Gitiles
Code Review Sign In
gerrit-public.fairphone.software / fp2-dev / platform / external / llvm / 6843344e1399eee254f39c3660e2b227c2511a9e / . / docs / LangRef.html
blob: 8ec11235dc2067495e23630833c63655d26a4893 [file] [log] [blame]
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]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="llvm.css" type="text/css">
11</head>
12
13<body>
14
15<div class="doc_title"> LLVM Language Reference Manual </div>
16<ol>
17 <li><a href="#abstract">Abstract</a></li>
18 <li><a href="#introduction">Introduction</a></li>
19 <li><a href="#identifiers">Identifiers</a></li>
20 <li><a href="#highlevel">High Level Structure</a>
21 <ol>
22 <li><a href="#modulestructure">Module Structure</a></li>
23 <li><a href="#linkage">Linkage Types</a></li>
24 <li><a href="#callingconv">Calling Conventions</a></li>
Chris Lattner5b6dc6e2009-01-11 20:53:49 +0000[diff] [blame]25 <li><a href="#namedtypes">Named Types</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]26 <li><a href="#globalvars">Global Variables</a></li>
27 <li><a href="#functionstructure">Functions</a></li>
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]28 <li><a href="#aliasstructure">Aliases</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]29 <li><a href="#paramattrs">Parameter Attributes</a></li>
Devang Patel008cd3e2008-09-26 23:51:19 +0000[diff] [blame]30 <li><a href="#fnattrs">Function Attributes</a></li>
Gordon Henriksen13fe5e32007-12-10 03:18:06 +0000[diff] [blame]31 <li><a href="#gc">Garbage Collector Names</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]32 <li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
33 <li><a href="#datalayout">Data Layout</a></li>
34 </ol>
35 </li>
36 <li><a href="#typesystem">Type System</a>
37 <ol>
Chris Lattner488772f2008-01-04 04:32:38 +0000[diff] [blame]38 <li><a href="#t_classifications">Type Classifications</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]39 <li><a href="#t_primitive">Primitive Types</a>
40 <ol>
Chris Lattner488772f2008-01-04 04:32:38 +0000[diff] [blame]41 <li><a href="#t_floating">Floating Point Types</a></li>
42 <li><a href="#t_void">Void Type</a></li>
43 <li><a href="#t_label">Label Type</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]44 </ol>
45 </li>
46 <li><a href="#t_derived">Derived Types</a>
47 <ol>
Chris Lattner251ab812007-12-18 06:18:21 +0000[diff] [blame]48 <li><a href="#t_integer">Integer Type</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]49 <li><a href="#t_array">Array Type</a></li>
50 <li><a href="#t_function">Function Type</a></li>
51 <li><a href="#t_pointer">Pointer Type</a></li>
52 <li><a href="#t_struct">Structure Type</a></li>
53 <li><a href="#t_pstruct">Packed Structure Type</a></li>
54 <li><a href="#t_vector">Vector Type</a></li>
55 <li><a href="#t_opaque">Opaque Type</a></li>
56 </ol>
57 </li>
Chris Lattner515195a2009-02-02 07:32:36 +0000[diff] [blame]58 <li><a href="#t_uprefs">Type Up-references</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]59 </ol>
60 </li>
61 <li><a href="#constants">Constants</a>
62 <ol>
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]63 <li><a href="#simpleconstants">Simple Constants</a></li>
Chris Lattner97063852009-02-28 18:32:25 +0000[diff] [blame]64 <li><a href="#complexconstants">Complex Constants</a></li>
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]65 <li><a href="#globalconstants">Global Variable and Function Addresses</a></li>
66 <li><a href="#undefvalues">Undefined Values</a></li>
67 <li><a href="#constantexprs">Constant Expressions</a></li>
Nick Lewycky4dcf8102009-04-04 07:22:01 +0000[diff] [blame]68 <li><a href="#metadata">Embedded Metadata</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]69 </ol>
70 </li>
71 <li><a href="#othervalues">Other Values</a>
72 <ol>
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]73 <li><a href="#inlineasm">Inline Assembler Expressions</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]74 </ol>
75 </li>
76 <li><a href="#instref">Instruction Reference</a>
77 <ol>
78 <li><a href="#terminators">Terminator Instructions</a>
79 <ol>
80 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
81 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
82 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
83 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
84 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
85 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
86 </ol>
87 </li>
88 <li><a href="#binaryops">Binary Operations</a>
89 <ol>
90 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
91 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
92 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
93 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
94 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
95 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
96 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
97 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
98 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
99 </ol>
100 </li>
101 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
102 <ol>
103 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
104 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
105 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
106 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
107 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
108 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
109 </ol>
110 </li>
111 <li><a href="#vectorops">Vector Operations</a>
112 <ol>
113 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
114 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
115 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
116 </ol>
117 </li>
Dan Gohman74d6faf2008-05-12 23:51:09 +0000[diff] [blame]118 <li><a href="#aggregateops">Aggregate Operations</a>
119 <ol>
120 <li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li>
121 <li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li>
122 </ol>
123 </li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]124 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
125 <ol>
126 <li><a href="#i_malloc">'<tt>malloc</tt>' Instruction</a></li>
127 <li><a href="#i_free">'<tt>free</tt>' Instruction</a></li>
128 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
129 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
130 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
131 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
132 </ol>
133 </li>
134 <li><a href="#convertops">Conversion Operations</a>
135 <ol>
136 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
137 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
138 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
139 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
140 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
141 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
142 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
143 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
144 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
145 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
146 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
147 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
148 </ol>
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]149 </li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]150 <li><a href="#otherops">Other Operations</a>
151 <ol>
152 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
153 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
Nate Begeman646fa482008-05-12 19:01:56 +0000[diff] [blame]154 <li><a href="#i_vicmp">'<tt>vicmp</tt>' Instruction</a></li>
155 <li><a href="#i_vfcmp">'<tt>vfcmp</tt>' Instruction</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]156 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
157 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
158 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
159 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
160 </ol>
161 </li>
162 </ol>
163 </li>
164 <li><a href="#intrinsics">Intrinsic Functions</a>
165 <ol>
166 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
167 <ol>
168 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
169 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
170 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
171 </ol>
172 </li>
173 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
174 <ol>
175 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
176 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
177 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
178 </ol>
179 </li>
180 <li><a href="#int_codegen">Code Generator Intrinsics</a>
181 <ol>
182 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
183 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
184 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
185 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
186 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
187 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
188 <li><a href="#int_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
189 </ol>
190 </li>
191 <li><a href="#int_libc">Standard C Library Intrinsics</a>
192 <ol>
193 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
194 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
195 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
196 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
197 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
Dan Gohman361079c2007-10-15 20:30:11 +0000[diff] [blame]198 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
199 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
200 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]201 </ol>
202 </li>
203 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
204 <ol>
205 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
206 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
207 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
208 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
209 <li><a href="#int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic </a></li>
210 <li><a href="#int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic </a></li>
211 </ol>
212 </li>
Bill Wendling3f8cebe2009-02-08 01:40:31 +0000[diff] [blame]213 <li><a href="#int_overflow">Arithmetic with Overflow Intrinsics</a>
214 <ol>
Bill Wendling3e1258b2009-02-08 04:04:40 +0000[diff] [blame]215 <li><a href="#int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt> Intrinsics</a></li>
216 <li><a href="#int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt> Intrinsics</a></li>
217 <li><a href="#int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt> Intrinsics</a></li>
218 <li><a href="#int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt> Intrinsics</a></li>
219 <li><a href="#int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt> Intrinsics</a></li>
Bill Wendlingbda98b62009-02-08 23:00:09 +0000[diff] [blame]220 <li><a href="#int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt> Intrinsics</a></li>
Bill Wendling3f8cebe2009-02-08 01:40:31 +0000[diff] [blame]221 </ol>
222 </li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]223 <li><a href="#int_debugger">Debugger intrinsics</a></li>
224 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
Duncan Sands7407a9f2007-09-11 14:10:23 +0000[diff] [blame]225 <li><a href="#int_trampoline">Trampoline Intrinsic</a>
Duncan Sands38947cd2007-07-27 12:58:54 +0000[diff] [blame]226 <ol>
227 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
Duncan Sands38947cd2007-07-27 12:58:54 +0000[diff] [blame]228 </ol>
229 </li>
Bill Wendling9127adb2008-11-18 22:10:53 +0000[diff] [blame]230 <li><a href="#int_atomics">Atomic intrinsics</a>
231 <ol>
232 <li><a href="#int_memory_barrier"><tt>llvm.memory_barrier</tt></a></li>
233 <li><a href="#int_atomic_cmp_swap"><tt>llvm.atomic.cmp.swap</tt></a></li>
234 <li><a href="#int_atomic_swap"><tt>llvm.atomic.swap</tt></a></li>
235 <li><a href="#int_atomic_load_add"><tt>llvm.atomic.load.add</tt></a></li>
236 <li><a href="#int_atomic_load_sub"><tt>llvm.atomic.load.sub</tt></a></li>
237 <li><a href="#int_atomic_load_and"><tt>llvm.atomic.load.and</tt></a></li>
238 <li><a href="#int_atomic_load_nand"><tt>llvm.atomic.load.nand</tt></a></li>
239 <li><a href="#int_atomic_load_or"><tt>llvm.atomic.load.or</tt></a></li>
240 <li><a href="#int_atomic_load_xor"><tt>llvm.atomic.load.xor</tt></a></li>
241 <li><a href="#int_atomic_load_max"><tt>llvm.atomic.load.max</tt></a></li>
242 <li><a href="#int_atomic_load_min"><tt>llvm.atomic.load.min</tt></a></li>
243 <li><a href="#int_atomic_load_umax"><tt>llvm.atomic.load.umax</tt></a></li>
244 <li><a href="#int_atomic_load_umin"><tt>llvm.atomic.load.umin</tt></a></li>
245 </ol>
246 </li>
Reid Spencerb043f672007-07-20 19:59:11 +0000[diff] [blame]247 <li><a href="#int_general">General intrinsics</a>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]248 <ol>
Reid Spencerb043f672007-07-20 19:59:11 +0000[diff] [blame]249 <li><a href="#int_var_annotation">
Bill Wendlinge4164592008-11-19 05:56:17 +0000[diff] [blame]250 '<tt>llvm.var.annotation</tt>' Intrinsic</a></li>
Tanya Lattnerb306a9e2007-09-21 22:59:12 +0000[diff] [blame]251 <li><a href="#int_annotation">
Bill Wendlinge4164592008-11-19 05:56:17 +0000[diff] [blame]252 '<tt>llvm.annotation.*</tt>' Intrinsic</a></li>
Anton Korobeynikove6e764f2008-01-15 22:31:34 +0000[diff] [blame]253 <li><a href="#int_trap">
Bill Wendlinge4164592008-11-19 05:56:17 +0000[diff] [blame]254 '<tt>llvm.trap</tt>' Intrinsic</a></li>
255 <li><a href="#int_stackprotector">
256 '<tt>llvm.stackprotector</tt>' Intrinsic</a></li>
Tanya Lattnerb306a9e2007-09-21 22:59:12 +0000[diff] [blame]257 </ol>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]258 </li>
259 </ol>
260 </li>
261</ol>
262
263<div class="doc_author">
264 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
265 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
266</div>
267
268<!-- *********************************************************************** -->
269<div class="doc_section"> <a name="abstract">Abstract </a></div>
270<!-- *********************************************************************** -->
271
272<div class="doc_text">
273<p>This document is a reference manual for the LLVM assembly language.
Bill Wendlinge7846a52008-08-05 22:29:16 +0000[diff] [blame]274LLVM is a Static Single Assignment (SSA) based representation that provides
Chris Lattner96451482008-08-05 18:29:16 +0000[diff] [blame]275type safety, low-level operations, flexibility, and the capability of
276representing 'all' high-level languages cleanly. It is the common code
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]277representation used throughout all phases of the LLVM compilation
278strategy.</p>
279</div>
280
281<!-- *********************************************************************** -->
282<div class="doc_section"> <a name="introduction">Introduction</a> </div>
283<!-- *********************************************************************** -->
284
285<div class="doc_text">
286
287<p>The LLVM code representation is designed to be used in three
288different forms: as an in-memory compiler IR, as an on-disk bitcode
289representation (suitable for fast loading by a Just-In-Time compiler),
290and as a human readable assembly language representation. This allows
291LLVM to provide a powerful intermediate representation for efficient
292compiler transformations and analysis, while providing a natural means
293to debug and visualize the transformations. The three different forms
294of LLVM are all equivalent. This document describes the human readable
295representation and notation.</p>
296
297<p>The LLVM representation aims to be light-weight and low-level
298while being expressive, typed, and extensible at the same time. It
299aims to be a "universal IR" of sorts, by being at a low enough level
300that high-level ideas may be cleanly mapped to it (similar to how
301microprocessors are "universal IR's", allowing many source languages to
302be mapped to them). By providing type information, LLVM can be used as
303the target of optimizations: for example, through pointer analysis, it
304can be proven that a C automatic variable is never accessed outside of
305the current function... allowing it to be promoted to a simple SSA
306value instead of a memory location.</p>
307
308</div>
309
310<!-- _______________________________________________________________________ -->
311<div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
312
313<div class="doc_text">
314
315<p>It is important to note that this document describes 'well formed'
316LLVM assembly language. There is a difference between what the parser
317accepts and what is considered 'well formed'. For example, the
318following instruction is syntactically okay, but not well formed:</p>
319
320<div class="doc_code">
321<pre>
322%x = <a href="#i_add">add</a> i32 1, %x
323</pre>
324</div>
325
326<p>...because the definition of <tt>%x</tt> does not dominate all of
327its uses. The LLVM infrastructure provides a verification pass that may
328be used to verify that an LLVM module is well formed. This pass is
329automatically run by the parser after parsing input assembly and by
330the optimizer before it outputs bitcode. The violations pointed out
331by the verifier pass indicate bugs in transformation passes or input to
332the parser.</p>
333</div>
334
Chris Lattnera83fdc02007-10-03 17:34:29 +0000[diff] [blame]335<!-- Describe the typesetting conventions here. -->
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]336
337<!-- *********************************************************************** -->
338<div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
339<!-- *********************************************************************** -->
340
341<div class="doc_text">
342
Reid Spencerc8245b02007-08-07 14:34:28 +0000[diff] [blame]343 <p>LLVM identifiers come in two basic types: global and local. Global
344 identifiers (functions, global variables) begin with the @ character. Local
345 identifiers (register names, types) begin with the % character. Additionally,
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]346 there are three different formats for identifiers, for different purposes:</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]347
348<ol>
Reid Spencerc8245b02007-08-07 14:34:28 +0000[diff] [blame]349 <li>Named values are represented as a string of characters with their prefix.
350 For example, %foo, @DivisionByZero, %a.really.long.identifier. The actual
351 regular expression used is '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]352 Identifiers which require other characters in their names can be surrounded
Daniel Dunbar73f9a772008-10-14 23:51:43 +0000[diff] [blame]353 with quotes. Special characters may be escaped using "\xx" where xx is the
354 ASCII code for the character in hexadecimal. In this way, any character can
355 be used in a name value, even quotes themselves.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]356
Reid Spencerc8245b02007-08-07 14:34:28 +0000[diff] [blame]357 <li>Unnamed values are represented as an unsigned numeric value with their
358 prefix. For example, %12, @2, %44.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]359
360 <li>Constants, which are described in a <a href="#constants">section about
361 constants</a>, below.</li>
362</ol>
363
Reid Spencerc8245b02007-08-07 14:34:28 +0000[diff] [blame]364<p>LLVM requires that values start with a prefix for two reasons: Compilers
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]365don't need to worry about name clashes with reserved words, and the set of
366reserved words may be expanded in the future without penalty. Additionally,
367unnamed identifiers allow a compiler to quickly come up with a temporary
368variable without having to avoid symbol table conflicts.</p>
369
370<p>Reserved words in LLVM are very similar to reserved words in other
371languages. There are keywords for different opcodes
372('<tt><a href="#i_add">add</a></tt>',
373 '<tt><a href="#i_bitcast">bitcast</a></tt>',
374 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
375href="#t_void">void</a></tt>', '<tt><a href="#t_primitive">i32</a></tt>', etc...),
376and others. These reserved words cannot conflict with variable names, because
Reid Spencerc8245b02007-08-07 14:34:28 +0000[diff] [blame]377none of them start with a prefix character ('%' or '@').</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]378
379<p>Here is an example of LLVM code to multiply the integer variable
380'<tt>%X</tt>' by 8:</p>
381
382<p>The easy way:</p>
383
384<div class="doc_code">
385<pre>
386%result = <a href="#i_mul">mul</a> i32 %X, 8
387</pre>
388</div>
389
390<p>After strength reduction:</p>
391
392<div class="doc_code">
393<pre>
394%result = <a href="#i_shl">shl</a> i32 %X, i8 3
395</pre>
396</div>
397
398<p>And the hard way:</p>
399
400<div class="doc_code">
401<pre>
402<a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
403<a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
404%result = <a href="#i_add">add</a> i32 %1, %1
405</pre>
406</div>
407
408<p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
409important lexical features of LLVM:</p>
410
411<ol>
412
413 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
414 line.</li>
415
416 <li>Unnamed temporaries are created when the result of a computation is not
417 assigned to a named value.</li>
418
419 <li>Unnamed temporaries are numbered sequentially</li>
420
421</ol>
422
423<p>...and it also shows a convention that we follow in this document. When
424demonstrating instructions, we will follow an instruction with a comment that
425defines the type and name of value produced. Comments are shown in italic
426text.</p>
427
428</div>
429
430<!-- *********************************************************************** -->
431<div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
432<!-- *********************************************************************** -->
433
434<!-- ======================================================================= -->
435<div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
436</div>
437
438<div class="doc_text">
439
440<p>LLVM programs are composed of "Module"s, each of which is a
441translation unit of the input programs. Each module consists of
442functions, global variables, and symbol table entries. Modules may be
443combined together with the LLVM linker, which merges function (and
444global variable) definitions, resolves forward declarations, and merges
445symbol table entries. Here is an example of the "hello world" module:</p>
446
447<div class="doc_code">
448<pre><i>; Declare the string constant as a global constant...</i>
449<a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a
450 href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
451
452<i>; External declaration of the puts function</i>
453<a href="#functionstructure">declare</a> i32 @puts(i8 *) <i>; i32(i8 *)* </i>
454
455<i>; Definition of main function</i>
456define i32 @main() { <i>; i32()* </i>
Dan Gohman01852382009-01-04 23:44:43 +0000[diff] [blame]457 <i>; Convert [13 x i8]* to i8 *...</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]458 %cast210 = <a
Dan Gohman01852382009-01-04 23:44:43 +0000[diff] [blame]459 href="#i_getelementptr">getelementptr</a> [13 x i8]* @.LC0, i64 0, i64 0 <i>; i8 *</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]460
461 <i>; Call puts function to write out the string to stdout...</i>
462 <a
463 href="#i_call">call</a> i32 @puts(i8 * %cast210) <i>; i32</i>
464 <a
465 href="#i_ret">ret</a> i32 0<br>}<br>
466</pre>
467</div>
468
469<p>This example is made up of a <a href="#globalvars">global variable</a>
470named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
471function, and a <a href="#functionstructure">function definition</a>
472for "<tt>main</tt>".</p>
473
474<p>In general, a module is made up of a list of global values,
475where both functions and global variables are global values. Global values are
476represented by a pointer to a memory location (in this case, a pointer to an
477array of char, and a pointer to a function), and have one of the following <a
478href="#linkage">linkage types</a>.</p>
479
480</div>
481
482<!-- ======================================================================= -->
483<div class="doc_subsection">
484 <a name="linkage">Linkage Types</a>
485</div>
486
487<div class="doc_text">
488
489<p>
490All Global Variables and Functions have one of the following types of linkage:
491</p>
492
493<dl>
494
Rafael Espindolaa168fc92009-01-15 20:18:42 +0000[diff] [blame]495 <dt><tt><b><a name="linkage_private">private</a></b></tt>: </dt>
496
497 <dd>Global values with private linkage are only directly accessible by
498 objects in the current module. In particular, linking code into a module with
499 an private global value may cause the private to be renamed as necessary to
500 avoid collisions. Because the symbol is private to the module, all
501 references can be updated. This doesn't show up in any symbol table in the
502 object file.
503 </dd>
504
Dale Johannesen96e7e092008-05-23 23:13:41 +0000[diff] [blame]505 <dt><tt><b><a name="linkage_internal">internal</a></b></tt>: </dt>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]506
Duncan Sandsa75223a2009-01-16 09:29:46 +0000[diff] [blame]507 <dd> Similar to private, but the value shows as a local symbol (STB_LOCAL in
Rafael Espindolaa168fc92009-01-15 20:18:42 +0000[diff] [blame]508 the case of ELF) in the object file. This corresponds to the notion of the
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]509 '<tt>static</tt>' keyword in C.
510 </dd>
511
Chris Lattner68433442009-04-13 05:44:34 +0000[diff] [blame^]512 <dt><tt><b><a name="available_externally">available_externally</a></b></tt>:
513 </dt>
514
515 <dd>Globals with "<tt>available_externally</tt>" linkage are never emitted
516 into the object file corresponding to the LLVM module. They exist to
517 allow inlining and other optimizations to take place given knowledge of the
518 definition of the global, which is known to be somewhere outside the module.
519 Globals with <tt>available_externally</tt> linkage are allowed to be discarded
520 at will, and are otherwise the same as <tt>linkonce_odr</tt>. This linkage
521 type is only allowed on definitions, not declarations.</dd>
522
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]523 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
524
525 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
526 the same name when linkage occurs. This is typically used to implement
527 inline functions, templates, or other code which must be generated in each
528 translation unit that uses it. Unreferenced <tt>linkonce</tt> globals are
529 allowed to be discarded.
530 </dd>
531
Dale Johannesen96e7e092008-05-23 23:13:41 +0000[diff] [blame]532 <dt><tt><b><a name="linkage_common">common</a></b></tt>: </dt>
533
534 <dd>"<tt>common</tt>" linkage is exactly the same as <tt>linkonce</tt>
535 linkage, except that unreferenced <tt>common</tt> globals may not be
536 discarded. This is used for globals that may be emitted in multiple
537 translation units, but that are not guaranteed to be emitted into every
538 translation unit that uses them. One example of this is tentative
539 definitions in C, such as "<tt>int X;</tt>" at global scope.
540 </dd>
541
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]542 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
543
Dale Johannesen96e7e092008-05-23 23:13:41 +0000[diff] [blame]544 <dd>"<tt>weak</tt>" linkage is the same as <tt>common</tt> linkage, except
545 that some targets may choose to emit different assembly sequences for them
546 for target-dependent reasons. This is used for globals that are declared
547 "weak" in C source code.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]548 </dd>
549
550 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
551
552 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
553 pointer to array type. When two global variables with appending linkage are
554 linked together, the two global arrays are appended together. This is the
555 LLVM, typesafe, equivalent of having the system linker append together
556 "sections" with identical names when .o files are linked.
557 </dd>
558
559 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt>: </dt>
Duncan Sands19d161f2009-03-07 15:45:40 +0000[diff] [blame]560
Chris Lattner96451482008-08-05 18:29:16 +0000[diff] [blame]561 <dd>The semantics of this linkage follow the ELF object file model: the
562 symbol is weak until linked, if not linked, the symbol becomes null instead
563 of being an undefined reference.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]564 </dd>
565
Duncan Sands19d161f2009-03-07 15:45:40 +0000[diff] [blame]566 <dt><tt><b><a name="linkage_linkonce">linkonce_odr</a></b></tt>: </dt>
Duncan Sands19d161f2009-03-07 15:45:40 +0000[diff] [blame]567 <dt><tt><b><a name="linkage_weak">weak_odr</a></b></tt>: </dt>
Chris Lattner68433442009-04-13 05:44:34 +0000[diff] [blame^]568 <dd>Some languages allow differing globals to be merged, such as two
Duncan Sands19d161f2009-03-07 15:45:40 +0000[diff] [blame]569 functions with different semantics. Other languages, such as <tt>C++</tt>,
570 ensure that only equivalent globals are ever merged (the "one definition
Chris Lattner68433442009-04-13 05:44:34 +0000[diff] [blame^]571 rule" - "ODR"). Such languages can use the <tt>linkonce_odr</tt>
Duncan Sandsb95df792009-03-11 20:14:15 +0000[diff] [blame]572 and <tt>weak_odr</tt> linkage types to indicate that the global will only
573 be merged with equivalent globals. These linkage types are otherwise the
574 same as their non-<tt>odr</tt> versions.
Duncan Sands19d161f2009-03-07 15:45:40 +0000[diff] [blame]575 </dd>
576
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]577 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
578
579 <dd>If none of the above identifiers are used, the global is externally
580 visible, meaning that it participates in linkage and can be used to resolve
581 external symbol references.
582 </dd>
583</dl>
584
585 <p>
586 The next two types of linkage are targeted for Microsoft Windows platform
587 only. They are designed to support importing (exporting) symbols from (to)
Chris Lattner96451482008-08-05 18:29:16 +0000[diff] [blame]588 DLLs (Dynamic Link Libraries).
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]589 </p>
590
591 <dl>
592 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt>: </dt>
593
594 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
595 or variable via a global pointer to a pointer that is set up by the DLL
596 exporting the symbol. On Microsoft Windows targets, the pointer name is
Dan Gohman5ec99832009-01-12 21:35:55 +0000[diff] [blame]597 formed by combining <code>__imp_</code> and the function or variable name.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]598 </dd>
599
600 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt>: </dt>
601
602 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
603 pointer to a pointer in a DLL, so that it can be referenced with the
604 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
Dan Gohman5ec99832009-01-12 21:35:55 +0000[diff] [blame]605 name is formed by combining <code>__imp_</code> and the function or variable
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]606 name.
607 </dd>
608
609</dl>
610
Dan Gohman4dfac702008-11-24 17:18:39 +0000[diff] [blame]611<p>For example, since the "<tt>.LC0</tt>"
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]612variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
613variable and was linked with this one, one of the two would be renamed,
614preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
615external (i.e., lacking any linkage declarations), they are accessible
616outside of the current module.</p>
617<p>It is illegal for a function <i>declaration</i>
Duncan Sands565f65d2009-03-11 08:08:06 +0000[diff] [blame]618to have any linkage type other than "externally visible", <tt>dllimport</tt>
619or <tt>extern_weak</tt>.</p>
Duncan Sands19d161f2009-03-07 15:45:40 +0000[diff] [blame]620<p>Aliases can have only <tt>external</tt>, <tt>internal</tt>, <tt>weak</tt>
621or <tt>weak_odr</tt> linkages.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]622</div>
623
624<!-- ======================================================================= -->
625<div class="doc_subsection">
626 <a name="callingconv">Calling Conventions</a>
627</div>
628
629<div class="doc_text">
630
631<p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
632and <a href="#i_invoke">invokes</a> can all have an optional calling convention
633specified for the call. The calling convention of any pair of dynamic
634caller/callee must match, or the behavior of the program is undefined. The
635following calling conventions are supported by LLVM, and more may be added in
636the future:</p>
637
638<dl>
639 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
640
641 <dd>This calling convention (the default if no other calling convention is
642 specified) matches the target C calling conventions. This calling convention
643 supports varargs function calls and tolerates some mismatch in the declared
644 prototype and implemented declaration of the function (as does normal C).
645 </dd>
646
647 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
648
649 <dd>This calling convention attempts to make calls as fast as possible
650 (e.g. by passing things in registers). This calling convention allows the
651 target to use whatever tricks it wants to produce fast code for the target,
Chris Lattner96451482008-08-05 18:29:16 +0000[diff] [blame]652 without having to conform to an externally specified ABI (Application Binary
653 Interface). Implementations of this convention should allow arbitrary
Arnold Schwaighofer07444922008-05-14 09:17:12 +0000[diff] [blame]654 <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> to be
655 supported. This calling convention does not support varargs and requires the
656 prototype of all callees to exactly match the prototype of the function
657 definition.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]658 </dd>
659
660 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
661
662 <dd>This calling convention attempts to make code in the caller as efficient
663 as possible under the assumption that the call is not commonly executed. As
664 such, these calls often preserve all registers so that the call does not break
665 any live ranges in the caller side. This calling convention does not support
666 varargs and requires the prototype of all callees to exactly match the
667 prototype of the function definition.
668 </dd>
669
670 <dt><b>"<tt>cc &lt;<em>n</em>&gt;</tt>" - Numbered convention</b>:</dt>
671
672 <dd>Any calling convention may be specified by number, allowing
673 target-specific calling conventions to be used. Target specific calling
674 conventions start at 64.
675 </dd>
676</dl>
677
678<p>More calling conventions can be added/defined on an as-needed basis, to
679support pascal conventions or any other well-known target-independent
680convention.</p>
681
682</div>
683
684<!-- ======================================================================= -->
685<div class="doc_subsection">
686 <a name="visibility">Visibility Styles</a>
687</div>
688
689<div class="doc_text">
690
691<p>
692All Global Variables and Functions have one of the following visibility styles:
693</p>
694
695<dl>
696 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
697
Chris Lattner96451482008-08-05 18:29:16 +0000[diff] [blame]698 <dd>On targets that use the ELF object file format, default visibility means
699 that the declaration is visible to other
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]700 modules and, in shared libraries, means that the declared entity may be
701 overridden. On Darwin, default visibility means that the declaration is
702 visible to other modules. Default visibility corresponds to "external
703 linkage" in the language.
704 </dd>
705
706 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
707
708 <dd>Two declarations of an object with hidden visibility refer to the same
709 object if they are in the same shared object. Usually, hidden visibility
710 indicates that the symbol will not be placed into the dynamic symbol table,
711 so no other module (executable or shared library) can reference it
712 directly.
713 </dd>
714
715 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
716
717 <dd>On ELF, protected visibility indicates that the symbol will be placed in
718 the dynamic symbol table, but that references within the defining module will
719 bind to the local symbol. That is, the symbol cannot be overridden by another
720 module.
721 </dd>
722</dl>
723
724</div>
725
726<!-- ======================================================================= -->
727<div class="doc_subsection">
Chris Lattner5b6dc6e2009-01-11 20:53:49 +0000[diff] [blame]728 <a name="namedtypes">Named Types</a>
729</div>
730
731<div class="doc_text">
732
733<p>LLVM IR allows you to specify name aliases for certain types. This can make
734it easier to read the IR and make the IR more condensed (particularly when
735recursive types are involved). An example of a name specification is:
736</p>
737
738<div class="doc_code">
739<pre>
740%mytype = type { %mytype*, i32 }
741</pre>
742</div>
743
744<p>You may give a name to any <a href="#typesystem">type</a> except "<a
745href="t_void">void</a>". Type name aliases may be used anywhere a type is
746expected with the syntax "%mytype".</p>
747
748<p>Note that type names are aliases for the structural type that they indicate,
749and that you can therefore specify multiple names for the same type. This often
750leads to confusing behavior when dumping out a .ll file. Since LLVM IR uses
751structural typing, the name is not part of the type. When printing out LLVM IR,
752the printer will pick <em>one name</em> to render all types of a particular
753shape. This means that if you have code where two different source types end up
754having the same LLVM type, that the dumper will sometimes print the "wrong" or
755unexpected type. This is an important design point and isn't going to
756change.</p>
757
758</div>
759
Chris Lattner5b6dc6e2009-01-11 20:53:49 +0000[diff] [blame]760<!-- ======================================================================= -->
761<div class="doc_subsection">
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]762 <a name="globalvars">Global Variables</a>
763</div>
764
765<div class="doc_text">
766
767<p>Global variables define regions of memory allocated at compilation time
768instead of run-time. Global variables may optionally be initialized, may have
769an explicit section to be placed in, and may have an optional explicit alignment
770specified. A variable may be defined as "thread_local", which means that it
771will not be shared by threads (each thread will have a separated copy of the
772variable). A variable may be defined as a global "constant," which indicates
773that the contents of the variable will <b>never</b> be modified (enabling better
774optimization, allowing the global data to be placed in the read-only section of
775an executable, etc). Note that variables that need runtime initialization
776cannot be marked "constant" as there is a store to the variable.</p>
777
778<p>
779LLVM explicitly allows <em>declarations</em> of global variables to be marked
780constant, even if the final definition of the global is not. This capability
781can be used to enable slightly better optimization of the program, but requires
782the language definition to guarantee that optimizations based on the
783'constantness' are valid for the translation units that do not include the
784definition.
785</p>
786
787<p>As SSA values, global variables define pointer values that are in
788scope (i.e. they dominate) all basic blocks in the program. Global
789variables always define a pointer to their "content" type because they
790describe a region of memory, and all memory objects in LLVM are
791accessed through pointers.</p>
792
Christopher Lambdd0049d2007-12-11 09:31:00 +0000[diff] [blame]793<p>A global variable may be declared to reside in a target-specifc numbered
794address space. For targets that support them, address spaces may affect how
795optimizations are performed and/or what target instructions are used to access
Christopher Lamb20a39e92007-12-12 08:44:39 +0000[diff] [blame]796the variable. The default address space is zero. The address space qualifier
797must precede any other attributes.</p>
Christopher Lambdd0049d2007-12-11 09:31:00 +0000[diff] [blame]798
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]799<p>LLVM allows an explicit section to be specified for globals. If the target
800supports it, it will emit globals to the section specified.</p>
801
802<p>An explicit alignment may be specified for a global. If not present, or if
803the alignment is set to zero, the alignment of the global is set by the target
804to whatever it feels convenient. If an explicit alignment is specified, the
805global is forced to have at least that much alignment. All alignments must be
806a power of 2.</p>
807
Christopher Lambdd0049d2007-12-11 09:31:00 +0000[diff] [blame]808<p>For example, the following defines a global in a numbered address space with
809an initializer, section, and alignment:</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]810
811<div class="doc_code">
812<pre>
Dan Gohman21ef02c2009-01-11 00:40:00 +0000[diff] [blame]813@G = addrspace(5) constant float 1.0, section "foo", align 4
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]814</pre>
815</div>
816
817</div>
818
819
820<!-- ======================================================================= -->
821<div class="doc_subsection">
822 <a name="functionstructure">Functions</a>
823</div>
824
825<div class="doc_text">
826
827<p>LLVM function definitions consist of the "<tt>define</tt>" keyord,
828an optional <a href="#linkage">linkage type</a>, an optional
829<a href="#visibility">visibility style</a>, an optional
830<a href="#callingconv">calling convention</a>, a return type, an optional
831<a href="#paramattrs">parameter attribute</a> for the return type, a function
832name, a (possibly empty) argument list (each with optional
Devang Patelac2fc272008-10-06 18:50:38 +0000[diff] [blame]833<a href="#paramattrs">parameter attributes</a>), optional
834<a href="#fnattrs">function attributes</a>, an optional section,
835an optional alignment, an optional <a href="#gc">garbage collector name</a>,
Chris Lattneree202542008-10-04 18:10:21 +0000[diff] [blame]836an opening curly brace, a list of basic blocks, and a closing curly brace.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]837
838LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
839optional <a href="#linkage">linkage type</a>, an optional
840<a href="#visibility">visibility style</a>, an optional
841<a href="#callingconv">calling convention</a>, a return type, an optional
842<a href="#paramattrs">parameter attribute</a> for the return type, a function
Gordon Henriksen13fe5e32007-12-10 03:18:06 +0000[diff] [blame]843name, a possibly empty list of arguments, an optional alignment, and an optional
Gordon Henriksend606f5b2007-12-10 03:30:21 +0000[diff] [blame]844<a href="#gc">garbage collector name</a>.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]845
Chris Lattner96451482008-08-05 18:29:16 +0000[diff] [blame]846<p>A function definition contains a list of basic blocks, forming the CFG
847(Control Flow Graph) for
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]848the function. Each basic block may optionally start with a label (giving the
849basic block a symbol table entry), contains a list of instructions, and ends
850with a <a href="#terminators">terminator</a> instruction (such as a branch or
851function return).</p>
852
853<p>The first basic block in a function is special in two ways: it is immediately
854executed on entrance to the function, and it is not allowed to have predecessor
855basic blocks (i.e. there can not be any branches to the entry block of a
856function). Because the block can have no predecessors, it also cannot have any
857<a href="#i_phi">PHI nodes</a>.</p>
858
859<p>LLVM allows an explicit section to be specified for functions. If the target
860supports it, it will emit functions to the section specified.</p>
861
862<p>An explicit alignment may be specified for a function. If not present, or if
863the alignment is set to zero, the alignment of the function is set by the target
864to whatever it feels convenient. If an explicit alignment is specified, the
865function is forced to have at least that much alignment. All alignments must be
866a power of 2.</p>
867
Devang Pateld0bfcc72008-10-07 17:48:33 +0000[diff] [blame]868 <h5>Syntax:</h5>
869
870<div class="doc_code">
Chris Lattner1e5c5cd02008-10-13 16:55:18 +0000[diff] [blame]871<tt>
872define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
873 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
874 &lt;ResultType&gt; @&lt;FunctionName&gt; ([argument list])
875 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
876 [<a href="#gc">gc</a>] { ... }
877</tt>
Devang Pateld0bfcc72008-10-07 17:48:33 +0000[diff] [blame]878</div>
879
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]880</div>
881
882
883<!-- ======================================================================= -->
884<div class="doc_subsection">
885 <a name="aliasstructure">Aliases</a>
886</div>
887<div class="doc_text">
888 <p>Aliases act as "second name" for the aliasee value (which can be either
Anton Korobeynikov96822812008-03-22 08:36:14 +0000[diff] [blame]889 function, global variable, another alias or bitcast of global value). Aliases
890 may have an optional <a href="#linkage">linkage type</a>, and an
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]891 optional <a href="#visibility">visibility style</a>.</p>
892
893 <h5>Syntax:</h5>
894
895<div class="doc_code">
896<pre>
Duncan Sandsd7bfabf2008-09-12 20:48:21 +0000[diff] [blame]897@&lt;Name&gt; = alias [Linkage] [Visibility] &lt;AliaseeTy&gt; @&lt;Aliasee&gt;
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]898</pre>
899</div>
900
901</div>
902
903
904
905<!-- ======================================================================= -->
906<div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
907<div class="doc_text">
908 <p>The return type and each parameter of a function type may have a set of
909 <i>parameter attributes</i> associated with them. Parameter attributes are
910 used to communicate additional information about the result or parameters of
Duncan Sandsf5588dc2007-11-27 13:23:08 +0000[diff] [blame]911 a function. Parameter attributes are considered to be part of the function,
912 not of the function type, so functions with different parameter attributes
913 can have the same function type.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]914
915 <p>Parameter attributes are simple keywords that follow the type specified. If
916 multiple parameter attributes are needed, they are space separated. For
917 example:</p>
918
919<div class="doc_code">
920<pre>
Nick Lewycky3022a742009-02-15 23:06:14 +0000[diff] [blame]921declare i32 @printf(i8* noalias nocapture, ...)
Chris Lattnerf33b8452008-10-04 18:33:34 +0000[diff] [blame]922declare i32 @atoi(i8 zeroext)
923declare signext i8 @returns_signed_char()
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]924</pre>
925</div>
926
Duncan Sandsf5588dc2007-11-27 13:23:08 +0000[diff] [blame]927 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
928 <tt>readonly</tt>) come immediately after the argument list.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]929
930 <p>Currently, only the following parameter attributes are defined:</p>
931 <dl>
Reid Spencerf234bed2007-07-19 23:13:04 +0000[diff] [blame]932 <dt><tt>zeroext</tt></dt>
Chris Lattnerf33b8452008-10-04 18:33:34 +0000[diff] [blame]933 <dd>This indicates to the code generator that the parameter or return value
934 should be zero-extended to a 32-bit value by the caller (for a parameter)
935 or the callee (for a return value).</dd>
Chris Lattner275e6be2008-01-11 06:20:47 +0000[diff] [blame]936
Reid Spencerf234bed2007-07-19 23:13:04 +0000[diff] [blame]937 <dt><tt>signext</tt></dt>
Chris Lattnerf33b8452008-10-04 18:33:34 +0000[diff] [blame]938 <dd>This indicates to the code generator that the parameter or return value
939 should be sign-extended to a 32-bit value by the caller (for a parameter)
940 or the callee (for a return value).</dd>
Chris Lattner275e6be2008-01-11 06:20:47 +0000[diff] [blame]941
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]942 <dt><tt>inreg</tt></dt>
Dale Johannesenc08a0e22008-09-25 20:47:45 +0000[diff] [blame]943 <dd>This indicates that this parameter or return value should be treated
944 in a special target-dependent fashion during while emitting code for a
945 function call or return (usually, by putting it in a register as opposed
Chris Lattnerf33b8452008-10-04 18:33:34 +0000[diff] [blame]946 to memory, though some targets use it to distinguish between two different
947 kinds of registers). Use of this attribute is target-specific.</dd>
Chris Lattner275e6be2008-01-11 06:20:47 +0000[diff] [blame]948
Duncan Sands7d94e5a2008-10-06 08:14:18 +0000[diff] [blame]949 <dt><tt><a name="byval">byval</a></tt></dt>
Chris Lattner04c86182008-01-15 04:34:22 +0000[diff] [blame]950 <dd>This indicates that the pointer parameter should really be passed by
951 value to the function. The attribute implies that a hidden copy of the
952 pointee is made between the caller and the callee, so the callee is unable
Chris Lattner6a9f3c42008-08-05 18:21:08 +0000[diff] [blame]953 to modify the value in the callee. This attribute is only valid on LLVM
Chris Lattner04c86182008-01-15 04:34:22 +0000[diff] [blame]954 pointer arguments. It is generally used to pass structs and arrays by
Duncan Sands7d94e5a2008-10-06 08:14:18 +0000[diff] [blame]955 value, but is also valid on pointers to scalars. The copy is considered to
956 belong to the caller not the callee (for example,
957 <tt><a href="#readonly">readonly</a></tt> functions should not write to
Devang Patelac2fc272008-10-06 18:50:38 +0000[diff] [blame]958 <tt>byval</tt> parameters). This is not a valid attribute for return
Chris Lattnere39c5bf2009-02-05 05:42:28 +0000[diff] [blame]959 values. The byval attribute also supports specifying an alignment with the
960 align attribute. This has a target-specific effect on the code generator
961 that usually indicates a desired alignment for the synthesized stack
962 slot.</dd>
Chris Lattner275e6be2008-01-11 06:20:47 +0000[diff] [blame]963
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]964 <dt><tt>sret</tt></dt>
Duncan Sands616cc032008-02-18 04:19:38 +0000[diff] [blame]965 <dd>This indicates that the pointer parameter specifies the address of a
966 structure that is the return value of the function in the source program.
Chris Lattnerf33b8452008-10-04 18:33:34 +0000[diff] [blame]967 This pointer must be guaranteed by the caller to be valid: loads and stores
968 to the structure may be assumed by the callee to not to trap. This may only
Devang Patelac2fc272008-10-06 18:50:38 +0000[diff] [blame]969 be applied to the first parameter. This is not a valid attribute for
970 return values. </dd>
Chris Lattner275e6be2008-01-11 06:20:47 +0000[diff] [blame]971
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]972 <dt><tt>noalias</tt></dt>
Nick Lewyckyff384472008-11-24 03:41:24 +0000[diff] [blame]973 <dd>This indicates that the pointer does not alias any global or any other
974 parameter. The caller is responsible for ensuring that this is the
Nick Lewyckya604a942008-11-24 05:00:44 +0000[diff] [blame]975 case. On a function return value, <tt>noalias</tt> additionally indicates
976 that the pointer does not alias any other pointers visible to the
Nick Lewycky40fb6f22008-12-19 06:39:12 +0000[diff] [blame]977 caller. For further details, please see the discussion of the NoAlias
978 response in
979 <a href="http://llvm.org/docs/AliasAnalysis.html#MustMayNo">alias
980 analysis</a>.</dd>
981
982 <dt><tt>nocapture</tt></dt>
983 <dd>This indicates that the callee does not make any copies of the pointer
984 that outlive the callee itself. This is not a valid attribute for return
985 values.</dd>
Chris Lattner275e6be2008-01-11 06:20:47 +0000[diff] [blame]986
Duncan Sands4ee46812007-07-27 19:57:41 +0000[diff] [blame]987 <dt><tt>nest</tt></dt>
Duncan Sandsf1a7d4c2008-07-08 09:27:25 +0000[diff] [blame]988 <dd>This indicates that the pointer parameter can be excised using the
Devang Patelac2fc272008-10-06 18:50:38 +0000[diff] [blame]989 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
990 attribute for return values.</dd>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]991 </dl>
992
993</div>
994
995<!-- ======================================================================= -->
996<div class="doc_subsection">
Gordon Henriksen13fe5e32007-12-10 03:18:06 +0000[diff] [blame]997 <a name="gc">Garbage Collector Names</a>
998</div>
999
1000<div class="doc_text">
1001<p>Each function may specify a garbage collector name, which is simply a
1002string.</p>
1003
1004<div class="doc_code"><pre
1005>define void @f() gc "name" { ...</pre></div>
1006
1007<p>The compiler declares the supported values of <i>name</i>. Specifying a
1008collector which will cause the compiler to alter its output in order to support
1009the named garbage collection algorithm.</p>
1010</div>
1011
1012<!-- ======================================================================= -->
1013<div class="doc_subsection">
Devang Patel008cd3e2008-09-26 23:51:19 +0000[diff] [blame]1014 <a name="fnattrs">Function Attributes</a>
Devang Pateld468f1c2008-09-04 23:05:13 +0000[diff] [blame]1015</div>
1016
1017<div class="doc_text">
Devang Patel008cd3e2008-09-26 23:51:19 +0000[diff] [blame]1018
1019<p>Function attributes are set to communicate additional information about
1020 a function. Function attributes are considered to be part of the function,
1021 not of the function type, so functions with different parameter attributes
1022 can have the same function type.</p>
1023
1024 <p>Function attributes are simple keywords that follow the type specified. If
1025 multiple attributes are needed, they are space separated. For
1026 example:</p>
Devang Pateld468f1c2008-09-04 23:05:13 +0000[diff] [blame]1027
1028<div class="doc_code">
Bill Wendling74d3eac2008-09-07 10:26:33 +0000[diff] [blame]1029<pre>
Devang Patel008cd3e2008-09-26 23:51:19 +0000[diff] [blame]1030define void @f() noinline { ... }
1031define void @f() alwaysinline { ... }
1032define void @f() alwaysinline optsize { ... }
1033define void @f() optsize
Bill Wendling74d3eac2008-09-07 10:26:33 +0000[diff] [blame]1034</pre>
Devang Pateld468f1c2008-09-04 23:05:13 +0000[diff] [blame]1035</div>
1036
Bill Wendling74d3eac2008-09-07 10:26:33 +0000[diff] [blame]1037<dl>
Devang Patel008cd3e2008-09-26 23:51:19 +0000[diff] [blame]1038<dt><tt>alwaysinline</tt></dt>
Chris Lattner82d70f92008-10-04 18:23:17 +0000[diff] [blame]1039<dd>This attribute indicates that the inliner should attempt to inline this
1040function into callers whenever possible, ignoring any active inlining size
1041threshold for this caller.</dd>
Bill Wendling74d3eac2008-09-07 10:26:33 +0000[diff] [blame]1042
Devang Patel008cd3e2008-09-26 23:51:19 +0000[diff] [blame]1043<dt><tt>noinline</tt></dt>
Chris Lattner82d70f92008-10-04 18:23:17 +0000[diff] [blame]1044<dd>This attribute indicates that the inliner should never inline this function
Chris Lattner377f3ae2008-10-05 17:14:59 +0000[diff] [blame]1045in any situation. This attribute may not be used together with the
Chris Lattner82d70f92008-10-04 18:23:17 +0000[diff] [blame]1046<tt>alwaysinline</tt> attribute.</dd>
Bill Wendling74d3eac2008-09-07 10:26:33 +0000[diff] [blame]1047
Devang Patel008cd3e2008-09-26 23:51:19 +0000[diff] [blame]1048<dt><tt>optsize</tt></dt>
Devang Patelc29099b2008-09-29 18:34:44 +0000[diff] [blame]1049<dd>This attribute suggests that optimization passes and code generator passes
Chris Lattner82d70f92008-10-04 18:23:17 +0000[diff] [blame]1050make choices that keep the code size of this function low, and otherwise do
1051optimizations specifically to reduce code size.</dd>
Bill Wendling74d3eac2008-09-07 10:26:33 +0000[diff] [blame]1052
Devang Patel008cd3e2008-09-26 23:51:19 +0000[diff] [blame]1053<dt><tt>noreturn</tt></dt>
Chris Lattner82d70f92008-10-04 18:23:17 +0000[diff] [blame]1054<dd>This function attribute indicates that the function never returns normally.
1055This produces undefined behavior at runtime if the function ever does
1056dynamically return.</dd>
Devang Patel008cd3e2008-09-26 23:51:19 +0000[diff] [blame]1057
1058<dt><tt>nounwind</tt></dt>
Chris Lattner82d70f92008-10-04 18:23:17 +0000[diff] [blame]1059<dd>This function attribute indicates that the function never returns with an
1060unwind or exceptional control flow. If the function does unwind, its runtime
1061behavior is undefined.</dd>
1062
1063<dt><tt>readnone</tt></dt>
Duncan Sands7d94e5a2008-10-06 08:14:18 +0000[diff] [blame]1064<dd>This attribute indicates that the function computes its result (or the
1065exception it throws) based strictly on its arguments, without dereferencing any
1066pointer arguments or otherwise accessing any mutable state (e.g. memory, control
1067registers, etc) visible to caller functions. It does not write through any
1068pointer arguments (including <tt><a href="#byval">byval</a></tt> arguments) and
1069never changes any state visible to callers.</dd>
Devang Patel008cd3e2008-09-26 23:51:19 +0000[diff] [blame]1070
Duncan Sands7d94e5a2008-10-06 08:14:18 +0000[diff] [blame]1071<dt><tt><a name="readonly">readonly</a></tt></dt>
1072<dd>This attribute indicates that the function does not write through any
1073pointer arguments (including <tt><a href="#byval">byval</a></tt> arguments)
1074or otherwise modify any state (e.g. memory, control registers, etc) visible to
1075caller functions. It may dereference pointer arguments and read state that may
1076be set in the caller. A readonly function always returns the same value (or
1077throws the same exception) when called with the same set of arguments and global
1078state.</dd>
Bill Wendlingdfaabba2008-11-13 01:02:51 +0000[diff] [blame]1079
1080<dt><tt><a name="ssp">ssp</a></tt></dt>
Bill Wendling34f70d82008-11-26 19:19:05 +0000[diff] [blame]1081<dd>This attribute indicates that the function should emit a stack smashing
Bill Wendlingdfaabba2008-11-13 01:02:51 +0000[diff] [blame]1082protector. It is in the form of a "canary"&mdash;a random value placed on the
1083stack before the local variables that's checked upon return from the function to
1084see if it has been overwritten. A heuristic is used to determine if a function
Bill Wendling34f70d82008-11-26 19:19:05 +0000[diff] [blame]1085needs stack protectors or not.
Bill Wendlingdfaabba2008-11-13 01:02:51 +0000[diff] [blame]1086
Bill Wendlingbe9ec3f2008-11-26 19:07:40 +0000[diff] [blame]1087<p>If a function that has an <tt>ssp</tt> attribute is inlined into a function
1088that doesn't have an <tt>ssp</tt> attribute, then the resulting function will
1089have an <tt>ssp</tt> attribute.</p></dd>
1090
1091<dt><tt>sspreq</tt></dt>
Bill Wendling34f70d82008-11-26 19:19:05 +0000[diff] [blame]1092<dd>This attribute indicates that the function should <em>always</em> emit a
Bill Wendlingdfaabba2008-11-13 01:02:51 +0000[diff] [blame]1093stack smashing protector. This overrides the <tt><a href="#ssp">ssp</a></tt>
Bill Wendling34f70d82008-11-26 19:19:05 +0000[diff] [blame]1094function attribute.
Bill Wendlingbe9ec3f2008-11-26 19:07:40 +0000[diff] [blame]1095
1096<p>If a function that has an <tt>sspreq</tt> attribute is inlined into a
1097function that doesn't have an <tt>sspreq</tt> attribute or which has
1098an <tt>ssp</tt> attribute, then the resulting function will have
1099an <tt>sspreq</tt> attribute.</p></dd>
Bill Wendling74d3eac2008-09-07 10:26:33 +0000[diff] [blame]1100</dl>
1101
Devang Pateld468f1c2008-09-04 23:05:13 +0000[diff] [blame]1102</div>
1103
1104<!-- ======================================================================= -->
1105<div class="doc_subsection">
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1106 <a name="moduleasm">Module-Level Inline Assembly</a>
1107</div>
1108
1109<div class="doc_text">
1110<p>
1111Modules may contain "module-level inline asm" blocks, which corresponds to the
1112GCC "file scope inline asm" blocks. These blocks are internally concatenated by
1113LLVM and treated as a single unit, but may be separated in the .ll file if
1114desired. The syntax is very simple:
1115</p>
1116
1117<div class="doc_code">
1118<pre>
1119module asm "inline asm code goes here"
1120module asm "more can go here"
1121</pre>
1122</div>
1123
1124<p>The strings can contain any character by escaping non-printable characters.
1125 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1126 for the number.
1127</p>
1128
1129<p>
1130 The inline asm code is simply printed to the machine code .s file when
1131 assembly code is generated.
1132</p>
1133</div>
1134
1135<!-- ======================================================================= -->
1136<div class="doc_subsection">
1137 <a name="datalayout">Data Layout</a>
1138</div>
1139
1140<div class="doc_text">
1141<p>A module may specify a target specific data layout string that specifies how
1142data is to be laid out in memory. The syntax for the data layout is simply:</p>
1143<pre> target datalayout = "<i>layout specification</i>"</pre>
1144<p>The <i>layout specification</i> consists of a list of specifications
1145separated by the minus sign character ('-'). Each specification starts with a
1146letter and may include other information after the letter to define some
1147aspect of the data layout. The specifications accepted are as follows: </p>
1148<dl>
1149 <dt><tt>E</tt></dt>
1150 <dd>Specifies that the target lays out data in big-endian form. That is, the
1151 bits with the most significance have the lowest address location.</dd>
1152 <dt><tt>e</tt></dt>
Chris Lattner96451482008-08-05 18:29:16 +0000[diff] [blame]1153 <dd>Specifies that the target lays out data in little-endian form. That is,
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1154 the bits with the least significance have the lowest address location.</dd>
1155 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1156 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1157 <i>preferred</i> alignments. All sizes are in bits. Specifying the <i>pref</i>
1158 alignment is optional. If omitted, the preceding <tt>:</tt> should be omitted
1159 too.</dd>
1160 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1161 <dd>This specifies the alignment for an integer type of a given bit
1162 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1163 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1164 <dd>This specifies the alignment for a vector type of a given bit
1165 <i>size</i>.</dd>
1166 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1167 <dd>This specifies the alignment for a floating point type of a given bit
1168 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
1169 (double).</dd>
1170 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1171 <dd>This specifies the alignment for an aggregate type of a given bit
1172 <i>size</i>.</dd>
1173</dl>
1174<p>When constructing the data layout for a given target, LLVM starts with a
1175default set of specifications which are then (possibly) overriden by the
1176specifications in the <tt>datalayout</tt> keyword. The default specifications
1177are given in this list:</p>
1178<ul>
1179 <li><tt>E</tt> - big endian</li>
1180 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
1181 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1182 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1183 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1184 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
Chris Lattner96451482008-08-05 18:29:16 +0000[diff] [blame]1185 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1186 alignment of 64-bits</li>
1187 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1188 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1189 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1190 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1191 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1192</ul>
Chris Lattner6a9f3c42008-08-05 18:21:08 +0000[diff] [blame]1193<p>When LLVM is determining the alignment for a given type, it uses the
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]1194following rules:</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1195<ol>
1196 <li>If the type sought is an exact match for one of the specifications, that
1197 specification is used.</li>
1198 <li>If no match is found, and the type sought is an integer type, then the
1199 smallest integer type that is larger than the bitwidth of the sought type is
1200 used. If none of the specifications are larger than the bitwidth then the the
1201 largest integer type is used. For example, given the default specifications
1202 above, the i7 type will use the alignment of i8 (next largest) while both
1203 i65 and i256 will use the alignment of i64 (largest specified).</li>
1204 <li>If no match is found, and the type sought is a vector type, then the
1205 largest vector type that is smaller than the sought vector type will be used
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]1206 as a fall back. This happens because &lt;128 x double&gt; can be implemented
1207 in terms of 64 &lt;2 x double&gt;, for example.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1208</ol>
1209</div>
1210
1211<!-- *********************************************************************** -->
1212<div class="doc_section"> <a name="typesystem">Type System</a> </div>
1213<!-- *********************************************************************** -->
1214
1215<div class="doc_text">
1216
1217<p>The LLVM type system is one of the most important features of the
1218intermediate representation. Being typed enables a number of
Chris Lattner96451482008-08-05 18:29:16 +0000[diff] [blame]1219optimizations to be performed on the intermediate representation directly,
1220without having to do
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1221extra analyses on the side before the transformation. A strong type
1222system makes it easier to read the generated code and enables novel
1223analyses and transformations that are not feasible to perform on normal
1224three address code representations.</p>
1225
1226</div>
1227
1228<!-- ======================================================================= -->
Chris Lattner488772f2008-01-04 04:32:38 +0000[diff] [blame]1229<div class="doc_subsection"> <a name="t_classifications">Type
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1230Classifications</a> </div>
1231<div class="doc_text">
Chris Lattner488772f2008-01-04 04:32:38 +0000[diff] [blame]1232<p>The types fall into a few useful
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1233classifications:</p>
1234
1235<table border="1" cellspacing="0" cellpadding="4">
1236 <tbody>
1237 <tr><th>Classification</th><th>Types</th></tr>
1238 <tr>
Chris Lattner488772f2008-01-04 04:32:38 +0000[diff] [blame]1239 <td><a href="#t_integer">integer</a></td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1240 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1241 </tr>
1242 <tr>
Chris Lattner488772f2008-01-04 04:32:38 +0000[diff] [blame]1243 <td><a href="#t_floating">floating point</a></td>
1244 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1245 </tr>
1246 <tr>
1247 <td><a name="t_firstclass">first class</a></td>
Chris Lattner488772f2008-01-04 04:32:38 +0000[diff] [blame]1248 <td><a href="#t_integer">integer</a>,
1249 <a href="#t_floating">floating point</a>,
1250 <a href="#t_pointer">pointer</a>,
Dan Gohmanf6237db2008-06-18 18:42:13 +0000[diff] [blame]1251 <a href="#t_vector">vector</a>,
Dan Gohman74d6faf2008-05-12 23:51:09 +0000[diff] [blame]1252 <a href="#t_struct">structure</a>,
1253 <a href="#t_array">array</a>,
Dan Gohmand13951e2008-05-23 22:50:26 +0000[diff] [blame]1254 <a href="#t_label">label</a>.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1255 </td>
1256 </tr>
Chris Lattner488772f2008-01-04 04:32:38 +0000[diff] [blame]1257 <tr>
1258 <td><a href="#t_primitive">primitive</a></td>
1259 <td><a href="#t_label">label</a>,
1260 <a href="#t_void">void</a>,
Chris Lattner488772f2008-01-04 04:32:38 +0000[diff] [blame]1261 <a href="#t_floating">floating point</a>.</td>
1262 </tr>
1263 <tr>
1264 <td><a href="#t_derived">derived</a></td>
1265 <td><a href="#t_integer">integer</a>,
1266 <a href="#t_array">array</a>,
1267 <a href="#t_function">function</a>,
1268 <a href="#t_pointer">pointer</a>,
1269 <a href="#t_struct">structure</a>,
1270 <a href="#t_pstruct">packed structure</a>,
1271 <a href="#t_vector">vector</a>,
1272 <a href="#t_opaque">opaque</a>.
Dan Gohman032ba852008-10-14 16:32:04 +0000[diff] [blame]1273 </td>
Chris Lattner488772f2008-01-04 04:32:38 +0000[diff] [blame]1274 </tr>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1275 </tbody>
1276</table>
1277
1278<p>The <a href="#t_firstclass">first class</a> types are perhaps the
1279most important. Values of these types are the only ones which can be
1280produced by instructions, passed as arguments, or used as operands to
Dan Gohman4f29e422008-05-23 21:53:15 +0000[diff] [blame]1281instructions.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1282</div>
1283
1284<!-- ======================================================================= -->
Chris Lattner488772f2008-01-04 04:32:38 +0000[diff] [blame]1285<div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
Chris Lattner86437612008-01-04 04:34:14 +0000[diff] [blame]1286
Chris Lattner488772f2008-01-04 04:32:38 +0000[diff] [blame]1287<div class="doc_text">
1288<p>The primitive types are the fundamental building blocks of the LLVM
1289system.</p>
1290
Chris Lattner86437612008-01-04 04:34:14 +0000[diff] [blame]1291</div>
1292
Chris Lattner488772f2008-01-04 04:32:38 +0000[diff] [blame]1293<!-- _______________________________________________________________________ -->
1294<div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1295
1296<div class="doc_text">
1297 <table>
1298 <tbody>
1299 <tr><th>Type</th><th>Description</th></tr>
1300 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1301 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1302 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1303 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1304 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1305 </tbody>
1306 </table>
1307</div>
1308
1309<!-- _______________________________________________________________________ -->
1310<div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1311
1312<div class="doc_text">
1313<h5>Overview:</h5>
1314<p>The void type does not represent any value and has no size.</p>
1315
1316<h5>Syntax:</h5>
1317
1318<pre>
1319 void
1320</pre>
1321</div>
1322
1323<!-- _______________________________________________________________________ -->
1324<div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1325
1326<div class="doc_text">
1327<h5>Overview:</h5>
1328<p>The label type represents code labels.</p>
1329
1330<h5>Syntax:</h5>
1331
1332<pre>
1333 label
1334</pre>
1335</div>
1336
1337
1338<!-- ======================================================================= -->
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1339<div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1340
1341<div class="doc_text">
1342
1343<p>The real power in LLVM comes from the derived types in the system.
1344This is what allows a programmer to represent arrays, functions,
1345pointers, and other useful types. Note that these derived types may be
1346recursive: For example, it is possible to have a two dimensional array.</p>
1347
1348</div>
1349
1350<!-- _______________________________________________________________________ -->
1351<div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1352
1353<div class="doc_text">
1354
1355<h5>Overview:</h5>
1356<p>The integer type is a very simple derived type that simply specifies an
1357arbitrary bit width for the integer type desired. Any bit width from 1 bit to
13582^23-1 (about 8 million) can be specified.</p>
1359
1360<h5>Syntax:</h5>
1361
1362<pre>
1363 iN
1364</pre>
1365
1366<p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1367value.</p>
1368
1369<h5>Examples:</h5>
1370<table class="layout">
Chris Lattner251ab812007-12-18 06:18:21 +0000[diff] [blame]1371 <tbody>
1372 <tr>
1373 <td><tt>i1</tt></td>
1374 <td>a single-bit integer.</td>
1375 </tr><tr>
1376 <td><tt>i32</tt></td>
1377 <td>a 32-bit integer.</td>
1378 </tr><tr>
1379 <td><tt>i1942652</tt></td>
1380 <td>a really big integer of over 1 million bits.</td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1381 </tr>
Chris Lattner251ab812007-12-18 06:18:21 +0000[diff] [blame]1382 </tbody>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1383</table>
djge93155c2009-01-24 15:58:40 +0000[diff] [blame]1384
1385<p>Note that the code generator does not yet support large integer types
1386to be used as function return types. The specific limit on how large a
1387return type the code generator can currently handle is target-dependent;
1388currently it's often 64 bits for 32-bit targets and 128 bits for 64-bit
1389targets.</p>
1390
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1391</div>
1392
1393<!-- _______________________________________________________________________ -->
1394<div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1395
1396<div class="doc_text">
1397
1398<h5>Overview:</h5>
1399
1400<p>The array type is a very simple derived type that arranges elements
1401sequentially in memory. The array type requires a size (number of
1402elements) and an underlying data type.</p>
1403
1404<h5>Syntax:</h5>
1405
1406<pre>
1407 [&lt;# elements&gt; x &lt;elementtype&gt;]
1408</pre>
1409
1410<p>The number of elements is a constant integer value; elementtype may
1411be any type with a size.</p>
1412
1413<h5>Examples:</h5>
1414<table class="layout">
1415 <tr class="layout">
Chris Lattner7311d222007-12-19 05:04:11 +0000[diff] [blame]1416 <td class="left"><tt>[40 x i32]</tt></td>
1417 <td class="left">Array of 40 32-bit integer values.</td>
1418 </tr>
1419 <tr class="layout">
1420 <td class="left"><tt>[41 x i32]</tt></td>
1421 <td class="left">Array of 41 32-bit integer values.</td>
1422 </tr>
1423 <tr class="layout">
1424 <td class="left"><tt>[4 x i8]</tt></td>
1425 <td class="left">Array of 4 8-bit integer values.</td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1426 </tr>
1427</table>
1428<p>Here are some examples of multidimensional arrays:</p>
1429<table class="layout">
1430 <tr class="layout">
Chris Lattner7311d222007-12-19 05:04:11 +0000[diff] [blame]1431 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1432 <td class="left">3x4 array of 32-bit integer values.</td>
1433 </tr>
1434 <tr class="layout">
1435 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1436 <td class="left">12x10 array of single precision floating point values.</td>
1437 </tr>
1438 <tr class="layout">
1439 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1440 <td class="left">2x3x4 array of 16-bit integer values.</td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1441 </tr>
1442</table>
1443
1444<p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1445length array. Normally, accesses past the end of an array are undefined in
1446LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
1447As a special case, however, zero length arrays are recognized to be variable
1448length. This allows implementation of 'pascal style arrays' with the LLVM
1449type "{ i32, [0 x float]}", for example.</p>
1450
djge93155c2009-01-24 15:58:40 +0000[diff] [blame]1451<p>Note that the code generator does not yet support large aggregate types
1452to be used as function return types. The specific limit on how large an
1453aggregate return type the code generator can currently handle is
1454target-dependent, and also dependent on the aggregate element types.</p>
1455
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1456</div>
1457
1458<!-- _______________________________________________________________________ -->
1459<div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1460<div class="doc_text">
Chris Lattner43030e72008-04-23 04:59:35 +0000[diff] [blame]1461
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1462<h5>Overview:</h5>
Chris Lattner43030e72008-04-23 04:59:35 +0000[diff] [blame]1463
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1464<p>The function type can be thought of as a function signature. It
Devang Pateld4ba41d2008-03-24 05:35:41 +0000[diff] [blame]1465consists of a return type and a list of formal parameter types. The
Chris Lattner43030e72008-04-23 04:59:35 +0000[diff] [blame]1466return type of a function type is a scalar type, a void type, or a struct type.
Devang Pateld5404c02008-03-24 20:52:42 +0000[diff] [blame]1467If the return type is a struct type then all struct elements must be of first
Chris Lattner43030e72008-04-23 04:59:35 +0000[diff] [blame]1468class types, and the struct must have at least one element.</p>
Devang Patela3cc5372008-03-10 20:49:15 +0000[diff] [blame]1469
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1470<h5>Syntax:</h5>
Chris Lattner43030e72008-04-23 04:59:35 +0000[diff] [blame]1471
1472<pre>
1473 &lt;returntype list&gt; (&lt;parameter list&gt;)
1474</pre>
1475
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1476<p>...where '<tt>&lt;parameter list&gt;</tt>' is a comma-separated list of type
1477specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1478which indicates that the function takes a variable number of arguments.
1479Variable argument functions can access their arguments with the <a
Devang Patela3cc5372008-03-10 20:49:15 +0000[diff] [blame]1480 href="#int_varargs">variable argument handling intrinsic</a> functions.
1481'<tt>&lt;returntype list&gt;</tt>' is a comma-separated list of
1482<a href="#t_firstclass">first class</a> type specifiers.</p>
Chris Lattner43030e72008-04-23 04:59:35 +0000[diff] [blame]1483
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1484<h5>Examples:</h5>
1485<table class="layout">
1486 <tr class="layout">
1487 <td class="left"><tt>i32 (i32)</tt></td>
1488 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1489 </td>
1490 </tr><tr class="layout">
Reid Spencerf234bed2007-07-19 23:13:04 +0000[diff] [blame]1491 <td class="left"><tt>float&nbsp;(i16&nbsp;signext,&nbsp;i32&nbsp;*)&nbsp;*
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1492 </tt></td>
1493 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1494 an <tt>i16</tt> that should be sign extended and a
1495 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1496 <tt>float</tt>.
1497 </td>
1498 </tr><tr class="layout">
1499 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1500 <td class="left">A vararg function that takes at least one
1501 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1502 which returns an integer. This is the signature for <tt>printf</tt> in
1503 LLVM.
1504 </td>
Devang Pateld4ba41d2008-03-24 05:35:41 +0000[diff] [blame]1505 </tr><tr class="layout">
1506 <td class="left"><tt>{i32, i32} (i32)</tt></td>
Misha Brukmanafc88b02008-11-27 06:41:20 +0000[diff] [blame]1507 <td class="left">A function taking an <tt>i32</tt>, returning two
1508 <tt>i32</tt> values as an aggregate of type <tt>{ i32, i32 }</tt>
Devang Pateld4ba41d2008-03-24 05:35:41 +0000[diff] [blame]1509 </td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1510 </tr>
1511</table>
1512
1513</div>
1514<!-- _______________________________________________________________________ -->
1515<div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1516<div class="doc_text">
1517<h5>Overview:</h5>
1518<p>The structure type is used to represent a collection of data members
1519together in memory. The packing of the field types is defined to match
1520the ABI of the underlying processor. The elements of a structure may
1521be any type that has a size.</p>
1522<p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1523and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1524field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1525instruction.</p>
1526<h5>Syntax:</h5>
1527<pre> { &lt;type list&gt; }<br></pre>
1528<h5>Examples:</h5>
1529<table class="layout">
1530 <tr class="layout">
1531 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1532 <td class="left">A triple of three <tt>i32</tt> values</td>
1533 </tr><tr class="layout">
1534 <td class="left"><tt>{&nbsp;float,&nbsp;i32&nbsp;(i32)&nbsp;*&nbsp;}</tt></td>
1535 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1536 second element is a <a href="#t_pointer">pointer</a> to a
1537 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1538 an <tt>i32</tt>.</td>
1539 </tr>
1540</table>
djge93155c2009-01-24 15:58:40 +0000[diff] [blame]1541
1542<p>Note that the code generator does not yet support large aggregate types
1543to be used as function return types. The specific limit on how large an
1544aggregate return type the code generator can currently handle is
1545target-dependent, and also dependent on the aggregate element types.</p>
1546
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1547</div>
1548
1549<!-- _______________________________________________________________________ -->
1550<div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1551</div>
1552<div class="doc_text">
1553<h5>Overview:</h5>
1554<p>The packed structure type is used to represent a collection of data members
1555together in memory. There is no padding between fields. Further, the alignment
1556of a packed structure is 1 byte. The elements of a packed structure may
1557be any type that has a size.</p>
1558<p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1559and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1560field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1561instruction.</p>
1562<h5>Syntax:</h5>
1563<pre> &lt; { &lt;type list&gt; } &gt; <br></pre>
1564<h5>Examples:</h5>
1565<table class="layout">
1566 <tr class="layout">
1567 <td class="left"><tt>&lt; { i32, i32, i32 } &gt;</tt></td>
1568 <td class="left">A triple of three <tt>i32</tt> values</td>
1569 </tr><tr class="layout">
Bill Wendling74d3eac2008-09-07 10:26:33 +0000[diff] [blame]1570 <td class="left">
1571<tt>&lt;&nbsp;{&nbsp;float,&nbsp;i32&nbsp;(i32)*&nbsp;}&nbsp;&gt;</tt></td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1572 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1573 second element is a <a href="#t_pointer">pointer</a> to a
1574 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1575 an <tt>i32</tt>.</td>
1576 </tr>
1577</table>
1578</div>
1579
1580<!-- _______________________________________________________________________ -->
1581<div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1582<div class="doc_text">
1583<h5>Overview:</h5>
1584<p>As in many languages, the pointer type represents a pointer or
Christopher Lambdd0049d2007-12-11 09:31:00 +0000[diff] [blame]1585reference to another object, which must live in memory. Pointer types may have
1586an optional address space attribute defining the target-specific numbered
1587address space where the pointed-to object resides. The default address space is
1588zero.</p>
Chris Lattner96edbd32009-02-08 19:53:29 +0000[diff] [blame]1589
1590<p>Note that LLVM does not permit pointers to void (<tt>void*</tt>) nor does
Chris Lattner0d879e02009-02-08 22:21:28 +0000[diff] [blame]1591it permit pointers to labels (<tt>label*</tt>). Use <tt>i8*</tt> instead.</p>
Chris Lattner96edbd32009-02-08 19:53:29 +0000[diff] [blame]1592
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1593<h5>Syntax:</h5>
1594<pre> &lt;type&gt; *<br></pre>
1595<h5>Examples:</h5>
1596<table class="layout">
1597 <tr class="layout">
Dan Gohman01852382009-01-04 23:44:43 +0000[diff] [blame]1598 <td class="left"><tt>[4 x i32]*</tt></td>
Chris Lattner7311d222007-12-19 05:04:11 +0000[diff] [blame]1599 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1600 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1601 </tr>
1602 <tr class="layout">
1603 <td class="left"><tt>i32 (i32 *) *</tt></td>
1604 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1605 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
Chris Lattner7311d222007-12-19 05:04:11 +0000[diff] [blame]1606 <tt>i32</tt>.</td>
1607 </tr>
1608 <tr class="layout">
1609 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1610 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1611 that resides in address space #5.</td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1612 </tr>
1613</table>
1614</div>
1615
1616<!-- _______________________________________________________________________ -->
1617<div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1618<div class="doc_text">
1619
1620<h5>Overview:</h5>
1621
1622<p>A vector type is a simple derived type that represents a vector
1623of elements. Vector types are used when multiple primitive data
1624are operated in parallel using a single instruction (SIMD).
1625A vector type requires a size (number of
1626elements) and an underlying primitive data type. Vectors must have a power
1627of two length (1, 2, 4, 8, 16 ...). Vector types are
1628considered <a href="#t_firstclass">first class</a>.</p>
1629
1630<h5>Syntax:</h5>
1631
1632<pre>
1633 &lt; &lt;# elements&gt; x &lt;elementtype&gt; &gt;
1634</pre>
1635
1636<p>The number of elements is a constant integer value; elementtype may
1637be any integer or floating point type.</p>
1638
1639<h5>Examples:</h5>
1640
1641<table class="layout">
1642 <tr class="layout">
Chris Lattner7311d222007-12-19 05:04:11 +0000[diff] [blame]1643 <td class="left"><tt>&lt;4 x i32&gt;</tt></td>
1644 <td class="left">Vector of 4 32-bit integer values.</td>
1645 </tr>
1646 <tr class="layout">
1647 <td class="left"><tt>&lt;8 x float&gt;</tt></td>
1648 <td class="left">Vector of 8 32-bit floating-point values.</td>
1649 </tr>
1650 <tr class="layout">
1651 <td class="left"><tt>&lt;2 x i64&gt;</tt></td>
1652 <td class="left">Vector of 2 64-bit integer values.</td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1653 </tr>
1654</table>
djge93155c2009-01-24 15:58:40 +0000[diff] [blame]1655
1656<p>Note that the code generator does not yet support large vector types
1657to be used as function return types. The specific limit on how large a
1658vector return type codegen can currently handle is target-dependent;
1659currently it's often a few times longer than a hardware vector register.</p>
1660
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1661</div>
1662
1663<!-- _______________________________________________________________________ -->
1664<div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1665<div class="doc_text">
1666
1667<h5>Overview:</h5>
1668
1669<p>Opaque types are used to represent unknown types in the system. This
Gordon Henriksenda0706e2007-10-14 00:34:53 +0000[diff] [blame]1670corresponds (for example) to the C notion of a forward declared structure type.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1671In LLVM, opaque types can eventually be resolved to any type (not just a
1672structure type).</p>
1673
1674<h5>Syntax:</h5>
1675
1676<pre>
1677 opaque
1678</pre>
1679
1680<h5>Examples:</h5>
1681
1682<table class="layout">
1683 <tr class="layout">
Chris Lattner7311d222007-12-19 05:04:11 +0000[diff] [blame]1684 <td class="left"><tt>opaque</tt></td>
1685 <td class="left">An opaque type.</td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1686 </tr>
1687</table>
1688</div>
1689
Chris Lattner515195a2009-02-02 07:32:36 +0000[diff] [blame]1690<!-- ======================================================================= -->
1691<div class="doc_subsection">
1692 <a name="t_uprefs">Type Up-references</a>
1693</div>
1694
1695<div class="doc_text">
1696<h5>Overview:</h5>
1697<p>
1698An "up reference" allows you to refer to a lexically enclosing type without
1699requiring it to have a name. For instance, a structure declaration may contain a
1700pointer to any of the types it is lexically a member of. Example of up
1701references (with their equivalent as named type declarations) include:</p>
1702
1703<pre>
Chris Lattner5ad632d2009-02-09 10:00:56 +0000[diff] [blame]1704 { \2 * } %x = type { %x* }
Chris Lattner515195a2009-02-02 07:32:36 +0000[diff] [blame]1705 { \2 }* %y = type { %y }*
1706 \1* %z = type %z*
1707</pre>
1708
1709<p>
1710An up reference is needed by the asmprinter for printing out cyclic types when
1711there is no declared name for a type in the cycle. Because the asmprinter does
1712not want to print out an infinite type string, it needs a syntax to handle
1713recursive types that have no names (all names are optional in llvm IR).
1714</p>
1715
1716<h5>Syntax:</h5>
1717<pre>
1718 \&lt;level&gt;
1719</pre>
1720
1721<p>
1722The level is the count of the lexical type that is being referred to.
1723</p>
1724
1725<h5>Examples:</h5>
1726
1727<table class="layout">
1728 <tr class="layout">
1729 <td class="left"><tt>\1*</tt></td>
1730 <td class="left">Self-referential pointer.</td>
1731 </tr>
1732 <tr class="layout">
1733 <td class="left"><tt>{ { \3*, i8 }, i32 }</tt></td>
1734 <td class="left">Recursive structure where the upref refers to the out-most
1735 structure.</td>
1736 </tr>
1737</table>
1738</div>
1739
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1740
1741<!-- *********************************************************************** -->
1742<div class="doc_section"> <a name="constants">Constants</a> </div>
1743<!-- *********************************************************************** -->
1744
1745<div class="doc_text">
1746
1747<p>LLVM has several different basic types of constants. This section describes
1748them all and their syntax.</p>
1749
1750</div>
1751
1752<!-- ======================================================================= -->
1753<div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1754
1755<div class="doc_text">
1756
1757<dl>
1758 <dt><b>Boolean constants</b></dt>
1759
1760 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1761 constants of the <tt><a href="#t_primitive">i1</a></tt> type.
1762 </dd>
1763
1764 <dt><b>Integer constants</b></dt>
1765
1766 <dd>Standard integers (such as '4') are constants of the <a
1767 href="#t_integer">integer</a> type. Negative numbers may be used with
1768 integer types.
1769 </dd>
1770
1771 <dt><b>Floating point constants</b></dt>
1772
1773 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1774 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
Chris Lattnerdc15b1d2008-04-01 18:45:27 +0000[diff] [blame]1775 notation (see below). The assembler requires the exact decimal value of
1776 a floating-point constant. For example, the assembler accepts 1.25 but
1777 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
1778 constants must have a <a href="#t_floating">floating point</a> type. </dd>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1779
1780 <dt><b>Null pointer constants</b></dt>
1781
1782 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1783 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1784
1785</dl>
1786
Dale Johannesenf82a52f2009-02-11 22:14:51 +0000[diff] [blame]1787<p>The one non-intuitive notation for constants is the hexadecimal form
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1788of floating point constants. For example, the form '<tt>double
17890x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
17904.5e+15</tt>'. The only time hexadecimal floating point constants are required
1791(and the only time that they are generated by the disassembler) is when a
1792floating point constant must be emitted but it cannot be represented as a
Dale Johannesenf82a52f2009-02-11 22:14:51 +0000[diff] [blame]1793decimal floating point number in a reasonable number of digits. For example,
1794NaN's, infinities, and other
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1795special values are represented in their IEEE hexadecimal format so that
1796assembly and disassembly do not cause any bits to change in the constants.</p>
Dale Johannesenf82a52f2009-02-11 22:14:51 +0000[diff] [blame]1797<p>When using the hexadecimal form, constants of types float and double are
1798represented using the 16-digit form shown above (which matches the IEEE754
1799representation for double); float values must, however, be exactly representable
1800as IEE754 single precision.
1801Hexadecimal format is always used for long
1802double, and there are three forms of long double. The 80-bit
1803format used by x86 is represented as <tt>0xK</tt>
1804followed by 20 hexadecimal digits.
1805The 128-bit format used by PowerPC (two adjacent doubles) is represented
1806by <tt>0xM</tt> followed by 32 hexadecimal digits. The IEEE 128-bit
1807format is represented
1808by <tt>0xL</tt> followed by 32 hexadecimal digits; no currently supported
1809target uses this format. Long doubles will only work if they match
1810the long double format on your target. All hexadecimal formats are big-endian
1811(sign bit at the left).</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1812</div>
1813
1814<!-- ======================================================================= -->
Chris Lattner97063852009-02-28 18:32:25 +0000[diff] [blame]1815<div class="doc_subsection">
1816<a name="aggregateconstants"> <!-- old anchor -->
1817<a name="complexconstants">Complex Constants</a></a>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1818</div>
1819
1820<div class="doc_text">
Chris Lattner97063852009-02-28 18:32:25 +0000[diff] [blame]1821<p>Complex constants are a (potentially recursive) combination of simple
1822constants and smaller complex constants.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1823
1824<dl>
1825 <dt><b>Structure constants</b></dt>
1826
1827 <dd>Structure constants are represented with notation similar to structure
1828 type definitions (a comma separated list of elements, surrounded by braces
Chris Lattner6c8de962007-12-25 20:34:52 +0000[diff] [blame]1829 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
1830 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>". Structure constants
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1831 must have <a href="#t_struct">structure type</a>, and the number and
1832 types of elements must match those specified by the type.
1833 </dd>
1834
1835 <dt><b>Array constants</b></dt>
1836
1837 <dd>Array constants are represented with notation similar to array type
1838 definitions (a comma separated list of elements, surrounded by square brackets
1839 (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74 ]</tt>". Array
1840 constants must have <a href="#t_array">array type</a>, and the number and
1841 types of elements must match those specified by the type.
1842 </dd>
1843
1844 <dt><b>Vector constants</b></dt>
1845
1846 <dd>Vector constants are represented with notation similar to vector type
1847 definitions (a comma separated list of elements, surrounded by
1848 less-than/greater-than's (<tt>&lt;&gt;</tt>)). For example: "<tt>&lt; i32 42,
1849 i32 11, i32 74, i32 100 &gt;</tt>". Vector constants must have <a
1850 href="#t_vector">vector type</a>, and the number and types of elements must
1851 match those specified by the type.
1852 </dd>
1853
1854 <dt><b>Zero initialization</b></dt>
1855
1856 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1857 value to zero of <em>any</em> type, including scalar and aggregate types.
1858 This is often used to avoid having to print large zero initializers (e.g. for
1859 large arrays) and is always exactly equivalent to using explicit zero
1860 initializers.
1861 </dd>
Nick Lewycky4dcf8102009-04-04 07:22:01 +0000[diff] [blame]1862
1863 <dt><b>Metadata node</b></dt>
1864
1865 <dd>A metadata node is a structure-like constant with the type of an empty
1866 struct. For example: "<tt>{ } !{ i32 0, { } !"test" }</tt>". Unlike other
1867 constants that are meant to be interpreted as part of the instruction stream,
1868 metadata is a place to attach additional information such as debug info.
1869 </dd>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1870</dl>
1871
1872</div>
1873
1874<!-- ======================================================================= -->
1875<div class="doc_subsection">
1876 <a name="globalconstants">Global Variable and Function Addresses</a>
1877</div>
1878
1879<div class="doc_text">
1880
1881<p>The addresses of <a href="#globalvars">global variables</a> and <a
1882href="#functionstructure">functions</a> are always implicitly valid (link-time)
1883constants. These constants are explicitly referenced when the <a
1884href="#identifiers">identifier for the global</a> is used and always have <a
1885href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1886file:</p>
1887
1888<div class="doc_code">
1889<pre>
1890@X = global i32 17
1891@Y = global i32 42
1892@Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1893</pre>
1894</div>
1895
1896</div>
1897
1898<!-- ======================================================================= -->
1899<div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1900<div class="doc_text">
1901 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1902 no specific value. Undefined values may be of any type and be used anywhere
1903 a constant is permitted.</p>
1904
1905 <p>Undefined values indicate to the compiler that the program is well defined
1906 no matter what value is used, giving the compiler more freedom to optimize.
1907 </p>
1908</div>
1909
1910<!-- ======================================================================= -->
1911<div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1912</div>
1913
1914<div class="doc_text">
1915
1916<p>Constant expressions are used to allow expressions involving other constants
1917to be used as constants. Constant expressions may be of any <a
1918href="#t_firstclass">first class</a> type and may involve any LLVM operation
1919that does not have side effects (e.g. load and call are not supported). The
1920following is the syntax for constant expressions:</p>
1921
1922<dl>
1923 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1924 <dd>Truncate a constant to another type. The bit size of CST must be larger
1925 than the bit size of TYPE. Both types must be integers.</dd>
1926
1927 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1928 <dd>Zero extend a constant to another type. The bit size of CST must be
1929 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1930
1931 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1932 <dd>Sign extend a constant to another type. The bit size of CST must be
1933 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1934
1935 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1936 <dd>Truncate a floating point constant to another floating point type. The
1937 size of CST must be larger than the size of TYPE. Both types must be
1938 floating point.</dd>
1939
1940 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1941 <dd>Floating point extend a constant to another type. The size of CST must be
1942 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1943
Reid Spencere6adee82007-07-31 14:40:14 +0000[diff] [blame]1944 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1945 <dd>Convert a floating point constant to the corresponding unsigned integer
Nate Begeman78246ca2007-11-17 03:58:34 +0000[diff] [blame]1946 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1947 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1948 of the same number of elements. If the value won't fit in the integer type,
1949 the results are undefined.</dd>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1950
1951 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
1952 <dd>Convert a floating point constant to the corresponding signed integer
Nate Begeman78246ca2007-11-17 03:58:34 +0000[diff] [blame]1953 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1954 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1955 of the same number of elements. If the value won't fit in the integer type,
1956 the results are undefined.</dd>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1957
1958 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
1959 <dd>Convert an unsigned integer constant to the corresponding floating point
Nate Begeman78246ca2007-11-17 03:58:34 +0000[diff] [blame]1960 constant. TYPE must be a scalar or vector floating point type. CST must be of
1961 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1962 of the same number of elements. If the value won't fit in the floating point
1963 type, the results are undefined.</dd>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1964
1965 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
1966 <dd>Convert a signed integer constant to the corresponding floating point
Nate Begeman78246ca2007-11-17 03:58:34 +0000[diff] [blame]1967 constant. TYPE must be a scalar or vector floating point type. CST must be of
1968 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1969 of the same number of elements. If the value won't fit in the floating point
1970 type, the results are undefined.</dd>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1971
1972 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
1973 <dd>Convert a pointer typed constant to the corresponding integer constant
1974 TYPE must be an integer type. CST must be of pointer type. The CST value is
1975 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1976
1977 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
1978 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1979 pointer type. CST must be of integer type. The CST value is zero extended,
1980 truncated, or unchanged to make it fit in a pointer size. This one is
1981 <i>really</i> dangerous!</dd>
1982
1983 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
Chris Lattner557bc5d2009-02-28 18:27:03 +0000[diff] [blame]1984 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
1985 are the same as those for the <a href="#i_bitcast">bitcast
1986 instruction</a>.</dd>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1987
1988 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1989
1990 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1991 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1992 instruction, the index list may have zero or more indexes, which are required
1993 to make sense for the type of "CSTPTR".</dd>
1994
1995 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1996
1997 <dd>Perform the <a href="#i_select">select operation</a> on
1998 constants.</dd>
1999
2000 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
2001 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
2002
2003 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
2004 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
2005
Nate Begeman646fa482008-05-12 19:01:56 +0000[diff] [blame]2006 <dt><b><tt>vicmp COND ( VAL1, VAL2 )</tt></b></dt>
2007 <dd>Performs the <a href="#i_vicmp">vicmp operation</a> on constants.</dd>
2008
2009 <dt><b><tt>vfcmp COND ( VAL1, VAL2 )</tt></b></dt>
2010 <dd>Performs the <a href="#i_vfcmp">vfcmp operation</a> on constants.</dd>
2011
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2012 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
2013
2014 <dd>Perform the <a href="#i_extractelement">extractelement
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]2015 operation</a> on constants.</dd>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2016
2017 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
2018
2019 <dd>Perform the <a href="#i_insertelement">insertelement
2020 operation</a> on constants.</dd>
2021
2022
2023 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
2024
2025 <dd>Perform the <a href="#i_shufflevector">shufflevector
2026 operation</a> on constants.</dd>
2027
2028 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
2029
2030 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2031 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
2032 binary</a> operations. The constraints on operands are the same as those for
2033 the corresponding instruction (e.g. no bitwise operations on floating point
2034 values are allowed).</dd>
2035</dl>
2036</div>
2037
Nick Lewycky4dcf8102009-04-04 07:22:01 +0000[diff] [blame]2038<!-- ======================================================================= -->
2039<div class="doc_subsection"><a name="metadata">Embedded Metadata</a>
2040</div>
2041
2042<div class="doc_text">
2043
2044<p>Embedded metadata provides a way to attach arbitrary data to the
2045instruction stream without affecting the behaviour of the program. There are
2046two metadata primitives, strings and nodes. All metadata has the type of an
2047empty struct and is identified in syntax by a preceding exclamation point
2048('<tt>!</tt>').
2049</p>
2050
2051<p>A metadata string is a string surrounded by double quotes. It can contain
2052any character by escaping non-printable characters with "\xx" where "xx" is
2053the two digit hex code. For example: "<tt>!"test\00"</tt>".
2054</p>
2055
2056<p>Metadata nodes are represented with notation similar to structure constants
2057(a comma separated list of elements, surrounded by braces and preceeded by an
2058exclamation point). For example: "<tt>!{ { } !"test\00", i32 10}</tt>".
2059</p>
2060
2061<p>Optimizations may rely on metadata to provide additional information about
2062the program that isn't available in the instructions, or that isn't easily
2063computable. Similarly, the code generator may expect a certain metadata format
2064to be used to express debugging information.</p>
2065</div>
2066
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2067<!-- *********************************************************************** -->
2068<div class="doc_section"> <a name="othervalues">Other Values</a> </div>
2069<!-- *********************************************************************** -->
2070
2071<!-- ======================================================================= -->
2072<div class="doc_subsection">
2073<a name="inlineasm">Inline Assembler Expressions</a>
2074</div>
2075
2076<div class="doc_text">
2077
2078<p>
2079LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
2080Module-Level Inline Assembly</a>) through the use of a special value. This
2081value represents the inline assembler as a string (containing the instructions
2082to emit), a list of operand constraints (stored as a string), and a flag that
2083indicates whether or not the inline asm expression has side effects. An example
2084inline assembler expression is:
2085</p>
2086
2087<div class="doc_code">
2088<pre>
2089i32 (i32) asm "bswap $0", "=r,r"
2090</pre>
2091</div>
2092
2093<p>
2094Inline assembler expressions may <b>only</b> be used as the callee operand of
2095a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
2096</p>
2097
2098<div class="doc_code">
2099<pre>
2100%X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
2101</pre>
2102</div>
2103
2104<p>
2105Inline asms with side effects not visible in the constraint list must be marked
2106as having side effects. This is done through the use of the
2107'<tt>sideeffect</tt>' keyword, like so:
2108</p>
2109
2110<div class="doc_code">
2111<pre>
2112call void asm sideeffect "eieio", ""()
2113</pre>
2114</div>
2115
2116<p>TODO: The format of the asm and constraints string still need to be
2117documented here. Constraints on what can be done (e.g. duplication, moving, etc
Chris Lattner1cdd64e2008-10-04 18:36:02 +0000[diff] [blame]2118need to be documented). This is probably best done by reference to another
2119document that covers inline asm from a holistic perspective.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2120</p>
2121
2122</div>
2123
2124<!-- *********************************************************************** -->
2125<div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
2126<!-- *********************************************************************** -->
2127
2128<div class="doc_text">
2129
2130<p>The LLVM instruction set consists of several different
2131classifications of instructions: <a href="#terminators">terminator
2132instructions</a>, <a href="#binaryops">binary instructions</a>,
2133<a href="#bitwiseops">bitwise binary instructions</a>, <a
2134 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
2135instructions</a>.</p>
2136
2137</div>
2138
2139<!-- ======================================================================= -->
2140<div class="doc_subsection"> <a name="terminators">Terminator
2141Instructions</a> </div>
2142
2143<div class="doc_text">
2144
2145<p>As mentioned <a href="#functionstructure">previously</a>, every
2146basic block in a program ends with a "Terminator" instruction, which
2147indicates which block should be executed after the current block is
2148finished. These terminator instructions typically yield a '<tt>void</tt>'
2149value: they produce control flow, not values (the one exception being
2150the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
2151<p>There are six different terminator instructions: the '<a
2152 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
2153instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
2154the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
2155 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
2156 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
2157
2158</div>
2159
2160<!-- _______________________________________________________________________ -->
2161<div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
2162Instruction</a> </div>
2163<div class="doc_text">
2164<h5>Syntax:</h5>
Dan Gohman3e700032008-10-04 19:00:07 +0000[diff] [blame]2165<pre>
2166 ret &lt;type&gt; &lt;value&gt; <i>; Return a value from a non-void function</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2167 ret void <i>; Return from void function</i>
2168</pre>
Chris Lattner43030e72008-04-23 04:59:35 +0000[diff] [blame]2169
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2170<h5>Overview:</h5>
Chris Lattner43030e72008-04-23 04:59:35 +0000[diff] [blame]2171
Dan Gohman3e700032008-10-04 19:00:07 +0000[diff] [blame]2172<p>The '<tt>ret</tt>' instruction is used to return control flow (and
2173optionally a value) from a function back to the caller.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2174<p>There are two forms of the '<tt>ret</tt>' instruction: one that
Dan Gohman3e700032008-10-04 19:00:07 +0000[diff] [blame]2175returns a value and then causes control flow, and one that just causes
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2176control flow to occur.</p>
Chris Lattner43030e72008-04-23 04:59:35 +0000[diff] [blame]2177
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2178<h5>Arguments:</h5>
Chris Lattner43030e72008-04-23 04:59:35 +0000[diff] [blame]2179
Dan Gohman3e700032008-10-04 19:00:07 +0000[diff] [blame]2180<p>The '<tt>ret</tt>' instruction optionally accepts a single argument,
2181the return value. The type of the return value must be a
2182'<a href="#t_firstclass">first class</a>' type.</p>
2183
2184<p>A function is not <a href="#wellformed">well formed</a> if
2185it it has a non-void return type and contains a '<tt>ret</tt>'
2186instruction with no return value or a return value with a type that
2187does not match its type, or if it has a void return type and contains
2188a '<tt>ret</tt>' instruction with a return value.</p>
Chris Lattner43030e72008-04-23 04:59:35 +0000[diff] [blame]2189
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2190<h5>Semantics:</h5>
Chris Lattner43030e72008-04-23 04:59:35 +0000[diff] [blame]2191
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2192<p>When the '<tt>ret</tt>' instruction is executed, control flow
2193returns back to the calling function's context. If the caller is a "<a
2194 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
2195the instruction after the call. If the caller was an "<a
2196 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
2197at the beginning of the "normal" destination block. If the instruction
2198returns a value, that value shall set the call or invoke instruction's
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]2199return value.</p>
Chris Lattner43030e72008-04-23 04:59:35 +0000[diff] [blame]2200
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2201<h5>Example:</h5>
Chris Lattner43030e72008-04-23 04:59:35 +0000[diff] [blame]2202
2203<pre>
2204 ret i32 5 <i>; Return an integer value of 5</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2205 ret void <i>; Return from a void function</i>
Bill Wendlingd163e2d2009-02-28 22:12:54 +0000[diff] [blame]2206 ret { i32, i8 } { i32 4, i8 2 } <i>; Return a struct of values 4 and 2</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2207</pre>
Dan Gohman60967192009-01-12 23:12:39 +0000[diff] [blame]2208
djge93155c2009-01-24 15:58:40 +0000[diff] [blame]2209<p>Note that the code generator does not yet fully support large
2210 return values. The specific sizes that are currently supported are
2211 dependent on the target. For integers, on 32-bit targets the limit
2212 is often 64 bits, and on 64-bit targets the limit is often 128 bits.
2213 For aggregate types, the current limits are dependent on the element
2214 types; for example targets are often limited to 2 total integer
2215 elements and 2 total floating-point elements.</p>
Dan Gohman60967192009-01-12 23:12:39 +0000[diff] [blame]2216
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2217</div>
2218<!-- _______________________________________________________________________ -->
2219<div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
2220<div class="doc_text">
2221<h5>Syntax:</h5>
2222<pre> br i1 &lt;cond&gt;, label &lt;iftrue&gt;, label &lt;iffalse&gt;<br> br label &lt;dest&gt; <i>; Unconditional branch</i>
2223</pre>
2224<h5>Overview:</h5>
2225<p>The '<tt>br</tt>' instruction is used to cause control flow to
2226transfer to a different basic block in the current function. There are
2227two forms of this instruction, corresponding to a conditional branch
2228and an unconditional branch.</p>
2229<h5>Arguments:</h5>
2230<p>The conditional branch form of the '<tt>br</tt>' instruction takes a
2231single '<tt>i1</tt>' value and two '<tt>label</tt>' values. The
2232unconditional form of the '<tt>br</tt>' instruction takes a single
2233'<tt>label</tt>' value as a target.</p>
2234<h5>Semantics:</h5>
2235<p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
2236argument is evaluated. If the value is <tt>true</tt>, control flows
2237to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
2238control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
2239<h5>Example:</h5>
2240<pre>Test:<br> %cond = <a href="#i_icmp">icmp</a> eq, i32 %a, %b<br> br i1 %cond, label %IfEqual, label %IfUnequal<br>IfEqual:<br> <a
2241 href="#i_ret">ret</a> i32 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> i32 0<br></pre>
2242</div>
2243<!-- _______________________________________________________________________ -->
2244<div class="doc_subsubsection">
2245 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
2246</div>
2247
2248<div class="doc_text">
2249<h5>Syntax:</h5>
2250
2251<pre>
2252 switch &lt;intty&gt; &lt;value&gt;, label &lt;defaultdest&gt; [ &lt;intty&gt; &lt;val&gt;, label &lt;dest&gt; ... ]
2253</pre>
2254
2255<h5>Overview:</h5>
2256
2257<p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
2258several different places. It is a generalization of the '<tt>br</tt>'
2259instruction, allowing a branch to occur to one of many possible
2260destinations.</p>
2261
2262
2263<h5>Arguments:</h5>
2264
2265<p>The '<tt>switch</tt>' instruction uses three parameters: an integer
2266comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
2267an array of pairs of comparison value constants and '<tt>label</tt>'s. The
2268table is not allowed to contain duplicate constant entries.</p>
2269
2270<h5>Semantics:</h5>
2271
2272<p>The <tt>switch</tt> instruction specifies a table of values and
2273destinations. When the '<tt>switch</tt>' instruction is executed, this
2274table is searched for the given value. If the value is found, control flow is
2275transfered to the corresponding destination; otherwise, control flow is
2276transfered to the default destination.</p>
2277
2278<h5>Implementation:</h5>
2279
2280<p>Depending on properties of the target machine and the particular
2281<tt>switch</tt> instruction, this instruction may be code generated in different
2282ways. For example, it could be generated as a series of chained conditional
2283branches or with a lookup table.</p>
2284
2285<h5>Example:</h5>
2286
2287<pre>
2288 <i>; Emulate a conditional br instruction</i>
2289 %Val = <a href="#i_zext">zext</a> i1 %value to i32
Dan Gohman01852382009-01-04 23:44:43 +0000[diff] [blame]2290 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2291
2292 <i>; Emulate an unconditional br instruction</i>
2293 switch i32 0, label %dest [ ]
2294
2295 <i>; Implement a jump table:</i>
Dan Gohman01852382009-01-04 23:44:43 +0000[diff] [blame]2296 switch i32 %val, label %otherwise [ i32 0, label %onzero
2297 i32 1, label %onone
2298 i32 2, label %ontwo ]
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2299</pre>
2300</div>
2301
2302<!-- _______________________________________________________________________ -->
2303<div class="doc_subsubsection">
2304 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
2305</div>
2306
2307<div class="doc_text">
2308
2309<h5>Syntax:</h5>
2310
2311<pre>
Devang Pateld0bfcc72008-10-07 17:48:33 +0000[diff] [blame]2312 &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>]
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2313 to label &lt;normal label&gt; unwind label &lt;exception label&gt;
2314</pre>
2315
2316<h5>Overview:</h5>
2317
2318<p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
2319function, with the possibility of control flow transfer to either the
2320'<tt>normal</tt>' label or the
2321'<tt>exception</tt>' label. If the callee function returns with the
2322"<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
2323"normal" label. If the callee (or any indirect callees) returns with the "<a
2324href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]2325continued at the dynamically nearest "exception" label.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2326
2327<h5>Arguments:</h5>
2328
2329<p>This instruction requires several arguments:</p>
2330
2331<ol>
2332 <li>
2333 The optional "cconv" marker indicates which <a href="#callingconv">calling
2334 convention</a> the call should use. If none is specified, the call defaults
2335 to using C calling conventions.
2336 </li>
Devang Patelac2fc272008-10-06 18:50:38 +0000[diff] [blame]2337
2338 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
2339 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>',
2340 and '<tt>inreg</tt>' attributes are valid here.</li>
2341
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2342 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
2343 function value being invoked. In most cases, this is a direct function
2344 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
2345 an arbitrary pointer to function value.
2346 </li>
2347
2348 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
2349 function to be invoked. </li>
2350
2351 <li>'<tt>function args</tt>': argument list whose types match the function
2352 signature argument types. If the function signature indicates the function
2353 accepts a variable number of arguments, the extra arguments can be
2354 specified. </li>
2355
2356 <li>'<tt>normal label</tt>': the label reached when the called function
2357 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
2358
2359 <li>'<tt>exception label</tt>': the label reached when a callee returns with
2360 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
2361
Devang Pateld0bfcc72008-10-07 17:48:33 +0000[diff] [blame]2362 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
Devang Patelac2fc272008-10-06 18:50:38 +0000[diff] [blame]2363 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
2364 '<tt>readnone</tt>' attributes are valid here.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2365</ol>
2366
2367<h5>Semantics:</h5>
2368
2369<p>This instruction is designed to operate as a standard '<tt><a
2370href="#i_call">call</a></tt>' instruction in most regards. The primary
2371difference is that it establishes an association with a label, which is used by
2372the runtime library to unwind the stack.</p>
2373
2374<p>This instruction is used in languages with destructors to ensure that proper
2375cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
2376exception. Additionally, this is important for implementation of
2377'<tt>catch</tt>' clauses in high-level languages that support them.</p>
2378
2379<h5>Example:</h5>
2380<pre>
Nick Lewyckya1c11a12008-03-16 07:18:12 +0000[diff] [blame]2381 %retval = invoke i32 @Test(i32 15) to label %Continue
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2382 unwind label %TestCleanup <i>; {i32}:retval set</i>
Nick Lewyckya1c11a12008-03-16 07:18:12 +0000[diff] [blame]2383 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2384 unwind label %TestCleanup <i>; {i32}:retval set</i>
2385</pre>
2386</div>
2387
2388
2389<!-- _______________________________________________________________________ -->
2390
2391<div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
2392Instruction</a> </div>
2393
2394<div class="doc_text">
2395
2396<h5>Syntax:</h5>
2397<pre>
2398 unwind
2399</pre>
2400
2401<h5>Overview:</h5>
2402
2403<p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
2404at the first callee in the dynamic call stack which used an <a
2405href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
2406primarily used to implement exception handling.</p>
2407
2408<h5>Semantics:</h5>
2409
Chris Lattner8b094fc2008-04-19 21:01:16 +0000[diff] [blame]2410<p>The '<tt>unwind</tt>' instruction causes execution of the current function to
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2411immediately halt. The dynamic call stack is then searched for the first <a
2412href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
2413execution continues at the "exceptional" destination block specified by the
2414<tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
2415dynamic call chain, undefined behavior results.</p>
2416</div>
2417
2418<!-- _______________________________________________________________________ -->
2419
2420<div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
2421Instruction</a> </div>
2422
2423<div class="doc_text">
2424
2425<h5>Syntax:</h5>
2426<pre>
2427 unreachable
2428</pre>
2429
2430<h5>Overview:</h5>
2431
2432<p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
2433instruction is used to inform the optimizer that a particular portion of the
2434code is not reachable. This can be used to indicate that the code after a
2435no-return function cannot be reached, and other facts.</p>
2436
2437<h5>Semantics:</h5>
2438
2439<p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
2440</div>
2441
2442
2443
2444<!-- ======================================================================= -->
2445<div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
2446<div class="doc_text">
2447<p>Binary operators are used to do most of the computation in a
Chris Lattnerab596d92008-04-01 18:47:32 +0000[diff] [blame]2448program. They require two operands of the same type, execute an operation on them, and
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2449produce a single value. The operands might represent
2450multiple data, as is the case with the <a href="#t_vector">vector</a> data type.
Chris Lattnerab596d92008-04-01 18:47:32 +0000[diff] [blame]2451The result value has the same type as its operands.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2452<p>There are several different binary operators:</p>
2453</div>
2454<!-- _______________________________________________________________________ -->
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2455<div class="doc_subsubsection">
2456 <a name="i_add">'<tt>add</tt>' Instruction</a>
2457</div>
2458
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2459<div class="doc_text">
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2460
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2461<h5>Syntax:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2462
2463<pre>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2464 &lt;result&gt; = add &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2465</pre>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2466
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2467<h5>Overview:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2468
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2469<p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2470
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2471<h5>Arguments:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2472
2473<p>The two arguments to the '<tt>add</tt>' instruction must be <a
2474 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>, or
2475 <a href="#t_vector">vector</a> values. Both arguments must have identical
2476 types.</p>
2477
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2478<h5>Semantics:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2479
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2480<p>The value produced is the integer or floating point sum of the two
2481operands.</p>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2482
Chris Lattner9aba1e22008-01-28 00:36:27 +0000[diff] [blame]2483<p>If an integer sum has unsigned overflow, the result returned is the
2484mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2485the result.</p>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2486
Chris Lattner9aba1e22008-01-28 00:36:27 +0000[diff] [blame]2487<p>Because LLVM integers use a two's complement representation, this
2488instruction is appropriate for both signed and unsigned integers.</p>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2489
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2490<h5>Example:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2491
2492<pre>
2493 &lt;result&gt; = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2494</pre>
2495</div>
2496<!-- _______________________________________________________________________ -->
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2497<div class="doc_subsubsection">
2498 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
2499</div>
2500
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2501<div class="doc_text">
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2502
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2503<h5>Syntax:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2504
2505<pre>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2506 &lt;result&gt; = sub &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2507</pre>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2508
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2509<h5>Overview:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2510
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2511<p>The '<tt>sub</tt>' instruction returns the difference of its two
2512operands.</p>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2513
2514<p>Note that the '<tt>sub</tt>' instruction is used to represent the
2515'<tt>neg</tt>' instruction present in most other intermediate
2516representations.</p>
2517
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2518<h5>Arguments:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2519
2520<p>The two arguments to the '<tt>sub</tt>' instruction must be <a
2521 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>,
2522 or <a href="#t_vector">vector</a> values. Both arguments must have identical
2523 types.</p>
2524
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2525<h5>Semantics:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2526
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2527<p>The value produced is the integer or floating point difference of
2528the two operands.</p>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2529
Chris Lattner9aba1e22008-01-28 00:36:27 +0000[diff] [blame]2530<p>If an integer difference has unsigned overflow, the result returned is the
2531mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2532the result.</p>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2533
Chris Lattner9aba1e22008-01-28 00:36:27 +0000[diff] [blame]2534<p>Because LLVM integers use a two's complement representation, this
2535instruction is appropriate for both signed and unsigned integers.</p>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2536
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2537<h5>Example:</h5>
2538<pre>
2539 &lt;result&gt; = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
2540 &lt;result&gt; = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
2541</pre>
2542</div>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2543
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2544<!-- _______________________________________________________________________ -->
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2545<div class="doc_subsubsection">
2546 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
2547</div>
2548
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2549<div class="doc_text">
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2550
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2551<h5>Syntax:</h5>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2552<pre> &lt;result&gt; = mul &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2553</pre>
2554<h5>Overview:</h5>
2555<p>The '<tt>mul</tt>' instruction returns the product of its two
2556operands.</p>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2557
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2558<h5>Arguments:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2559
2560<p>The two arguments to the '<tt>mul</tt>' instruction must be <a
2561href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>,
2562or <a href="#t_vector">vector</a> values. Both arguments must have identical
2563types.</p>
2564
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2565<h5>Semantics:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2566
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2567<p>The value produced is the integer or floating point product of the
2568two operands.</p>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2569
Chris Lattner9aba1e22008-01-28 00:36:27 +0000[diff] [blame]2570<p>If the result of an integer multiplication has unsigned overflow,
2571the result returned is the mathematical result modulo
25722<sup>n</sup>, where n is the bit width of the result.</p>
2573<p>Because LLVM integers use a two's complement representation, and the
2574result is the same width as the operands, this instruction returns the
2575correct result for both signed and unsigned integers. If a full product
2576(e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands
2577should be sign-extended or zero-extended as appropriate to the
2578width of the full product.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2579<h5>Example:</h5>
2580<pre> &lt;result&gt; = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
2581</pre>
2582</div>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2583
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2584<!-- _______________________________________________________________________ -->
2585<div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
2586</a></div>
2587<div class="doc_text">
2588<h5>Syntax:</h5>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2589<pre> &lt;result&gt; = udiv &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2590</pre>
2591<h5>Overview:</h5>
2592<p>The '<tt>udiv</tt>' instruction returns the quotient of its two
2593operands.</p>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2594
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2595<h5>Arguments:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2596
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2597<p>The two arguments to the '<tt>udiv</tt>' instruction must be
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2598<a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2599values. Both arguments must have identical types.</p>
2600
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2601<h5>Semantics:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2602
Chris Lattner9aba1e22008-01-28 00:36:27 +0000[diff] [blame]2603<p>The value produced is the unsigned integer quotient of the two operands.</p>
2604<p>Note that unsigned integer division and signed integer division are distinct
2605operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
2606<p>Division by zero leads to undefined behavior.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2607<h5>Example:</h5>
2608<pre> &lt;result&gt; = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2609</pre>
2610</div>
2611<!-- _______________________________________________________________________ -->
2612<div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
2613</a> </div>
2614<div class="doc_text">
2615<h5>Syntax:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2616<pre>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2617 &lt;result&gt; = sdiv &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2618</pre>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2619
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2620<h5>Overview:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2621
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2622<p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
2623operands.</p>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2624
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2625<h5>Arguments:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2626
2627<p>The two arguments to the '<tt>sdiv</tt>' instruction must be
2628<a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2629values. Both arguments must have identical types.</p>
2630
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2631<h5>Semantics:</h5>
Chris Lattnerdc15b1d2008-04-01 18:45:27 +0000[diff] [blame]2632<p>The value produced is the signed integer quotient of the two operands rounded towards zero.</p>
Chris Lattner9aba1e22008-01-28 00:36:27 +0000[diff] [blame]2633<p>Note that signed integer division and unsigned integer division are distinct
2634operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
2635<p>Division by zero leads to undefined behavior. Overflow also leads to
2636undefined behavior; this is a rare case, but can occur, for example,
2637by doing a 32-bit division of -2147483648 by -1.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2638<h5>Example:</h5>
2639<pre> &lt;result&gt; = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2640</pre>
2641</div>
2642<!-- _______________________________________________________________________ -->
2643<div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
2644Instruction</a> </div>
2645<div class="doc_text">
2646<h5>Syntax:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2647<pre>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2648 &lt;result&gt; = fdiv &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2649</pre>
2650<h5>Overview:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2651
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2652<p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
2653operands.</p>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2654
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2655<h5>Arguments:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2656
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2657<p>The two arguments to the '<tt>fdiv</tt>' instruction must be
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2658<a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a>
2659of floating point values. Both arguments must have identical types.</p>
2660
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2661<h5>Semantics:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2662
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2663<p>The value produced is the floating point quotient of the two operands.</p>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2664
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2665<h5>Example:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2666
2667<pre>
2668 &lt;result&gt; = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2669</pre>
2670</div>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2671
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2672<!-- _______________________________________________________________________ -->
2673<div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
2674</div>
2675<div class="doc_text">
2676<h5>Syntax:</h5>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2677<pre> &lt;result&gt; = urem &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2678</pre>
2679<h5>Overview:</h5>
2680<p>The '<tt>urem</tt>' instruction returns the remainder from the
2681unsigned division of its two arguments.</p>
2682<h5>Arguments:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2683<p>The two arguments to the '<tt>urem</tt>' instruction must be
2684<a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2685values. Both arguments must have identical types.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2686<h5>Semantics:</h5>
2687<p>This instruction returns the unsigned integer <i>remainder</i> of a division.
Chris Lattnerdc15b1d2008-04-01 18:45:27 +0000[diff] [blame]2688This instruction always performs an unsigned division to get the remainder.</p>
Chris Lattner9aba1e22008-01-28 00:36:27 +0000[diff] [blame]2689<p>Note that unsigned integer remainder and signed integer remainder are
2690distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
2691<p>Taking the remainder of a division by zero leads to undefined behavior.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2692<h5>Example:</h5>
2693<pre> &lt;result&gt; = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2694</pre>
2695
2696</div>
2697<!-- _______________________________________________________________________ -->
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2698<div class="doc_subsubsection">
2699 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
2700</div>
2701
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2702<div class="doc_text">
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2703
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2704<h5>Syntax:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2705
2706<pre>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2707 &lt;result&gt; = srem &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2708</pre>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2709
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2710<h5>Overview:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2711
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2712<p>The '<tt>srem</tt>' instruction returns the remainder from the
Dan Gohman3e3fd8c2007-11-05 23:35:22 +0000[diff] [blame]2713signed division of its two operands. This instruction can also take
2714<a href="#t_vector">vector</a> versions of the values in which case
2715the elements must be integers.</p>
Chris Lattner08497ce2008-01-04 04:33:49 +0000[diff] [blame]2716
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2717<h5>Arguments:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2718
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2719<p>The two arguments to the '<tt>srem</tt>' instruction must be
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2720<a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2721values. Both arguments must have identical types.</p>
2722
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2723<h5>Semantics:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2724
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2725<p>This instruction returns the <i>remainder</i> of a division (where the result
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2726has the same sign as the dividend, <tt>op1</tt>), not the <i>modulo</i>
2727operator (where the result has the same sign as the divisor, <tt>op2</tt>) of
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2728a value. For more information about the difference, see <a
2729 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
2730Math Forum</a>. For a table of how this is implemented in various languages,
2731please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
2732Wikipedia: modulo operation</a>.</p>
Chris Lattner9aba1e22008-01-28 00:36:27 +0000[diff] [blame]2733<p>Note that signed integer remainder and unsigned integer remainder are
2734distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
2735<p>Taking the remainder of a division by zero leads to undefined behavior.
2736Overflow also leads to undefined behavior; this is a rare case, but can occur,
2737for example, by taking the remainder of a 32-bit division of -2147483648 by -1.
2738(The remainder doesn't actually overflow, but this rule lets srem be
2739implemented using instructions that return both the result of the division
2740and the remainder.)</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2741<h5>Example:</h5>
2742<pre> &lt;result&gt; = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2743</pre>
2744
2745</div>
2746<!-- _______________________________________________________________________ -->
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2747<div class="doc_subsubsection">
2748 <a name="i_frem">'<tt>frem</tt>' Instruction</a> </div>
2749
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2750<div class="doc_text">
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2751
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2752<h5>Syntax:</h5>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2753<pre> &lt;result&gt; = frem &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2754</pre>
2755<h5>Overview:</h5>
2756<p>The '<tt>frem</tt>' instruction returns the remainder from the
2757division of its two operands.</p>
2758<h5>Arguments:</h5>
2759<p>The two arguments to the '<tt>frem</tt>' instruction must be
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2760<a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a>
2761of floating point values. Both arguments must have identical types.</p>
2762
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2763<h5>Semantics:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2764
Chris Lattnerdc15b1d2008-04-01 18:45:27 +0000[diff] [blame]2765<p>This instruction returns the <i>remainder</i> of a division.
2766The remainder has the same sign as the dividend.</p>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2767
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2768<h5>Example:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2769
2770<pre>
2771 &lt;result&gt; = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2772</pre>
2773</div>
2774
2775<!-- ======================================================================= -->
2776<div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
2777Operations</a> </div>
2778<div class="doc_text">
2779<p>Bitwise binary operators are used to do various forms of
2780bit-twiddling in a program. They are generally very efficient
2781instructions and can commonly be strength reduced from other
Chris Lattnerdc15b1d2008-04-01 18:45:27 +0000[diff] [blame]2782instructions. They require two operands of the same type, execute an operation on them,
2783and produce a single value. The resulting value is the same type as its operands.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2784</div>
2785
2786<!-- _______________________________________________________________________ -->
2787<div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
2788Instruction</a> </div>
2789<div class="doc_text">
2790<h5>Syntax:</h5>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2791<pre> &lt;result&gt; = shl &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2792</pre>
Chris Lattnerd939d9f2007-10-03 21:01:14 +0000[diff] [blame]2793
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2794<h5>Overview:</h5>
Chris Lattnerd939d9f2007-10-03 21:01:14 +0000[diff] [blame]2795
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2796<p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2797the left a specified number of bits.</p>
Chris Lattnerd939d9f2007-10-03 21:01:14 +0000[diff] [blame]2798
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2799<h5>Arguments:</h5>
Chris Lattnerd939d9f2007-10-03 21:01:14 +0000[diff] [blame]2800
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2801<p>Both arguments to the '<tt>shl</tt>' instruction must be the same <a
Nate Begemanbb1ce942008-07-29 15:49:41 +0000[diff] [blame]2802 href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2803type. '<tt>op2</tt>' is treated as an unsigned value.</p>
Chris Lattnerd939d9f2007-10-03 21:01:14 +0000[diff] [blame]2804
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2805<h5>Semantics:</h5>
Chris Lattnerd939d9f2007-10-03 21:01:14 +0000[diff] [blame]2806
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2807<p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod 2<sup>n</sup>,
2808where n is the width of the result. If <tt>op2</tt> is (statically or dynamically) negative or
Mon P Wangb0f51822008-12-10 08:55:09 +0000[diff] [blame]2809equal to or larger than the number of bits in <tt>op1</tt>, the result is undefined.
2810If the arguments are vectors, each vector element of <tt>op1</tt> is shifted by the
2811corresponding shift amount in <tt>op2</tt>.</p>
Chris Lattnerd939d9f2007-10-03 21:01:14 +0000[diff] [blame]2812
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2813<h5>Example:</h5><pre>
2814 &lt;result&gt; = shl i32 4, %var <i>; yields {i32}: 4 &lt;&lt; %var</i>
2815 &lt;result&gt; = shl i32 4, 2 <i>; yields {i32}: 16</i>
2816 &lt;result&gt; = shl i32 1, 10 <i>; yields {i32}: 1024</i>
Chris Lattnerd939d9f2007-10-03 21:01:14 +0000[diff] [blame]2817 &lt;result&gt; = shl i32 1, 32 <i>; undefined</i>
Mon P Wang9901e732008-12-09 05:46:39 +0000[diff] [blame]2818 &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>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2819</pre>
2820</div>
2821<!-- _______________________________________________________________________ -->
2822<div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
2823Instruction</a> </div>
2824<div class="doc_text">
2825<h5>Syntax:</h5>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2826<pre> &lt;result&gt; = lshr &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2827</pre>
2828
2829<h5>Overview:</h5>
2830<p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
2831operand shifted to the right a specified number of bits with zero fill.</p>
2832
2833<h5>Arguments:</h5>
2834<p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
Nate Begemanbb1ce942008-07-29 15:49:41 +0000[diff] [blame]2835<a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2836type. '<tt>op2</tt>' is treated as an unsigned value.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2837
2838<h5>Semantics:</h5>
Chris Lattnerd939d9f2007-10-03 21:01:14 +0000[diff] [blame]2839
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2840<p>This instruction always performs a logical shift right operation. The most
2841significant bits of the result will be filled with zero bits after the
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2842shift. If <tt>op2</tt> is (statically or dynamically) equal to or larger than
Mon P Wangb0f51822008-12-10 08:55:09 +0000[diff] [blame]2843the number of bits in <tt>op1</tt>, the result is undefined. If the arguments are
2844vectors, each vector element of <tt>op1</tt> is shifted by the corresponding shift
2845amount in <tt>op2</tt>.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2846
2847<h5>Example:</h5>
2848<pre>
2849 &lt;result&gt; = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
2850 &lt;result&gt; = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
2851 &lt;result&gt; = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
2852 &lt;result&gt; = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
Chris Lattnerd939d9f2007-10-03 21:01:14 +0000[diff] [blame]2853 &lt;result&gt; = lshr i32 1, 32 <i>; undefined</i>
Mon P Wang9901e732008-12-09 05:46:39 +0000[diff] [blame]2854 &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>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2855</pre>
2856</div>
2857
2858<!-- _______________________________________________________________________ -->
2859<div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
2860Instruction</a> </div>
2861<div class="doc_text">
2862
2863<h5>Syntax:</h5>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2864<pre> &lt;result&gt; = ashr &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2865</pre>
2866
2867<h5>Overview:</h5>
2868<p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
2869operand shifted to the right a specified number of bits with sign extension.</p>
2870
2871<h5>Arguments:</h5>
2872<p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
Nate Begemanbb1ce942008-07-29 15:49:41 +0000[diff] [blame]2873<a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2874type. '<tt>op2</tt>' is treated as an unsigned value.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2875
2876<h5>Semantics:</h5>
2877<p>This instruction always performs an arithmetic shift right operation,
2878The most significant bits of the result will be filled with the sign bit
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2879of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
Mon P Wangb0f51822008-12-10 08:55:09 +0000[diff] [blame]2880larger than the number of bits in <tt>op1</tt>, the result is undefined. If the
2881arguments are vectors, each vector element of <tt>op1</tt> is shifted by the
2882corresponding shift amount in <tt>op2</tt>.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2883
2884<h5>Example:</h5>
2885<pre>
2886 &lt;result&gt; = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
2887 &lt;result&gt; = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
2888 &lt;result&gt; = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
2889 &lt;result&gt; = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
Chris Lattnerd939d9f2007-10-03 21:01:14 +0000[diff] [blame]2890 &lt;result&gt; = ashr i32 1, 32 <i>; undefined</i>
Mon P Wang9901e732008-12-09 05:46:39 +0000[diff] [blame]2891 &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>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2892</pre>
2893</div>
2894
2895<!-- _______________________________________________________________________ -->
2896<div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
2897Instruction</a> </div>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2898
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2899<div class="doc_text">
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2900
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2901<h5>Syntax:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2902
2903<pre>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2904 &lt;result&gt; = and &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2905</pre>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2906
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2907<h5>Overview:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2908
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2909<p>The '<tt>and</tt>' instruction returns the bitwise logical and of
2910its two operands.</p>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2911
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2912<h5>Arguments:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2913
2914<p>The two arguments to the '<tt>and</tt>' instruction must be
2915<a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2916values. Both arguments must have identical types.</p>
2917
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2918<h5>Semantics:</h5>
2919<p>The truth table used for the '<tt>and</tt>' instruction is:</p>
2920<p> </p>
Bill Wendling1e19eac2008-09-07 10:29:20 +0000[diff] [blame]2921<div>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2922<table border="1" cellspacing="0" cellpadding="4">
2923 <tbody>
2924 <tr>
2925 <td>In0</td>
2926 <td>In1</td>
2927 <td>Out</td>
2928 </tr>
2929 <tr>
2930 <td>0</td>
2931 <td>0</td>
2932 <td>0</td>
2933 </tr>
2934 <tr>
2935 <td>0</td>
2936 <td>1</td>
2937 <td>0</td>
2938 </tr>
2939 <tr>
2940 <td>1</td>
2941 <td>0</td>
2942 <td>0</td>
2943 </tr>
2944 <tr>
2945 <td>1</td>
2946 <td>1</td>
2947 <td>1</td>
2948 </tr>
2949 </tbody>
2950</table>
2951</div>
2952<h5>Example:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2953<pre>
2954 &lt;result&gt; = and i32 4, %var <i>; yields {i32}:result = 4 &amp; %var</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2955 &lt;result&gt; = and i32 15, 40 <i>; yields {i32}:result = 8</i>
2956 &lt;result&gt; = and i32 4, 8 <i>; yields {i32}:result = 0</i>
2957</pre>
2958</div>
2959<!-- _______________________________________________________________________ -->
2960<div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
2961<div class="doc_text">
2962<h5>Syntax:</h5>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2963<pre> &lt;result&gt; = or &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2964</pre>
2965<h5>Overview:</h5>
2966<p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
2967or of its two operands.</p>
2968<h5>Arguments:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2969
2970<p>The two arguments to the '<tt>or</tt>' instruction must be
2971<a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2972values. Both arguments must have identical types.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2973<h5>Semantics:</h5>
2974<p>The truth table used for the '<tt>or</tt>' instruction is:</p>
2975<p> </p>
Bill Wendling1e19eac2008-09-07 10:29:20 +0000[diff] [blame]2976<div>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2977<table border="1" cellspacing="0" cellpadding="4">
2978 <tbody>
2979 <tr>
2980 <td>In0</td>
2981 <td>In1</td>
2982 <td>Out</td>
2983 </tr>
2984 <tr>
2985 <td>0</td>
2986 <td>0</td>
2987 <td>0</td>
2988 </tr>
2989 <tr>
2990 <td>0</td>
2991 <td>1</td>
2992 <td>1</td>
2993 </tr>
2994 <tr>
2995 <td>1</td>
2996 <td>0</td>
2997 <td>1</td>
2998 </tr>
2999 <tr>
3000 <td>1</td>
3001 <td>1</td>
3002 <td>1</td>
3003 </tr>
3004 </tbody>
3005</table>
3006</div>
3007<h5>Example:</h5>
3008<pre> &lt;result&gt; = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
3009 &lt;result&gt; = or i32 15, 40 <i>; yields {i32}:result = 47</i>
3010 &lt;result&gt; = or i32 4, 8 <i>; yields {i32}:result = 12</i>
3011</pre>
3012</div>
3013<!-- _______________________________________________________________________ -->
3014<div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
3015Instruction</a> </div>
3016<div class="doc_text">
3017<h5>Syntax:</h5>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]3018<pre> &lt;result&gt; = xor &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3019</pre>
3020<h5>Overview:</h5>
3021<p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
3022or of its two operands. The <tt>xor</tt> is used to implement the
3023"one's complement" operation, which is the "~" operator in C.</p>
3024<h5>Arguments:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]3025<p>The two arguments to the '<tt>xor</tt>' instruction must be
3026<a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3027values. Both arguments must have identical types.</p>
3028
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3029<h5>Semantics:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]3030
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3031<p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
3032<p> </p>
Bill Wendling1e19eac2008-09-07 10:29:20 +0000[diff] [blame]3033<div>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3034<table border="1" cellspacing="0" cellpadding="4">
3035 <tbody>
3036 <tr>
3037 <td>In0</td>
3038 <td>In1</td>
3039 <td>Out</td>
3040 </tr>
3041 <tr>
3042 <td>0</td>
3043 <td>0</td>
3044 <td>0</td>
3045 </tr>
3046 <tr>
3047 <td>0</td>
3048 <td>1</td>
3049 <td>1</td>
3050 </tr>
3051 <tr>
3052 <td>1</td>
3053 <td>0</td>
3054 <td>1</td>
3055 </tr>
3056 <tr>
3057 <td>1</td>
3058 <td>1</td>
3059 <td>0</td>
3060 </tr>
3061 </tbody>
3062</table>
3063</div>
3064<p> </p>
3065<h5>Example:</h5>
3066<pre> &lt;result&gt; = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
3067 &lt;result&gt; = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
3068 &lt;result&gt; = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
3069 &lt;result&gt; = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
3070</pre>
3071</div>
3072
3073<!-- ======================================================================= -->
3074<div class="doc_subsection">
3075 <a name="vectorops">Vector Operations</a>
3076</div>
3077
3078<div class="doc_text">
3079
3080<p>LLVM supports several instructions to represent vector operations in a
3081target-independent manner. These instructions cover the element-access and
3082vector-specific operations needed to process vectors effectively. While LLVM
3083does directly support these vector operations, many sophisticated algorithms
3084will want to use target-specific intrinsics to take full advantage of a specific
3085target.</p>
3086
3087</div>
3088
3089<!-- _______________________________________________________________________ -->
3090<div class="doc_subsubsection">
3091 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
3092</div>
3093
3094<div class="doc_text">
3095
3096<h5>Syntax:</h5>
3097
3098<pre>
3099 &lt;result&gt; = extractelement &lt;n x &lt;ty&gt;&gt; &lt;val&gt;, i32 &lt;idx&gt; <i>; yields &lt;ty&gt;</i>
3100</pre>
3101
3102<h5>Overview:</h5>
3103
3104<p>
3105The '<tt>extractelement</tt>' instruction extracts a single scalar
3106element from a vector at a specified index.
3107</p>
3108
3109
3110<h5>Arguments:</h5>
3111
3112<p>
3113The first operand of an '<tt>extractelement</tt>' instruction is a
3114value of <a href="#t_vector">vector</a> type. The second operand is
3115an index indicating the position from which to extract the element.
3116The index may be a variable.</p>
3117
3118<h5>Semantics:</h5>
3119
3120<p>
3121The result is a scalar of the same type as the element type of
3122<tt>val</tt>. Its value is the value at position <tt>idx</tt> of
3123<tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
3124results are undefined.
3125</p>
3126
3127<h5>Example:</h5>
3128
3129<pre>
3130 %result = extractelement &lt;4 x i32&gt; %vec, i32 0 <i>; yields i32</i>
3131</pre>
3132</div>
3133
3134
3135<!-- _______________________________________________________________________ -->
3136<div class="doc_subsubsection">
3137 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
3138</div>
3139
3140<div class="doc_text">
3141
3142<h5>Syntax:</h5>
3143
3144<pre>
Dan Gohmanbcc3c502008-05-12 23:38:42 +0000[diff] [blame]3145 &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>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3146</pre>
3147
3148<h5>Overview:</h5>
3149
3150<p>
3151The '<tt>insertelement</tt>' instruction inserts a scalar
3152element into a vector at a specified index.
3153</p>
3154
3155
3156<h5>Arguments:</h5>
3157
3158<p>
3159The first operand of an '<tt>insertelement</tt>' instruction is a
3160value of <a href="#t_vector">vector</a> type. The second operand is a
3161scalar value whose type must equal the element type of the first
3162operand. The third operand is an index indicating the position at
3163which to insert the value. The index may be a variable.</p>
3164
3165<h5>Semantics:</h5>
3166
3167<p>
3168The result is a vector of the same type as <tt>val</tt>. Its
3169element values are those of <tt>val</tt> except at position
3170<tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
3171exceeds the length of <tt>val</tt>, the results are undefined.
3172</p>
3173
3174<h5>Example:</h5>
3175
3176<pre>
3177 %result = insertelement &lt;4 x i32&gt; %vec, i32 1, i32 0 <i>; yields &lt;4 x i32&gt;</i>
3178</pre>
3179</div>
3180
3181<!-- _______________________________________________________________________ -->
3182<div class="doc_subsubsection">
3183 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
3184</div>
3185
3186<div class="doc_text">
3187
3188<h5>Syntax:</h5>
3189
3190<pre>
Mon P Wangbff5d9c2008-11-10 04:46:22 +0000[diff] [blame]3191 &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>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3192</pre>
3193
3194<h5>Overview:</h5>
3195
3196<p>
3197The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
Mon P Wangbff5d9c2008-11-10 04:46:22 +0000[diff] [blame]3198from two input vectors, returning a vector with the same element type as
3199the input and length that is the same as the shuffle mask.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3200</p>
3201
3202<h5>Arguments:</h5>
3203
3204<p>
Mon P Wangbff5d9c2008-11-10 04:46:22 +0000[diff] [blame]3205The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
3206with types that match each other. The third argument is a shuffle mask whose
3207element type is always 'i32'. The result of the instruction is a vector whose
3208length is the same as the shuffle mask and whose element type is the same as
3209the element type of the first two operands.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3210</p>
3211
3212<p>
3213The shuffle mask operand is required to be a constant vector with either
3214constant integer or undef values.
3215</p>
3216
3217<h5>Semantics:</h5>
3218
3219<p>
3220The elements of the two input vectors are numbered from left to right across
3221both of the vectors. The shuffle mask operand specifies, for each element of
Mon P Wangbff5d9c2008-11-10 04:46:22 +0000[diff] [blame]3222the result vector, which element of the two input vectors the result element
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3223gets. The element selector may be undef (meaning "don't care") and the second
3224operand may be undef if performing a shuffle from only one vector.
3225</p>
3226
3227<h5>Example:</h5>
3228
3229<pre>
3230 %result = shufflevector &lt;4 x i32&gt; %v1, &lt;4 x i32&gt; %v2,
3231 &lt;4 x i32&gt; &lt;i32 0, i32 4, i32 1, i32 5&gt; <i>; yields &lt;4 x i32&gt;</i>
3232 %result = shufflevector &lt;4 x i32&gt; %v1, &lt;4 x i32&gt; undef,
3233 &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.
Mon P Wangbff5d9c2008-11-10 04:46:22 +0000[diff] [blame]3234 %result = shufflevector &lt;8 x i32&gt; %v1, &lt;8 x i32&gt; undef,
3235 &lt;4 x i32&gt; &lt;i32 0, i32 1, i32 2, i32 3&gt; <i>; yields &lt;4 x i32&gt;</i>
3236 %result = shufflevector &lt;4 x i32&gt; %v1, &lt;4 x i32&gt; %v2,
3237 &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>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3238</pre>
3239</div>
3240
3241
3242<!-- ======================================================================= -->
3243<div class="doc_subsection">
Dan Gohman74d6faf2008-05-12 23:51:09 +0000[diff] [blame]3244 <a name="aggregateops">Aggregate Operations</a>
3245</div>
3246
3247<div class="doc_text">
3248
3249<p>LLVM supports several instructions for working with aggregate values.
3250</p>
3251
3252</div>
3253
3254<!-- _______________________________________________________________________ -->
3255<div class="doc_subsubsection">
3256 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
3257</div>
3258
3259<div class="doc_text">
3260
3261<h5>Syntax:</h5>
3262
3263<pre>
3264 &lt;result&gt; = extractvalue &lt;aggregate type&gt; &lt;val&gt;, &lt;idx&gt;{, &lt;idx&gt;}*
3265</pre>
3266
3267<h5>Overview:</h5>
3268
3269<p>
Dan Gohman6c6dea02008-05-13 18:16:06 +0000[diff] [blame]3270The '<tt>extractvalue</tt>' instruction extracts the value of a struct field
3271or array element from an aggregate value.
Dan Gohman74d6faf2008-05-12 23:51:09 +0000[diff] [blame]3272</p>
3273
3274
3275<h5>Arguments:</h5>
3276
3277<p>
3278The first operand of an '<tt>extractvalue</tt>' instruction is a
3279value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a>
Dan Gohman6c6dea02008-05-13 18:16:06 +0000[diff] [blame]3280type. The operands are constant indices to specify which value to extract
Dan Gohmane5febe42008-05-31 00:58:22 +0000[diff] [blame]3281in a similar manner as indices in a
Dan Gohman74d6faf2008-05-12 23:51:09 +0000[diff] [blame]3282'<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
3283</p>
3284
3285<h5>Semantics:</h5>
3286
3287<p>
3288The result is the value at the position in the aggregate specified by
3289the index operands.
3290</p>
3291
3292<h5>Example:</h5>
3293
3294<pre>
Dan Gohmane5febe42008-05-31 00:58:22 +0000[diff] [blame]3295 %result = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
Dan Gohman74d6faf2008-05-12 23:51:09 +0000[diff] [blame]3296</pre>
3297</div>
3298
3299
3300<!-- _______________________________________________________________________ -->
3301<div class="doc_subsubsection">
3302 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
3303</div>
3304
3305<div class="doc_text">
3306
3307<h5>Syntax:</h5>
3308
3309<pre>
Dan Gohmane5febe42008-05-31 00:58:22 +0000[diff] [blame]3310 &lt;result&gt; = insertvalue &lt;aggregate type&gt; &lt;val&gt;, &lt;ty&gt; &lt;val&gt;, &lt;idx&gt; <i>; yields &lt;n x &lt;ty&gt;&gt;</i>
Dan Gohman74d6faf2008-05-12 23:51:09 +0000[diff] [blame]3311</pre>
3312
3313<h5>Overview:</h5>
3314
3315<p>
3316The '<tt>insertvalue</tt>' instruction inserts a value
Dan Gohman6c6dea02008-05-13 18:16:06 +0000[diff] [blame]3317into a struct field or array element in an aggregate.
Dan Gohman74d6faf2008-05-12 23:51:09 +0000[diff] [blame]3318</p>
3319
3320
3321<h5>Arguments:</h5>
3322
3323<p>
3324The first operand of an '<tt>insertvalue</tt>' instruction is a
3325value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a> type.
3326The second operand is a first-class value to insert.
Dan Gohman4f29e422008-05-23 21:53:15 +0000[diff] [blame]3327The following operands are constant indices
Dan Gohmane5febe42008-05-31 00:58:22 +0000[diff] [blame]3328indicating the position at which to insert the value in a similar manner as
Dan Gohman6c6dea02008-05-13 18:16:06 +0000[diff] [blame]3329indices in a
Dan Gohman74d6faf2008-05-12 23:51:09 +0000[diff] [blame]3330'<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
3331The value to insert must have the same type as the value identified
Dan Gohman6c6dea02008-05-13 18:16:06 +0000[diff] [blame]3332by the indices.
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]3333</p>
Dan Gohman74d6faf2008-05-12 23:51:09 +0000[diff] [blame]3334
3335<h5>Semantics:</h5>
3336
3337<p>
3338The result is an aggregate of the same type as <tt>val</tt>. Its
3339value is that of <tt>val</tt> except that the value at the position
Dan Gohman6c6dea02008-05-13 18:16:06 +0000[diff] [blame]3340specified by the indices is that of <tt>elt</tt>.
Dan Gohman74d6faf2008-05-12 23:51:09 +0000[diff] [blame]3341</p>
3342
3343<h5>Example:</h5>
3344
3345<pre>
Dan Gohmanb1aab4e2008-06-23 15:26:37 +0000[diff] [blame]3346 %result = insertvalue {i32, float} %agg, i32 1, 0 <i>; yields {i32, float}</i>
Dan Gohman74d6faf2008-05-12 23:51:09 +0000[diff] [blame]3347</pre>
3348</div>
3349
3350
3351<!-- ======================================================================= -->
3352<div class="doc_subsection">
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3353 <a name="memoryops">Memory Access and Addressing Operations</a>
3354</div>
3355
3356<div class="doc_text">
3357
3358<p>A key design point of an SSA-based representation is how it
3359represents memory. In LLVM, no memory locations are in SSA form, which
3360makes things very simple. This section describes how to read, write,
3361allocate, and free memory in LLVM.</p>
3362
3363</div>
3364
3365<!-- _______________________________________________________________________ -->
3366<div class="doc_subsubsection">
3367 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
3368</div>
3369
3370<div class="doc_text">
3371
3372<h5>Syntax:</h5>
3373
3374<pre>
3375 &lt;result&gt; = malloc &lt;type&gt;[, i32 &lt;NumElements&gt;][, align &lt;alignment&gt;] <i>; yields {type*}:result</i>
3376</pre>
3377
3378<h5>Overview:</h5>
3379
3380<p>The '<tt>malloc</tt>' instruction allocates memory from the system
Christopher Lambcfe00962007-12-17 01:00:21 +0000[diff] [blame]3381heap and returns a pointer to it. The object is always allocated in the generic
3382address space (address space zero).</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3383
3384<h5>Arguments:</h5>
3385
3386<p>The '<tt>malloc</tt>' instruction allocates
3387<tt>sizeof(&lt;type&gt;)*NumElements</tt>
3388bytes of memory from the operating system and returns a pointer of the
3389appropriate type to the program. If "NumElements" is specified, it is the
Gabor Greif5082cf42008-02-09 22:24:34 +0000[diff] [blame]3390number of elements allocated, otherwise "NumElements" is defaulted to be one.
Chris Lattner10368b62008-04-02 00:38:26 +0000[diff] [blame]3391If a constant alignment is specified, the value result of the allocation is guaranteed to
Gabor Greif5082cf42008-02-09 22:24:34 +0000[diff] [blame]3392be aligned to at least that boundary. If not specified, or if zero, the target can
3393choose to align the allocation on any convenient boundary.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3394
3395<p>'<tt>type</tt>' must be a sized type.</p>
3396
3397<h5>Semantics:</h5>
3398
3399<p>Memory is allocated using the system "<tt>malloc</tt>" function, and
Nick Lewyckyff384472008-11-24 03:41:24 +0000[diff] [blame]3400a pointer is returned. The result of a zero byte allocation is undefined. The
Chris Lattner8b094fc2008-04-19 21:01:16 +0000[diff] [blame]3401result is null if there is insufficient memory available.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3402
3403<h5>Example:</h5>
3404
3405<pre>
Dan Gohmanf54f50a2009-01-04 23:49:44 +0000[diff] [blame]3406 %array = malloc [4 x i8] <i>; yields {[%4 x i8]*}:array</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3407
3408 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
3409 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
3410 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
3411 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
3412 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
3413</pre>
Dan Gohman60967192009-01-12 23:12:39 +0000[diff] [blame]3414
3415<p>Note that the code generator does not yet respect the
3416 alignment value.</p>
3417
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3418</div>
3419
3420<!-- _______________________________________________________________________ -->
3421<div class="doc_subsubsection">
3422 <a name="i_free">'<tt>free</tt>' Instruction</a>
3423</div>
3424
3425<div class="doc_text">
3426
3427<h5>Syntax:</h5>
3428
3429<pre>
Dan Gohmanf54f50a2009-01-04 23:49:44 +0000[diff] [blame]3430 free &lt;type&gt; &lt;value&gt; <i>; yields {void}</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3431</pre>
3432
3433<h5>Overview:</h5>
3434
3435<p>The '<tt>free</tt>' instruction returns memory back to the unused
3436memory heap to be reallocated in the future.</p>
3437
3438<h5>Arguments:</h5>
3439
3440<p>'<tt>value</tt>' shall be a pointer value that points to a value
3441that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
3442instruction.</p>
3443
3444<h5>Semantics:</h5>
3445
3446<p>Access to the memory pointed to by the pointer is no longer defined
Chris Lattner329d6532008-04-19 22:41:32 +0000[diff] [blame]3447after this instruction executes. If the pointer is null, the operation
3448is a noop.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3449
3450<h5>Example:</h5>
3451
3452<pre>
Dan Gohmanf54f50a2009-01-04 23:49:44 +0000[diff] [blame]3453 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3454 free [4 x i8]* %array
3455</pre>
3456</div>
3457
3458<!-- _______________________________________________________________________ -->
3459<div class="doc_subsubsection">
3460 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
3461</div>
3462
3463<div class="doc_text">
3464
3465<h5>Syntax:</h5>
3466
3467<pre>
3468 &lt;result&gt; = alloca &lt;type&gt;[, i32 &lt;NumElements&gt;][, align &lt;alignment&gt;] <i>; yields {type*}:result</i>
3469</pre>
3470
3471<h5>Overview:</h5>
3472
3473<p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
3474currently executing function, to be automatically released when this function
Christopher Lambcfe00962007-12-17 01:00:21 +0000[diff] [blame]3475returns to its caller. The object is always allocated in the generic address
3476space (address space zero).</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3477
3478<h5>Arguments:</h5>
3479
3480<p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(&lt;type&gt;)*NumElements</tt>
3481bytes of memory on the runtime stack, returning a pointer of the
Gabor Greif5082cf42008-02-09 22:24:34 +0000[diff] [blame]3482appropriate type to the program. If "NumElements" is specified, it is the
3483number of elements allocated, otherwise "NumElements" is defaulted to be one.
Chris Lattner10368b62008-04-02 00:38:26 +0000[diff] [blame]3484If a constant alignment is specified, the value result of the allocation is guaranteed
Gabor Greif5082cf42008-02-09 22:24:34 +0000[diff] [blame]3485to be aligned to at least that boundary. If not specified, or if zero, the target
3486can choose to align the allocation on any convenient boundary.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3487
3488<p>'<tt>type</tt>' may be any sized type.</p>
3489
3490<h5>Semantics:</h5>
3491
Chris Lattner8b094fc2008-04-19 21:01:16 +0000[diff] [blame]3492<p>Memory is allocated; a pointer is returned. The operation is undefiend if
3493there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3494memory is automatically released when the function returns. The '<tt>alloca</tt>'
3495instruction is commonly used to represent automatic variables that must
3496have an address available. When the function returns (either with the <tt><a
3497 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
Chris Lattner10368b62008-04-02 00:38:26 +0000[diff] [blame]3498instructions), the memory is reclaimed. Allocating zero bytes
3499is legal, but the result is undefined.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3500
3501<h5>Example:</h5>
3502
3503<pre>
Dan Gohmanf54f50a2009-01-04 23:49:44 +0000[diff] [blame]3504 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
3505 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
3506 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
3507 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3508</pre>
3509</div>
3510
3511<!-- _______________________________________________________________________ -->
3512<div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
3513Instruction</a> </div>
3514<div class="doc_text">
3515<h5>Syntax:</h5>
3516<pre> &lt;result&gt; = load &lt;ty&gt;* &lt;pointer&gt;[, align &lt;alignment&gt;]<br> &lt;result&gt; = volatile load &lt;ty&gt;* &lt;pointer&gt;[, align &lt;alignment&gt;]<br></pre>
3517<h5>Overview:</h5>
3518<p>The '<tt>load</tt>' instruction is used to read from memory.</p>
3519<h5>Arguments:</h5>
3520<p>The argument to the '<tt>load</tt>' instruction specifies the memory
3521address from which to load. The pointer must point to a <a
3522 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
3523marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
3524the number or order of execution of this <tt>load</tt> with other
3525volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
3526instructions. </p>
Chris Lattner21003c82008-01-06 21:04:43 +0000[diff] [blame]3527<p>
Chris Lattner10368b62008-04-02 00:38:26 +0000[diff] [blame]3528The optional constant "align" argument specifies the alignment of the operation
Chris Lattner21003c82008-01-06 21:04:43 +0000[diff] [blame]3529(that is, the alignment of the memory address). A value of 0 or an
3530omitted "align" argument means that the operation has the preferential
3531alignment for the target. It is the responsibility of the code emitter
3532to ensure that the alignment information is correct. Overestimating
3533the alignment results in an undefined behavior. Underestimating the
3534alignment may produce less efficient code. An alignment of 1 is always
3535safe.
3536</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3537<h5>Semantics:</h5>
Duncan Sandsfe87fff2009-03-22 11:33:16 +0000[diff] [blame]3538<p>The location of memory pointed to is loaded. If the value being loaded
3539is of scalar type then the number of bytes read does not exceed the minimum
3540number of bytes needed to hold all bits of the type. For example, loading an
3541<tt>i24</tt> reads at most three bytes. When loading a value of a type like
3542<tt>i20</tt> with a size that is not an integral number of bytes, the result
3543is undefined if the value was not originally written using a store of the
3544same type.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3545<h5>Examples:</h5>
3546<pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3547 <a
3548 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
3549 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
3550</pre>
3551</div>
3552<!-- _______________________________________________________________________ -->
3553<div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
3554Instruction</a> </div>
3555<div class="doc_text">
3556<h5>Syntax:</h5>
3557<pre> store &lt;ty&gt; &lt;value&gt;, &lt;ty&gt;* &lt;pointer&gt;[, align &lt;alignment&gt;] <i>; yields {void}</i>
3558 volatile store &lt;ty&gt; &lt;value&gt;, &lt;ty&gt;* &lt;pointer&gt;[, align &lt;alignment&gt;] <i>; yields {void}</i>
3559</pre>
3560<h5>Overview:</h5>
3561<p>The '<tt>store</tt>' instruction is used to write to memory.</p>
3562<h5>Arguments:</h5>
3563<p>There are two arguments to the '<tt>store</tt>' instruction: a value
3564to store and an address at which to store it. The type of the '<tt>&lt;pointer&gt;</tt>'
Chris Lattner10368b62008-04-02 00:38:26 +0000[diff] [blame]3565operand must be a pointer to the <a href="#t_firstclass">first class</a> type
3566of the '<tt>&lt;value&gt;</tt>'
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3567operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
3568optimizer is not allowed to modify the number or order of execution of
3569this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
3570 href="#i_store">store</a></tt> instructions.</p>
Chris Lattner21003c82008-01-06 21:04:43 +0000[diff] [blame]3571<p>
Chris Lattner10368b62008-04-02 00:38:26 +0000[diff] [blame]3572The optional constant "align" argument specifies the alignment of the operation
Chris Lattner21003c82008-01-06 21:04:43 +0000[diff] [blame]3573(that is, the alignment of the memory address). A value of 0 or an
3574omitted "align" argument means that the operation has the preferential
3575alignment for the target. It is the responsibility of the code emitter
3576to ensure that the alignment information is correct. Overestimating
3577the alignment results in an undefined behavior. Underestimating the
3578alignment may produce less efficient code. An alignment of 1 is always
3579safe.
3580</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3581<h5>Semantics:</h5>
3582<p>The contents of memory are updated to contain '<tt>&lt;value&gt;</tt>'
Duncan Sandsfe87fff2009-03-22 11:33:16 +0000[diff] [blame]3583at the location specified by the '<tt>&lt;pointer&gt;</tt>' operand.
3584If '<tt>&lt;value&gt;</tt>' is of scalar type then the number of bytes
3585written does not exceed the minimum number of bytes needed to hold all
3586bits of the type. For example, storing an <tt>i24</tt> writes at most
3587three bytes. When writing a value of a type like <tt>i20</tt> with a
3588size that is not an integral number of bytes, it is unspecified what
3589happens to the extra bits that do not belong to the type, but they will
3590typically be overwritten.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3591<h5>Example:</h5>
3592<pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
Bill Wendling63ffa142007-10-22 05:10:05 +0000[diff] [blame]3593 store i32 3, i32* %ptr <i>; yields {void}</i>
3594 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3595</pre>
3596</div>
3597
3598<!-- _______________________________________________________________________ -->
3599<div class="doc_subsubsection">
3600 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
3601</div>
3602
3603<div class="doc_text">
3604<h5>Syntax:</h5>
3605<pre>
Matthijs Kooijman32a080e2008-10-13 13:44:15 +0000[diff] [blame]3606 &lt;result&gt; = getelementptr &lt;pty&gt;* &lt;ptrval&gt;{, &lt;ty&gt; &lt;idx&gt;}*
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3607</pre>
3608
3609<h5>Overview:</h5>
3610
3611<p>
3612The '<tt>getelementptr</tt>' instruction is used to get the address of a
Matthijs Kooijman32a080e2008-10-13 13:44:15 +0000[diff] [blame]3613subelement of an aggregate data structure. It performs address calculation only
3614and does not access memory.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3615
3616<h5>Arguments:</h5>
3617
Matthijs Kooijman32a080e2008-10-13 13:44:15 +0000[diff] [blame]3618<p>The first argument is always a pointer, and forms the basis of the
3619calculation. The remaining arguments are indices, that indicate which of the
3620elements of the aggregate object are indexed. The interpretation of each index
3621is dependent on the type being indexed into. The first index always indexes the
3622pointer value given as the first argument, the second index indexes a value of
3623the type pointed to (not necessarily the value directly pointed to, since the
3624first index can be non-zero), etc. The first type indexed into must be a pointer
3625value, subsequent types can be arrays, vectors and structs. Note that subsequent
3626types being indexed into can never be pointers, since that would require loading
3627the pointer before continuing calculation.</p>
3628
3629<p>The type of each index argument depends on the type it is indexing into.
3630When indexing into a (packed) structure, only <tt>i32</tt> integer
3631<b>constants</b> are allowed. When indexing into an array, pointer or vector,
3632only integers of 32 or 64 bits are allowed (also non-constants). 32-bit values
3633will be sign extended to 64-bits if required.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3634
3635<p>For example, let's consider a C code fragment and how it gets
3636compiled to LLVM:</p>
3637
3638<div class="doc_code">
3639<pre>
3640struct RT {
3641 char A;
3642 int B[10][20];
3643 char C;
3644};
3645struct ST {
3646 int X;
3647 double Y;
3648 struct RT Z;
3649};
3650
3651int *foo(struct ST *s) {
3652 return &amp;s[1].Z.B[5][13];
3653}
3654</pre>
3655</div>
3656
3657<p>The LLVM code generated by the GCC frontend is:</p>
3658
3659<div class="doc_code">
3660<pre>
Chris Lattner5b6dc6e2009-01-11 20:53:49 +0000[diff] [blame]3661%RT = <a href="#namedtypes">type</a> { i8 , [10 x [20 x i32]], i8 }
3662%ST = <a href="#namedtypes">type</a> { i32, double, %RT }
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3663
3664define i32* %foo(%ST* %s) {
3665entry:
3666 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
3667 ret i32* %reg
3668}
3669</pre>
3670</div>
3671
3672<h5>Semantics:</h5>
3673
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3674<p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
3675type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
3676}</tt>' type, a structure. The second index indexes into the third element of
3677the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
3678i8 }</tt>' type, another structure. The third index indexes into the second
3679element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
3680array. The two dimensions of the array are subscripted into, yielding an
3681'<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
3682to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
3683
3684<p>Note that it is perfectly legal to index partially through a
3685structure, returning a pointer to an inner element. Because of this,
3686the LLVM code for the given testcase is equivalent to:</p>
3687
3688<pre>
3689 define i32* %foo(%ST* %s) {
3690 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
3691 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
3692 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
3693 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
3694 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
3695 ret i32* %t5
3696 }
3697</pre>
3698
Chris Lattner50609942009-03-09 20:55:18 +0000[diff] [blame]3699<p>Note that it is undefined to access an array out of bounds: array
3700and pointer indexes must always be within the defined bounds of the
3701array type when accessed with an instruction that dereferences the
3702pointer (e.g. a load or store instruction). The one exception for
3703this rule is zero length arrays. These arrays are defined to be
3704accessible as variable length arrays, which requires access beyond the
3705zero'th element.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3706
3707<p>The getelementptr instruction is often confusing. For some more insight
3708into how it works, see <a href="GetElementPtr.html">the getelementptr
3709FAQ</a>.</p>
3710
3711<h5>Example:</h5>
3712
3713<pre>
3714 <i>; yields [12 x i8]*:aptr</i>
Matthijs Kooijman32a080e2008-10-13 13:44:15 +0000[diff] [blame]3715 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
3716 <i>; yields i8*:vptr</i>
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]3717 %vptr = getelementptr {i32, &lt;2 x i8&gt;}* %svptr, i64 0, i32 1, i32 1
Matthijs Kooijman32a080e2008-10-13 13:44:15 +0000[diff] [blame]3718 <i>; yields i8*:eptr</i>
3719 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3720</pre>
3721</div>
3722
3723<!-- ======================================================================= -->
3724<div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
3725</div>
3726<div class="doc_text">
3727<p>The instructions in this category are the conversion instructions (casting)
3728which all take a single operand and a type. They perform various bit conversions
3729on the operand.</p>
3730</div>
3731
3732<!-- _______________________________________________________________________ -->
3733<div class="doc_subsubsection">
3734 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
3735</div>
3736<div class="doc_text">
3737
3738<h5>Syntax:</h5>
3739<pre>
3740 &lt;result&gt; = trunc &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3741</pre>
3742
3743<h5>Overview:</h5>
3744<p>
3745The '<tt>trunc</tt>' instruction truncates its operand to the type <tt>ty2</tt>.
3746</p>
3747
3748<h5>Arguments:</h5>
3749<p>
3750The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
3751be an <a href="#t_integer">integer</a> type, and a type that specifies the size
3752and type of the result, which must be an <a href="#t_integer">integer</a>
3753type. The bit size of <tt>value</tt> must be larger than the bit size of
3754<tt>ty2</tt>. Equal sized types are not allowed.</p>
3755
3756<h5>Semantics:</h5>
3757<p>
3758The '<tt>trunc</tt>' instruction truncates the high order bits in <tt>value</tt>
3759and converts the remaining bits to <tt>ty2</tt>. Since the source size must be
3760larger than the destination size, <tt>trunc</tt> cannot be a <i>no-op cast</i>.
3761It will always truncate bits.</p>
3762
3763<h5>Example:</h5>
3764<pre>
3765 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
3766 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
3767 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
3768</pre>
3769</div>
3770
3771<!-- _______________________________________________________________________ -->
3772<div class="doc_subsubsection">
3773 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
3774</div>
3775<div class="doc_text">
3776
3777<h5>Syntax:</h5>
3778<pre>
3779 &lt;result&gt; = zext &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3780</pre>
3781
3782<h5>Overview:</h5>
3783<p>The '<tt>zext</tt>' instruction zero extends its operand to type
3784<tt>ty2</tt>.</p>
3785
3786
3787<h5>Arguments:</h5>
3788<p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
3789<a href="#t_integer">integer</a> type, and a type to cast it to, which must
3790also be of <a href="#t_integer">integer</a> type. The bit size of the
3791<tt>value</tt> must be smaller than the bit size of the destination type,
3792<tt>ty2</tt>.</p>
3793
3794<h5>Semantics:</h5>
3795<p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
3796bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
3797
3798<p>When zero extending from i1, the result will always be either 0 or 1.</p>
3799
3800<h5>Example:</h5>
3801<pre>
3802 %X = zext i32 257 to i64 <i>; yields i64:257</i>
3803 %Y = zext i1 true to i32 <i>; yields i32:1</i>
3804</pre>
3805</div>
3806
3807<!-- _______________________________________________________________________ -->
3808<div class="doc_subsubsection">
3809 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
3810</div>
3811<div class="doc_text">
3812
3813<h5>Syntax:</h5>
3814<pre>
3815 &lt;result&gt; = sext &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3816</pre>
3817
3818<h5>Overview:</h5>
3819<p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
3820
3821<h5>Arguments:</h5>
3822<p>
3823The '<tt>sext</tt>' instruction takes a value to cast, which must be of
3824<a href="#t_integer">integer</a> type, and a type to cast it to, which must
3825also be of <a href="#t_integer">integer</a> type. The bit size of the
3826<tt>value</tt> must be smaller than the bit size of the destination type,
3827<tt>ty2</tt>.</p>
3828
3829<h5>Semantics:</h5>
3830<p>
3831The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
3832bit (highest order bit) of the <tt>value</tt> until it reaches the bit size of
3833the type <tt>ty2</tt>.</p>
3834
3835<p>When sign extending from i1, the extension always results in -1 or 0.</p>
3836
3837<h5>Example:</h5>
3838<pre>
3839 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
3840 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
3841</pre>
3842</div>
3843
3844<!-- _______________________________________________________________________ -->
3845<div class="doc_subsubsection">
3846 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
3847</div>
3848
3849<div class="doc_text">
3850
3851<h5>Syntax:</h5>
3852
3853<pre>
3854 &lt;result&gt; = fptrunc &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3855</pre>
3856
3857<h5>Overview:</h5>
3858<p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
3859<tt>ty2</tt>.</p>
3860
3861
3862<h5>Arguments:</h5>
3863<p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
3864 point</a> value to cast and a <a href="#t_floating">floating point</a> type to
3865cast it to. The size of <tt>value</tt> must be larger than the size of
3866<tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
3867<i>no-op cast</i>.</p>
3868
3869<h5>Semantics:</h5>
3870<p> The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
3871<a href="#t_floating">floating point</a> type to a smaller
3872<a href="#t_floating">floating point</a> type. If the value cannot fit within
3873the destination type, <tt>ty2</tt>, then the results are undefined.</p>
3874
3875<h5>Example:</h5>
3876<pre>
3877 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
3878 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
3879</pre>
3880</div>
3881
3882<!-- _______________________________________________________________________ -->
3883<div class="doc_subsubsection">
3884 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
3885</div>
3886<div class="doc_text">
3887
3888<h5>Syntax:</h5>
3889<pre>
3890 &lt;result&gt; = fpext &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3891</pre>
3892
3893<h5>Overview:</h5>
3894<p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
3895floating point value.</p>
3896
3897<h5>Arguments:</h5>
3898<p>The '<tt>fpext</tt>' instruction takes a
3899<a href="#t_floating">floating point</a> <tt>value</tt> to cast,
3900and a <a href="#t_floating">floating point</a> type to cast it to. The source
3901type must be smaller than the destination type.</p>
3902
3903<h5>Semantics:</h5>
3904<p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
3905<a href="#t_floating">floating point</a> type to a larger
3906<a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
3907used to make a <i>no-op cast</i> because it always changes bits. Use
3908<tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
3909
3910<h5>Example:</h5>
3911<pre>
3912 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
3913 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
3914</pre>
3915</div>
3916
3917<!-- _______________________________________________________________________ -->
3918<div class="doc_subsubsection">
3919 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
3920</div>
3921<div class="doc_text">
3922
3923<h5>Syntax:</h5>
3924<pre>
Reid Spencere6adee82007-07-31 14:40:14 +0000[diff] [blame]3925 &lt;result&gt; = fptoui &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3926</pre>
3927
3928<h5>Overview:</h5>
Reid Spencere6adee82007-07-31 14:40:14 +0000[diff] [blame]3929<p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3930unsigned integer equivalent of type <tt>ty2</tt>.
3931</p>
3932
3933<h5>Arguments:</h5>
Reid Spencere6adee82007-07-31 14:40:14 +0000[diff] [blame]3934<p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
Nate Begeman78246ca2007-11-17 03:58:34 +0000[diff] [blame]3935scalar or vector <a href="#t_floating">floating point</a> value, and a type
3936to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3937type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3938vector integer type with the same number of elements as <tt>ty</tt></p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3939
3940<h5>Semantics:</h5>
Reid Spencere6adee82007-07-31 14:40:14 +0000[diff] [blame]3941<p> The '<tt>fptoui</tt>' instruction converts its
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3942<a href="#t_floating">floating point</a> operand into the nearest (rounding
3943towards zero) unsigned integer value. If the value cannot fit in <tt>ty2</tt>,
3944the results are undefined.</p>
3945
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3946<h5>Example:</h5>
3947<pre>
Reid Spencere6adee82007-07-31 14:40:14 +0000[diff] [blame]3948 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
Chris Lattner681f1e82007-09-22 03:17:52 +0000[diff] [blame]3949 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
Reid Spencere6adee82007-07-31 14:40:14 +0000[diff] [blame]3950 %X = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3951</pre>
3952</div>
3953
3954<!-- _______________________________________________________________________ -->
3955<div class="doc_subsubsection">
3956 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
3957</div>
3958<div class="doc_text">
3959
3960<h5>Syntax:</h5>
3961<pre>
3962 &lt;result&gt; = fptosi &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3963</pre>
3964
3965<h5>Overview:</h5>
3966<p>The '<tt>fptosi</tt>' instruction converts
3967<a href="#t_floating">floating point</a> <tt>value</tt> to type <tt>ty2</tt>.
3968</p>
3969
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3970<h5>Arguments:</h5>
3971<p> The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
Nate Begeman78246ca2007-11-17 03:58:34 +0000[diff] [blame]3972scalar or vector <a href="#t_floating">floating point</a> value, and a type
3973to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3974type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3975vector integer type with the same number of elements as <tt>ty</tt></p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3976
3977<h5>Semantics:</h5>
3978<p>The '<tt>fptosi</tt>' instruction converts its
3979<a href="#t_floating">floating point</a> operand into the nearest (rounding
3980towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
3981the results are undefined.</p>
3982
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3983<h5>Example:</h5>
3984<pre>
3985 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
Chris Lattner681f1e82007-09-22 03:17:52 +0000[diff] [blame]3986 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3987 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
3988</pre>
3989</div>
3990
3991<!-- _______________________________________________________________________ -->
3992<div class="doc_subsubsection">
3993 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
3994</div>
3995<div class="doc_text">
3996
3997<h5>Syntax:</h5>
3998<pre>
3999 &lt;result&gt; = uitofp &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
4000</pre>
4001
4002<h5>Overview:</h5>
4003<p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
4004integer and converts that value to the <tt>ty2</tt> type.</p>
4005
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4006<h5>Arguments:</h5>
Nate Begeman78246ca2007-11-17 03:58:34 +0000[diff] [blame]4007<p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
4008scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
4009to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
4010type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
4011floating point type with the same number of elements as <tt>ty</tt></p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4012
4013<h5>Semantics:</h5>
4014<p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
4015integer quantity and converts it to the corresponding floating point value. If
4016the value cannot fit in the floating point value, the results are undefined.</p>
4017
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4018<h5>Example:</h5>
4019<pre>
4020 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]4021 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4022</pre>
4023</div>
4024
4025<!-- _______________________________________________________________________ -->
4026<div class="doc_subsubsection">
4027 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
4028</div>
4029<div class="doc_text">
4030
4031<h5>Syntax:</h5>
4032<pre>
4033 &lt;result&gt; = sitofp &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
4034</pre>
4035
4036<h5>Overview:</h5>
4037<p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed
4038integer and converts that value to the <tt>ty2</tt> type.</p>
4039
4040<h5>Arguments:</h5>
Nate Begeman78246ca2007-11-17 03:58:34 +0000[diff] [blame]4041<p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
4042scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
4043to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
4044type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
4045floating point type with the same number of elements as <tt>ty</tt></p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4046
4047<h5>Semantics:</h5>
4048<p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed
4049integer quantity and converts it to the corresponding floating point value. If
4050the value cannot fit in the floating point value, the results are undefined.</p>
4051
4052<h5>Example:</h5>
4053<pre>
4054 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]4055 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4056</pre>
4057</div>
4058
4059<!-- _______________________________________________________________________ -->
4060<div class="doc_subsubsection">
4061 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
4062</div>
4063<div class="doc_text">
4064
4065<h5>Syntax:</h5>
4066<pre>
4067 &lt;result&gt; = ptrtoint &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
4068</pre>
4069
4070<h5>Overview:</h5>
4071<p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
4072the integer type <tt>ty2</tt>.</p>
4073
4074<h5>Arguments:</h5>
4075<p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
4076must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]4077<tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4078
4079<h5>Semantics:</h5>
4080<p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
4081<tt>ty2</tt> by interpreting the pointer value as an integer and either
4082truncating or zero extending that value to the size of the integer type. If
4083<tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
4084<tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
4085are the same size, then nothing is done (<i>no-op cast</i>) other than a type
4086change.</p>
4087
4088<h5>Example:</h5>
4089<pre>
4090 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
4091 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
4092</pre>
4093</div>
4094
4095<!-- _______________________________________________________________________ -->
4096<div class="doc_subsubsection">
4097 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
4098</div>
4099<div class="doc_text">
4100
4101<h5>Syntax:</h5>
4102<pre>
4103 &lt;result&gt; = inttoptr &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
4104</pre>
4105
4106<h5>Overview:</h5>
4107<p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to
4108a pointer type, <tt>ty2</tt>.</p>
4109
4110<h5>Arguments:</h5>
4111<p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
4112value to cast, and a type to cast it to, which must be a
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]4113<a href="#t_pointer">pointer</a> type.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4114
4115<h5>Semantics:</h5>
4116<p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
4117<tt>ty2</tt> by applying either a zero extension or a truncation depending on
4118the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
4119size of a pointer then a truncation is done. If <tt>value</tt> is smaller than
4120the size of a pointer then a zero extension is done. If they are the same size,
4121nothing is done (<i>no-op cast</i>).</p>
4122
4123<h5>Example:</h5>
4124<pre>
4125 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
4126 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
4127 %Y = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
4128</pre>
4129</div>
4130
4131<!-- _______________________________________________________________________ -->
4132<div class="doc_subsubsection">
4133 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
4134</div>
4135<div class="doc_text">
4136
4137<h5>Syntax:</h5>
4138<pre>
4139 &lt;result&gt; = bitcast &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
4140</pre>
4141
4142<h5>Overview:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]4143
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4144<p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
4145<tt>ty2</tt> without changing any bits.</p>
4146
4147<h5>Arguments:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]4148
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4149<p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be
Dan Gohman7305fa02008-09-08 16:45:59 +0000[diff] [blame]4150a non-aggregate first class value, and a type to cast it to, which must also be
4151a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes of
4152<tt>value</tt>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4153and the destination type, <tt>ty2</tt>, must be identical. If the source
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]4154type is a pointer, the destination type must also be a pointer. This
4155instruction supports bitwise conversion of vectors to integers and to vectors
4156of other types (as long as they have the same size).</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4157
4158<h5>Semantics:</h5>
4159<p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
4160<tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
4161this conversion. The conversion is done as if the <tt>value</tt> had been
4162stored to memory and read back as type <tt>ty2</tt>. Pointer types may only be
4163converted to other pointer types with this instruction. To convert pointers to
4164other types, use the <a href="#i_inttoptr">inttoptr</a> or
4165<a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
4166
4167<h5>Example:</h5>
4168<pre>
4169 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
4170 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]4171 %Z = bitcast &lt;2 x int&gt; %V to i64; <i>; yields i64: %V</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4172</pre>
4173</div>
4174
4175<!-- ======================================================================= -->
4176<div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
4177<div class="doc_text">
4178<p>The instructions in this category are the "miscellaneous"
4179instructions, which defy better classification.</p>
4180</div>
4181
4182<!-- _______________________________________________________________________ -->
4183<div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
4184</div>
4185<div class="doc_text">
4186<h5>Syntax:</h5>
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]4187<pre> &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>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4188</pre>
4189<h5>Overview:</h5>
Dan Gohmanb60ca3c2008-09-09 01:02:47 +0000[diff] [blame]4190<p>The '<tt>icmp</tt>' instruction returns a boolean value or
4191a vector of boolean values based on comparison
4192of its two integer, integer vector, or pointer operands.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4193<h5>Arguments:</h5>
4194<p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
4195the condition code indicating the kind of comparison to perform. It is not
4196a value, just a keyword. The possible condition code are:
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]4197</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4198<ol>
4199 <li><tt>eq</tt>: equal</li>
4200 <li><tt>ne</tt>: not equal </li>
4201 <li><tt>ugt</tt>: unsigned greater than</li>
4202 <li><tt>uge</tt>: unsigned greater or equal</li>
4203 <li><tt>ult</tt>: unsigned less than</li>
4204 <li><tt>ule</tt>: unsigned less or equal</li>
4205 <li><tt>sgt</tt>: signed greater than</li>
4206 <li><tt>sge</tt>: signed greater or equal</li>
4207 <li><tt>slt</tt>: signed less than</li>
4208 <li><tt>sle</tt>: signed less or equal</li>
4209</ol>
4210<p>The remaining two arguments must be <a href="#t_integer">integer</a> or
Dan Gohmanb60ca3c2008-09-09 01:02:47 +0000[diff] [blame]4211<a href="#t_pointer">pointer</a>
4212or integer <a href="#t_vector">vector</a> typed.
4213They must also be identical types.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4214<h5>Semantics:</h5>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]4215<p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4216the condition code given as <tt>cond</tt>. The comparison performed always
Dan Gohmanb60ca3c2008-09-09 01:02:47 +0000[diff] [blame]4217yields either an <a href="#t_primitive"><tt>i1</tt></a> or vector of <tt>i1</tt> result, as follows:
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]4218</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4219<ol>
4220 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
4221 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
4222 </li>
4223 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]4224 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4225 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]4226 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4227 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]4228 <tt>true</tt> if <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4229 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]4230 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4231 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]4232 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4233 <li><tt>sgt</tt>: interprets the operands as signed values and yields
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]4234 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4235 <li><tt>sge</tt>: interprets the operands as signed values and yields
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]4236 <tt>true</tt> if <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4237 <li><tt>slt</tt>: interprets the operands as signed values and yields
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]4238 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4239 <li><tt>sle</tt>: interprets the operands as signed values and yields
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]4240 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4241</ol>
4242<p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
4243values are compared as if they were integers.</p>
Dan Gohmanb60ca3c2008-09-09 01:02:47 +0000[diff] [blame]4244<p>If the operands are integer vectors, then they are compared
4245element by element. The result is an <tt>i1</tt> vector with
4246the same number of elements as the values being compared.
4247Otherwise, the result is an <tt>i1</tt>.
4248</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4249
4250<h5>Example:</h5>
4251<pre> &lt;result&gt; = icmp eq i32 4, 5 <i>; yields: result=false</i>
4252 &lt;result&gt; = icmp ne float* %X, %X <i>; yields: result=false</i>
4253 &lt;result&gt; = icmp ult i16 4, 5 <i>; yields: result=true</i>
4254 &lt;result&gt; = icmp sgt i16 4, 5 <i>; yields: result=false</i>
4255 &lt;result&gt; = icmp ule i16 -4, 5 <i>; yields: result=false</i>
4256 &lt;result&gt; = icmp sge i16 4, 5 <i>; yields: result=false</i>
4257</pre>
Dan Gohmana53eb382009-01-22 01:39:38 +0000[diff] [blame]4258
4259<p>Note that the code generator does not yet support vector types with
4260 the <tt>icmp</tt> instruction.</p>
4261
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4262</div>
4263
4264<!-- _______________________________________________________________________ -->
4265<div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
4266</div>
4267<div class="doc_text">
4268<h5>Syntax:</h5>
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]4269<pre> &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>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4270</pre>
4271<h5>Overview:</h5>
Dan Gohmanb60ca3c2008-09-09 01:02:47 +0000[diff] [blame]4272<p>The '<tt>fcmp</tt>' instruction returns a boolean value
4273or vector of boolean values based on comparison
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]4274of its operands.</p>
Dan Gohmanb60ca3c2008-09-09 01:02:47 +0000[diff] [blame]4275<p>
4276If the operands are floating point scalars, then the result
4277type is a boolean (<a href="#t_primitive"><tt>i1</tt></a>).
4278</p>
4279<p>If the operands are floating point vectors, then the result type
4280is a vector of boolean with the same number of elements as the
4281operands being compared.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4282<h5>Arguments:</h5>
4283<p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
4284the condition code indicating the kind of comparison to perform. It is not
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]4285a value, just a keyword. The possible condition code are:</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4286<ol>
4287 <li><tt>false</tt>: no comparison, always returns false</li>
4288 <li><tt>oeq</tt>: ordered and equal</li>
4289 <li><tt>ogt</tt>: ordered and greater than </li>
4290 <li><tt>oge</tt>: ordered and greater than or equal</li>
4291 <li><tt>olt</tt>: ordered and less than </li>
4292 <li><tt>ole</tt>: ordered and less than or equal</li>
4293 <li><tt>one</tt>: ordered and not equal</li>
4294 <li><tt>ord</tt>: ordered (no nans)</li>
4295 <li><tt>ueq</tt>: unordered or equal</li>
4296 <li><tt>ugt</tt>: unordered or greater than </li>
4297 <li><tt>uge</tt>: unordered or greater than or equal</li>
4298 <li><tt>ult</tt>: unordered or less than </li>
4299 <li><tt>ule</tt>: unordered or less than or equal</li>
4300 <li><tt>une</tt>: unordered or not equal</li>
4301 <li><tt>uno</tt>: unordered (either nans)</li>
4302 <li><tt>true</tt>: no comparison, always returns true</li>
4303</ol>
4304<p><i>Ordered</i> means that neither operand is a QNAN while
4305<i>unordered</i> means that either operand may be a QNAN.</p>
Dan Gohmanb60ca3c2008-09-09 01:02:47 +0000[diff] [blame]4306<p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be
4307either a <a href="#t_floating">floating point</a> type
4308or a <a href="#t_vector">vector</a> of floating point type.
4309They must have identical types.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4310<h5>Semantics:</h5>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]4311<p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
Dan Gohmanb60ca3c2008-09-09 01:02:47 +0000[diff] [blame]4312according to the condition code given as <tt>cond</tt>.
4313If the operands are vectors, then the vectors are compared
4314element by element.
4315Each comparison performed
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]4316always yields an <a href="#t_primitive">i1</a> result, as follows:</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4317<ol>
4318 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
4319 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]4320 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4321 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]4322 <tt>op1</tt> is greather than <tt>op2</tt>.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4323 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]4324 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4325 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]4326 <tt>op1</tt> is less than <tt>op2</tt>.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4327 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]4328 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4329 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]4330 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4331 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
4332 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]4333 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4334 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]4335 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4336 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]4337 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4338 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]4339 <tt>op1</tt> is less than <tt>op2</tt>.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4340 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]4341 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4342 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]4343 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4344 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
4345 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
4346</ol>
4347
4348<h5>Example:</h5>
4349<pre> &lt;result&gt; = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
Dan Gohmanb60ca3c2008-09-09 01:02:47 +0000[diff] [blame]4350 &lt;result&gt; = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
4351 &lt;result&gt; = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
4352 &lt;result&gt; = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4353</pre>
Dan Gohmana53eb382009-01-22 01:39:38 +0000[diff] [blame]4354
4355<p>Note that the code generator does not yet support vector types with
4356 the <tt>fcmp</tt> instruction.</p>
4357
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4358</div>
4359
4360<!-- _______________________________________________________________________ -->
Nate Begeman646fa482008-05-12 19:01:56 +0000[diff] [blame]4361<div class="doc_subsubsection">
4362 <a name="i_vicmp">'<tt>vicmp</tt>' Instruction</a>
4363</div>
4364<div class="doc_text">
4365<h5>Syntax:</h5>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]4366<pre> &lt;result&gt; = vicmp &lt;cond&gt; &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt; <i>; yields {ty}:result</i>
Nate Begeman646fa482008-05-12 19:01:56 +0000[diff] [blame]4367</pre>
4368<h5>Overview:</h5>
4369<p>The '<tt>vicmp</tt>' instruction returns an integer vector value based on
4370element-wise comparison of its two integer vector operands.</p>
4371<h5>Arguments:</h5>
4372<p>The '<tt>vicmp</tt>' instruction takes three operands. The first operand is
4373the condition code indicating the kind of comparison to perform. It is not
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]4374a value, just a keyword. The possible condition code are:</p>
Nate Begeman646fa482008-05-12 19:01:56 +0000[diff] [blame]4375<ol>
4376 <li><tt>eq</tt>: equal</li>
4377 <li><tt>ne</tt>: not equal </li>
4378 <li><tt>ugt</tt>: unsigned greater than</li>
4379 <li><tt>uge</tt>: unsigned greater or equal</li>
4380 <li><tt>ult</tt>: unsigned less than</li>
4381 <li><tt>ule</tt>: unsigned less or equal</li>
4382 <li><tt>sgt</tt>: signed greater than</li>
4383 <li><tt>sge</tt>: signed greater or equal</li>
4384 <li><tt>slt</tt>: signed less than</li>
4385 <li><tt>sle</tt>: signed less or equal</li>
4386</ol>
Dan Gohmanb60ca3c2008-09-09 01:02:47 +0000[diff] [blame]4387<p>The remaining two arguments must be <a href="#t_vector">vector</a> or
Nate Begeman646fa482008-05-12 19:01:56 +0000[diff] [blame]4388<a href="#t_integer">integer</a> typed. They must also be identical types.</p>
4389<h5>Semantics:</h5>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]4390<p>The '<tt>vicmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
Nate Begeman646fa482008-05-12 19:01:56 +0000[diff] [blame]4391according to the condition code given as <tt>cond</tt>. The comparison yields a
4392<a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, of
4393identical type as the values being compared. The most significant bit in each
4394element is 1 if the element-wise comparison evaluates to true, and is 0
4395otherwise. All other bits of the result are undefined. The condition codes
4396are evaluated identically to the <a href="#i_icmp">'<tt>icmp</tt>'
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]4397instruction</a>.</p>
Nate Begeman646fa482008-05-12 19:01:56 +0000[diff] [blame]4398
4399<h5>Example:</h5>
4400<pre>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]4401 &lt;result&gt; = vicmp eq &lt;2 x i32&gt; &lt; i32 4, i32 0&gt;, &lt; i32 5, i32 0&gt; <i>; yields: result=&lt;2 x i32&gt; &lt; i32 0, i32 -1 &gt;</i>
4402 &lt;result&gt; = vicmp ult &lt;2 x i8 &gt; &lt; i8 1, i8 2&gt;, &lt; i8 2, i8 2 &gt; <i>; yields: result=&lt;2 x i8&gt; &lt; i8 -1, i8 0 &gt;</i>
Nate Begeman646fa482008-05-12 19:01:56 +0000[diff] [blame]4403</pre>
4404</div>
4405
4406<!-- _______________________________________________________________________ -->
4407<div class="doc_subsubsection">
4408 <a name="i_vfcmp">'<tt>vfcmp</tt>' Instruction</a>
4409</div>
4410<div class="doc_text">
4411<h5>Syntax:</h5>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]4412<pre> &lt;result&gt; = vfcmp &lt;cond&gt; &lt;ty&gt; &lt;op1&gt;, &lt;op2&gt;</pre>
Nate Begeman646fa482008-05-12 19:01:56 +0000[diff] [blame]4413<h5>Overview:</h5>
4414<p>The '<tt>vfcmp</tt>' instruction returns an integer vector value based on
4415element-wise comparison of its two floating point vector operands. The output
4416elements have the same width as the input elements.</p>
4417<h5>Arguments:</h5>
4418<p>The '<tt>vfcmp</tt>' instruction takes three operands. The first operand is
4419the condition code indicating the kind of comparison to perform. It is not
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]4420a value, just a keyword. The possible condition code are:</p>
Nate Begeman646fa482008-05-12 19:01:56 +0000[diff] [blame]4421<ol>
4422 <li><tt>false</tt>: no comparison, always returns false</li>
4423 <li><tt>oeq</tt>: ordered and equal</li>
4424 <li><tt>ogt</tt>: ordered and greater than </li>
4425 <li><tt>oge</tt>: ordered and greater than or equal</li>
4426 <li><tt>olt</tt>: ordered and less than </li>
4427 <li><tt>ole</tt>: ordered and less than or equal</li>
4428 <li><tt>one</tt>: ordered and not equal</li>
4429 <li><tt>ord</tt>: ordered (no nans)</li>
4430 <li><tt>ueq</tt>: unordered or equal</li>
4431 <li><tt>ugt</tt>: unordered or greater than </li>
4432 <li><tt>uge</tt>: unordered or greater than or equal</li>
4433 <li><tt>ult</tt>: unordered or less than </li>
4434 <li><tt>ule</tt>: unordered or less than or equal</li>
4435 <li><tt>une</tt>: unordered or not equal</li>
4436 <li><tt>uno</tt>: unordered (either nans)</li>
4437 <li><tt>true</tt>: no comparison, always returns true</li>
4438</ol>
4439<p>The remaining two arguments must be <a href="#t_vector">vector</a> of
4440<a href="#t_floating">floating point</a> typed. They must also be identical
4441types.</p>
4442<h5>Semantics:</h5>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]4443<p>The '<tt>vfcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
Nate Begeman646fa482008-05-12 19:01:56 +0000[diff] [blame]4444according to the condition code given as <tt>cond</tt>. The comparison yields a
4445<a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, with
4446an identical number of elements as the values being compared, and each element
4447having identical with to the width of the floating point elements. The most
4448significant bit in each element is 1 if the element-wise comparison evaluates to
4449true, and is 0 otherwise. All other bits of the result are undefined. The
4450condition codes are evaluated identically to the
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]4451<a href="#i_fcmp">'<tt>fcmp</tt>' instruction</a>.</p>
Nate Begeman646fa482008-05-12 19:01:56 +0000[diff] [blame]4452
4453<h5>Example:</h5>
4454<pre>
Chris Lattner1e5c5cd02008-10-13 16:55:18 +0000[diff] [blame]4455 <i>; yields: result=&lt;2 x i32&gt; &lt; i32 0, i32 -1 &gt;</i>
4456 &lt;result&gt; = vfcmp oeq &lt;2 x float&gt; &lt; float 4, float 0 &gt;, &lt; float 5, float 0 &gt;
4457
4458 <i>; yields: result=&lt;2 x i64&gt; &lt; i64 -1, i64 0 &gt;</i>
4459 &lt;result&gt; = vfcmp ult &lt;2 x double&gt; &lt; double 1, double 2 &gt;, &lt; double 2, double 2&gt;
Nate Begeman646fa482008-05-12 19:01:56 +0000[diff] [blame]4460</pre>
4461</div>
4462
4463<!-- _______________________________________________________________________ -->
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]4464<div class="doc_subsubsection">
4465 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
4466</div>
4467
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4468<div class="doc_text">
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]4469
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4470<h5>Syntax:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]4471
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4472<pre> &lt;result&gt; = phi &lt;ty&gt; [ &lt;val0&gt;, &lt;label0&gt;], ...<br></pre>
4473<h5>Overview:</h5>
4474<p>The '<tt>phi</tt>' instruction is used to implement the &#966; node in
4475the SSA graph representing the function.</p>
4476<h5>Arguments:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]4477
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4478<p>The type of the incoming values is specified with the first type
4479field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
4480as arguments, with one pair for each predecessor basic block of the
4481current block. Only values of <a href="#t_firstclass">first class</a>
4482type may be used as the value arguments to the PHI node. Only labels
4483may be used as the label arguments.</p>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]4484
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4485<p>There must be no non-phi instructions between the start of a basic
4486block and the PHI instructions: i.e. PHI instructions must be first in
4487a basic block.</p>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]4488
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4489<h5>Semantics:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]4490
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4491<p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
4492specified by the pair corresponding to the predecessor basic block that executed
4493just prior to the current block.</p>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]4494
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4495<h5>Example:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]4496<pre>
4497Loop: ; Infinite loop that counts from 0 on up...
4498 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
4499 %nextindvar = add i32 %indvar, 1
4500 br label %Loop
4501</pre>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4502</div>
4503
4504<!-- _______________________________________________________________________ -->
4505<div class="doc_subsubsection">
4506 <a name="i_select">'<tt>select</tt>' Instruction</a>
4507</div>
4508
4509<div class="doc_text">
4510
4511<h5>Syntax:</h5>
4512
4513<pre>
Dan Gohmanb60ca3c2008-09-09 01:02:47 +0000[diff] [blame]4514 &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>
4515
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]4516 <i>selty</i> is either i1 or {&lt;N x i1&gt;}
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4517</pre>
4518
4519<h5>Overview:</h5>
4520
4521<p>
4522The '<tt>select</tt>' instruction is used to choose one value based on a
4523condition, without branching.
4524</p>
4525
4526
4527<h5>Arguments:</h5>
4528
4529<p>
Dan Gohmanb60ca3c2008-09-09 01:02:47 +0000[diff] [blame]4530The '<tt>select</tt>' instruction requires an 'i1' value or
4531a vector of 'i1' values indicating the
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]4532condition, and two values of the same <a href="#t_firstclass">first class</a>
Dan Gohmanb60ca3c2008-09-09 01:02:47 +0000[diff] [blame]4533type. If the val1/val2 are vectors and
4534the condition is a scalar, then entire vectors are selected, not
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]4535individual elements.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4536</p>
4537
4538<h5>Semantics:</h5>
4539
4540<p>
Dan Gohmanb60ca3c2008-09-09 01:02:47 +0000[diff] [blame]4541If the condition is an i1 and it evaluates to 1, the instruction returns the first
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4542value argument; otherwise, it returns the second value argument.
4543</p>
Dan Gohmanb60ca3c2008-09-09 01:02:47 +0000[diff] [blame]4544<p>
4545If the condition is a vector of i1, then the value arguments must
4546be vectors of the same size, and the selection is done element
4547by element.
4548</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4549
4550<h5>Example:</h5>
4551
4552<pre>
4553 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
4554</pre>
Dan Gohmana53eb382009-01-22 01:39:38 +0000[diff] [blame]4555
4556<p>Note that the code generator does not yet support conditions
4557 with vector type.</p>
4558
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4559</div>
4560
4561
4562<!-- _______________________________________________________________________ -->
4563<div class="doc_subsubsection">
4564 <a name="i_call">'<tt>call</tt>' Instruction</a>
4565</div>
4566
4567<div class="doc_text">
4568
4569<h5>Syntax:</h5>
4570<pre>
Devang Pateld0bfcc72008-10-07 17:48:33 +0000[diff] [blame]4571 &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>]
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4572</pre>
4573
4574<h5>Overview:</h5>
4575
4576<p>The '<tt>call</tt>' instruction represents a simple function call.</p>
4577
4578<h5>Arguments:</h5>
4579
4580<p>This instruction requires several arguments:</p>
4581
4582<ol>
4583 <li>
4584 <p>The optional "tail" marker indicates whether the callee function accesses
4585 any allocas or varargs in the caller. If the "tail" marker is present, the
4586 function call is eligible for tail call optimization. Note that calls may
4587 be marked "tail" even if they do not occur before a <a
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]4588 href="#i_ret"><tt>ret</tt></a> instruction.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4589 </li>
4590 <li>
4591 <p>The optional "cconv" marker indicates which <a href="#callingconv">calling
4592 convention</a> the call should use. If none is specified, the call defaults
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]4593 to using C calling conventions.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4594 </li>
Devang Patelac2fc272008-10-06 18:50:38 +0000[diff] [blame]4595
4596 <li>
4597 <p>The optional <a href="#paramattrs">Parameter Attributes</a> list for
4598 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>',
4599 and '<tt>inreg</tt>' attributes are valid here.</p>
4600 </li>
4601
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4602 <li>
Nick Lewycky93082fc2007-09-08 13:57:50 +0000[diff] [blame]4603 <p>'<tt>ty</tt>': the type of the call instruction itself which is also
4604 the type of the return value. Functions that return no value are marked
4605 <tt><a href="#t_void">void</a></tt>.</p>
4606 </li>
4607 <li>
4608 <p>'<tt>fnty</tt>': shall be the signature of the pointer to function
4609 value being invoked. The argument types must match the types implied by
4610 this signature. This type can be omitted if the function is not varargs
4611 and if the function type does not return a pointer to a function.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4612 </li>
4613 <li>
4614 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
4615 be invoked. In most cases, this is a direct function invocation, but
4616 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
4617 to function value.</p>
4618 </li>
4619 <li>
4620 <p>'<tt>function args</tt>': argument list whose types match the
4621 function signature argument types. All arguments must be of
4622 <a href="#t_firstclass">first class</a> type. If the function signature
4623 indicates the function accepts a variable number of arguments, the extra
4624 arguments can be specified.</p>
4625 </li>
Devang Patelac2fc272008-10-06 18:50:38 +0000[diff] [blame]4626 <li>
Devang Pateld0bfcc72008-10-07 17:48:33 +0000[diff] [blame]4627 <p>The optional <a href="#fnattrs">function attributes</a> list. Only
Devang Patelac2fc272008-10-06 18:50:38 +0000[diff] [blame]4628 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
4629 '<tt>readnone</tt>' attributes are valid here.</p>
4630 </li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4631</ol>
4632
4633<h5>Semantics:</h5>
4634
4635<p>The '<tt>call</tt>' instruction is used to cause control flow to
4636transfer to a specified function, with its incoming arguments bound to
4637the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
4638instruction in the called function, control flow continues with the
4639instruction after the function call, and the return value of the
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]4640function is bound to the result argument.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4641
4642<h5>Example:</h5>
4643
4644<pre>
Nick Lewycky93082fc2007-09-08 13:57:50 +0000[diff] [blame]4645 %retval = call i32 @test(i32 %argc)
Chris Lattner5e893ef2008-03-21 17:24:17 +0000[diff] [blame]4646 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42) <i>; yields i32</i>
4647 %X = tail call i32 @foo() <i>; yields i32</i>
4648 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
4649 call void %foo(i8 97 signext)
Devang Patela3cc5372008-03-10 20:49:15 +0000[diff] [blame]4650
4651 %struct.A = type { i32, i8 }
Devang Patelac2fc272008-10-06 18:50:38 +0000[diff] [blame]4652 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
Dan Gohman3e700032008-10-04 19:00:07 +0000[diff] [blame]4653 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
4654 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
Chris Lattnerac454b32008-10-08 06:26:11 +0000[diff] [blame]4655 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
Matthijs Kooijman2c4e05a2008-10-07 10:03:45 +0000[diff] [blame]4656 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4657</pre>
4658
4659</div>
4660
4661<!-- _______________________________________________________________________ -->
4662<div class="doc_subsubsection">
4663 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
4664</div>
4665
4666<div class="doc_text">
4667
4668<h5>Syntax:</h5>
4669
4670<pre>
4671 &lt;resultval&gt; = va_arg &lt;va_list*&gt; &lt;arglist&gt;, &lt;argty&gt;
4672</pre>
4673
4674<h5>Overview:</h5>
4675
4676<p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
4677the "variable argument" area of a function call. It is used to implement the
4678<tt>va_arg</tt> macro in C.</p>
4679
4680<h5>Arguments:</h5>
4681
4682<p>This instruction takes a <tt>va_list*</tt> value and the type of
4683the argument. It returns a value of the specified argument type and
4684increments the <tt>va_list</tt> to point to the next argument. The
4685actual type of <tt>va_list</tt> is target specific.</p>
4686
4687<h5>Semantics:</h5>
4688
4689<p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
4690type from the specified <tt>va_list</tt> and causes the
4691<tt>va_list</tt> to point to the next argument. For more information,
4692see the variable argument handling <a href="#int_varargs">Intrinsic
4693Functions</a>.</p>
4694
4695<p>It is legal for this instruction to be called in a function which does not
4696take a variable number of arguments, for example, the <tt>vfprintf</tt>
4697function.</p>
4698
4699<p><tt>va_arg</tt> is an LLVM instruction instead of an <a
4700href="#intrinsics">intrinsic function</a> because it takes a type as an
4701argument.</p>
4702
4703<h5>Example:</h5>
4704
4705<p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
4706
Dan Gohman60967192009-01-12 23:12:39 +0000[diff] [blame]4707<p>Note that the code generator does not yet fully support va_arg
4708 on many targets. Also, it does not currently support va_arg with
4709 aggregate types on any target.</p>
4710
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4711</div>
4712
4713<!-- *********************************************************************** -->
4714<div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
4715<!-- *********************************************************************** -->
4716
4717<div class="doc_text">
4718
4719<p>LLVM supports the notion of an "intrinsic function". These functions have
4720well known names and semantics and are required to follow certain restrictions.
4721Overall, these intrinsics represent an extension mechanism for the LLVM
4722language that does not require changing all of the transformations in LLVM when
4723adding to the language (or the bitcode reader/writer, the parser, etc...).</p>
4724
4725<p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
4726prefix is reserved in LLVM for intrinsic names; thus, function names may not
4727begin with this prefix. Intrinsic functions must always be external functions:
4728you cannot define the body of intrinsic functions. Intrinsic functions may
4729only be used in call or invoke instructions: it is illegal to take the address
4730of an intrinsic function. Additionally, because intrinsic functions are part
4731of the LLVM language, it is required if any are added that they be documented
4732here.</p>
4733
Chandler Carrutha228e392007-08-04 01:51:18 +0000[diff] [blame]4734<p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
4735a family of functions that perform the same operation but on different data
4736types. Because LLVM can represent over 8 million different integer types,
4737overloading is used commonly to allow an intrinsic function to operate on any
4738integer type. One or more of the argument types or the result type can be
4739overloaded to accept any integer type. Argument types may also be defined as
4740exactly matching a previous argument's type or the result type. This allows an
4741intrinsic function which accepts multiple arguments, but needs all of them to
4742be of the same type, to only be overloaded with respect to a single argument or
4743the result.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4744
Chandler Carrutha228e392007-08-04 01:51:18 +0000[diff] [blame]4745<p>Overloaded intrinsics will have the names of its overloaded argument types
4746encoded into its function name, each preceded by a period. Only those types
4747which are overloaded result in a name suffix. Arguments whose type is matched
4748against another type do not. For example, the <tt>llvm.ctpop</tt> function can
4749take an integer of any width and returns an integer of exactly the same integer
4750width. This leads to a family of functions such as
4751<tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29 %val)</tt>.
4752Only one type, the return type, is overloaded, and only one type suffix is
4753required. Because the argument's type is matched against the return type, it
4754does not require its own name suffix.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4755
4756<p>To learn how to add an intrinsic function, please see the
4757<a href="ExtendingLLVM.html">Extending LLVM Guide</a>.
4758</p>
4759
4760</div>
4761
4762<!-- ======================================================================= -->
4763<div class="doc_subsection">
4764 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
4765</div>
4766
4767<div class="doc_text">
4768
4769<p>Variable argument support is defined in LLVM with the <a
4770 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
4771intrinsic functions. These functions are related to the similarly
4772named macros defined in the <tt>&lt;stdarg.h&gt;</tt> header file.</p>
4773
4774<p>All of these functions operate on arguments that use a
4775target-specific value type "<tt>va_list</tt>". The LLVM assembly
4776language reference manual does not define what this type is, so all
4777transformations should be prepared to handle these functions regardless of
4778the type used.</p>
4779
4780<p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
4781instruction and the variable argument handling intrinsic functions are
4782used.</p>
4783
4784<div class="doc_code">
4785<pre>
4786define i32 @test(i32 %X, ...) {
4787 ; Initialize variable argument processing
4788 %ap = alloca i8*
4789 %ap2 = bitcast i8** %ap to i8*
4790 call void @llvm.va_start(i8* %ap2)
4791
4792 ; Read a single integer argument
4793 %tmp = va_arg i8** %ap, i32
4794
4795 ; Demonstrate usage of llvm.va_copy and llvm.va_end
4796 %aq = alloca i8*
4797 %aq2 = bitcast i8** %aq to i8*
4798 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
4799 call void @llvm.va_end(i8* %aq2)
4800
4801 ; Stop processing of arguments.
4802 call void @llvm.va_end(i8* %ap2)
4803 ret i32 %tmp
4804}
4805
4806declare void @llvm.va_start(i8*)
4807declare void @llvm.va_copy(i8*, i8*)
4808declare void @llvm.va_end(i8*)
4809</pre>
4810</div>
4811
4812</div>
4813
4814<!-- _______________________________________________________________________ -->
4815<div class="doc_subsubsection">
4816 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
4817</div>
4818
4819
4820<div class="doc_text">
4821<h5>Syntax:</h5>
4822<pre> declare void %llvm.va_start(i8* &lt;arglist&gt;)<br></pre>
4823<h5>Overview:</h5>
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]4824<p>The '<tt>llvm.va_start</tt>' intrinsic initializes
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4825<tt>*&lt;arglist&gt;</tt> for subsequent use by <tt><a
4826href="#i_va_arg">va_arg</a></tt>.</p>
4827
4828<h5>Arguments:</h5>
4829
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]4830<p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4831
4832<h5>Semantics:</h5>
4833
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]4834<p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4835macro available in C. In a target-dependent way, it initializes the
4836<tt>va_list</tt> element to which the argument points, so that the next call to
4837<tt>va_arg</tt> will produce the first variable argument passed to the function.
4838Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
4839last argument of the function as the compiler can figure that out.</p>
4840
4841</div>
4842
4843<!-- _______________________________________________________________________ -->
4844<div class="doc_subsubsection">
4845 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
4846</div>
4847
4848<div class="doc_text">
4849<h5>Syntax:</h5>
4850<pre> declare void @llvm.va_end(i8* &lt;arglist&gt;)<br></pre>
4851<h5>Overview:</h5>
4852
4853<p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*&lt;arglist&gt;</tt>,
4854which has been initialized previously with <tt><a href="#int_va_start">llvm.va_start</a></tt>
4855or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
4856
4857<h5>Arguments:</h5>
4858
4859<p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
4860
4861<h5>Semantics:</h5>
4862
4863<p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
4864macro available in C. In a target-dependent way, it destroys the
4865<tt>va_list</tt> element to which the argument points. Calls to <a
4866href="#int_va_start"><tt>llvm.va_start</tt></a> and <a href="#int_va_copy">
4867<tt>llvm.va_copy</tt></a> must be matched exactly with calls to
4868<tt>llvm.va_end</tt>.</p>
4869
4870</div>
4871
4872<!-- _______________________________________________________________________ -->
4873<div class="doc_subsubsection">
4874 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
4875</div>
4876
4877<div class="doc_text">
4878
4879<h5>Syntax:</h5>
4880
4881<pre>
4882 declare void @llvm.va_copy(i8* &lt;destarglist&gt;, i8* &lt;srcarglist&gt;)
4883</pre>
4884
4885<h5>Overview:</h5>
4886
4887<p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
4888from the source argument list to the destination argument list.</p>
4889
4890<h5>Arguments:</h5>
4891
4892<p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
4893The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
4894
4895
4896<h5>Semantics:</h5>
4897
4898<p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
4899macro available in C. In a target-dependent way, it copies the source
4900<tt>va_list</tt> element into the destination <tt>va_list</tt> element. This
4901intrinsic is necessary because the <tt><a href="#int_va_start">
4902llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
4903example, memory allocation.</p>
4904
4905</div>
4906
4907<!-- ======================================================================= -->
4908<div class="doc_subsection">
4909 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
4910</div>
4911
4912<div class="doc_text">
4913
4914<p>
4915LLVM support for <a href="GarbageCollection.html">Accurate Garbage
Chris Lattner96451482008-08-05 18:29:16 +0000[diff] [blame]4916Collection</a> (GC) requires the implementation and generation of these
4917intrinsics.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4918These intrinsics allow identification of <a href="#int_gcroot">GC roots on the
4919stack</a>, as well as garbage collector implementations that require <a
4920href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a> barriers.
4921Front-ends for type-safe garbage collected languages should generate these
4922intrinsics to make use of the LLVM garbage collectors. For more details, see <a
4923href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
4924</p>
Christopher Lambcfe00962007-12-17 01:00:21 +0000[diff] [blame]4925
4926<p>The garbage collection intrinsics only operate on objects in the generic
4927 address space (address space zero).</p>
4928
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4929</div>
4930
4931<!-- _______________________________________________________________________ -->
4932<div class="doc_subsubsection">
4933 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
4934</div>
4935
4936<div class="doc_text">
4937
4938<h5>Syntax:</h5>
4939
4940<pre>
Chris Lattner38bd5dd2007-09-21 17:30:40 +0000[diff] [blame]4941 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4942</pre>
4943
4944<h5>Overview:</h5>
4945
4946<p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
4947the code generator, and allows some metadata to be associated with it.</p>
4948
4949<h5>Arguments:</h5>
4950
4951<p>The first argument specifies the address of a stack object that contains the
4952root pointer. The second pointer (which must be either a constant or a global
4953value address) contains the meta-data to be associated with the root.</p>
4954
4955<h5>Semantics:</h5>
4956
Chris Lattnera7d94ba2008-04-24 05:59:56 +0000[diff] [blame]4957<p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4958location. At compile-time, the code generator generates information to allow
Gordon Henriksen40393542007-12-25 02:31:26 +0000[diff] [blame]4959the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
4960intrinsic may only be used in a function which <a href="#gc">specifies a GC
4961algorithm</a>.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4962
4963</div>
4964
4965
4966<!-- _______________________________________________________________________ -->
4967<div class="doc_subsubsection">
4968 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
4969</div>
4970
4971<div class="doc_text">
4972
4973<h5>Syntax:</h5>
4974
4975<pre>
Chris Lattner38bd5dd2007-09-21 17:30:40 +0000[diff] [blame]4976 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4977</pre>
4978
4979<h5>Overview:</h5>
4980
4981<p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
4982locations, allowing garbage collector implementations that require read
4983barriers.</p>
4984
4985<h5>Arguments:</h5>
4986
4987<p>The second argument is the address to read from, which should be an address
4988allocated from the garbage collector. The first object is a pointer to the
4989start of the referenced object, if needed by the language runtime (otherwise
4990null).</p>
4991
4992<h5>Semantics:</h5>
4993
4994<p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
4995instruction, but may be replaced with substantially more complex code by the
Gordon Henriksen40393542007-12-25 02:31:26 +0000[diff] [blame]4996garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
4997may only be used in a function which <a href="#gc">specifies a GC
4998algorithm</a>.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4999
5000</div>
5001
5002
5003<!-- _______________________________________________________________________ -->
5004<div class="doc_subsubsection">
5005 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
5006</div>
5007
5008<div class="doc_text">
5009
5010<h5>Syntax:</h5>
5011
5012<pre>
Chris Lattner38bd5dd2007-09-21 17:30:40 +0000[diff] [blame]5013 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5014</pre>
5015
5016<h5>Overview:</h5>
5017
5018<p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
5019locations, allowing garbage collector implementations that require write
5020barriers (such as generational or reference counting collectors).</p>
5021
5022<h5>Arguments:</h5>
5023
5024<p>The first argument is the reference to store, the second is the start of the
5025object to store it to, and the third is the address of the field of Obj to
5026store to. If the runtime does not require a pointer to the object, Obj may be
5027null.</p>
5028
5029<h5>Semantics:</h5>
5030
5031<p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
5032instruction, but may be replaced with substantially more complex code by the
Gordon Henriksen40393542007-12-25 02:31:26 +0000[diff] [blame]5033garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
5034may only be used in a function which <a href="#gc">specifies a GC
5035algorithm</a>.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5036
5037</div>
5038
5039
5040
5041<!-- ======================================================================= -->
5042<div class="doc_subsection">
5043 <a name="int_codegen">Code Generator Intrinsics</a>
5044</div>
5045
5046<div class="doc_text">
5047<p>
5048These intrinsics are provided by LLVM to expose special features that may only
5049be implemented with code generator support.
5050</p>
5051
5052</div>
5053
5054<!-- _______________________________________________________________________ -->
5055<div class="doc_subsubsection">
5056 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
5057</div>
5058
5059<div class="doc_text">
5060
5061<h5>Syntax:</h5>
5062<pre>
5063 declare i8 *@llvm.returnaddress(i32 &lt;level&gt;)
5064</pre>
5065
5066<h5>Overview:</h5>
5067
5068<p>
5069The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
5070target-specific value indicating the return address of the current function
5071or one of its callers.
5072</p>
5073
5074<h5>Arguments:</h5>
5075
5076<p>
5077The argument to this intrinsic indicates which function to return the address
5078for. Zero indicates the calling function, one indicates its caller, etc. The
5079argument is <b>required</b> to be a constant integer value.
5080</p>
5081
5082<h5>Semantics:</h5>
5083
5084<p>
5085The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
5086the return address of the specified call frame, or zero if it cannot be
5087identified. The value returned by this intrinsic is likely to be incorrect or 0
5088for arguments other than zero, so it should only be used for debugging purposes.
5089</p>
5090
5091<p>
5092Note that calling this intrinsic does not prevent function inlining or other
5093aggressive transformations, so the value returned may not be that of the obvious
5094source-language caller.
5095</p>
5096</div>
5097
5098
5099<!-- _______________________________________________________________________ -->
5100<div class="doc_subsubsection">
5101 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
5102</div>
5103
5104<div class="doc_text">
5105
5106<h5>Syntax:</h5>
5107<pre>
Chris Lattner38bd5dd2007-09-21 17:30:40 +0000[diff] [blame]5108 declare i8 *@llvm.frameaddress(i32 &lt;level&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5109</pre>
5110
5111<h5>Overview:</h5>
5112
5113<p>
5114The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
5115target-specific frame pointer value for the specified stack frame.
5116</p>
5117
5118<h5>Arguments:</h5>
5119
5120<p>
5121The argument to this intrinsic indicates which function to return the frame
5122pointer for. Zero indicates the calling function, one indicates its caller,
5123etc. The argument is <b>required</b> to be a constant integer value.
5124</p>
5125
5126<h5>Semantics:</h5>
5127
5128<p>
5129The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
5130the frame address of the specified call frame, or zero if it cannot be
5131identified. The value returned by this intrinsic is likely to be incorrect or 0
5132for arguments other than zero, so it should only be used for debugging purposes.
5133</p>
5134
5135<p>
5136Note that calling this intrinsic does not prevent function inlining or other
5137aggressive transformations, so the value returned may not be that of the obvious
5138source-language caller.
5139</p>
5140</div>
5141
5142<!-- _______________________________________________________________________ -->
5143<div class="doc_subsubsection">
5144 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
5145</div>
5146
5147<div class="doc_text">
5148
5149<h5>Syntax:</h5>
5150<pre>
Chris Lattner38bd5dd2007-09-21 17:30:40 +0000[diff] [blame]5151 declare i8 *@llvm.stacksave()
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5152</pre>
5153
5154<h5>Overview:</h5>
5155
5156<p>
5157The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
5158the function stack, for use with <a href="#int_stackrestore">
5159<tt>llvm.stackrestore</tt></a>. This is useful for implementing language
5160features like scoped automatic variable sized arrays in C99.
5161</p>
5162
5163<h5>Semantics:</h5>
5164
5165<p>
5166This intrinsic returns a opaque pointer value that can be passed to <a
5167href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
5168<tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
5169<tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
5170state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
5171practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
5172that were allocated after the <tt>llvm.stacksave</tt> was executed.
5173</p>
5174
5175</div>
5176
5177<!-- _______________________________________________________________________ -->
5178<div class="doc_subsubsection">
5179 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
5180</div>
5181
5182<div class="doc_text">
5183
5184<h5>Syntax:</h5>
5185<pre>
5186 declare void @llvm.stackrestore(i8 * %ptr)
5187</pre>
5188
5189<h5>Overview:</h5>
5190
5191<p>
5192The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
5193the function stack to the state it was in when the corresponding <a
5194href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
5195useful for implementing language features like scoped automatic variable sized
5196arrays in C99.
5197</p>
5198
5199<h5>Semantics:</h5>
5200
5201<p>
5202See the description for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.
5203</p>
5204
5205</div>
5206
5207
5208<!-- _______________________________________________________________________ -->
5209<div class="doc_subsubsection">
5210 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
5211</div>
5212
5213<div class="doc_text">
5214
5215<h5>Syntax:</h5>
5216<pre>
Chris Lattner38bd5dd2007-09-21 17:30:40 +0000[diff] [blame]5217 declare void @llvm.prefetch(i8* &lt;address&gt;, i32 &lt;rw&gt;, i32 &lt;locality&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5218</pre>
5219
5220<h5>Overview:</h5>
5221
5222
5223<p>
5224The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
5225a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
5226no
5227effect on the behavior of the program but can change its performance
5228characteristics.
5229</p>
5230
5231<h5>Arguments:</h5>
5232
5233<p>
5234<tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
5235determining if the fetch should be for a read (0) or write (1), and
5236<tt>locality</tt> is a temporal locality specifier ranging from (0) - no
5237locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
5238<tt>locality</tt> arguments must be constant integers.
5239</p>
5240
5241<h5>Semantics:</h5>
5242
5243<p>
5244This intrinsic does not modify the behavior of the program. In particular,
5245prefetches cannot trap and do not produce a value. On targets that support this
5246intrinsic, the prefetch can provide hints to the processor cache for better
5247performance.
5248</p>
5249
5250</div>
5251
5252<!-- _______________________________________________________________________ -->
5253<div class="doc_subsubsection">
5254 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
5255</div>
5256
5257<div class="doc_text">
5258
5259<h5>Syntax:</h5>
5260<pre>
Chris Lattner38bd5dd2007-09-21 17:30:40 +0000[diff] [blame]5261 declare void @llvm.pcmarker(i32 &lt;id&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5262</pre>
5263
5264<h5>Overview:</h5>
5265
5266
5267<p>
5268The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
Chris Lattner96451482008-08-05 18:29:16 +0000[diff] [blame]5269(PC) in a region of
5270code to simulators and other tools. The method is target specific, but it is
5271expected that the marker will use exported symbols to transmit the PC of the
5272marker.
5273The marker makes no guarantees that it will remain with any specific instruction
5274after optimizations. It is possible that the presence of a marker will inhibit
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5275optimizations. The intended use is to be inserted after optimizations to allow
5276correlations of simulation runs.
5277</p>
5278
5279<h5>Arguments:</h5>
5280
5281<p>
5282<tt>id</tt> is a numerical id identifying the marker.
5283</p>
5284
5285<h5>Semantics:</h5>
5286
5287<p>
5288This intrinsic does not modify the behavior of the program. Backends that do not
5289support this intrinisic may ignore it.
5290</p>
5291
5292</div>
5293
5294<!-- _______________________________________________________________________ -->
5295<div class="doc_subsubsection">
5296 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
5297</div>
5298
5299<div class="doc_text">
5300
5301<h5>Syntax:</h5>
5302<pre>
5303 declare i64 @llvm.readcyclecounter( )
5304</pre>
5305
5306<h5>Overview:</h5>
5307
5308
5309<p>
5310The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
5311counter register (or similar low latency, high accuracy clocks) on those targets
5312that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
5313As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
5314should only be used for small timings.
5315</p>
5316
5317<h5>Semantics:</h5>
5318
5319<p>
5320When directly supported, reading the cycle counter should not modify any memory.
5321Implementations are allowed to either return a application specific value or a
5322system wide value. On backends without support, this is lowered to a constant 0.
5323</p>
5324
5325</div>
5326
5327<!-- ======================================================================= -->
5328<div class="doc_subsection">
5329 <a name="int_libc">Standard C Library Intrinsics</a>
5330</div>
5331
5332<div class="doc_text">
5333<p>
5334LLVM provides intrinsics for a few important standard C library functions.
5335These intrinsics allow source-language front-ends to pass information about the
5336alignment of the pointer arguments to the code generator, providing opportunity
5337for more efficient code generation.
5338</p>
5339
5340</div>
5341
5342<!-- _______________________________________________________________________ -->
5343<div class="doc_subsubsection">
5344 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
5345</div>
5346
5347<div class="doc_text">
5348
5349<h5>Syntax:</h5>
Chris Lattner82c2e432008-11-21 16:42:48 +0000[diff] [blame]5350<p>This is an overloaded intrinsic. You can use llvm.memcpy on any integer bit
5351width. Not all targets support all bit widths however.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5352<pre>
Chris Lattner82c2e432008-11-21 16:42:48 +0000[diff] [blame]5353 declare void @llvm.memcpy.i8(i8 * &lt;dest&gt;, i8 * &lt;src&gt;,
5354 i8 &lt;len&gt;, i32 &lt;align&gt;)
5355 declare void @llvm.memcpy.i16(i8 * &lt;dest&gt;, i8 * &lt;src&gt;,
5356 i16 &lt;len&gt;, i32 &lt;align&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5357 declare void @llvm.memcpy.i32(i8 * &lt;dest&gt;, i8 * &lt;src&gt;,
5358 i32 &lt;len&gt;, i32 &lt;align&gt;)
5359 declare void @llvm.memcpy.i64(i8 * &lt;dest&gt;, i8 * &lt;src&gt;,
5360 i64 &lt;len&gt;, i32 &lt;align&gt;)
5361</pre>
5362
5363<h5>Overview:</h5>
5364
5365<p>
5366The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
5367location to the destination location.
5368</p>
5369
5370<p>
5371Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
5372intrinsics do not return a value, and takes an extra alignment argument.
5373</p>
5374
5375<h5>Arguments:</h5>
5376
5377<p>
5378The first argument is a pointer to the destination, the second is a pointer to
5379the source. The third argument is an integer argument
5380specifying the number of bytes to copy, and the fourth argument is the alignment
5381of the source and destination locations.
5382</p>
5383
5384<p>
5385If the call to this intrinisic has an alignment value that is not 0 or 1, then
5386the caller guarantees that both the source and destination pointers are aligned
5387to that boundary.
5388</p>
5389
5390<h5>Semantics:</h5>
5391
5392<p>
5393The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
5394location to the destination location, which are not allowed to overlap. It
5395copies "len" bytes of memory over. If the argument is known to be aligned to
5396some boundary, this can be specified as the fourth argument, otherwise it should
5397be set to 0 or 1.
5398</p>
5399</div>
5400
5401
5402<!-- _______________________________________________________________________ -->
5403<div class="doc_subsubsection">
5404 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
5405</div>
5406
5407<div class="doc_text">
5408
5409<h5>Syntax:</h5>
Chris Lattner82c2e432008-11-21 16:42:48 +0000[diff] [blame]5410<p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
5411width. Not all targets support all bit widths however.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5412<pre>
Chris Lattner82c2e432008-11-21 16:42:48 +0000[diff] [blame]5413 declare void @llvm.memmove.i8(i8 * &lt;dest&gt;, i8 * &lt;src&gt;,
5414 i8 &lt;len&gt;, i32 &lt;align&gt;)
5415 declare void @llvm.memmove.i16(i8 * &lt;dest&gt;, i8 * &lt;src&gt;,
5416 i16 &lt;len&gt;, i32 &lt;align&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5417 declare void @llvm.memmove.i32(i8 * &lt;dest&gt;, i8 * &lt;src&gt;,
5418 i32 &lt;len&gt;, i32 &lt;align&gt;)
5419 declare void @llvm.memmove.i64(i8 * &lt;dest&gt;, i8 * &lt;src&gt;,
5420 i64 &lt;len&gt;, i32 &lt;align&gt;)
5421</pre>
5422
5423<h5>Overview:</h5>
5424
5425<p>
5426The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
5427location to the destination location. It is similar to the
Chris Lattnerdba16ea2008-01-06 19:51:52 +0000[diff] [blame]5428'<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to overlap.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5429</p>
5430
5431<p>
5432Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
5433intrinsics do not return a value, and takes an extra alignment argument.
5434</p>
5435
5436<h5>Arguments:</h5>
5437
5438<p>
5439The first argument is a pointer to the destination, the second is a pointer to
5440the source. The third argument is an integer argument
5441specifying the number of bytes to copy, and the fourth argument is the alignment
5442of the source and destination locations.
5443</p>
5444
5445<p>
5446If the call to this intrinisic has an alignment value that is not 0 or 1, then
5447the caller guarantees that the source and destination pointers are aligned to
5448that boundary.
5449</p>
5450
5451<h5>Semantics:</h5>
5452
5453<p>
5454The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
5455location to the destination location, which may overlap. It
5456copies "len" bytes of memory over. If the argument is known to be aligned to
5457some boundary, this can be specified as the fourth argument, otherwise it should
5458be set to 0 or 1.
5459</p>
5460</div>
5461
5462
5463<!-- _______________________________________________________________________ -->
5464<div class="doc_subsubsection">
5465 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
5466</div>
5467
5468<div class="doc_text">
5469
5470<h5>Syntax:</h5>
Chris Lattner82c2e432008-11-21 16:42:48 +0000[diff] [blame]5471<p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
5472width. Not all targets support all bit widths however.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5473<pre>
Chris Lattner82c2e432008-11-21 16:42:48 +0000[diff] [blame]5474 declare void @llvm.memset.i8(i8 * &lt;dest&gt;, i8 &lt;val&gt;,
5475 i8 &lt;len&gt;, i32 &lt;align&gt;)
5476 declare void @llvm.memset.i16(i8 * &lt;dest&gt;, i8 &lt;val&gt;,
5477 i16 &lt;len&gt;, i32 &lt;align&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5478 declare void @llvm.memset.i32(i8 * &lt;dest&gt;, i8 &lt;val&gt;,
5479 i32 &lt;len&gt;, i32 &lt;align&gt;)
5480 declare void @llvm.memset.i64(i8 * &lt;dest&gt;, i8 &lt;val&gt;,
5481 i64 &lt;len&gt;, i32 &lt;align&gt;)
5482</pre>
5483
5484<h5>Overview:</h5>
5485
5486<p>
5487The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
5488byte value.
5489</p>
5490
5491<p>
5492Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
5493does not return a value, and takes an extra alignment argument.
5494</p>
5495
5496<h5>Arguments:</h5>
5497
5498<p>
5499The first argument is a pointer to the destination to fill, the second is the
5500byte value to fill it with, the third argument is an integer
5501argument specifying the number of bytes to fill, and the fourth argument is the
5502known alignment of destination location.
5503</p>
5504
5505<p>
5506If the call to this intrinisic has an alignment value that is not 0 or 1, then
5507the caller guarantees that the destination pointer is aligned to that boundary.
5508</p>
5509
5510<h5>Semantics:</h5>
5511
5512<p>
5513The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
5514the
5515destination location. If the argument is known to be aligned to some boundary,
5516this can be specified as the fourth argument, otherwise it should be set to 0 or
55171.
5518</p>
5519</div>
5520
5521
5522<!-- _______________________________________________________________________ -->
5523<div class="doc_subsubsection">
5524 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
5525</div>
5526
5527<div class="doc_text">
5528
5529<h5>Syntax:</h5>
Dale Johannesenf9adbb62007-10-02 17:47:38 +0000[diff] [blame]5530<p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
Dan Gohman361079c2007-10-15 20:30:11 +0000[diff] [blame]5531floating point or vector of floating point type. Not all targets support all
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]5532types however.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5533<pre>
Dale Johannesenf9adbb62007-10-02 17:47:38 +0000[diff] [blame]5534 declare float @llvm.sqrt.f32(float %Val)
5535 declare double @llvm.sqrt.f64(double %Val)
5536 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
5537 declare fp128 @llvm.sqrt.f128(fp128 %Val)
5538 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5539</pre>
5540
5541<h5>Overview:</h5>
5542
5543<p>
5544The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
Dan Gohman361079c2007-10-15 20:30:11 +0000[diff] [blame]5545returning the same value as the libm '<tt>sqrt</tt>' functions would. Unlike
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5546<tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
Chris Lattnerf914a002008-01-29 07:00:44 +0000[diff] [blame]5547negative numbers other than -0.0 (which allows for better optimization, because
5548there is no need to worry about errno being set). <tt>llvm.sqrt(-0.0)</tt> is
5549defined to return -0.0 like IEEE sqrt.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5550</p>
5551
5552<h5>Arguments:</h5>
5553
5554<p>
5555The argument and return value are floating point numbers of the same type.
5556</p>
5557
5558<h5>Semantics:</h5>
5559
5560<p>
5561This function returns the sqrt of the specified operand if it is a nonnegative
5562floating point number.
5563</p>
5564</div>
5565
5566<!-- _______________________________________________________________________ -->
5567<div class="doc_subsubsection">
5568 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
5569</div>
5570
5571<div class="doc_text">
5572
5573<h5>Syntax:</h5>
Dale Johannesenf9adbb62007-10-02 17:47:38 +0000[diff] [blame]5574<p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
Dan Gohman361079c2007-10-15 20:30:11 +0000[diff] [blame]5575floating point or vector of floating point type. Not all targets support all
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]5576types however.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5577<pre>
Dale Johannesenf9adbb62007-10-02 17:47:38 +0000[diff] [blame]5578 declare float @llvm.powi.f32(float %Val, i32 %power)
5579 declare double @llvm.powi.f64(double %Val, i32 %power)
5580 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
5581 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
5582 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5583</pre>
5584
5585<h5>Overview:</h5>
5586
5587<p>
5588The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
5589specified (positive or negative) power. The order of evaluation of
Dan Gohman361079c2007-10-15 20:30:11 +0000[diff] [blame]5590multiplications is not defined. When a vector of floating point type is
5591used, the second argument remains a scalar integer value.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5592</p>
5593
5594<h5>Arguments:</h5>
5595
5596<p>
5597The second argument is an integer power, and the first is a value to raise to
5598that power.
5599</p>
5600
5601<h5>Semantics:</h5>
5602
5603<p>
5604This function returns the first value raised to the second power with an
5605unspecified sequence of rounding operations.</p>
5606</div>
5607
Dan Gohman361079c2007-10-15 20:30:11 +0000[diff] [blame]5608<!-- _______________________________________________________________________ -->
5609<div class="doc_subsubsection">
5610 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
5611</div>
5612
5613<div class="doc_text">
5614
5615<h5>Syntax:</h5>
5616<p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
5617floating point or vector of floating point type. Not all targets support all
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]5618types however.</p>
Dan Gohman361079c2007-10-15 20:30:11 +0000[diff] [blame]5619<pre>
5620 declare float @llvm.sin.f32(float %Val)
5621 declare double @llvm.sin.f64(double %Val)
5622 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
5623 declare fp128 @llvm.sin.f128(fp128 %Val)
5624 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
5625</pre>
5626
5627<h5>Overview:</h5>
5628
5629<p>
5630The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.
5631</p>
5632
5633<h5>Arguments:</h5>
5634
5635<p>
5636The argument and return value are floating point numbers of the same type.
5637</p>
5638
5639<h5>Semantics:</h5>
5640
5641<p>
5642This function returns the sine of the specified operand, returning the
5643same values as the libm <tt>sin</tt> functions would, and handles error
Dan Gohmaneaba92e2007-10-17 18:05:13 +0000[diff] [blame]5644conditions in the same way.</p>
Dan Gohman361079c2007-10-15 20:30:11 +0000[diff] [blame]5645</div>
5646
5647<!-- _______________________________________________________________________ -->
5648<div class="doc_subsubsection">
5649 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
5650</div>
5651
5652<div class="doc_text">
5653
5654<h5>Syntax:</h5>
5655<p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
5656floating point or vector of floating point type. Not all targets support all
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]5657types however.</p>
Dan Gohman361079c2007-10-15 20:30:11 +0000[diff] [blame]5658<pre>
5659 declare float @llvm.cos.f32(float %Val)
5660 declare double @llvm.cos.f64(double %Val)
5661 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
5662 declare fp128 @llvm.cos.f128(fp128 %Val)
5663 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
5664</pre>
5665
5666<h5>Overview:</h5>
5667
5668<p>
5669The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.
5670</p>
5671
5672<h5>Arguments:</h5>
5673
5674<p>
5675The argument and return value are floating point numbers of the same type.
5676</p>
5677
5678<h5>Semantics:</h5>
5679
5680<p>
5681This function returns the cosine of the specified operand, returning the
5682same values as the libm <tt>cos</tt> functions would, and handles error
Dan Gohmaneaba92e2007-10-17 18:05:13 +0000[diff] [blame]5683conditions in the same way.</p>
Dan Gohman361079c2007-10-15 20:30:11 +0000[diff] [blame]5684</div>
5685
5686<!-- _______________________________________________________________________ -->
5687<div class="doc_subsubsection">
5688 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
5689</div>
5690
5691<div class="doc_text">
5692
5693<h5>Syntax:</h5>
5694<p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
5695floating point or vector of floating point type. Not all targets support all
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]5696types however.</p>
Dan Gohman361079c2007-10-15 20:30:11 +0000[diff] [blame]5697<pre>
5698 declare float @llvm.pow.f32(float %Val, float %Power)
5699 declare double @llvm.pow.f64(double %Val, double %Power)
5700 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
5701 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
5702 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
5703</pre>
5704
5705<h5>Overview:</h5>
5706
5707<p>
5708The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
5709specified (positive or negative) power.
5710</p>
5711
5712<h5>Arguments:</h5>
5713
5714<p>
5715The second argument is a floating point power, and the first is a value to
5716raise to that power.
5717</p>
5718
5719<h5>Semantics:</h5>
5720
5721<p>
5722This function returns the first value raised to the second power,
5723returning the
5724same values as the libm <tt>pow</tt> functions would, and handles error
Dan Gohmaneaba92e2007-10-17 18:05:13 +0000[diff] [blame]5725conditions in the same way.</p>
Dan Gohman361079c2007-10-15 20:30:11 +0000[diff] [blame]5726</div>
5727
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5728
5729<!-- ======================================================================= -->
5730<div class="doc_subsection">
5731 <a name="int_manip">Bit Manipulation Intrinsics</a>
5732</div>
5733
5734<div class="doc_text">
5735<p>
5736LLVM provides intrinsics for a few important bit manipulation operations.
5737These allow efficient code generation for some algorithms.
5738</p>
5739
5740</div>
5741
5742<!-- _______________________________________________________________________ -->
5743<div class="doc_subsubsection">
5744 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
5745</div>
5746
5747<div class="doc_text">
5748
5749<h5>Syntax:</h5>
5750<p>This is an overloaded intrinsic function. You can use bswap on any integer
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]5751type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5752<pre>
Chandler Carrutha228e392007-08-04 01:51:18 +0000[diff] [blame]5753 declare i16 @llvm.bswap.i16(i16 &lt;id&gt;)
5754 declare i32 @llvm.bswap.i32(i32 &lt;id&gt;)
5755 declare i64 @llvm.bswap.i64(i64 &lt;id&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5756</pre>
5757
5758<h5>Overview:</h5>
5759
5760<p>
5761The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
5762values with an even number of bytes (positive multiple of 16 bits). These are
5763useful for performing operations on data that is not in the target's native
5764byte order.
5765</p>
5766
5767<h5>Semantics:</h5>
5768
5769<p>
Chandler Carrutha228e392007-08-04 01:51:18 +0000[diff] [blame]5770The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5771and low byte of the input i16 swapped. Similarly, the <tt>llvm.bswap.i32</tt>
5772intrinsic returns an i32 value that has the four bytes of the input i32
5773swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the returned
Chandler Carrutha228e392007-08-04 01:51:18 +0000[diff] [blame]5774i32 will have its bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i48</tt>,
5775<tt>llvm.bswap.i64</tt> and other intrinsics extend this concept to
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5776additional even-byte lengths (6 bytes, 8 bytes and more, respectively).
5777</p>
5778
5779</div>
5780
5781<!-- _______________________________________________________________________ -->
5782<div class="doc_subsubsection">
5783 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
5784</div>
5785
5786<div class="doc_text">
5787
5788<h5>Syntax:</h5>
5789<p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]5790width. Not all targets support all bit widths however.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5791<pre>
Bill Wendling3f8cebe2009-02-08 01:40:31 +0000[diff] [blame]5792 declare i8 @llvm.ctpop.i8(i8 &lt;src&gt;)
Chandler Carrutha228e392007-08-04 01:51:18 +0000[diff] [blame]5793 declare i16 @llvm.ctpop.i16(i16 &lt;src&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5794 declare i32 @llvm.ctpop.i32(i32 &lt;src&gt;)
Chandler Carrutha228e392007-08-04 01:51:18 +0000[diff] [blame]5795 declare i64 @llvm.ctpop.i64(i64 &lt;src&gt;)
5796 declare i256 @llvm.ctpop.i256(i256 &lt;src&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5797</pre>
5798
5799<h5>Overview:</h5>
5800
5801<p>
5802The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
5803value.
5804</p>
5805
5806<h5>Arguments:</h5>
5807
5808<p>
5809The only argument is the value to be counted. The argument may be of any
5810integer type. The return type must match the argument type.
5811</p>
5812
5813<h5>Semantics:</h5>
5814
5815<p>
5816The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
5817</p>
5818</div>
5819
5820<!-- _______________________________________________________________________ -->
5821<div class="doc_subsubsection">
5822 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
5823</div>
5824
5825<div class="doc_text">
5826
5827<h5>Syntax:</h5>
5828<p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]5829integer bit width. Not all targets support all bit widths however.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5830<pre>
Chandler Carrutha228e392007-08-04 01:51:18 +0000[diff] [blame]5831 declare i8 @llvm.ctlz.i8 (i8 &lt;src&gt;)
5832 declare i16 @llvm.ctlz.i16(i16 &lt;src&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5833 declare i32 @llvm.ctlz.i32(i32 &lt;src&gt;)
Chandler Carrutha228e392007-08-04 01:51:18 +0000[diff] [blame]5834 declare i64 @llvm.ctlz.i64(i64 &lt;src&gt;)
5835 declare i256 @llvm.ctlz.i256(i256 &lt;src&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5836</pre>
5837
5838<h5>Overview:</h5>
5839
5840<p>
5841The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
5842leading zeros in a variable.
5843</p>
5844
5845<h5>Arguments:</h5>
5846
5847<p>
5848The only argument is the value to be counted. The argument may be of any
5849integer type. The return type must match the argument type.
5850</p>
5851
5852<h5>Semantics:</h5>
5853
5854<p>
5855The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
5856in a variable. If the src == 0 then the result is the size in bits of the type
5857of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.
5858</p>
5859</div>
5860
5861
5862
5863<!-- _______________________________________________________________________ -->
5864<div class="doc_subsubsection">
5865 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
5866</div>
5867
5868<div class="doc_text">
5869
5870<h5>Syntax:</h5>
5871<p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]5872integer bit width. Not all targets support all bit widths however.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5873<pre>
Chandler Carrutha228e392007-08-04 01:51:18 +0000[diff] [blame]5874 declare i8 @llvm.cttz.i8 (i8 &lt;src&gt;)
5875 declare i16 @llvm.cttz.i16(i16 &lt;src&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5876 declare i32 @llvm.cttz.i32(i32 &lt;src&gt;)
Chandler Carrutha228e392007-08-04 01:51:18 +0000[diff] [blame]5877 declare i64 @llvm.cttz.i64(i64 &lt;src&gt;)
5878 declare i256 @llvm.cttz.i256(i256 &lt;src&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5879</pre>
5880
5881<h5>Overview:</h5>
5882
5883<p>
5884The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
5885trailing zeros.
5886</p>
5887
5888<h5>Arguments:</h5>
5889
5890<p>
5891The only argument is the value to be counted. The argument may be of any
5892integer type. The return type must match the argument type.
5893</p>
5894
5895<h5>Semantics:</h5>
5896
5897<p>
5898The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
5899in a variable. If the src == 0 then the result is the size in bits of the type
5900of src. For example, <tt>llvm.cttz(2) = 1</tt>.
5901</p>
5902</div>
5903
5904<!-- _______________________________________________________________________ -->
5905<div class="doc_subsubsection">
5906 <a name="int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic</a>
5907</div>
5908
5909<div class="doc_text">
5910
5911<h5>Syntax:</h5>
5912<p>This is an overloaded intrinsic. You can use <tt>llvm.part.select</tt>
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]5913on any integer bit width.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5914<pre>
Chandler Carrutha228e392007-08-04 01:51:18 +0000[diff] [blame]5915 declare i17 @llvm.part.select.i17 (i17 %val, i32 %loBit, i32 %hiBit)
5916 declare i29 @llvm.part.select.i29 (i29 %val, i32 %loBit, i32 %hiBit)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5917</pre>
5918
5919<h5>Overview:</h5>
5920<p>The '<tt>llvm.part.select</tt>' family of intrinsic functions selects a
5921range of bits from an integer value and returns them in the same bit width as
5922the original value.</p>
5923
5924<h5>Arguments:</h5>
5925<p>The first argument, <tt>%val</tt> and the result may be integer types of
5926any bit width but they must have the same bit width. The second and third
5927arguments must be <tt>i32</tt> type since they specify only a bit index.</p>
5928
5929<h5>Semantics:</h5>
5930<p>The operation of the '<tt>llvm.part.select</tt>' intrinsic has two modes
5931of operation: forwards and reverse. If <tt>%loBit</tt> is greater than
5932<tt>%hiBits</tt> then the intrinsic operates in reverse mode. Otherwise it
5933operates in forward mode.</p>
5934<p>In forward mode, this intrinsic is the equivalent of shifting <tt>%val</tt>
5935right by <tt>%loBit</tt> bits and then ANDing it with a mask with
5936only the <tt>%hiBit - %loBit</tt> bits set, as follows:</p>
5937<ol>
5938 <li>The <tt>%val</tt> is shifted right (LSHR) by the number of bits specified
5939 by <tt>%loBits</tt>. This normalizes the value to the low order bits.</li>
5940 <li>The <tt>%loBits</tt> value is subtracted from the <tt>%hiBits</tt> value
5941 to determine the number of bits to retain.</li>
5942 <li>A mask of the retained bits is created by shifting a -1 value.</li>
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]5943 <li>The mask is ANDed with <tt>%val</tt> to produce the result.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5944</ol>
5945<p>In reverse mode, a similar computation is made except that the bits are
5946returned in the reverse order. So, for example, if <tt>X</tt> has the value
5947<tt>i16 0x0ACF (101011001111)</tt> and we apply
5948<tt>part.select(i16 X, 8, 3)</tt> to it, we get back the value
5949<tt>i16 0x0026 (000000100110)</tt>.</p>
5950</div>
5951
5952<div class="doc_subsubsection">
5953 <a name="int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic</a>
5954</div>
5955
5956<div class="doc_text">
5957
5958<h5>Syntax:</h5>
5959<p>This is an overloaded intrinsic. You can use <tt>llvm.part.set</tt>
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]5960on any integer bit width.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5961<pre>
Chandler Carrutha228e392007-08-04 01:51:18 +0000[diff] [blame]5962 declare i17 @llvm.part.set.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
5963 declare i29 @llvm.part.set.i29.i9 (i29 %val, i9 %repl, i32 %lo, i32 %hi)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5964</pre>
5965
5966<h5>Overview:</h5>
5967<p>The '<tt>llvm.part.set</tt>' family of intrinsic functions replaces a range
5968of bits in an integer value with another integer value. It returns the integer
5969with the replaced bits.</p>
5970
5971<h5>Arguments:</h5>
Bill Wendling3f8cebe2009-02-08 01:40:31 +0000[diff] [blame]5972<p>The first argument, <tt>%val</tt>, and the result may be integer types of
5973any bit width, but they must have the same bit width. <tt>%val</tt> is the value
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5974whose bits will be replaced. The second argument, <tt>%repl</tt> may be an
5975integer of any bit width. The third and fourth arguments must be <tt>i32</tt>
5976type since they specify only a bit index.</p>
5977
5978<h5>Semantics:</h5>
5979<p>The operation of the '<tt>llvm.part.set</tt>' intrinsic has two modes
5980of operation: forwards and reverse. If <tt>%lo</tt> is greater than
5981<tt>%hi</tt> then the intrinsic operates in reverse mode. Otherwise it
5982operates in forward mode.</p>
Bill Wendling3f8cebe2009-02-08 01:40:31 +0000[diff] [blame]5983
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5984<p>For both modes, the <tt>%repl</tt> value is prepared for use by either
5985truncating it down to the size of the replacement area or zero extending it
5986up to that size.</p>
Bill Wendling3f8cebe2009-02-08 01:40:31 +0000[diff] [blame]5987
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5988<p>In forward mode, the bits between <tt>%lo</tt> and <tt>%hi</tt> (inclusive)
5989are replaced with corresponding bits from <tt>%repl</tt>. That is the 0th bit
5990in <tt>%repl</tt> replaces the <tt>%lo</tt>th bit in <tt>%val</tt> and etc. up
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]5991to the <tt>%hi</tt>th bit.</p>
Bill Wendling3f8cebe2009-02-08 01:40:31 +0000[diff] [blame]5992
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5993<p>In reverse mode, a similar computation is made except that the bits are
5994reversed. That is, the <tt>0</tt>th bit in <tt>%repl</tt> replaces the
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]5995<tt>%hi</tt> bit in <tt>%val</tt> and etc. down to the <tt>%lo</tt>th bit.</p>
Bill Wendling3f8cebe2009-02-08 01:40:31 +0000[diff] [blame]5996
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5997<h5>Examples:</h5>
Bill Wendling3f8cebe2009-02-08 01:40:31 +0000[diff] [blame]5998
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5999<pre>
6000 llvm.part.set(0xFFFF, 0, 4, 7) -&gt; 0xFF0F
6001 llvm.part.set(0xFFFF, 0, 7, 4) -&gt; 0xFF0F
6002 llvm.part.set(0xFFFF, 1, 7, 4) -&gt; 0xFF8F
6003 llvm.part.set(0xFFFF, F, 8, 3) -&gt; 0xFFE7
6004 llvm.part.set(0xFFFF, 0, 3, 8) -&gt; 0xFE07
6005</pre>
Bill Wendling3f8cebe2009-02-08 01:40:31 +0000[diff] [blame]6006
6007</div>
6008
Bill Wendling3e1258b2009-02-08 04:04:40 +0000[diff] [blame]6009<!-- ======================================================================= -->
6010<div class="doc_subsection">
6011 <a name="int_overflow">Arithmetic with Overflow Intrinsics</a>
6012</div>
6013
6014<div class="doc_text">
6015<p>
6016LLVM provides intrinsics for some arithmetic with overflow operations.
6017</p>
6018
6019</div>
6020
Bill Wendling3f8cebe2009-02-08 01:40:31 +0000[diff] [blame]6021<!-- _______________________________________________________________________ -->
6022<div class="doc_subsubsection">
Bill Wendling3e1258b2009-02-08 04:04:40 +0000[diff] [blame]6023 <a name="int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt>' Intrinsics</a>
Bill Wendling3f8cebe2009-02-08 01:40:31 +0000[diff] [blame]6024</div>
6025
6026<div class="doc_text">
6027
6028<h5>Syntax:</h5>
6029
6030<p>This is an overloaded intrinsic. You can use <tt>llvm.sadd.with.overflow</tt>
Bill Wendling3e1258b2009-02-08 04:04:40 +0000[diff] [blame]6031on any integer bit width.</p>
Bill Wendling3f8cebe2009-02-08 01:40:31 +0000[diff] [blame]6032
6033<pre>
6034 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
6035 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6036 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
6037</pre>
6038
6039<h5>Overview:</h5>
6040
6041<p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
6042a signed addition of the two arguments, and indicate whether an overflow
6043occurred during the signed summation.</p>
6044
6045<h5>Arguments:</h5>
6046
6047<p>The arguments (%a and %b) and the first element of the result structure may
6048be of integer types of any bit width, but they must have the same bit width. The
6049second element of the result structure must be of type <tt>i1</tt>. <tt>%a</tt>
6050and <tt>%b</tt> are the two values that will undergo signed addition.</p>
6051
6052<h5>Semantics:</h5>
6053
6054<p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
6055a signed addition of the two variables. They return a structure &mdash; the
6056first element of which is the signed summation, and the second element of which
6057is a bit specifying if the signed summation resulted in an overflow.</p>
6058
6059<h5>Examples:</h5>
6060<pre>
6061 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6062 %sum = extractvalue {i32, i1} %res, 0
6063 %obit = extractvalue {i32, i1} %res, 1
6064 br i1 %obit, label %overflow, label %normal
6065</pre>
6066
6067</div>
6068
6069<!-- _______________________________________________________________________ -->
6070<div class="doc_subsubsection">
Bill Wendling3e1258b2009-02-08 04:04:40 +0000[diff] [blame]6071 <a name="int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt>' Intrinsics</a>
Bill Wendling3f8cebe2009-02-08 01:40:31 +0000[diff] [blame]6072</div>
6073
6074<div class="doc_text">
6075
6076<h5>Syntax:</h5>
6077
6078<p>This is an overloaded intrinsic. You can use <tt>llvm.uadd.with.overflow</tt>
Bill Wendling3e1258b2009-02-08 04:04:40 +0000[diff] [blame]6079on any integer bit width.</p>
Bill Wendling3f8cebe2009-02-08 01:40:31 +0000[diff] [blame]6080
6081<pre>
6082 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
6083 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6084 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
6085</pre>
6086
6087<h5>Overview:</h5>
6088
6089<p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
6090an unsigned addition of the two arguments, and indicate whether a carry occurred
6091during the unsigned summation.</p>
6092
6093<h5>Arguments:</h5>
6094
6095<p>The arguments (%a and %b) and the first element of the result structure may
6096be of integer types of any bit width, but they must have the same bit width. The
6097second element of the result structure must be of type <tt>i1</tt>. <tt>%a</tt>
6098and <tt>%b</tt> are the two values that will undergo unsigned addition.</p>
6099
6100<h5>Semantics:</h5>
6101
6102<p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
6103an unsigned addition of the two arguments. They return a structure &mdash; the
6104first element of which is the sum, and the second element of which is a bit
6105specifying if the unsigned summation resulted in a carry.</p>
6106
6107<h5>Examples:</h5>
6108<pre>
6109 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6110 %sum = extractvalue {i32, i1} %res, 0
6111 %obit = extractvalue {i32, i1} %res, 1
6112 br i1 %obit, label %carry, label %normal
6113</pre>
6114
6115</div>
6116
6117<!-- _______________________________________________________________________ -->
6118<div class="doc_subsubsection">
Bill Wendling3e1258b2009-02-08 04:04:40 +0000[diff] [blame]6119 <a name="int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt>' Intrinsics</a>
Bill Wendling3f8cebe2009-02-08 01:40:31 +0000[diff] [blame]6120</div>
6121
6122<div class="doc_text">
6123
6124<h5>Syntax:</h5>
6125
6126<p>This is an overloaded intrinsic. You can use <tt>llvm.ssub.with.overflow</tt>
Bill Wendling3e1258b2009-02-08 04:04:40 +0000[diff] [blame]6127on any integer bit width.</p>
Bill Wendling3f8cebe2009-02-08 01:40:31 +0000[diff] [blame]6128
6129<pre>
6130 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
6131 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6132 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
6133</pre>
6134
6135<h5>Overview:</h5>
6136
6137<p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
6138a signed subtraction of the two arguments, and indicate whether an overflow
6139occurred during the signed subtraction.</p>
6140
6141<h5>Arguments:</h5>
6142
6143<p>The arguments (%a and %b) and the first element of the result structure may
6144be of integer types of any bit width, but they must have the same bit width. The
6145second element of the result structure must be of type <tt>i1</tt>. <tt>%a</tt>
6146and <tt>%b</tt> are the two values that will undergo signed subtraction.</p>
6147
6148<h5>Semantics:</h5>
6149
6150<p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
6151a signed subtraction of the two arguments. They return a structure &mdash; the
6152first element of which is the subtraction, and the second element of which is a bit
6153specifying if the signed subtraction resulted in an overflow.</p>
6154
6155<h5>Examples:</h5>
6156<pre>
6157 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6158 %sum = extractvalue {i32, i1} %res, 0
6159 %obit = extractvalue {i32, i1} %res, 1
6160 br i1 %obit, label %overflow, label %normal
6161</pre>
6162
6163</div>
6164
6165<!-- _______________________________________________________________________ -->
6166<div class="doc_subsubsection">
Bill Wendling3e1258b2009-02-08 04:04:40 +0000[diff] [blame]6167 <a name="int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt>' Intrinsics</a>
Bill Wendling3f8cebe2009-02-08 01:40:31 +0000[diff] [blame]6168</div>
6169
6170<div class="doc_text">
6171
6172<h5>Syntax:</h5>
6173
6174<p>This is an overloaded intrinsic. You can use <tt>llvm.usub.with.overflow</tt>
Bill Wendling3e1258b2009-02-08 04:04:40 +0000[diff] [blame]6175on any integer bit width.</p>
Bill Wendling3f8cebe2009-02-08 01:40:31 +0000[diff] [blame]6176
6177<pre>
6178 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
6179 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6180 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
6181</pre>
6182
6183<h5>Overview:</h5>
6184
6185<p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
6186an unsigned subtraction of the two arguments, and indicate whether an overflow
6187occurred during the unsigned subtraction.</p>
6188
6189<h5>Arguments:</h5>
6190
6191<p>The arguments (%a and %b) and the first element of the result structure may
6192be of integer types of any bit width, but they must have the same bit width. The
6193second element of the result structure must be of type <tt>i1</tt>. <tt>%a</tt>
6194and <tt>%b</tt> are the two values that will undergo unsigned subtraction.</p>
6195
6196<h5>Semantics:</h5>
6197
6198<p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
6199an unsigned subtraction of the two arguments. They return a structure &mdash; the
6200first element of which is the subtraction, and the second element of which is a bit
6201specifying if the unsigned subtraction resulted in an overflow.</p>
6202
6203<h5>Examples:</h5>
6204<pre>
6205 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6206 %sum = extractvalue {i32, i1} %res, 0
6207 %obit = extractvalue {i32, i1} %res, 1
6208 br i1 %obit, label %overflow, label %normal
6209</pre>
6210
6211</div>
6212
6213<!-- _______________________________________________________________________ -->
6214<div class="doc_subsubsection">
Bill Wendling3e1258b2009-02-08 04:04:40 +0000[diff] [blame]6215 <a name="int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt>' Intrinsics</a>
Bill Wendling3f8cebe2009-02-08 01:40:31 +0000[diff] [blame]6216</div>
6217
6218<div class="doc_text">
6219
6220<h5>Syntax:</h5>
6221
6222<p>This is an overloaded intrinsic. You can use <tt>llvm.smul.with.overflow</tt>
Bill Wendling3e1258b2009-02-08 04:04:40 +0000[diff] [blame]6223on any integer bit width.</p>
Bill Wendling3f8cebe2009-02-08 01:40:31 +0000[diff] [blame]6224
6225<pre>
6226 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
6227 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6228 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
6229</pre>
6230
6231<h5>Overview:</h5>
6232
6233<p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
6234a signed multiplication of the two arguments, and indicate whether an overflow
6235occurred during the signed multiplication.</p>
6236
6237<h5>Arguments:</h5>
6238
6239<p>The arguments (%a and %b) and the first element of the result structure may
6240be of integer types of any bit width, but they must have the same bit width. The
6241second element of the result structure must be of type <tt>i1</tt>. <tt>%a</tt>
6242and <tt>%b</tt> are the two values that will undergo signed multiplication.</p>
6243
6244<h5>Semantics:</h5>
6245
6246<p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
6247a signed multiplication of the two arguments. They return a structure &mdash;
6248the first element of which is the multiplication, and the second element of
6249which is a bit specifying if the signed multiplication resulted in an
6250overflow.</p>
6251
6252<h5>Examples:</h5>
6253<pre>
6254 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6255 %sum = extractvalue {i32, i1} %res, 0
6256 %obit = extractvalue {i32, i1} %res, 1
6257 br i1 %obit, label %overflow, label %normal
6258</pre>
6259
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]6260</div>
6261
Bill Wendlingbda98b62009-02-08 23:00:09 +0000[diff] [blame]6262<!-- _______________________________________________________________________ -->
6263<div class="doc_subsubsection">
6264 <a name="int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt>' Intrinsics</a>
6265</div>
6266
6267<div class="doc_text">
6268
6269<h5>Syntax:</h5>
6270
6271<p>This is an overloaded intrinsic. You can use <tt>llvm.umul.with.overflow</tt>
6272on any integer bit width.</p>
6273
6274<pre>
6275 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
6276 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6277 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
6278</pre>
6279
6280<h5>Overview:</h5>
6281
6282<p><i><b>Warning:</b> '<tt>llvm.umul.with.overflow</tt>' is badly broken. It is
6283actively being fixed, but it should not currently be used!</i></p>
6284
6285<p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
6286a unsigned multiplication of the two arguments, and indicate whether an overflow
6287occurred during the unsigned multiplication.</p>
6288
6289<h5>Arguments:</h5>
6290
6291<p>The arguments (%a and %b) and the first element of the result structure may
6292be of integer types of any bit width, but they must have the same bit width. The
6293second element of the result structure must be of type <tt>i1</tt>. <tt>%a</tt>
6294and <tt>%b</tt> are the two values that will undergo unsigned
6295multiplication.</p>
6296
6297<h5>Semantics:</h5>
6298
6299<p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
6300an unsigned multiplication of the two arguments. They return a structure &mdash;
6301the first element of which is the multiplication, and the second element of
6302which is a bit specifying if the unsigned multiplication resulted in an
6303overflow.</p>
6304
6305<h5>Examples:</h5>
6306<pre>
6307 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6308 %sum = extractvalue {i32, i1} %res, 0
6309 %obit = extractvalue {i32, i1} %res, 1
6310 br i1 %obit, label %overflow, label %normal
6311</pre>
6312
6313</div>
6314
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]6315<!-- ======================================================================= -->
6316<div class="doc_subsection">
6317 <a name="int_debugger">Debugger Intrinsics</a>
6318</div>
6319
6320<div class="doc_text">
6321<p>
6322The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
6323are described in the <a
6324href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
6325Debugging</a> document.
6326</p>
6327</div>
6328
6329
6330<!-- ======================================================================= -->
6331<div class="doc_subsection">
6332 <a name="int_eh">Exception Handling Intrinsics</a>
6333</div>
6334
6335<div class="doc_text">
6336<p> The LLVM exception handling intrinsics (which all start with
6337<tt>llvm.eh.</tt> prefix), are described in the <a
6338href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
6339Handling</a> document. </p>
6340</div>
6341
6342<!-- ======================================================================= -->
6343<div class="doc_subsection">
Duncan Sands7407a9f2007-09-11 14:10:23 +0000[diff] [blame]6344 <a name="int_trampoline">Trampoline Intrinsic</a>
Duncan Sands38947cd2007-07-27 12:58:54 +0000[diff] [blame]6345</div>
6346
6347<div class="doc_text">
6348<p>
Duncan Sands7407a9f2007-09-11 14:10:23 +0000[diff] [blame]6349 This intrinsic makes it possible to excise one parameter, marked with
Duncan Sands38947cd2007-07-27 12:58:54 +0000[diff] [blame]6350 the <tt>nest</tt> attribute, from a function. The result is a callable
6351 function pointer lacking the nest parameter - the caller does not need
6352 to provide a value for it. Instead, the value to use is stored in
6353 advance in a "trampoline", a block of memory usually allocated
6354 on the stack, which also contains code to splice the nest value into the
6355 argument list. This is used to implement the GCC nested function address
6356 extension.
6357</p>
6358<p>
6359 For example, if the function is
6360 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
Bill Wendlinge262b4c2007-09-22 09:23:55 +0000[diff] [blame]6361 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as follows:</p>
Duncan Sands38947cd2007-07-27 12:58:54 +0000[diff] [blame]6362<pre>
Duncan Sands7407a9f2007-09-11 14:10:23 +0000[diff] [blame]6363 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
6364 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
6365 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
6366 %fp = bitcast i8* %p to i32 (i32, i32)*
Duncan Sands38947cd2007-07-27 12:58:54 +0000[diff] [blame]6367</pre>
Bill Wendlinge262b4c2007-09-22 09:23:55 +0000[diff] [blame]6368 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
6369 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
Duncan Sands38947cd2007-07-27 12:58:54 +0000[diff] [blame]6370</div>
6371
6372<!-- _______________________________________________________________________ -->
6373<div class="doc_subsubsection">
6374 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
6375</div>
6376<div class="doc_text">
6377<h5>Syntax:</h5>
6378<pre>
Duncan Sands7407a9f2007-09-11 14:10:23 +0000[diff] [blame]6379declare i8* @llvm.init.trampoline(i8* &lt;tramp&gt;, i8* &lt;func&gt;, i8* &lt;nval&gt;)
Duncan Sands38947cd2007-07-27 12:58:54 +0000[diff] [blame]6380</pre>
6381<h5>Overview:</h5>
6382<p>
Duncan Sands7407a9f2007-09-11 14:10:23 +0000[diff] [blame]6383 This fills the memory pointed to by <tt>tramp</tt> with code
6384 and returns a function pointer suitable for executing it.
Duncan Sands38947cd2007-07-27 12:58:54 +0000[diff] [blame]6385</p>
6386<h5>Arguments:</h5>
6387<p>
6388 The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
6389 pointers. The <tt>tramp</tt> argument must point to a sufficiently large
6390 and sufficiently aligned block of memory; this memory is written to by the
Duncan Sands35012212007-08-22 23:39:54 +0000[diff] [blame]6391 intrinsic. Note that the size and the alignment are target-specific - LLVM
6392 currently provides no portable way of determining them, so a front-end that
6393 generates this intrinsic needs to have some target-specific knowledge.
6394 The <tt>func</tt> argument must hold a function bitcast to an <tt>i8*</tt>.
Duncan Sands38947cd2007-07-27 12:58:54 +0000[diff] [blame]6395</p>
6396<h5>Semantics:</h5>
6397<p>
6398 The block of memory pointed to by <tt>tramp</tt> is filled with target
Duncan Sands7407a9f2007-09-11 14:10:23 +0000[diff] [blame]6399 dependent code, turning it into a function. A pointer to this function is
6400 returned, but needs to be bitcast to an
Duncan Sands38947cd2007-07-27 12:58:54 +0000[diff] [blame]6401 <a href="#int_trampoline">appropriate function pointer type</a>
Duncan Sands7407a9f2007-09-11 14:10:23 +0000[diff] [blame]6402 before being called. The new function's signature is the same as that of
6403 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
6404 removed. At most one such <tt>nest</tt> argument is allowed, and it must be
6405 of pointer type. Calling the new function is equivalent to calling
6406 <tt>func</tt> with the same argument list, but with <tt>nval</tt> used for the
6407 missing <tt>nest</tt> argument. If, after calling
6408 <tt>llvm.init.trampoline</tt>, the memory pointed to by <tt>tramp</tt> is
6409 modified, then the effect of any later call to the returned function pointer is
6410 undefined.
Duncan Sands38947cd2007-07-27 12:58:54 +0000[diff] [blame]6411</p>
6412</div>
6413
6414<!-- ======================================================================= -->
6415<div class="doc_subsection">
Andrew Lenharth785610d2008-02-16 01:24:58 +0000[diff] [blame]6416 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
6417</div>
6418
6419<div class="doc_text">
6420<p>
6421 These intrinsic functions expand the "universal IR" of LLVM to represent
6422 hardware constructs for atomic operations and memory synchronization. This
6423 provides an interface to the hardware, not an interface to the programmer. It
Chris Lattner96451482008-08-05 18:29:16 +0000[diff] [blame]6424 is aimed at a low enough level to allow any programming models or APIs
6425 (Application Programming Interfaces) which
Andrew Lenharth785610d2008-02-16 01:24:58 +0000[diff] [blame]6426 need atomic behaviors to map cleanly onto it. It is also modeled primarily on
6427 hardware behavior. Just as hardware provides a "universal IR" for source
6428 languages, it also provides a starting point for developing a "universal"
6429 atomic operation and synchronization IR.
6430</p>
6431<p>
6432 These do <em>not</em> form an API such as high-level threading libraries,
6433 software transaction memory systems, atomic primitives, and intrinsic
6434 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
6435 application libraries. The hardware interface provided by LLVM should allow
6436 a clean implementation of all of these APIs and parallel programming models.
6437 No one model or paradigm should be selected above others unless the hardware
6438 itself ubiquitously does so.
6439
6440</p>
6441</div>
6442
6443<!-- _______________________________________________________________________ -->
6444<div class="doc_subsubsection">
6445 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
6446</div>
6447<div class="doc_text">
6448<h5>Syntax:</h5>
6449<pre>
6450declare void @llvm.memory.barrier( i1 &lt;ll&gt;, i1 &lt;ls&gt;, i1 &lt;sl&gt;, i1 &lt;ss&gt;,
6451i1 &lt;device&gt; )
6452
6453</pre>
6454<h5>Overview:</h5>
6455<p>
6456 The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
6457 specific pairs of memory access types.
6458</p>
6459<h5>Arguments:</h5>
6460<p>
6461 The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
6462 The first four arguments enables a specific barrier as listed below. The fith
6463 argument specifies that the barrier applies to io or device or uncached memory.
6464
6465</p>
6466 <ul>
6467 <li><tt>ll</tt>: load-load barrier</li>
6468 <li><tt>ls</tt>: load-store barrier</li>
6469 <li><tt>sl</tt>: store-load barrier</li>
6470 <li><tt>ss</tt>: store-store barrier</li>
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]6471 <li><tt>device</tt>: barrier applies to device and uncached memory also.</li>
Andrew Lenharth785610d2008-02-16 01:24:58 +0000[diff] [blame]6472 </ul>
6473<h5>Semantics:</h5>
6474<p>
6475 This intrinsic causes the system to enforce some ordering constraints upon
6476 the loads and stores of the program. This barrier does not indicate
6477 <em>when</em> any events will occur, it only enforces an <em>order</em> in
6478 which they occur. For any of the specified pairs of load and store operations
6479 (f.ex. load-load, or store-load), all of the first operations preceding the
6480 barrier will complete before any of the second operations succeeding the
6481 barrier begin. Specifically the semantics for each pairing is as follows:
6482</p>
6483 <ul>
6484 <li><tt>ll</tt>: All loads before the barrier must complete before any load
6485 after the barrier begins.</li>
6486
6487 <li><tt>ls</tt>: All loads before the barrier must complete before any
6488 store after the barrier begins.</li>
6489 <li><tt>ss</tt>: All stores before the barrier must complete before any
6490 store after the barrier begins.</li>
6491 <li><tt>sl</tt>: All stores before the barrier must complete before any
6492 load after the barrier begins.</li>
6493 </ul>
6494<p>
6495 These semantics are applied with a logical "and" behavior when more than one
6496 is enabled in a single memory barrier intrinsic.
6497</p>
6498<p>
6499 Backends may implement stronger barriers than those requested when they do not
6500 support as fine grained a barrier as requested. Some architectures do not
6501 need all types of barriers and on such architectures, these become noops.
6502</p>
6503<h5>Example:</h5>
6504<pre>
6505%ptr = malloc i32
6506 store i32 4, %ptr
6507
6508%result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
6509 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
6510 <i>; guarantee the above finishes</i>
6511 store i32 8, %ptr <i>; before this begins</i>
6512</pre>
6513</div>
6514
Andrew Lenharthe44f3902008-02-21 06:45:13 +0000[diff] [blame]6515<!-- _______________________________________________________________________ -->
6516<div class="doc_subsubsection">
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6517 <a name="int_atomic_cmp_swap">'<tt>llvm.atomic.cmp.swap.*</tt>' Intrinsic</a>
Andrew Lenharthe44f3902008-02-21 06:45:13 +0000[diff] [blame]6518</div>
6519<div class="doc_text">
6520<h5>Syntax:</h5>
6521<p>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6522 This is an overloaded intrinsic. You can use <tt>llvm.atomic.cmp.swap</tt> on
6523 any integer bit width and for different address spaces. Not all targets
6524 support all bit widths however.</p>
Andrew Lenharthe44f3902008-02-21 06:45:13 +0000[diff] [blame]6525
6526<pre>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6527declare i8 @llvm.atomic.cmp.swap.i8.p0i8( i8* &lt;ptr&gt;, i8 &lt;cmp&gt;, i8 &lt;val&gt; )
6528declare i16 @llvm.atomic.cmp.swap.i16.p0i16( i16* &lt;ptr&gt;, i16 &lt;cmp&gt;, i16 &lt;val&gt; )
6529declare i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* &lt;ptr&gt;, i32 &lt;cmp&gt;, i32 &lt;val&gt; )
6530declare i64 @llvm.atomic.cmp.swap.i64.p0i64( i64* &lt;ptr&gt;, i64 &lt;cmp&gt;, i64 &lt;val&gt; )
Andrew Lenharthe44f3902008-02-21 06:45:13 +0000[diff] [blame]6531
6532</pre>
6533<h5>Overview:</h5>
6534<p>
6535 This loads a value in memory and compares it to a given value. If they are
6536 equal, it stores a new value into the memory.
6537</p>
6538<h5>Arguments:</h5>
6539<p>
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6540 The <tt>llvm.atomic.cmp.swap</tt> intrinsic takes three arguments. The result as
Andrew Lenharthe44f3902008-02-21 06:45:13 +0000[diff] [blame]6541 well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
6542 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
6543 this integer type. While any bit width integer may be used, targets may only
6544 lower representations they support in hardware.
6545
6546</p>
6547<h5>Semantics:</h5>
6548<p>
6549 This entire intrinsic must be executed atomically. It first loads the value
6550 in memory pointed to by <tt>ptr</tt> and compares it with the value
6551 <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the memory. The
6552 loaded value is yielded in all cases. This provides the equivalent of an
6553 atomic compare-and-swap operation within the SSA framework.
6554</p>
6555<h5>Examples:</h5>
6556
6557<pre>
6558%ptr = malloc i32
6559 store i32 4, %ptr
6560
6561%val1 = add i32 4, 4
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6562%result1 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 4, %val1 )
Andrew Lenharthe44f3902008-02-21 06:45:13 +0000[diff] [blame]6563 <i>; yields {i32}:result1 = 4</i>
6564%stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
6565%memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
6566
6567%val2 = add i32 1, 1
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6568%result2 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 5, %val2 )
Andrew Lenharthe44f3902008-02-21 06:45:13 +0000[diff] [blame]6569 <i>; yields {i32}:result2 = 8</i>
6570%stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
6571
6572%memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
6573</pre>
6574</div>
6575
6576<!-- _______________________________________________________________________ -->
6577<div class="doc_subsubsection">
6578 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
6579</div>
6580<div class="doc_text">
6581<h5>Syntax:</h5>
6582
6583<p>
6584 This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
6585 integer bit width. Not all targets support all bit widths however.</p>
6586<pre>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6587declare i8 @llvm.atomic.swap.i8.p0i8( i8* &lt;ptr&gt;, i8 &lt;val&gt; )
6588declare i16 @llvm.atomic.swap.i16.p0i16( i16* &lt;ptr&gt;, i16 &lt;val&gt; )
6589declare i32 @llvm.atomic.swap.i32.p0i32( i32* &lt;ptr&gt;, i32 &lt;val&gt; )
6590declare i64 @llvm.atomic.swap.i64.p0i64( i64* &lt;ptr&gt;, i64 &lt;val&gt; )
Andrew Lenharthe44f3902008-02-21 06:45:13 +0000[diff] [blame]6591
6592</pre>
6593<h5>Overview:</h5>
6594<p>
6595 This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
6596 the value from memory. It then stores the value in <tt>val</tt> in the memory
6597 at <tt>ptr</tt>.
6598</p>
6599<h5>Arguments:</h5>
6600
6601<p>
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6602 The <tt>llvm.atomic.swap</tt> intrinsic takes two arguments. Both the
Andrew Lenharthe44f3902008-02-21 06:45:13 +0000[diff] [blame]6603 <tt>val</tt> argument and the result must be integers of the same bit width.
6604 The first argument, <tt>ptr</tt>, must be a pointer to a value of this
6605 integer type. The targets may only lower integer representations they
6606 support.
6607</p>
6608<h5>Semantics:</h5>
6609<p>
6610 This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
6611 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
6612 equivalent of an atomic swap operation within the SSA framework.
6613
6614</p>
6615<h5>Examples:</h5>
6616<pre>
6617%ptr = malloc i32
6618 store i32 4, %ptr
6619
6620%val1 = add i32 4, 4
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6621%result1 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val1 )
Andrew Lenharthe44f3902008-02-21 06:45:13 +0000[diff] [blame]6622 <i>; yields {i32}:result1 = 4</i>
6623%stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
6624%memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
6625
6626%val2 = add i32 1, 1
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6627%result2 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val2 )
Andrew Lenharthe44f3902008-02-21 06:45:13 +0000[diff] [blame]6628 <i>; yields {i32}:result2 = 8</i>
6629
6630%stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
6631%memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
6632</pre>
6633</div>
6634
6635<!-- _______________________________________________________________________ -->
6636<div class="doc_subsubsection">
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6637 <a name="int_atomic_load_add">'<tt>llvm.atomic.load.add.*</tt>' Intrinsic</a>
Andrew Lenharthe44f3902008-02-21 06:45:13 +0000[diff] [blame]6638
6639</div>
6640<div class="doc_text">
6641<h5>Syntax:</h5>
6642<p>
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6643 This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.add</tt> on any
Andrew Lenharthe44f3902008-02-21 06:45:13 +0000[diff] [blame]6644 integer bit width. Not all targets support all bit widths however.</p>
6645<pre>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6646declare i8 @llvm.atomic.load.add.i8..p0i8( i8* &lt;ptr&gt;, i8 &lt;delta&gt; )
6647declare i16 @llvm.atomic.load.add.i16..p0i16( i16* &lt;ptr&gt;, i16 &lt;delta&gt; )
6648declare i32 @llvm.atomic.load.add.i32..p0i32( i32* &lt;ptr&gt;, i32 &lt;delta&gt; )
6649declare i64 @llvm.atomic.load.add.i64..p0i64( i64* &lt;ptr&gt;, i64 &lt;delta&gt; )
Andrew Lenharthe44f3902008-02-21 06:45:13 +0000[diff] [blame]6650
6651</pre>
6652<h5>Overview:</h5>
6653<p>
6654 This intrinsic adds <tt>delta</tt> to the value stored in memory at
6655 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
6656</p>
6657<h5>Arguments:</h5>
6658<p>
6659
6660 The intrinsic takes two arguments, the first a pointer to an integer value
6661 and the second an integer value. The result is also an integer value. These
6662 integer types can have any bit width, but they must all have the same bit
6663 width. The targets may only lower integer representations they support.
6664</p>
6665<h5>Semantics:</h5>
6666<p>
6667 This intrinsic does a series of operations atomically. It first loads the
6668 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
6669 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
6670</p>
6671
6672<h5>Examples:</h5>
6673<pre>
6674%ptr = malloc i32
6675 store i32 4, %ptr
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6676%result1 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 4 )
Andrew Lenharthe44f3902008-02-21 06:45:13 +0000[diff] [blame]6677 <i>; yields {i32}:result1 = 4</i>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6678%result2 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 2 )
Andrew Lenharthe44f3902008-02-21 06:45:13 +0000[diff] [blame]6679 <i>; yields {i32}:result2 = 8</i>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6680%result3 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 5 )
Andrew Lenharthe44f3902008-02-21 06:45:13 +0000[diff] [blame]6681 <i>; yields {i32}:result3 = 10</i>
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6682%memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
Andrew Lenharthe44f3902008-02-21 06:45:13 +0000[diff] [blame]6683</pre>
6684</div>
6685
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6686<!-- _______________________________________________________________________ -->
6687<div class="doc_subsubsection">
6688 <a name="int_atomic_load_sub">'<tt>llvm.atomic.load.sub.*</tt>' Intrinsic</a>
6689
6690</div>
6691<div class="doc_text">
6692<h5>Syntax:</h5>
6693<p>
6694 This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.sub</tt> on
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6695 any integer bit width and for different address spaces. Not all targets
6696 support all bit widths however.</p>
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6697<pre>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6698declare i8 @llvm.atomic.load.sub.i8.p0i32( i8* &lt;ptr&gt;, i8 &lt;delta&gt; )
6699declare i16 @llvm.atomic.load.sub.i16.p0i32( i16* &lt;ptr&gt;, i16 &lt;delta&gt; )
6700declare i32 @llvm.atomic.load.sub.i32.p0i32( i32* &lt;ptr&gt;, i32 &lt;delta&gt; )
6701declare i64 @llvm.atomic.load.sub.i64.p0i32( i64* &lt;ptr&gt;, i64 &lt;delta&gt; )
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6702
6703</pre>
6704<h5>Overview:</h5>
6705<p>
6706 This intrinsic subtracts <tt>delta</tt> to the value stored in memory at
6707 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
6708</p>
6709<h5>Arguments:</h5>
6710<p>
6711
6712 The intrinsic takes two arguments, the first a pointer to an integer value
6713 and the second an integer value. The result is also an integer value. These
6714 integer types can have any bit width, but they must all have the same bit
6715 width. The targets may only lower integer representations they support.
6716</p>
6717<h5>Semantics:</h5>
6718<p>
6719 This intrinsic does a series of operations atomically. It first loads the
6720 value stored at <tt>ptr</tt>. It then subtracts <tt>delta</tt>, stores the
6721 result to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
6722</p>
6723
6724<h5>Examples:</h5>
6725<pre>
6726%ptr = malloc i32
6727 store i32 8, %ptr
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6728%result1 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 4 )
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6729 <i>; yields {i32}:result1 = 8</i>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6730%result2 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 2 )
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6731 <i>; yields {i32}:result2 = 4</i>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6732%result3 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 5 )
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6733 <i>; yields {i32}:result3 = 2</i>
6734%memval1 = load i32* %ptr <i>; yields {i32}:memval1 = -3</i>
6735</pre>
6736</div>
6737
6738<!-- _______________________________________________________________________ -->
6739<div class="doc_subsubsection">
6740 <a name="int_atomic_load_and">'<tt>llvm.atomic.load.and.*</tt>' Intrinsic</a><br>
6741 <a name="int_atomic_load_nand">'<tt>llvm.atomic.load.nand.*</tt>' Intrinsic</a><br>
6742 <a name="int_atomic_load_or">'<tt>llvm.atomic.load.or.*</tt>' Intrinsic</a><br>
6743 <a name="int_atomic_load_xor">'<tt>llvm.atomic.load.xor.*</tt>' Intrinsic</a><br>
6744
6745</div>
6746<div class="doc_text">
6747<h5>Syntax:</h5>
6748<p>
6749 These are overloaded intrinsics. You can use <tt>llvm.atomic.load_and</tt>,
6750 <tt>llvm.atomic.load_nand</tt>, <tt>llvm.atomic.load_or</tt>, and
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6751 <tt>llvm.atomic.load_xor</tt> on any integer bit width and for different
6752 address spaces. Not all targets support all bit widths however.</p>
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6753<pre>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6754declare i8 @llvm.atomic.load.and.i8.p0i8( i8* &lt;ptr&gt;, i8 &lt;delta&gt; )
6755declare i16 @llvm.atomic.load.and.i16.p0i16( i16* &lt;ptr&gt;, i16 &lt;delta&gt; )
6756declare i32 @llvm.atomic.load.and.i32.p0i32( i32* &lt;ptr&gt;, i32 &lt;delta&gt; )
6757declare i64 @llvm.atomic.load.and.i64.p0i64( i64* &lt;ptr&gt;, i64 &lt;delta&gt; )
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6758
6759</pre>
6760
6761<pre>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6762declare i8 @llvm.atomic.load.or.i8.p0i8( i8* &lt;ptr&gt;, i8 &lt;delta&gt; )
6763declare i16 @llvm.atomic.load.or.i16.p0i16( i16* &lt;ptr&gt;, i16 &lt;delta&gt; )
6764declare i32 @llvm.atomic.load.or.i32.p0i32( i32* &lt;ptr&gt;, i32 &lt;delta&gt; )
6765declare i64 @llvm.atomic.load.or.i64.p0i64( i64* &lt;ptr&gt;, i64 &lt;delta&gt; )
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6766
6767</pre>
6768
6769<pre>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6770declare i8 @llvm.atomic.load.nand.i8.p0i32( i8* &lt;ptr&gt;, i8 &lt;delta&gt; )
6771declare i16 @llvm.atomic.load.nand.i16.p0i32( i16* &lt;ptr&gt;, i16 &lt;delta&gt; )
6772declare i32 @llvm.atomic.load.nand.i32.p0i32( i32* &lt;ptr&gt;, i32 &lt;delta&gt; )
6773declare i64 @llvm.atomic.load.nand.i64.p0i32( i64* &lt;ptr&gt;, i64 &lt;delta&gt; )
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6774
6775</pre>
6776
6777<pre>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6778declare i8 @llvm.atomic.load.xor.i8.p0i32( i8* &lt;ptr&gt;, i8 &lt;delta&gt; )
6779declare i16 @llvm.atomic.load.xor.i16.p0i32( i16* &lt;ptr&gt;, i16 &lt;delta&gt; )
6780declare i32 @llvm.atomic.load.xor.i32.p0i32( i32* &lt;ptr&gt;, i32 &lt;delta&gt; )
6781declare i64 @llvm.atomic.load.xor.i64.p0i32( i64* &lt;ptr&gt;, i64 &lt;delta&gt; )
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6782
6783</pre>
6784<h5>Overview:</h5>
6785<p>
6786 These intrinsics bitwise the operation (and, nand, or, xor) <tt>delta</tt> to
6787 the value stored in memory at <tt>ptr</tt>. It yields the original value
6788 at <tt>ptr</tt>.
6789</p>
6790<h5>Arguments:</h5>
6791<p>
6792
6793 These intrinsics take two arguments, the first a pointer to an integer value
6794 and the second an integer value. The result is also an integer value. These
6795 integer types can have any bit width, but they must all have the same bit
6796 width. The targets may only lower integer representations they support.
6797</p>
6798<h5>Semantics:</h5>
6799<p>
6800 These intrinsics does a series of operations atomically. They first load the
6801 value stored at <tt>ptr</tt>. They then do the bitwise operation
6802 <tt>delta</tt>, store the result to <tt>ptr</tt>. They yield the original
6803 value stored at <tt>ptr</tt>.
6804</p>
6805
6806<h5>Examples:</h5>
6807<pre>
6808%ptr = malloc i32
6809 store i32 0x0F0F, %ptr
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6810%result0 = call i32 @llvm.atomic.load.nand.i32.p0i32( i32* %ptr, i32 0xFF )
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6811 <i>; yields {i32}:result0 = 0x0F0F</i>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6812%result1 = call i32 @llvm.atomic.load.and.i32.p0i32( i32* %ptr, i32 0xFF )
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6813 <i>; yields {i32}:result1 = 0xFFFFFFF0</i>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6814%result2 = call i32 @llvm.atomic.load.or.i32.p0i32( i32* %ptr, i32 0F )
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6815 <i>; yields {i32}:result2 = 0xF0</i>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6816%result3 = call i32 @llvm.atomic.load.xor.i32.p0i32( i32* %ptr, i32 0F )
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6817 <i>; yields {i32}:result3 = FF</i>
6818%memval1 = load i32* %ptr <i>; yields {i32}:memval1 = F0</i>
6819</pre>
6820</div>
6821
6822
6823<!-- _______________________________________________________________________ -->
6824<div class="doc_subsubsection">
6825 <a name="int_atomic_load_max">'<tt>llvm.atomic.load.max.*</tt>' Intrinsic</a><br>
6826 <a name="int_atomic_load_min">'<tt>llvm.atomic.load.min.*</tt>' Intrinsic</a><br>
6827 <a name="int_atomic_load_umax">'<tt>llvm.atomic.load.umax.*</tt>' Intrinsic</a><br>
6828 <a name="int_atomic_load_umin">'<tt>llvm.atomic.load.umin.*</tt>' Intrinsic</a><br>
6829
6830</div>
6831<div class="doc_text">
6832<h5>Syntax:</h5>
6833<p>
6834 These are overloaded intrinsics. You can use <tt>llvm.atomic.load_max</tt>,
6835 <tt>llvm.atomic.load_min</tt>, <tt>llvm.atomic.load_umax</tt>, and
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6836 <tt>llvm.atomic.load_umin</tt> on any integer bit width and for different
6837 address spaces. Not all targets
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6838 support all bit widths however.</p>
6839<pre>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6840declare i8 @llvm.atomic.load.max.i8.p0i8( i8* &lt;ptr&gt;, i8 &lt;delta&gt; )
6841declare i16 @llvm.atomic.load.max.i16.p0i16( i16* &lt;ptr&gt;, i16 &lt;delta&gt; )
6842declare i32 @llvm.atomic.load.max.i32.p0i32( i32* &lt;ptr&gt;, i32 &lt;delta&gt; )
6843declare i64 @llvm.atomic.load.max.i64.p0i64( i64* &lt;ptr&gt;, i64 &lt;delta&gt; )
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6844
6845</pre>
6846
6847<pre>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6848declare i8 @llvm.atomic.load.min.i8.p0i8( i8* &lt;ptr&gt;, i8 &lt;delta&gt; )
6849declare i16 @llvm.atomic.load.min.i16.p0i16( i16* &lt;ptr&gt;, i16 &lt;delta&gt; )
6850declare i32 @llvm.atomic.load.min.i32..p0i32( i32* &lt;ptr&gt;, i32 &lt;delta&gt; )
6851declare i64 @llvm.atomic.load.min.i64..p0i64( i64* &lt;ptr&gt;, i64 &lt;delta&gt; )
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6852
6853</pre>
6854
6855<pre>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6856declare i8 @llvm.atomic.load.umax.i8.p0i8( i8* &lt;ptr&gt;, i8 &lt;delta&gt; )
6857declare i16 @llvm.atomic.load.umax.i16.p0i16( i16* &lt;ptr&gt;, i16 &lt;delta&gt; )
6858declare i32 @llvm.atomic.load.umax.i32.p0i32( i32* &lt;ptr&gt;, i32 &lt;delta&gt; )
6859declare i64 @llvm.atomic.load.umax.i64.p0i64( i64* &lt;ptr&gt;, i64 &lt;delta&gt; )
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6860
6861</pre>
6862
6863<pre>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6864declare i8 @llvm.atomic.load.umin.i8..p0i8( i8* &lt;ptr&gt;, i8 &lt;delta&gt; )
6865declare i16 @llvm.atomic.load.umin.i16.p0i16( i16* &lt;ptr&gt;, i16 &lt;delta&gt; )
6866declare i32 @llvm.atomic.load.umin.i32..p0i32( i32* &lt;ptr&gt;, i32 &lt;delta&gt; )
6867declare i64 @llvm.atomic.load.umin.i64..p0i64( i64* &lt;ptr&gt;, i64 &lt;delta&gt; )
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6868
6869</pre>
6870<h5>Overview:</h5>
6871<p>
6872 These intrinsics takes the signed or unsigned minimum or maximum of
6873 <tt>delta</tt> and the value stored in memory at <tt>ptr</tt>. It yields the
6874 original value at <tt>ptr</tt>.
6875</p>
6876<h5>Arguments:</h5>
6877<p>
6878
6879 These intrinsics take two arguments, the first a pointer to an integer value
6880 and the second an integer value. The result is also an integer value. These
6881 integer types can have any bit width, but they must all have the same bit
6882 width. The targets may only lower integer representations they support.
6883</p>
6884<h5>Semantics:</h5>
6885<p>
6886 These intrinsics does a series of operations atomically. They first load the
6887 value stored at <tt>ptr</tt>. They then do the signed or unsigned min or max
6888 <tt>delta</tt> and the value, store the result to <tt>ptr</tt>. They yield
6889 the original value stored at <tt>ptr</tt>.
6890</p>
6891
6892<h5>Examples:</h5>
6893<pre>
6894%ptr = malloc i32
6895 store i32 7, %ptr
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6896%result0 = call i32 @llvm.atomic.load.min.i32.p0i32( i32* %ptr, i32 -2 )
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6897 <i>; yields {i32}:result0 = 7</i>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6898%result1 = call i32 @llvm.atomic.load.max.i32.p0i32( i32* %ptr, i32 8 )
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6899 <i>; yields {i32}:result1 = -2</i>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6900%result2 = call i32 @llvm.atomic.load.umin.i32.p0i32( i32* %ptr, i32 10 )
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6901 <i>; yields {i32}:result2 = 8</i>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6902%result3 = call i32 @llvm.atomic.load.umax.i32.p0i32( i32* %ptr, i32 30 )
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6903 <i>; yields {i32}:result3 = 8</i>
6904%memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 30</i>
6905</pre>
6906</div>
Andrew Lenharth785610d2008-02-16 01:24:58 +0000[diff] [blame]6907
6908<!-- ======================================================================= -->
6909<div class="doc_subsection">
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]6910 <a name="int_general">General Intrinsics</a>
6911</div>
6912
6913<div class="doc_text">
6914<p> This class of intrinsics is designed to be generic and has
6915no specific purpose. </p>
6916</div>
6917
6918<!-- _______________________________________________________________________ -->
6919<div class="doc_subsubsection">
6920 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
6921</div>
6922
6923<div class="doc_text">
6924
6925<h5>Syntax:</h5>
6926<pre>
6927 declare void @llvm.var.annotation(i8* &lt;val&gt;, i8* &lt;str&gt;, i8* &lt;str&gt;, i32 &lt;int&gt; )
6928</pre>
6929
6930<h5>Overview:</h5>
6931
6932<p>
6933The '<tt>llvm.var.annotation</tt>' intrinsic
6934</p>
6935
6936<h5>Arguments:</h5>
6937
6938<p>
6939The first argument is a pointer to a value, the second is a pointer to a
6940global string, the third is a pointer to a global string which is the source
6941file name, and the last argument is the line number.
6942</p>
6943
6944<h5>Semantics:</h5>
6945
6946<p>
Anton Korobeynikove6e764f2008-01-15 22:31:34 +0000[diff] [blame]6947This intrinsic allows annotation of local variables with arbitrary strings.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]6948This can be useful for special purpose optimizations that want to look for these
Anton Korobeynikove6e764f2008-01-15 22:31:34 +0000[diff] [blame]6949annotations. These have no other defined use, they are ignored by code
6950generation and optimization.
6951</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]6952</div>
6953
Tanya Lattnerb306a9e2007-09-21 22:59:12 +0000[diff] [blame]6954<!-- _______________________________________________________________________ -->
6955<div class="doc_subsubsection">
Tanya Lattnerc9869b12007-09-21 23:57:59 +0000[diff] [blame]6956 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
Tanya Lattnerb306a9e2007-09-21 22:59:12 +0000[diff] [blame]6957</div>
6958
6959<div class="doc_text">
6960
6961<h5>Syntax:</h5>
Tanya Lattnere545be72007-09-21 23:56:27 +0000[diff] [blame]6962<p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
6963any integer bit width.
6964</p>
Tanya Lattnerb306a9e2007-09-21 22:59:12 +0000[diff] [blame]6965<pre>
Tanya Lattner09161fe2007-09-22 00:03:01 +0000[diff] [blame]6966 declare i8 @llvm.annotation.i8(i8 &lt;val&gt;, i8* &lt;str&gt;, i8* &lt;str&gt;, i32 &lt;int&gt; )
6967 declare i16 @llvm.annotation.i16(i16 &lt;val&gt;, i8* &lt;str&gt;, i8* &lt;str&gt;, i32 &lt;int&gt; )
6968 declare i32 @llvm.annotation.i32(i32 &lt;val&gt;, i8* &lt;str&gt;, i8* &lt;str&gt;, i32 &lt;int&gt; )
6969 declare i64 @llvm.annotation.i64(i64 &lt;val&gt;, i8* &lt;str&gt;, i8* &lt;str&gt;, i32 &lt;int&gt; )
6970 declare i256 @llvm.annotation.i256(i256 &lt;val&gt;, i8* &lt;str&gt;, i8* &lt;str&gt;, i32 &lt;int&gt; )
Tanya Lattnerb306a9e2007-09-21 22:59:12 +0000[diff] [blame]6971</pre>
6972
6973<h5>Overview:</h5>
Tanya Lattnere545be72007-09-21 23:56:27 +0000[diff] [blame]6974
6975<p>
6976The '<tt>llvm.annotation</tt>' intrinsic.
Tanya Lattnerb306a9e2007-09-21 22:59:12 +0000[diff] [blame]6977</p>
6978
6979<h5>Arguments:</h5>
6980
6981<p>
6982The first argument is an integer value (result of some expression),
6983the second is a pointer to a global string, the third is a pointer to a global
6984string which is the source file name, and the last argument is the line number.
Tanya Lattnere545be72007-09-21 23:56:27 +0000[diff] [blame]6985It returns the value of the first argument.
Tanya Lattnerb306a9e2007-09-21 22:59:12 +0000[diff] [blame]6986</p>
6987
6988<h5>Semantics:</h5>
6989
6990<p>
6991This intrinsic allows annotations to be put on arbitrary expressions
6992with arbitrary strings. This can be useful for special purpose optimizations
6993that want to look for these annotations. These have no other defined use, they
6994are ignored by code generation and optimization.
Dan Gohman2672f3e2008-10-14 16:51:45 +0000[diff] [blame]6995</p>
Tanya Lattnerb306a9e2007-09-21 22:59:12 +0000[diff] [blame]6996</div>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]6997
Anton Korobeynikove6e764f2008-01-15 22:31:34 +0000[diff] [blame]6998<!-- _______________________________________________________________________ -->
6999<div class="doc_subsubsection">
7000 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
7001</div>
7002
7003<div class="doc_text">
7004
7005<h5>Syntax:</h5>
7006<pre>
7007 declare void @llvm.trap()
7008</pre>
7009
7010<h5>Overview:</h5>
7011
7012<p>
7013The '<tt>llvm.trap</tt>' intrinsic
7014</p>
7015
7016<h5>Arguments:</h5>
7017
7018<p>
7019None
7020</p>
7021
7022<h5>Semantics:</h5>
7023
7024<p>
7025This intrinsics is lowered to the target dependent trap instruction. If the
7026target does not have a trap instruction, this intrinsic will be lowered to the
7027call of the abort() function.
7028</p>
7029</div>
7030
Bill Wendlinge4164592008-11-19 05:56:17 +0000[diff] [blame]7031<!-- _______________________________________________________________________ -->
7032<div class="doc_subsubsection">
Misha Brukman5dd7f4d2008-11-22 23:55:29 +0000[diff] [blame]7033 <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a>
Bill Wendlinge4164592008-11-19 05:56:17 +0000[diff] [blame]7034</div>
7035<div class="doc_text">
7036<h5>Syntax:</h5>
7037<pre>
7038declare void @llvm.stackprotector( i8* &lt;guard&gt;, i8** &lt;slot&gt; )
7039
7040</pre>
7041<h5>Overview:</h5>
7042<p>
7043 The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and stores
7044 it onto the stack at <tt>slot</tt>. The stack slot is adjusted to ensure that
7045 it is placed on the stack before local variables.
7046</p>
7047<h5>Arguments:</h5>
7048<p>
7049 The <tt>llvm.stackprotector</tt> intrinsic requires two pointer arguments. The
7050 first argument is the value loaded from the stack guard
7051 <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt> that
7052 has enough space to hold the value of the guard.
7053</p>
7054<h5>Semantics:</h5>
7055<p>
7056 This intrinsic causes the prologue/epilogue inserter to force the position of
7057 the <tt>AllocaInst</tt> stack slot to be before local variables on the
7058 stack. This is to ensure that if a local variable on the stack is overwritten,
7059 it will destroy the value of the guard. When the function exits, the guard on
7060 the stack is checked against the original guard. If they're different, then
7061 the program aborts by calling the <tt>__stack_chk_fail()</tt> function.
7062</p>
7063</div>
7064
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]7065<!-- *********************************************************************** -->
7066<hr>
7067<address>
7068 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
Misha Brukman947321d2008-12-11 17:34:48 +0000[diff] [blame]7069 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]7070 <a href="http://validator.w3.org/check/referer"><img
Misha Brukman947321d2008-12-11 17:34:48 +0000[diff] [blame]7071 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]7072
7073 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
7074 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
7075 Last modified: $Date$
7076</address>
Chris Lattner08497ce2008-01-04 04:33:49 +0000[diff] [blame]7077
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]7078</body>
7079</html>