  declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
7838  declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7839  declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
7840</pre>
7841
7842<h5>Overview:</h5>
7843<p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
7844   an unsigned addition of the two arguments, and indicate whether a carry
7845   occurred during the unsigned summation.</p>
7846
7847<h5>Arguments:</h5>
7848<p>The arguments (%a and %b) and the first element of the result structure may
7849   be of integer types of any bit width, but they must have the same bit
7850   width. The second element of the result structure must be of
7851   type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7852   undergo unsigned addition.</p>
7853
7854<h5>Semantics:</h5>
7855<p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
7856   an unsigned addition of the two arguments. They return a structure &mdash;
7857   the first element of which is the sum, and the second element of which is a
7858   bit specifying if the unsigned summation resulted in a carry.</p>
7859
7860<h5>Examples:</h5>
7861<pre>
7862  %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7863  %sum = extractvalue {i32, i1} %res, 0
7864  %obit = extractvalue {i32, i1} %res, 1
7865  br i1 %obit, label %carry, label %normal
7866</pre>
7867
7868</div>
7869
7870<!-- _______________________________________________________________________ -->
7871<h4>
7872  <a name="int_ssub_overflow">
7873    '<tt>llvm.ssub.with.overflow.*</tt>' Intrinsics
7874  </a>
7875</h4>
7876
7877<div>
7878
7879<h5>Syntax:</h5>
7880<p>This is an overloaded intrinsic. You can use <tt>llvm.ssub.with.overflow</tt>
7881   on any integer bit width.</p>
7882
7883<pre>
7884  declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
7885  declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7886  declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
7887</pre>
7888
7889<h5>Overview:</h5>
7890<p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
7891   a signed subtraction of the two arguments, and indicate whether an overflow
7892   occurred during the signed subtraction.</p>
7893
7894<h5>Arguments:</h5>
7895<p>The arguments (%a and %b) and the first element of the result structure may
7896   be of integer types of any bit width, but they must have the same bit
7897   width. The second element of the result structure must be of
7898   type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7899   undergo signed subtraction.</p>
7900
7901<h5>Semantics:</h5>
7902<p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
7903   a signed subtraction of the two arguments. They return a structure &mdash;
7904   the first element of which is the subtraction, and the second element of
7905   which is a bit specifying if the signed subtraction resulted in an
7906   overflow.</p>
7907
7908<h5>Examples:</h5>
7909<pre>
7910  %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
7911  %sum = extractvalue {i32, i1} %res, 0
7912  %obit = extractvalue {i32, i1} %res, 1
7913  br i1 %obit, label %overflow, label %normal
7914</pre>
7915
7916</div>
7917
7918<!-- _______________________________________________________________________ -->
7919<h4>
7920  <a name="int_usub_overflow">
7921    '<tt>llvm.usub.with.overflow.*</tt>' Intrinsics
7922  </a>
7923</h4>
7924
7925<div>
7926
7927<h5>Syntax:</h5>
7928<p>This is an overloaded intrinsic. You can use <tt>llvm.usub.with.overflow</tt>
7929   on any integer bit width.</p>
7930
7931<pre>
7932  declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
7933  declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7934  declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
7935</pre>
7936
7937<h5>Overview:</h5>
7938<p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
7939   an unsigned subtraction of the two arguments, and indicate whether an
7940   overflow occurred during the unsigned subtraction.</p>
7941
7942<h5>Arguments:</h5>
7943<p>The arguments (%a and %b) and the first element of the result structure may
7944   be of integer types of any bit width, but they must have the same bit
7945   width. The second element of the result structure must be of
7946   type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7947   undergo unsigned subtraction.</p>
7948
7949<h5>Semantics:</h5>
7950<p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
7951   an unsigned subtraction of the two arguments. They return a structure &mdash;
7952   the first element of which is the subtraction, and the second element of
7953   which is a bit specifying if the unsigned subtraction resulted in an
7954   overflow.</p>
7955
7956<h5>Examples:</h5>
7957<pre>
7958  %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
7959  %sum = extractvalue {i32, i1} %res, 0
7960  %obit = extractvalue {i32, i1} %res, 1
7961  br i1 %obit, label %overflow, label %normal
7962</pre>
7963
7964</div>
7965
7966<!-- _______________________________________________________________________ -->
7967<h4>
7968  <a name="int_smul_overflow">
7969    '<tt>llvm.smul.with.overflow.*</tt>' Intrinsics
7970  </a>
7971</h4>
7972
7973<div>
7974
7975<h5>Syntax:</h5>
7976<p>This is an overloaded intrinsic. You can use <tt>llvm.smul.with.overflow</tt>
7977   on any integer bit width.</p>
7978
7979<pre>
7980  declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
7981  declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
7982  declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
7983</pre>
7984
7985<h5>Overview:</h5>
7986
7987<p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
7988   a signed multiplication of the two arguments, and indicate whether an
7989   overflow occurred during the signed multiplication.</p>
7990
7991<h5>Arguments:</h5>
7992<p>The arguments (%a and %b) and the first element of the result structure may
7993   be of integer types of any bit width, but they must have the same bit
7994   width. The second element of the result structure must be of
7995   type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
7996   undergo signed multiplication.</p>
7997
7998<h5>Semantics:</h5>
7999<p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
8000   a signed multiplication of the two arguments. They return a structure &mdash;
8001   the first element of which is the multiplication, and the second element of
8002   which is a bit specifying if the signed multiplication resulted in an
8003   overflow.</p>
8004
8005<h5>Examples:</h5>
8006<pre>
8007  %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
8008  %sum = extractvalue {i32, i1} %res, 0
8009  %obit = extractvalue {i32, i1} %res, 1
8010  br i1 %obit, label %overflow, label %normal
8011</pre>
8012
8013</div>
8014
8015<!-- _______________________________________________________________________ -->
8016<h4>
8017  <a name="int_umul_overflow">
8018    '<tt>llvm.umul.with.overflow.*</tt>' Intrinsics
8019  </a>
8020</h4>
8021
8022<div>
8023
8024<h5>Syntax:</h5>
8025<p>This is an overloaded intrinsic. You can use <tt>llvm.umul.with.overflow</tt>
8026   on any integer bit width.</p>
8027
8028<pre>
8029  declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
8030  declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8031  declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
8032</pre>
8033
8034<h5>Overview:</h5>
8035<p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
8036   a unsigned multiplication of the two arguments, and indicate whether an
8037   overflow occurred during the unsigned multiplication.</p>
8038
8039<h5>Arguments:</h5>
8040<p>The arguments (%a and %b) and the first element of the result structure may
8041   be of integer types of any bit width, but they must have the same bit
8042   width. The second element of the result structure must be of
8043   type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
8044   undergo unsigned multiplication.</p>
8045
8046<h5>Semantics:</h5>
8047<p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
8048   an unsigned multiplication of the two arguments. They return a structure
8049   &mdash; the first element of which is the multiplication, and the second
8050   element of which is a bit specifying if the unsigned multiplication resulted
8051   in an overflow.</p>
8052
8053<h5>Examples:</h5>
8054<pre>
8055  %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
8056  %sum = extractvalue {i32, i1} %res, 0
8057  %obit = extractvalue {i32, i1} %res, 1
8058  br i1 %obit, label %overflow, label %normal
8059</pre>
8060
8061</div>
8062
8063</div>
8064
8065<!-- ======================================================================= -->
8066<h3>
8067  <a name="spec_arithmetic">Specialised Arithmetic Intrinsics</a>
8068</h3>
8069
8070<!-- _______________________________________________________________________ -->
8071
8072<h4>
8073  <a name="fmuladd">'<tt>llvm.fmuladd.*</tt>' Intrinsic</a>
8074</h4>
8075
8076<div>
8077
8078<h5>Syntax:</h5>
8079<pre>
8080  declare float @llvm.fmuladd.f32(float %a, float %b, float %c)
8081  declare double @llvm.fmuladd.f64(double %a, double %b, double %c)
8082</pre>
8083
8084<h5>Overview:</h5>
8085<p>The '<tt>llvm.fmuladd.*</tt>' intrinsic functions represent multiply-add
8086expressions that can be fused if the code generator determines that the fused
8087expression would be legal and efficient.</p>
8088
8089<h5>Arguments:</h5>
8090<p>The '<tt>llvm.fmuladd.*</tt>' intrinsics each take three arguments: two
8091multiplicands, a and b, and an addend c.</p>
8092
8093<h5>Semantics:</h5>
8094<p>The expression:</p>
8095<pre>
8096  %0 = call float @llvm.fmuladd.f32(%a, %b, %c)
8097</pre>
8098<p>is equivalent to the expression a * b + c, except that rounding will not be
8099performed between the multiplication and addition steps if the code generator
8100fuses the operations. Fusion is not guaranteed, even if the target platform
8101supports it. If a fused multiply-add is required the corresponding llvm.fma.*
8102intrinsic function should be used instead.</p>
8103
8104<h5>Examples:</h5>
8105<pre>
8106  %r2 = call float @llvm.fmuladd.f32(float %a, float %b, float %c) ; yields {float}:r2 = (a * b) + c
8107</pre>
8108
8109</div>
8110
8111<!-- ======================================================================= -->
8112<h3>
8113  <a name="int_fp16">Half Precision Floating Point Intrinsics</a>
8114</h3>
8115
8116<div>
8117
8118<p>For most target platforms, half precision floating point is a storage-only
8119   format. This means that it is
8120   a dense encoding (in memory) but does not support computation in the
8121   format.</p>
8122
8123<p>This means that code must first load the half-precision floating point
8124   value as an i16, then convert it to float with <a
8125   href="#int_convert_from_fp16"><tt>llvm.convert.from.fp16</tt></a>.
8126   Computation can then be performed on the float value (including extending to
8127   double etc).  To store the value back to memory, it is first converted to
8128   float if needed, then converted to i16 with
8129   <a href="#int_convert_to_fp16"><tt>llvm.convert.to.fp16</tt></a>, then
8130   storing as an i16 value.</p>
8131
8132<!-- _______________________________________________________________________ -->
8133<h4>
8134  <a name="int_convert_to_fp16">
8135    '<tt>llvm.convert.to.fp16</tt>' Intrinsic
8136  </a>
8137</h4>
8138
8139<div>
8140
8141<h5>Syntax:</h5>
8142<pre>
8143  declare i16 @llvm.convert.to.fp16(f32 %a)
8144</pre>
8145
8146<h5>Overview:</h5>
8147<p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
8148   a conversion from single precision floating point format to half precision
8149   floating point format.</p>
8150
8151<h5>Arguments:</h5>
8152<p>The intrinsic function contains single argument - the value to be
8153   converted.</p>
8154
8155<h5>Semantics:</h5>
8156<p>The '<tt>llvm.convert.to.fp16</tt>' intrinsic function performs
8157   a conversion from single precision floating point format to half precision
8158   floating point format. The return value is an <tt>i16</tt> which
8159   contains the converted number.</p>
8160
8161<h5>Examples:</h5>
8162<pre>
8163  %res = call i16 @llvm.convert.to.fp16(f32 %a)
8164  store i16 %res, i16* @x, align 2
8165</pre>
8166
8167</div>
8168
8169<!-- _______________________________________________________________________ -->
8170<h4>
8171  <a name="int_convert_from_fp16">
8172    '<tt>llvm.convert.from.fp16</tt>' Intrinsic
8173  </a>
8174</h4>
8175
8176<div>
8177
8178<h5>Syntax:</h5>
8179<pre>
8180  declare f32 @llvm.convert.from.fp16(i16 %a)
8181</pre>
8182
8183<h5>Overview:</h5>
8184<p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs
8185   a conversion from half precision floating point format to single precision
8186   floating point format.</p>
8187
8188<h5>Arguments:</h5>
8189<p>The intrinsic function contains single argument - the value to be
8190   converted.</p>
8191
8192<h5>Semantics:</h5>
8193<p>The '<tt>llvm.convert.from.fp16</tt>' intrinsic function performs a
8194   conversion from half single precision floating point format to single
8195   precision floating point format. The input half-float value is represented by
8196   an <tt>i16</tt> value.</p>
8197
8198<h5>Examples:</h5>
8199<pre>
8200  %a = load i16* @x, align 2
8201  %res = call f32 @llvm.convert.from.fp16(i16 %a)
8202</pre>
8203
8204</div>
8205
8206</div>
8207
8208<!-- ======================================================================= -->
8209<h3>
8210  <a name="int_debugger">Debugger Intrinsics</a>
8211</h3>
8212
8213<div>
8214
8215<p>The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt>
8216   prefix), are described in
8217   the <a href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source
8218   Level Debugging</a> document.</p>
8219
8220</div>
8221
8222<!-- ======================================================================= -->
8223<h3>
8224  <a name="int_eh">Exception Handling Intrinsics</a>
8225</h3>
8226
8227<div>
8228
8229<p>The LLVM exception handling intrinsics (which all start with
8230   <tt>llvm.eh.</tt> prefix), are described in
8231   the <a href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
8232   Handling</a> document.</p>
8233
8234</div>
8235
8236<!-- ======================================================================= -->
8237<h3>
8238  <a name="int_trampoline">Trampoline Intrinsics</a>
8239</h3>
8240
8241<div>
8242
8243<p>These intrinsics make it possible to excise one parameter, marked with
8244   the <a href="#nest"><tt>nest</tt></a> attribute, from a function.
8245   The result is a callable
8246   function pointer lacking the nest parameter - the caller does not need to
8247   provide a value for it.  Instead, the value to use is stored in advance in a
8248   "trampoline", a block of memory usually allocated on the stack, which also
8249   contains code to splice the nest value into the argument list.  This is used
8250   to implement the GCC nested function address extension.</p>
8251
8252<p>For example, if the function is
8253   <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
8254   pointer has signature <tt>i32 (i32, i32)*</tt>.  It can be created as
8255   follows:</p>
8256
8257<pre class="doc_code">
8258  %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
8259  %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
8260  call i8* @llvm.init.trampoline(i8* %tramp1, i8* bitcast (i32 (i8*, i32, i32)* @f to i8*), i8* %nval)
8261  %p = call i8* @llvm.adjust.trampoline(i8* %tramp1)
8262  %fp = bitcast i8* %p to i32 (i32, i32)*
8263</pre>
8264
8265<p>The call <tt>%val = call i32 %fp(i32 %x, i32 %y)</tt> is then equivalent
8266   to <tt>%val = call i32 %f(i8* %nval, i32 %x, i32 %y)</tt>.</p>
8267
8268<!-- _______________________________________________________________________ -->
8269<h4>
8270  <a name="int_it">
8271    '<tt>llvm.init.trampoline</tt>' Intrinsic
8272  </a>
8273</h4>
8274
8275<div>
8276
8277<h5>Syntax:</h5>
8278<pre>
8279  declare void @llvm.init.trampoline(i8* &lt;tramp&gt;, i8* &lt;func&gt;, i8* &lt;nval&gt;)
8280</pre>
8281
8282<h5>Overview:</h5>
8283<p>This fills the memory pointed to by <tt>tramp</tt> with executable code,
8284   turning it into a trampoline.</p>
8285
8286<h5>Arguments:</h5>
8287<p>The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
8288   pointers.  The <tt>tramp</tt> argument must point to a sufficiently large and
8289   sufficiently aligned block of memory; this memory is written to by the
8290   intrinsic.  Note that the size and the alignment are target-specific - LLVM
8291   currently provides no portable way of determining them, so a front-end that
8292   generates this intrinsic needs to have some target-specific knowledge.
8293   The <tt>func</tt> argument must hold a function bitcast to
8294   an <tt>i8*</tt>.</p>
8295
8296<h5>Semantics:</h5>
8297<p>The block of memory pointed to by <tt>tramp</tt> is filled with target
8298   dependent code, turning it into a function.  Then <tt>tramp</tt> needs to be
8299   passed to <a href="#int_at">llvm.adjust.trampoline</a> to get a pointer
8300   which can be <a href="#int_trampoline">bitcast (to a new function) and
8301   called</a>.  The new function's signature is the same as that of
8302   <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
8303   removed.  At most one such <tt>nest</tt> argument is allowed, and it must be of
8304   pointer type.  Calling the new function is equivalent to calling <tt>func</tt>
8305   with the same argument list, but with <tt>nval</tt> used for the missing
8306   <tt>nest</tt> argument.  If, after calling <tt>llvm.init.trampoline</tt>, the
8307   memory pointed to by <tt>tramp</tt> is modified, then the effect of any later call
8308   to the returned function pointer is undefined.</p>
8309</div>
8310
8311<!-- _______________________________________________________________________ -->
8312<h4>
8313  <a name="int_at">
8314    '<tt>llvm.adjust.trampoline</tt>' Intrinsic
8315  </a>
8316</h4>
8317
8318<div>
8319
8320<h5>Syntax:</h5>
8321<pre>
8322  declare i8* @llvm.adjust.trampoline(i8* &lt;tramp&gt;)
8323</pre>
8324
8325<h5>Overview:</h5>
8326<p>This performs any required machine-specific adjustment to the address of a
8327   trampoline (passed as <tt>tramp</tt>).</p>
8328
8329<h5>Arguments:</h5>
8330<p><tt>tramp</tt> must point to a block of memory which already has trampoline code
8331   filled in by a previous call to <a href="#int_it"><tt>llvm.init.trampoline</tt>
8332   </a>.</p>
8333
8334<h5>Semantics:</h5>
8335<p>On some architectures the address of the code to be executed needs to be
8336   different to the address where the trampoline is actually stored.  This
8337   intrinsic returns the executable address corresponding to <tt>tramp</tt>
8338   after performing the required machine specific adjustments.
8339   The pointer returned can then be <a href="#int_trampoline"> bitcast and
8340   executed</a>.
8341</p>
8342
8343</div>
8344
8345</div>
8346
8347<!-- ======================================================================= -->
8348<h3>
8349  <a name="int_memorymarkers">Memory Use Markers</a>
8350</h3>
8351
8352<div>
8353
8354<p>This class of intrinsics exists to information about the lifetime of memory
8355   objects and ranges where variables are immutable.</p>
8356
8357<!-- _______________________________________________________________________ -->
8358<h4>
8359  <a name="int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a>
8360</h4>
8361
8362<div>
8363
8364<h5>Syntax:</h5>
8365<pre>
8366  declare void @llvm.lifetime.start(i64 &lt;size&gt;, i8* nocapture &lt;ptr&gt;)
8367</pre>
8368
8369<h5>Overview:</h5>
8370<p>The '<tt>llvm.lifetime.start</tt>' intrinsic specifies the start of a memory
8371   object's lifetime.</p>
8372
8373<h5>Arguments:</h5>
8374<p>The first argument is a constant integer representing the size of the
8375   object, or -1 if it is variable sized.  The second argument is a pointer to
8376   the object.</p>
8377
8378<h5>Semantics:</h5>
8379<p>This intrinsic indicates that before this point in the code, the value of the
8380   memory pointed to by <tt>ptr</tt> is dead.  This means that it is known to
8381   never be used and has an undefined value.  A load from the pointer that
8382   precedes this intrinsic can be replaced with
8383   <tt>'<a href="#undefvalues">undef</a>'</tt>.</p>
8384
8385</div>
8386
8387<!-- _______________________________________________________________________ -->
8388<h4>
8389  <a name="int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a>
8390</h4>
8391
8392<div>
8393
8394<h5>Syntax:</h5>
8395<pre>
8396  declare void @llvm.lifetime.end(i64 &lt;size&gt;, i8* nocapture &lt;ptr&gt;)
8397</pre>
8398
8399<h5>Overview:</h5>
8400<p>The '<tt>llvm.lifetime.end</tt>' intrinsic specifies the end of a memory
8401   object's lifetime.</p>
8402
8403<h5>Arguments:</h5>
8404<p>The first argument is a constant integer representing the size of the
8405   object, or -1 if it is variable sized.  The second argument is a pointer to
8406   the object.</p>
8407
8408<h5>Semantics:</h5>
8409<p>This intrinsic indicates that after this point in the code, the value of the
8410   memory pointed to by <tt>ptr</tt> is dead.  This means that it is known to
8411   never be used and has an undefined value.  Any stores into the memory object
8412   following this intrinsic may be removed as dead.
8413
8414</div>
8415
8416<!-- _______________________________________________________________________ -->
8417<h4>
8418  <a name="int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a>
8419</h4>
8420
8421<div>
8422
8423<h5>Syntax:</h5>
8424<pre>
8425  declare {}* @llvm.invariant.start(i64 &lt;size&gt;, i8* nocapture &lt;ptr&gt;)
8426</pre>
8427
8428<h5>Overview:</h5>
8429<p>The '<tt>llvm.invariant.start</tt>' intrinsic specifies that the contents of
8430   a memory object will not change.</p>
8431
8432<h5>Arguments:</h5>
8433<p>The first argument is a constant integer representing the size of the
8434   object, or -1 if it is variable sized.  The second argument is a pointer to
8435   the object.</p>
8436
8437<h5>Semantics:</h5>
8438<p>This intrinsic indicates that until an <tt>llvm.invariant.end</tt> that uses
8439   the return value, the referenced memory location is constant and
8440   unchanging.</p>
8441
8442</div>
8443
8444<!-- _______________________________________________________________________ -->
8445<h4>
8446  <a name="int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a>
8447</h4>
8448
8449<div>
8450
8451<h5>Syntax:</h5>
8452<pre>
8453  declare void @llvm.invariant.end({}* &lt;start&gt;, i64 &lt;size&gt;, i8* nocapture &lt;ptr&gt;)
8454</pre>
8455
8456<h5>Overview:</h5>
8457<p>The '<tt>llvm.invariant.end</tt>' intrinsic specifies that the contents of
8458   a memory object are mutable.</p>
8459
8460<h5>Arguments:</h5>
8461<p>The first argument is the matching <tt>llvm.invariant.start</tt> intrinsic.
8462   The second argument is a constant integer representing the size of the
8463   object, or -1 if it is variable sized and the third argument is a pointer
8464   to the object.</p>
8465
8466<h5>Semantics:</h5>
8467<p>This intrinsic indicates that the memory is mutable again.</p>
8468
8469</div>
8470
8471</div>
8472
8473<!-- ======================================================================= -->
8474<h3>
8475  <a name="int_general">General Intrinsics</a>
8476</h3>
8477
8478<div>
8479
8480<p>This class of intrinsics is designed to be generic and has no specific
8481   purpose.</p>
8482
8483<!-- _______________________________________________________________________ -->
8484<h4>
8485  <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
8486</h4>
8487
8488<div>
8489
8490<h5>Syntax:</h5>
8491<pre>
8492  declare void @llvm.var.annotation(i8* &lt;val&gt;, i8* &lt;str&gt;, i8* &lt;str&gt;, i32  &lt;int&gt;)
8493</pre>
8494
8495<h5>Overview:</h5>
8496<p>The '<tt>llvm.var.annotation</tt>' intrinsic.</p>
8497
8498<h5>Arguments:</h5>
8499<p>The first argument is a pointer to a value, the second is a pointer to a
8500   global string, the third is a pointer to a global string which is the source
8501   file name, and the last argument is the line number.</p>
8502
8503<h5>Semantics:</h5>
8504<p>This intrinsic allows annotation of local variables with arbitrary strings.
8505   This can be useful for special purpose optimizations that want to look for
8506   these annotations.  These have no other defined use; they are ignored by code
8507   generation and optimization.</p>
8508
8509</div>
8510
8511<!-- _______________________________________________________________________ -->
8512<h4>
8513  <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
8514</h4>
8515
8516<div>
8517
8518<h5>Syntax:</h5>
8519<p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
8520   any integer bit width.</p>
8521
8522<pre>
8523  declare i8 @llvm.annotation.i8(i8 &lt;val&gt;, i8* &lt;str&gt;, i8* &lt;str&gt;, i32  &lt;int&gt;)
8524  declare i16 @llvm.annotation.i16(i16 &lt;val&gt;, i8* &lt;str&gt;, i8* &lt;str&gt;, i32  &lt;int&gt;)
8525  declare i32 @llvm.annotation.i32(i32 &lt;val&gt;, i8* &lt;str&gt;, i8* &lt;str&gt;, i32  &lt;int&gt;)
8526  declare i64 @llvm.annotation.i64(i64 &lt;val&gt;, i8* &lt;str&gt;, i8* &lt;str&gt;, i32  &lt;int&gt;)
8527  declare i256 @llvm.annotation.i256(i256 &lt;val&gt;, i8* &lt;str&gt;, i8* &lt;str&gt;, i32  &lt;int&gt;)
8528</pre>
8529
8530<h5>Overview:</h5>
8531<p>The '<tt>llvm.annotation</tt>' intrinsic.</p>
8532
8533<h5>Arguments:</h5>
8534<p>The first argument is an integer value (result of some expression), the
8535   second is a pointer to a global string, the third is a pointer to a global
8536   string which is the source file name, and the last argument is the line
8537   number.  It returns the value of the first argument.</p>
8538
8539<h5>Semantics:</h5>
8540<p>This intrinsic allows annotations to be put on arbitrary expressions with
8541   arbitrary strings.  This can be useful for special purpose optimizations that
8542   want to look for these annotations.  These have no other defined use; they
8543   are ignored by code generation and optimization.</p>
8544
8545</div>
8546
8547<!-- _______________________________________________________________________ -->
8548<h4>
8549  <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
8550</h4>
8551
8552<div>
8553
8554<h5>Syntax:</h5>
8555<pre>
8556  declare void @llvm.trap() noreturn nounwind
8557</pre>
8558
8559<h5>Overview:</h5>
8560<p>The '<tt>llvm.trap</tt>' intrinsic.</p>
8561
8562<h5>Arguments:</h5>
8563<p>None.</p>
8564
8565<h5>Semantics:</h5>
8566<p>This intrinsic is lowered to the target dependent trap instruction. If the
8567   target does not have a trap instruction, this intrinsic will be lowered to
8568   a call of the <tt>abort()</tt> function.</p>
8569
8570</div>
8571
8572<!-- _______________________________________________________________________ -->
8573<h4>
8574  <a name="int_debugtrap">'<tt>llvm.debugtrap</tt>' Intrinsic</a>
8575</h4>
8576
8577<div>
8578
8579<h5>Syntax:</h5>
8580<pre>
8581  declare void @llvm.debugtrap() nounwind
8582</pre>
8583
8584<h5>Overview:</h5>
8585<p>The '<tt>llvm.debugtrap</tt>' intrinsic.</p>
8586
8587<h5>Arguments:</h5>
8588<p>None.</p>
8589
8590<h5>Semantics:</h5>
8591<p>This intrinsic is lowered to code which is intended to cause an execution
8592   trap with the intention of requesting the attention of a debugger.</p>
8593
8594</div>
8595
8596<!-- _______________________________________________________________________ -->
8597<h4>
8598  <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a>
8599</h4>
8600
8601<div>
8602
8603<h5>Syntax:</h5>
8604<pre>
8605  declare void @llvm.stackprotector(i8* &lt;guard&gt;, i8** &lt;slot&gt;)
8606</pre>
8607
8608<h5>Overview:</h5>
8609<p>The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and
8610   stores it onto the stack at <tt>slot</tt>. The stack slot is adjusted to
8611   ensure that it is placed on the stack before local variables.</p>
8612
8613<h5>Arguments:</h5>
8614<p>The <tt>llvm.stackprotector</tt> intrinsic requires two pointer
8615   arguments. The first argument is the value loaded from the stack
8616   guard <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt>
8617   that has enough space to hold the value of the guard.</p>
8618
8619<h5>Semantics:</h5>
8620<p>This intrinsic causes the prologue/epilogue inserter to force the position of
8621   the <tt>AllocaInst</tt> stack slot to be before local variables on the
8622   stack. This is to ensure that if a local variable on the stack is
8623   overwritten, it will destroy the value of the guard. When the function exits,
8624   the guard on the stack is checked against the original guard. If they are
8625   different, then the program aborts by calling the <tt>__stack_chk_fail()</tt>
8626   function.</p>
8627
8628</div>
8629
8630<!-- _______________________________________________________________________ -->
8631<h4>
8632  <a name="int_objectsize">'<tt>llvm.objectsize</tt>' Intrinsic</a>
8633</h4>
8634
8635<div>
8636
8637<h5>Syntax:</h5>
8638<pre>
8639  declare i32 @llvm.objectsize.i32(i8* &lt;object&gt;, i1 &lt;min&gt;)
8640  declare i64 @llvm.objectsize.i64(i8* &lt;object&gt;, i1 &lt;min&gt;)
8641</pre>
8642
8643<h5>Overview:</h5>
8644<p>The <tt>llvm.objectsize</tt> intrinsic is designed to provide information to
8645   the optimizers to determine at compile time whether a) an operation (like
8646   memcpy) will overflow a buffer that corresponds to an object, or b) that a
8647   runtime check for overflow isn't necessary. An object in this context means
8648   an allocation of a specific class, structure, array, or other object.</p>
8649
8650<h5>Arguments:</h5>
8651<p>The <tt>llvm.objectsize</tt> intrinsic takes two arguments. The first
8652   argument is a pointer to or into the <tt>object</tt>. The second argument
8653   is a boolean and determines whether <tt>llvm.objectsize</tt> returns 0 (if
8654   true) or -1 (if false) when the object size is unknown.
8655   The second argument only accepts constants.</p>
8656
8657<h5>Semantics:</h5>
8658<p>The <tt>llvm.objectsize</tt> intrinsic is lowered to a constant representing
8659   the size of the object concerned. If the size cannot be determined at compile
8660   time, <tt>llvm.objectsize</tt> returns <tt>i32/i64 -1 or 0</tt>
8661   (depending on the <tt>min</tt> argument).</p>
8662
8663</div>
8664<!-- _______________________________________________________________________ -->
8665<h4>
8666  <a name="int_expect">'<tt>llvm.expect</tt>' Intrinsic</a>
8667</h4>
8668
8669<div>
8670
8671<h5>Syntax:</h5>
8672<pre>
8673  declare i32 @llvm.expect.i32(i32 &lt;val&gt;, i32 &lt;expected_val&gt;)
8674  declare i64 @llvm.expect.i64(i64 &lt;val&gt;, i64 &lt;expected_val&gt;)
8675</pre>
8676
8677<h5>Overview:</h5>
8678<p>The <tt>llvm.expect</tt> intrinsic provides information about expected (the
8679   most probable) value of <tt>val</tt>, which can be used by optimizers.</p>
8680
8681<h5>Arguments:</h5>
8682<p>The <tt>llvm.expect</tt> intrinsic takes two arguments. The first
8683   argument is a value. The second argument is an expected value, this needs to
8684   be a constant value, variables are not allowed.</p>
8685
8686<h5>Semantics:</h5>
8687<p>This intrinsic is lowered to the <tt>val</tt>.</p>
8688</div>
8689
8690<!-- _______________________________________________________________________ -->
8691<h4>
8692  <a name="int_donothing">'<tt>llvm.donothing</tt>' Intrinsic</a>
8693</h4>
8694
8695<div>
8696
8697<h5>Syntax:</h5>
8698<pre>
8699  declare void @llvm.donothing() nounwind readnone
8700</pre>
8701
8702<h5>Overview:</h5>
8703<p>The <tt>llvm.donothing</tt> intrinsic doesn't perform any operation. It's the
8704only intrinsic that can be called with an invoke instruction.</p>
8705
8706<h5>Arguments:</h5>
8707<p>None.</p>
8708
8709<h5>Semantics:</h5>
8710<p>This intrinsic does nothing, and it's removed by optimizers and ignored by
8711codegen.</p>
8712</div>
8713
8714</div>
8715
8716</div>
8717<!-- *********************************************************************** -->
8718<hr>
8719<address>
8720  <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
8721  src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
8722  <a href="http://validator.w3.org/check/referer"><img
8723  src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
8724
8725  <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
8726  <a href="http://llvm.org/">The LLVM Compiler Infrastructure</a><br>
8727  Last modified: $Date$
8728</address>
8729
8730</body>
8731</html>
8732