Gitiles
Code Review Sign In
gerrit-public.fairphone.software / fp2-dev / platform / external / llvm / 2c4e05a903d0df0353de5ab29bb315de55698c3e / . / docs / LangRef.html
blob: ad8ee8e174fd32d551fd89df6de81f4ef0f11e62 [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>
25 <li><a href="#globalvars">Global Variables</a></li>
26 <li><a href="#functionstructure">Functions</a></li>
27 <li><a href="#aliasstructure">Aliases</a>
28 <li><a href="#paramattrs">Parameter Attributes</a></li>
Devang Patel008cd3e2008-09-26 23:51:19 +0000[diff] [blame]29 <li><a href="#fnattrs">Function Attributes</a></li>
Gordon Henriksen13fe5e32007-12-10 03:18:06 +0000[diff] [blame]30 <li><a href="#gc">Garbage Collector Names</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]31 <li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
32 <li><a href="#datalayout">Data Layout</a></li>
33 </ol>
34 </li>
35 <li><a href="#typesystem">Type System</a>
36 <ol>
Chris Lattner488772f2008-01-04 04:32:38 +0000[diff] [blame]37 <li><a href="#t_classifications">Type Classifications</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]38 <li><a href="#t_primitive">Primitive Types</a>
39 <ol>
Chris Lattner488772f2008-01-04 04:32:38 +0000[diff] [blame]40 <li><a href="#t_floating">Floating Point Types</a></li>
41 <li><a href="#t_void">Void Type</a></li>
42 <li><a href="#t_label">Label Type</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]43 </ol>
44 </li>
45 <li><a href="#t_derived">Derived Types</a>
46 <ol>
Chris Lattner251ab812007-12-18 06:18:21 +0000[diff] [blame]47 <li><a href="#t_integer">Integer Type</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]48 <li><a href="#t_array">Array Type</a></li>
49 <li><a href="#t_function">Function Type</a></li>
50 <li><a href="#t_pointer">Pointer Type</a></li>
51 <li><a href="#t_struct">Structure Type</a></li>
52 <li><a href="#t_pstruct">Packed Structure Type</a></li>
53 <li><a href="#t_vector">Vector Type</a></li>
54 <li><a href="#t_opaque">Opaque Type</a></li>
55 </ol>
56 </li>
57 </ol>
58 </li>
59 <li><a href="#constants">Constants</a>
60 <ol>
61 <li><a href="#simpleconstants">Simple Constants</a>
62 <li><a href="#aggregateconstants">Aggregate Constants</a>
63 <li><a href="#globalconstants">Global Variable and Function Addresses</a>
64 <li><a href="#undefvalues">Undefined Values</a>
65 <li><a href="#constantexprs">Constant Expressions</a>
66 </ol>
67 </li>
68 <li><a href="#othervalues">Other Values</a>
69 <ol>
70 <li><a href="#inlineasm">Inline Assembler Expressions</a>
71 </ol>
72 </li>
73 <li><a href="#instref">Instruction Reference</a>
74 <ol>
75 <li><a href="#terminators">Terminator Instructions</a>
76 <ol>
77 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
78 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
79 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
80 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
81 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
82 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
83 </ol>
84 </li>
85 <li><a href="#binaryops">Binary Operations</a>
86 <ol>
87 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
88 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
89 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
90 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
91 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
92 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
93 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
94 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
95 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
96 </ol>
97 </li>
98 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
99 <ol>
100 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
101 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
102 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
103 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
104 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
105 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
106 </ol>
107 </li>
108 <li><a href="#vectorops">Vector Operations</a>
109 <ol>
110 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
111 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
112 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
113 </ol>
114 </li>
Dan Gohman74d6faf2008-05-12 23:51:09 +0000[diff] [blame]115 <li><a href="#aggregateops">Aggregate Operations</a>
116 <ol>
117 <li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li>
118 <li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li>
119 </ol>
120 </li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]121 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
122 <ol>
123 <li><a href="#i_malloc">'<tt>malloc</tt>' Instruction</a></li>
124 <li><a href="#i_free">'<tt>free</tt>' Instruction</a></li>
125 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
126 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
127 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
128 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
129 </ol>
130 </li>
131 <li><a href="#convertops">Conversion Operations</a>
132 <ol>
133 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
134 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
135 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
136 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
137 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
138 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
139 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
140 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
141 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
142 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
143 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
144 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
145 </ol>
146 <li><a href="#otherops">Other Operations</a>
147 <ol>
148 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
149 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
Nate Begeman646fa482008-05-12 19:01:56 +0000[diff] [blame]150 <li><a href="#i_vicmp">'<tt>vicmp</tt>' Instruction</a></li>
151 <li><a href="#i_vfcmp">'<tt>vfcmp</tt>' Instruction</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]152 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
153 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
154 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
155 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
156 </ol>
157 </li>
158 </ol>
159 </li>
160 <li><a href="#intrinsics">Intrinsic Functions</a>
161 <ol>
162 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
163 <ol>
164 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
165 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
166 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
167 </ol>
168 </li>
169 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
170 <ol>
171 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
172 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
173 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
174 </ol>
175 </li>
176 <li><a href="#int_codegen">Code Generator Intrinsics</a>
177 <ol>
178 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
179 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
180 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
181 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
182 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
183 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
184 <li><a href="#int_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
185 </ol>
186 </li>
187 <li><a href="#int_libc">Standard C Library Intrinsics</a>
188 <ol>
189 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
190 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
191 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
192 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
193 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
Dan Gohman361079c2007-10-15 20:30:11 +0000[diff] [blame]194 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
195 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
196 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]197 </ol>
198 </li>
199 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
200 <ol>
201 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
202 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
203 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
204 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
205 <li><a href="#int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic </a></li>
206 <li><a href="#int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic </a></li>
207 </ol>
208 </li>
209 <li><a href="#int_debugger">Debugger intrinsics</a></li>
210 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
Duncan Sands7407a9f2007-09-11 14:10:23 +0000[diff] [blame]211 <li><a href="#int_trampoline">Trampoline Intrinsic</a>
Duncan Sands38947cd2007-07-27 12:58:54 +0000[diff] [blame]212 <ol>
213 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
Duncan Sands38947cd2007-07-27 12:58:54 +0000[diff] [blame]214 </ol>
215 </li>
Andrew Lenharth785610d2008-02-16 01:24:58 +0000[diff] [blame]216 <li><a href="#int_atomics">Atomic intrinsics</a>
217 <ol>
Andrew Lenharthe44f3902008-02-21 06:45:13 +0000[diff] [blame]218 <li><a href="#int_memory_barrier"><tt>llvm.memory_barrier</tt></a></li>
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]219 <li><a href="#int_atomic_cmp_swap"><tt>llvm.atomic.cmp.swap</tt></a></li>
Andrew Lenharthe44f3902008-02-21 06:45:13 +0000[diff] [blame]220 <li><a href="#int_atomic_swap"><tt>llvm.atomic.swap</tt></a></li>
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]221 <li><a href="#int_atomic_load_add"><tt>llvm.atomic.load.add</tt></a></li>
222 <li><a href="#int_atomic_load_sub"><tt>llvm.atomic.load.sub</tt></a></li>
223 <li><a href="#int_atomic_load_and"><tt>llvm.atomic.load.and</tt></a></li>
224 <li><a href="#int_atomic_load_nand"><tt>llvm.atomic.load.nand</tt></a></li>
225 <li><a href="#int_atomic_load_or"><tt>llvm.atomic.load.or</tt></a></li>
226 <li><a href="#int_atomic_load_xor"><tt>llvm.atomic.load.xor</tt></a></li>
227 <li><a href="#int_atomic_load_max"><tt>llvm.atomic.load.max</tt></a></li>
228 <li><a href="#int_atomic_load_min"><tt>llvm.atomic.load.min</tt></a></li>
229 <li><a href="#int_atomic_load_umax"><tt>llvm.atomic.load.umax</tt></a></li>
230 <li><a href="#int_atomic_load_umin"><tt>llvm.atomic.load.umin</tt></a></li>
Andrew Lenharth785610d2008-02-16 01:24:58 +0000[diff] [blame]231 </ol>
232 </li>
Reid Spencerb043f672007-07-20 19:59:11 +0000[diff] [blame]233 <li><a href="#int_general">General intrinsics</a>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]234 <ol>
Reid Spencerb043f672007-07-20 19:59:11 +0000[diff] [blame]235 <li><a href="#int_var_annotation">
Tanya Lattner51369f32007-09-22 00:01:26 +0000[diff] [blame]236 <tt>llvm.var.annotation</tt>' Intrinsic</a></li>
Tanya Lattnerb306a9e2007-09-21 22:59:12 +0000[diff] [blame]237 <li><a href="#int_annotation">
Tanya Lattner51369f32007-09-22 00:01:26 +0000[diff] [blame]238 <tt>llvm.annotation.*</tt>' Intrinsic</a></li>
Anton Korobeynikove6e764f2008-01-15 22:31:34 +0000[diff] [blame]239 <li><a href="#int_trap">
240 <tt>llvm.trap</tt>' Intrinsic</a></li>
Tanya Lattnerb306a9e2007-09-21 22:59:12 +0000[diff] [blame]241 </ol>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]242 </li>
243 </ol>
244 </li>
245</ol>
246
247<div class="doc_author">
248 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
249 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
250</div>
251
252<!-- *********************************************************************** -->
253<div class="doc_section"> <a name="abstract">Abstract </a></div>
254<!-- *********************************************************************** -->
255
256<div class="doc_text">
257<p>This document is a reference manual for the LLVM assembly language.
Bill Wendlinge7846a52008-08-05 22:29:16 +0000[diff] [blame]258LLVM is a Static Single Assignment (SSA) based representation that provides
Chris Lattner96451482008-08-05 18:29:16 +0000[diff] [blame]259type safety, low-level operations, flexibility, and the capability of
260representing 'all' high-level languages cleanly. It is the common code
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]261representation used throughout all phases of the LLVM compilation
262strategy.</p>
263</div>
264
265<!-- *********************************************************************** -->
266<div class="doc_section"> <a name="introduction">Introduction</a> </div>
267<!-- *********************************************************************** -->
268
269<div class="doc_text">
270
271<p>The LLVM code representation is designed to be used in three
272different forms: as an in-memory compiler IR, as an on-disk bitcode
273representation (suitable for fast loading by a Just-In-Time compiler),
274and as a human readable assembly language representation. This allows
275LLVM to provide a powerful intermediate representation for efficient
276compiler transformations and analysis, while providing a natural means
277to debug and visualize the transformations. The three different forms
278of LLVM are all equivalent. This document describes the human readable
279representation and notation.</p>
280
281<p>The LLVM representation aims to be light-weight and low-level
282while being expressive, typed, and extensible at the same time. It
283aims to be a "universal IR" of sorts, by being at a low enough level
284that high-level ideas may be cleanly mapped to it (similar to how
285microprocessors are "universal IR's", allowing many source languages to
286be mapped to them). By providing type information, LLVM can be used as
287the target of optimizations: for example, through pointer analysis, it
288can be proven that a C automatic variable is never accessed outside of
289the current function... allowing it to be promoted to a simple SSA
290value instead of a memory location.</p>
291
292</div>
293
294<!-- _______________________________________________________________________ -->
295<div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
296
297<div class="doc_text">
298
299<p>It is important to note that this document describes 'well formed'
300LLVM assembly language. There is a difference between what the parser
301accepts and what is considered 'well formed'. For example, the
302following instruction is syntactically okay, but not well formed:</p>
303
304<div class="doc_code">
305<pre>
306%x = <a href="#i_add">add</a> i32 1, %x
307</pre>
308</div>
309
310<p>...because the definition of <tt>%x</tt> does not dominate all of
311its uses. The LLVM infrastructure provides a verification pass that may
312be used to verify that an LLVM module is well formed. This pass is
313automatically run by the parser after parsing input assembly and by
314the optimizer before it outputs bitcode. The violations pointed out
315by the verifier pass indicate bugs in transformation passes or input to
316the parser.</p>
317</div>
318
Chris Lattnera83fdc02007-10-03 17:34:29 +0000[diff] [blame]319<!-- Describe the typesetting conventions here. -->
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]320
321<!-- *********************************************************************** -->
322<div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
323<!-- *********************************************************************** -->
324
325<div class="doc_text">
326
Reid Spencerc8245b02007-08-07 14:34:28 +0000[diff] [blame]327 <p>LLVM identifiers come in two basic types: global and local. Global
328 identifiers (functions, global variables) begin with the @ character. Local
329 identifiers (register names, types) begin with the % character. Additionally,
330 there are three different formats for identifiers, for different purposes:
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]331
332<ol>
Reid Spencerc8245b02007-08-07 14:34:28 +0000[diff] [blame]333 <li>Named values are represented as a string of characters with their prefix.
334 For example, %foo, @DivisionByZero, %a.really.long.identifier. The actual
335 regular expression used is '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]336 Identifiers which require other characters in their names can be surrounded
Reid Spencerc8245b02007-08-07 14:34:28 +0000[diff] [blame]337 with quotes. In this way, anything except a <tt>&quot;</tt> character can
338 be used in a named value.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]339
Reid Spencerc8245b02007-08-07 14:34:28 +0000[diff] [blame]340 <li>Unnamed values are represented as an unsigned numeric value with their
341 prefix. For example, %12, @2, %44.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]342
343 <li>Constants, which are described in a <a href="#constants">section about
344 constants</a>, below.</li>
345</ol>
346
Reid Spencerc8245b02007-08-07 14:34:28 +0000[diff] [blame]347<p>LLVM requires that values start with a prefix for two reasons: Compilers
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]348don't need to worry about name clashes with reserved words, and the set of
349reserved words may be expanded in the future without penalty. Additionally,
350unnamed identifiers allow a compiler to quickly come up with a temporary
351variable without having to avoid symbol table conflicts.</p>
352
353<p>Reserved words in LLVM are very similar to reserved words in other
354languages. There are keywords for different opcodes
355('<tt><a href="#i_add">add</a></tt>',
356 '<tt><a href="#i_bitcast">bitcast</a></tt>',
357 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
358href="#t_void">void</a></tt>', '<tt><a href="#t_primitive">i32</a></tt>', etc...),
359and others. These reserved words cannot conflict with variable names, because
Reid Spencerc8245b02007-08-07 14:34:28 +0000[diff] [blame]360none of them start with a prefix character ('%' or '@').</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]361
362<p>Here is an example of LLVM code to multiply the integer variable
363'<tt>%X</tt>' by 8:</p>
364
365<p>The easy way:</p>
366
367<div class="doc_code">
368<pre>
369%result = <a href="#i_mul">mul</a> i32 %X, 8
370</pre>
371</div>
372
373<p>After strength reduction:</p>
374
375<div class="doc_code">
376<pre>
377%result = <a href="#i_shl">shl</a> i32 %X, i8 3
378</pre>
379</div>
380
381<p>And the hard way:</p>
382
383<div class="doc_code">
384<pre>
385<a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
386<a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
387%result = <a href="#i_add">add</a> i32 %1, %1
388</pre>
389</div>
390
391<p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
392important lexical features of LLVM:</p>
393
394<ol>
395
396 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
397 line.</li>
398
399 <li>Unnamed temporaries are created when the result of a computation is not
400 assigned to a named value.</li>
401
402 <li>Unnamed temporaries are numbered sequentially</li>
403
404</ol>
405
406<p>...and it also shows a convention that we follow in this document. When
407demonstrating instructions, we will follow an instruction with a comment that
408defines the type and name of value produced. Comments are shown in italic
409text.</p>
410
411</div>
412
413<!-- *********************************************************************** -->
414<div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
415<!-- *********************************************************************** -->
416
417<!-- ======================================================================= -->
418<div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
419</div>
420
421<div class="doc_text">
422
423<p>LLVM programs are composed of "Module"s, each of which is a
424translation unit of the input programs. Each module consists of
425functions, global variables, and symbol table entries. Modules may be
426combined together with the LLVM linker, which merges function (and
427global variable) definitions, resolves forward declarations, and merges
428symbol table entries. Here is an example of the "hello world" module:</p>
429
430<div class="doc_code">
431<pre><i>; Declare the string constant as a global constant...</i>
432<a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a
433 href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
434
435<i>; External declaration of the puts function</i>
436<a href="#functionstructure">declare</a> i32 @puts(i8 *) <i>; i32(i8 *)* </i>
437
438<i>; Definition of main function</i>
439define i32 @main() { <i>; i32()* </i>
440 <i>; Convert [13x i8 ]* to i8 *...</i>
441 %cast210 = <a
442 href="#i_getelementptr">getelementptr</a> [13 x i8 ]* @.LC0, i64 0, i64 0 <i>; i8 *</i>
443
444 <i>; Call puts function to write out the string to stdout...</i>
445 <a
446 href="#i_call">call</a> i32 @puts(i8 * %cast210) <i>; i32</i>
447 <a
448 href="#i_ret">ret</a> i32 0<br>}<br>
449</pre>
450</div>
451
452<p>This example is made up of a <a href="#globalvars">global variable</a>
453named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
454function, and a <a href="#functionstructure">function definition</a>
455for "<tt>main</tt>".</p>
456
457<p>In general, a module is made up of a list of global values,
458where both functions and global variables are global values. Global values are
459represented by a pointer to a memory location (in this case, a pointer to an
460array of char, and a pointer to a function), and have one of the following <a
461href="#linkage">linkage types</a>.</p>
462
463</div>
464
465<!-- ======================================================================= -->
466<div class="doc_subsection">
467 <a name="linkage">Linkage Types</a>
468</div>
469
470<div class="doc_text">
471
472<p>
473All Global Variables and Functions have one of the following types of linkage:
474</p>
475
476<dl>
477
Dale Johannesen96e7e092008-05-23 23:13:41 +0000[diff] [blame]478 <dt><tt><b><a name="linkage_internal">internal</a></b></tt>: </dt>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]479
480 <dd>Global values with internal linkage are only directly accessible by
481 objects in the current module. In particular, linking code into a module with
482 an internal global value may cause the internal to be renamed as necessary to
483 avoid collisions. Because the symbol is internal to the module, all
484 references can be updated. This corresponds to the notion of the
485 '<tt>static</tt>' keyword in C.
486 </dd>
487
488 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
489
490 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
491 the same name when linkage occurs. This is typically used to implement
492 inline functions, templates, or other code which must be generated in each
493 translation unit that uses it. Unreferenced <tt>linkonce</tt> globals are
494 allowed to be discarded.
495 </dd>
496
Dale Johannesen96e7e092008-05-23 23:13:41 +0000[diff] [blame]497 <dt><tt><b><a name="linkage_common">common</a></b></tt>: </dt>
498
499 <dd>"<tt>common</tt>" linkage is exactly the same as <tt>linkonce</tt>
500 linkage, except that unreferenced <tt>common</tt> globals may not be
501 discarded. This is used for globals that may be emitted in multiple
502 translation units, but that are not guaranteed to be emitted into every
503 translation unit that uses them. One example of this is tentative
504 definitions in C, such as "<tt>int X;</tt>" at global scope.
505 </dd>
506
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]507 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
508
Dale Johannesen96e7e092008-05-23 23:13:41 +0000[diff] [blame]509 <dd>"<tt>weak</tt>" linkage is the same as <tt>common</tt> linkage, except
510 that some targets may choose to emit different assembly sequences for them
511 for target-dependent reasons. This is used for globals that are declared
512 "weak" in C source code.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]513 </dd>
514
515 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
516
517 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
518 pointer to array type. When two global variables with appending linkage are
519 linked together, the two global arrays are appended together. This is the
520 LLVM, typesafe, equivalent of having the system linker append together
521 "sections" with identical names when .o files are linked.
522 </dd>
523
524 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt>: </dt>
Chris Lattner96451482008-08-05 18:29:16 +0000[diff] [blame]525 <dd>The semantics of this linkage follow the ELF object file model: the
526 symbol is weak until linked, if not linked, the symbol becomes null instead
527 of being an undefined reference.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]528 </dd>
529
530 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
531
532 <dd>If none of the above identifiers are used, the global is externally
533 visible, meaning that it participates in linkage and can be used to resolve
534 external symbol references.
535 </dd>
536</dl>
537
538 <p>
539 The next two types of linkage are targeted for Microsoft Windows platform
540 only. They are designed to support importing (exporting) symbols from (to)
Chris Lattner96451482008-08-05 18:29:16 +0000[diff] [blame]541 DLLs (Dynamic Link Libraries).
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]542 </p>
543
544 <dl>
545 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt>: </dt>
546
547 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
548 or variable via a global pointer to a pointer that is set up by the DLL
549 exporting the symbol. On Microsoft Windows targets, the pointer name is
550 formed by combining <code>_imp__</code> and the function or variable name.
551 </dd>
552
553 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt>: </dt>
554
555 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
556 pointer to a pointer in a DLL, so that it can be referenced with the
557 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
558 name is formed by combining <code>_imp__</code> and the function or variable
559 name.
560 </dd>
561
562</dl>
563
564<p><a name="linkage_external"></a>For example, since the "<tt>.LC0</tt>"
565variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
566variable and was linked with this one, one of the two would be renamed,
567preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
568external (i.e., lacking any linkage declarations), they are accessible
569outside of the current module.</p>
570<p>It is illegal for a function <i>declaration</i>
571to have any linkage type other than "externally visible", <tt>dllimport</tt>,
572or <tt>extern_weak</tt>.</p>
573<p>Aliases can have only <tt>external</tt>, <tt>internal</tt> and <tt>weak</tt>
574linkages.
575</div>
576
577<!-- ======================================================================= -->
578<div class="doc_subsection">
579 <a name="callingconv">Calling Conventions</a>
580</div>
581
582<div class="doc_text">
583
584<p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
585and <a href="#i_invoke">invokes</a> can all have an optional calling convention
586specified for the call. The calling convention of any pair of dynamic
587caller/callee must match, or the behavior of the program is undefined. The
588following calling conventions are supported by LLVM, and more may be added in
589the future:</p>
590
591<dl>
592 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
593
594 <dd>This calling convention (the default if no other calling convention is
595 specified) matches the target C calling conventions. This calling convention
596 supports varargs function calls and tolerates some mismatch in the declared
597 prototype and implemented declaration of the function (as does normal C).
598 </dd>
599
600 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
601
602 <dd>This calling convention attempts to make calls as fast as possible
603 (e.g. by passing things in registers). This calling convention allows the
604 target to use whatever tricks it wants to produce fast code for the target,
Chris Lattner96451482008-08-05 18:29:16 +0000[diff] [blame]605 without having to conform to an externally specified ABI (Application Binary
606 Interface). Implementations of this convention should allow arbitrary
Arnold Schwaighofer07444922008-05-14 09:17:12 +0000[diff] [blame]607 <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> to be
608 supported. This calling convention does not support varargs and requires the
609 prototype of all callees to exactly match the prototype of the function
610 definition.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]611 </dd>
612
613 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
614
615 <dd>This calling convention attempts to make code in the caller as efficient
616 as possible under the assumption that the call is not commonly executed. As
617 such, these calls often preserve all registers so that the call does not break
618 any live ranges in the caller side. This calling convention does not support
619 varargs and requires the prototype of all callees to exactly match the
620 prototype of the function definition.
621 </dd>
622
623 <dt><b>"<tt>cc &lt;<em>n</em>&gt;</tt>" - Numbered convention</b>:</dt>
624
625 <dd>Any calling convention may be specified by number, allowing
626 target-specific calling conventions to be used. Target specific calling
627 conventions start at 64.
628 </dd>
629</dl>
630
631<p>More calling conventions can be added/defined on an as-needed basis, to
632support pascal conventions or any other well-known target-independent
633convention.</p>
634
635</div>
636
637<!-- ======================================================================= -->
638<div class="doc_subsection">
639 <a name="visibility">Visibility Styles</a>
640</div>
641
642<div class="doc_text">
643
644<p>
645All Global Variables and Functions have one of the following visibility styles:
646</p>
647
648<dl>
649 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
650
Chris Lattner96451482008-08-05 18:29:16 +0000[diff] [blame]651 <dd>On targets that use the ELF object file format, default visibility means
652 that the declaration is visible to other
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]653 modules and, in shared libraries, means that the declared entity may be
654 overridden. On Darwin, default visibility means that the declaration is
655 visible to other modules. Default visibility corresponds to "external
656 linkage" in the language.
657 </dd>
658
659 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
660
661 <dd>Two declarations of an object with hidden visibility refer to the same
662 object if they are in the same shared object. Usually, hidden visibility
663 indicates that the symbol will not be placed into the dynamic symbol table,
664 so no other module (executable or shared library) can reference it
665 directly.
666 </dd>
667
668 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
669
670 <dd>On ELF, protected visibility indicates that the symbol will be placed in
671 the dynamic symbol table, but that references within the defining module will
672 bind to the local symbol. That is, the symbol cannot be overridden by another
673 module.
674 </dd>
675</dl>
676
677</div>
678
679<!-- ======================================================================= -->
680<div class="doc_subsection">
681 <a name="globalvars">Global Variables</a>
682</div>
683
684<div class="doc_text">
685
686<p>Global variables define regions of memory allocated at compilation time
687instead of run-time. Global variables may optionally be initialized, may have
688an explicit section to be placed in, and may have an optional explicit alignment
689specified. A variable may be defined as "thread_local", which means that it
690will not be shared by threads (each thread will have a separated copy of the
691variable). A variable may be defined as a global "constant," which indicates
692that the contents of the variable will <b>never</b> be modified (enabling better
693optimization, allowing the global data to be placed in the read-only section of
694an executable, etc). Note that variables that need runtime initialization
695cannot be marked "constant" as there is a store to the variable.</p>
696
697<p>
698LLVM explicitly allows <em>declarations</em> of global variables to be marked
699constant, even if the final definition of the global is not. This capability
700can be used to enable slightly better optimization of the program, but requires
701the language definition to guarantee that optimizations based on the
702'constantness' are valid for the translation units that do not include the
703definition.
704</p>
705
706<p>As SSA values, global variables define pointer values that are in
707scope (i.e. they dominate) all basic blocks in the program. Global
708variables always define a pointer to their "content" type because they
709describe a region of memory, and all memory objects in LLVM are
710accessed through pointers.</p>
711
Christopher Lambdd0049d2007-12-11 09:31:00 +0000[diff] [blame]712<p>A global variable may be declared to reside in a target-specifc numbered
713address space. For targets that support them, address spaces may affect how
714optimizations are performed and/or what target instructions are used to access
Christopher Lamb20a39e92007-12-12 08:44:39 +0000[diff] [blame]715the variable. The default address space is zero. The address space qualifier
716must precede any other attributes.</p>
Christopher Lambdd0049d2007-12-11 09:31:00 +0000[diff] [blame]717
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]718<p>LLVM allows an explicit section to be specified for globals. If the target
719supports it, it will emit globals to the section specified.</p>
720
721<p>An explicit alignment may be specified for a global. If not present, or if
722the alignment is set to zero, the alignment of the global is set by the target
723to whatever it feels convenient. If an explicit alignment is specified, the
724global is forced to have at least that much alignment. All alignments must be
725a power of 2.</p>
726
Christopher Lambdd0049d2007-12-11 09:31:00 +0000[diff] [blame]727<p>For example, the following defines a global in a numbered address space with
728an initializer, section, and alignment:</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]729
730<div class="doc_code">
731<pre>
Christopher Lambdd0049d2007-12-11 09:31:00 +0000[diff] [blame]732@G = constant float 1.0 addrspace(5), section "foo", align 4
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]733</pre>
734</div>
735
736</div>
737
738
739<!-- ======================================================================= -->
740<div class="doc_subsection">
741 <a name="functionstructure">Functions</a>
742</div>
743
744<div class="doc_text">
745
746<p>LLVM function definitions consist of the "<tt>define</tt>" keyord,
747an optional <a href="#linkage">linkage type</a>, an optional
748<a href="#visibility">visibility style</a>, an optional
749<a href="#callingconv">calling convention</a>, a return type, an optional
750<a href="#paramattrs">parameter attribute</a> for the return type, a function
751name, a (possibly empty) argument list (each with optional
Devang Patelac2fc272008-10-06 18:50:38 +0000[diff] [blame]752<a href="#paramattrs">parameter attributes</a>), optional
753<a href="#fnattrs">function attributes</a>, an optional section,
754an optional alignment, an optional <a href="#gc">garbage collector name</a>,
Chris Lattneree202542008-10-04 18:10:21 +0000[diff] [blame]755an opening curly brace, a list of basic blocks, and a closing curly brace.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]756
757LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
758optional <a href="#linkage">linkage type</a>, an optional
759<a href="#visibility">visibility style</a>, an optional
760<a href="#callingconv">calling convention</a>, a return type, an optional
761<a href="#paramattrs">parameter attribute</a> for the return type, a function
Gordon Henriksen13fe5e32007-12-10 03:18:06 +0000[diff] [blame]762name, a possibly empty list of arguments, an optional alignment, and an optional
Gordon Henriksend606f5b2007-12-10 03:30:21 +0000[diff] [blame]763<a href="#gc">garbage collector name</a>.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]764
Chris Lattner96451482008-08-05 18:29:16 +0000[diff] [blame]765<p>A function definition contains a list of basic blocks, forming the CFG
766(Control Flow Graph) for
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]767the function. Each basic block may optionally start with a label (giving the
768basic block a symbol table entry), contains a list of instructions, and ends
769with a <a href="#terminators">terminator</a> instruction (such as a branch or
770function return).</p>
771
772<p>The first basic block in a function is special in two ways: it is immediately
773executed on entrance to the function, and it is not allowed to have predecessor
774basic blocks (i.e. there can not be any branches to the entry block of a
775function). Because the block can have no predecessors, it also cannot have any
776<a href="#i_phi">PHI nodes</a>.</p>
777
778<p>LLVM allows an explicit section to be specified for functions. If the target
779supports it, it will emit functions to the section specified.</p>
780
781<p>An explicit alignment may be specified for a function. If not present, or if
782the alignment is set to zero, the alignment of the function is set by the target
783to whatever it feels convenient. If an explicit alignment is specified, the
784function is forced to have at least that much alignment. All alignments must be
785a power of 2.</p>
786
787</div>
788
789
790<!-- ======================================================================= -->
791<div class="doc_subsection">
792 <a name="aliasstructure">Aliases</a>
793</div>
794<div class="doc_text">
795 <p>Aliases act as "second name" for the aliasee value (which can be either
Anton Korobeynikov96822812008-03-22 08:36:14 +0000[diff] [blame]796 function, global variable, another alias or bitcast of global value). Aliases
797 may have an optional <a href="#linkage">linkage type</a>, and an
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]798 optional <a href="#visibility">visibility style</a>.</p>
799
800 <h5>Syntax:</h5>
801
802<div class="doc_code">
803<pre>
Duncan Sandsd7bfabf2008-09-12 20:48:21 +0000[diff] [blame]804@&lt;Name&gt; = alias [Linkage] [Visibility] &lt;AliaseeTy&gt; @&lt;Aliasee&gt;
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]805</pre>
806</div>
807
808</div>
809
810
811
812<!-- ======================================================================= -->
813<div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
814<div class="doc_text">
815 <p>The return type and each parameter of a function type may have a set of
816 <i>parameter attributes</i> associated with them. Parameter attributes are
817 used to communicate additional information about the result or parameters of
Duncan Sandsf5588dc2007-11-27 13:23:08 +0000[diff] [blame]818 a function. Parameter attributes are considered to be part of the function,
819 not of the function type, so functions with different parameter attributes
820 can have the same function type.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]821
822 <p>Parameter attributes are simple keywords that follow the type specified. If
823 multiple parameter attributes are needed, they are space separated. For
824 example:</p>
825
826<div class="doc_code">
827<pre>
Devang Patel008cd3e2008-09-26 23:51:19 +0000[diff] [blame]828declare i32 @printf(i8* noalias , ...)
Chris Lattnerf33b8452008-10-04 18:33:34 +0000[diff] [blame]829declare i32 @atoi(i8 zeroext)
830declare signext i8 @returns_signed_char()
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]831</pre>
832</div>
833
Duncan Sandsf5588dc2007-11-27 13:23:08 +0000[diff] [blame]834 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
835 <tt>readonly</tt>) come immediately after the argument list.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]836
837 <p>Currently, only the following parameter attributes are defined:</p>
838 <dl>
Reid Spencerf234bed2007-07-19 23:13:04 +0000[diff] [blame]839 <dt><tt>zeroext</tt></dt>
Chris Lattnerf33b8452008-10-04 18:33:34 +0000[diff] [blame]840 <dd>This indicates to the code generator that the parameter or return value
841 should be zero-extended to a 32-bit value by the caller (for a parameter)
842 or the callee (for a return value).</dd>
Chris Lattner275e6be2008-01-11 06:20:47 +0000[diff] [blame]843
Reid Spencerf234bed2007-07-19 23:13:04 +0000[diff] [blame]844 <dt><tt>signext</tt></dt>
Chris Lattnerf33b8452008-10-04 18:33:34 +0000[diff] [blame]845 <dd>This indicates to the code generator that the parameter or return value
846 should be sign-extended to a 32-bit value by the caller (for a parameter)
847 or the callee (for a return value).</dd>
Chris Lattner275e6be2008-01-11 06:20:47 +0000[diff] [blame]848
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]849 <dt><tt>inreg</tt></dt>
Dale Johannesenc08a0e22008-09-25 20:47:45 +0000[diff] [blame]850 <dd>This indicates that this parameter or return value should be treated
851 in a special target-dependent fashion during while emitting code for a
852 function call or return (usually, by putting it in a register as opposed
Chris Lattnerf33b8452008-10-04 18:33:34 +0000[diff] [blame]853 to memory, though some targets use it to distinguish between two different
854 kinds of registers). Use of this attribute is target-specific.</dd>
Chris Lattner275e6be2008-01-11 06:20:47 +0000[diff] [blame]855
Duncan Sands7d94e5a2008-10-06 08:14:18 +0000[diff] [blame]856 <dt><tt><a name="byval">byval</a></tt></dt>
Chris Lattner04c86182008-01-15 04:34:22 +0000[diff] [blame]857 <dd>This indicates that the pointer parameter should really be passed by
858 value to the function. The attribute implies that a hidden copy of the
859 pointee is made between the caller and the callee, so the callee is unable
Chris Lattner6a9f3c42008-08-05 18:21:08 +0000[diff] [blame]860 to modify the value in the callee. This attribute is only valid on LLVM
Chris Lattner04c86182008-01-15 04:34:22 +0000[diff] [blame]861 pointer arguments. It is generally used to pass structs and arrays by
Duncan Sands7d94e5a2008-10-06 08:14:18 +0000[diff] [blame]862 value, but is also valid on pointers to scalars. The copy is considered to
863 belong to the caller not the callee (for example,
864 <tt><a href="#readonly">readonly</a></tt> functions should not write to
Devang Patelac2fc272008-10-06 18:50:38 +0000[diff] [blame]865 <tt>byval</tt> parameters). This is not a valid attribute for return
866 values. </dd>
Chris Lattner275e6be2008-01-11 06:20:47 +0000[diff] [blame]867
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]868 <dt><tt>sret</tt></dt>
Duncan Sands616cc032008-02-18 04:19:38 +0000[diff] [blame]869 <dd>This indicates that the pointer parameter specifies the address of a
870 structure that is the return value of the function in the source program.
Chris Lattnerf33b8452008-10-04 18:33:34 +0000[diff] [blame]871 This pointer must be guaranteed by the caller to be valid: loads and stores
872 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]873 be applied to the first parameter. This is not a valid attribute for
874 return values. </dd>
Chris Lattner275e6be2008-01-11 06:20:47 +0000[diff] [blame]875
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]876 <dt><tt>noalias</tt></dt>
Owen Andersonc4fc4cd2008-02-18 04:09:01 +0000[diff] [blame]877 <dd>This indicates that the parameter does not alias any global or any other
878 parameter. The caller is responsible for ensuring that this is the case,
Devang Patelac2fc272008-10-06 18:50:38 +0000[diff] [blame]879 usually by placing the value in a stack allocation. This is not a valid
880 attribute for return values.</dd>
Chris Lattner275e6be2008-01-11 06:20:47 +0000[diff] [blame]881
Duncan Sands4ee46812007-07-27 19:57:41 +0000[diff] [blame]882 <dt><tt>nest</tt></dt>
Duncan Sandsf1a7d4c2008-07-08 09:27:25 +0000[diff] [blame]883 <dd>This indicates that the pointer parameter can be excised using the
Devang Patelac2fc272008-10-06 18:50:38 +0000[diff] [blame]884 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
885 attribute for return values.</dd>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]886 </dl>
887
888</div>
889
890<!-- ======================================================================= -->
891<div class="doc_subsection">
Gordon Henriksen13fe5e32007-12-10 03:18:06 +0000[diff] [blame]892 <a name="gc">Garbage Collector Names</a>
893</div>
894
895<div class="doc_text">
896<p>Each function may specify a garbage collector name, which is simply a
897string.</p>
898
899<div class="doc_code"><pre
900>define void @f() gc "name" { ...</pre></div>
901
902<p>The compiler declares the supported values of <i>name</i>. Specifying a
903collector which will cause the compiler to alter its output in order to support
904the named garbage collection algorithm.</p>
905</div>
906
907<!-- ======================================================================= -->
908<div class="doc_subsection">
Devang Patel008cd3e2008-09-26 23:51:19 +0000[diff] [blame]909 <a name="fnattrs">Function Attributes</a>
Devang Pateld468f1c2008-09-04 23:05:13 +0000[diff] [blame]910</div>
911
912<div class="doc_text">
Devang Patel008cd3e2008-09-26 23:51:19 +0000[diff] [blame]913
914<p>Function attributes are set to communicate additional information about
915 a function. Function attributes are considered to be part of the function,
916 not of the function type, so functions with different parameter attributes
917 can have the same function type.</p>
918
919 <p>Function attributes are simple keywords that follow the type specified. If
920 multiple attributes are needed, they are space separated. For
921 example:</p>
Devang Pateld468f1c2008-09-04 23:05:13 +0000[diff] [blame]922
923<div class="doc_code">
Bill Wendling74d3eac2008-09-07 10:26:33 +0000[diff] [blame]924<pre>
Devang Patel008cd3e2008-09-26 23:51:19 +0000[diff] [blame]925define void @f() noinline { ... }
926define void @f() alwaysinline { ... }
927define void @f() alwaysinline optsize { ... }
928define void @f() optsize
Bill Wendling74d3eac2008-09-07 10:26:33 +0000[diff] [blame]929</pre>
Devang Pateld468f1c2008-09-04 23:05:13 +0000[diff] [blame]930</div>
931
Bill Wendling74d3eac2008-09-07 10:26:33 +0000[diff] [blame]932<dl>
Devang Patel008cd3e2008-09-26 23:51:19 +0000[diff] [blame]933<dt><tt>alwaysinline</tt></dt>
Chris Lattner82d70f92008-10-04 18:23:17 +0000[diff] [blame]934<dd>This attribute indicates that the inliner should attempt to inline this
935function into callers whenever possible, ignoring any active inlining size
936threshold for this caller.</dd>
Bill Wendling74d3eac2008-09-07 10:26:33 +0000[diff] [blame]937
Devang Patel008cd3e2008-09-26 23:51:19 +0000[diff] [blame]938<dt><tt>noinline</tt></dt>
Chris Lattner82d70f92008-10-04 18:23:17 +0000[diff] [blame]939<dd>This attribute indicates that the inliner should never inline this function
Chris Lattner377f3ae2008-10-05 17:14:59 +0000[diff] [blame]940in any situation. This attribute may not be used together with the
Chris Lattner82d70f92008-10-04 18:23:17 +0000[diff] [blame]941<tt>alwaysinline</tt> attribute.</dd>
Bill Wendling74d3eac2008-09-07 10:26:33 +0000[diff] [blame]942
Devang Patel008cd3e2008-09-26 23:51:19 +0000[diff] [blame]943<dt><tt>optsize</tt></dt>
Devang Patelc29099b2008-09-29 18:34:44 +0000[diff] [blame]944<dd>This attribute suggests that optimization passes and code generator passes
Chris Lattner82d70f92008-10-04 18:23:17 +0000[diff] [blame]945make choices that keep the code size of this function low, and otherwise do
946optimizations specifically to reduce code size.</dd>
Bill Wendling74d3eac2008-09-07 10:26:33 +0000[diff] [blame]947
Devang Patel008cd3e2008-09-26 23:51:19 +0000[diff] [blame]948<dt><tt>noreturn</tt></dt>
Chris Lattner82d70f92008-10-04 18:23:17 +0000[diff] [blame]949<dd>This function attribute indicates that the function never returns normally.
950This produces undefined behavior at runtime if the function ever does
951dynamically return.</dd>
Devang Patel008cd3e2008-09-26 23:51:19 +0000[diff] [blame]952
953<dt><tt>nounwind</tt></dt>
Chris Lattner82d70f92008-10-04 18:23:17 +0000[diff] [blame]954<dd>This function attribute indicates that the function never returns with an
955unwind or exceptional control flow. If the function does unwind, its runtime
956behavior is undefined.</dd>
957
958<dt><tt>readnone</tt></dt>
Duncan Sands7d94e5a2008-10-06 08:14:18 +0000[diff] [blame]959<dd>This attribute indicates that the function computes its result (or the
960exception it throws) based strictly on its arguments, without dereferencing any
961pointer arguments or otherwise accessing any mutable state (e.g. memory, control
962registers, etc) visible to caller functions. It does not write through any
963pointer arguments (including <tt><a href="#byval">byval</a></tt> arguments) and
964never changes any state visible to callers.</dd>
Devang Patel008cd3e2008-09-26 23:51:19 +0000[diff] [blame]965
Duncan Sands7d94e5a2008-10-06 08:14:18 +0000[diff] [blame]966<dt><tt><a name="readonly">readonly</a></tt></dt>
967<dd>This attribute indicates that the function does not write through any
968pointer arguments (including <tt><a href="#byval">byval</a></tt> arguments)
969or otherwise modify any state (e.g. memory, control registers, etc) visible to
970caller functions. It may dereference pointer arguments and read state that may
971be set in the caller. A readonly function always returns the same value (or
972throws the same exception) when called with the same set of arguments and global
973state.</dd>
Bill Wendling74d3eac2008-09-07 10:26:33 +0000[diff] [blame]974</dl>
975
Devang Pateld468f1c2008-09-04 23:05:13 +0000[diff] [blame]976</div>
977
978<!-- ======================================================================= -->
979<div class="doc_subsection">
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]980 <a name="moduleasm">Module-Level Inline Assembly</a>
981</div>
982
983<div class="doc_text">
984<p>
985Modules may contain "module-level inline asm" blocks, which corresponds to the
986GCC "file scope inline asm" blocks. These blocks are internally concatenated by
987LLVM and treated as a single unit, but may be separated in the .ll file if
988desired. The syntax is very simple:
989</p>
990
991<div class="doc_code">
992<pre>
993module asm "inline asm code goes here"
994module asm "more can go here"
995</pre>
996</div>
997
998<p>The strings can contain any character by escaping non-printable characters.
999 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1000 for the number.
1001</p>
1002
1003<p>
1004 The inline asm code is simply printed to the machine code .s file when
1005 assembly code is generated.
1006</p>
1007</div>
1008
1009<!-- ======================================================================= -->
1010<div class="doc_subsection">
1011 <a name="datalayout">Data Layout</a>
1012</div>
1013
1014<div class="doc_text">
1015<p>A module may specify a target specific data layout string that specifies how
1016data is to be laid out in memory. The syntax for the data layout is simply:</p>
1017<pre> target datalayout = "<i>layout specification</i>"</pre>
1018<p>The <i>layout specification</i> consists of a list of specifications
1019separated by the minus sign character ('-'). Each specification starts with a
1020letter and may include other information after the letter to define some
1021aspect of the data layout. The specifications accepted are as follows: </p>
1022<dl>
1023 <dt><tt>E</tt></dt>
1024 <dd>Specifies that the target lays out data in big-endian form. That is, the
1025 bits with the most significance have the lowest address location.</dd>
1026 <dt><tt>e</tt></dt>
Chris Lattner96451482008-08-05 18:29:16 +0000[diff] [blame]1027 <dd>Specifies that the target lays out data in little-endian form. That is,
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1028 the bits with the least significance have the lowest address location.</dd>
1029 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1030 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1031 <i>preferred</i> alignments. All sizes are in bits. Specifying the <i>pref</i>
1032 alignment is optional. If omitted, the preceding <tt>:</tt> should be omitted
1033 too.</dd>
1034 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1035 <dd>This specifies the alignment for an integer type of a given bit
1036 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1037 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1038 <dd>This specifies the alignment for a vector type of a given bit
1039 <i>size</i>.</dd>
1040 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1041 <dd>This specifies the alignment for a floating point type of a given bit
1042 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
1043 (double).</dd>
1044 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1045 <dd>This specifies the alignment for an aggregate type of a given bit
1046 <i>size</i>.</dd>
1047</dl>
1048<p>When constructing the data layout for a given target, LLVM starts with a
1049default set of specifications which are then (possibly) overriden by the
1050specifications in the <tt>datalayout</tt> keyword. The default specifications
1051are given in this list:</p>
1052<ul>
1053 <li><tt>E</tt> - big endian</li>
1054 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
1055 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1056 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1057 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1058 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
Chris Lattner96451482008-08-05 18:29:16 +0000[diff] [blame]1059 <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]1060 alignment of 64-bits</li>
1061 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1062 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1063 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1064 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1065 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1066</ul>
Chris Lattner6a9f3c42008-08-05 18:21:08 +0000[diff] [blame]1067<p>When LLVM is determining the alignment for a given type, it uses the
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1068following rules:
1069<ol>
1070 <li>If the type sought is an exact match for one of the specifications, that
1071 specification is used.</li>
1072 <li>If no match is found, and the type sought is an integer type, then the
1073 smallest integer type that is larger than the bitwidth of the sought type is
1074 used. If none of the specifications are larger than the bitwidth then the the
1075 largest integer type is used. For example, given the default specifications
1076 above, the i7 type will use the alignment of i8 (next largest) while both
1077 i65 and i256 will use the alignment of i64 (largest specified).</li>
1078 <li>If no match is found, and the type sought is a vector type, then the
1079 largest vector type that is smaller than the sought vector type will be used
1080 as a fall back. This happens because <128 x double> can be implemented in
1081 terms of 64 <2 x double>, for example.</li>
1082</ol>
1083</div>
1084
1085<!-- *********************************************************************** -->
1086<div class="doc_section"> <a name="typesystem">Type System</a> </div>
1087<!-- *********************************************************************** -->
1088
1089<div class="doc_text">
1090
1091<p>The LLVM type system is one of the most important features of the
1092intermediate representation. Being typed enables a number of
Chris Lattner96451482008-08-05 18:29:16 +0000[diff] [blame]1093optimizations to be performed on the intermediate representation directly,
1094without having to do
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1095extra analyses on the side before the transformation. A strong type
1096system makes it easier to read the generated code and enables novel
1097analyses and transformations that are not feasible to perform on normal
1098three address code representations.</p>
1099
1100</div>
1101
1102<!-- ======================================================================= -->
Chris Lattner488772f2008-01-04 04:32:38 +0000[diff] [blame]1103<div class="doc_subsection"> <a name="t_classifications">Type
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1104Classifications</a> </div>
1105<div class="doc_text">
Chris Lattner488772f2008-01-04 04:32:38 +0000[diff] [blame]1106<p>The types fall into a few useful
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1107classifications:</p>
1108
1109<table border="1" cellspacing="0" cellpadding="4">
1110 <tbody>
1111 <tr><th>Classification</th><th>Types</th></tr>
1112 <tr>
Chris Lattner488772f2008-01-04 04:32:38 +0000[diff] [blame]1113 <td><a href="#t_integer">integer</a></td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1114 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1115 </tr>
1116 <tr>
Chris Lattner488772f2008-01-04 04:32:38 +0000[diff] [blame]1117 <td><a href="#t_floating">floating point</a></td>
1118 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1119 </tr>
1120 <tr>
1121 <td><a name="t_firstclass">first class</a></td>
Chris Lattner488772f2008-01-04 04:32:38 +0000[diff] [blame]1122 <td><a href="#t_integer">integer</a>,
1123 <a href="#t_floating">floating point</a>,
1124 <a href="#t_pointer">pointer</a>,
Dan Gohmanf6237db2008-06-18 18:42:13 +0000[diff] [blame]1125 <a href="#t_vector">vector</a>,
Dan Gohman74d6faf2008-05-12 23:51:09 +0000[diff] [blame]1126 <a href="#t_struct">structure</a>,
1127 <a href="#t_array">array</a>,
Dan Gohmand13951e2008-05-23 22:50:26 +0000[diff] [blame]1128 <a href="#t_label">label</a>.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1129 </td>
1130 </tr>
Chris Lattner488772f2008-01-04 04:32:38 +0000[diff] [blame]1131 <tr>
1132 <td><a href="#t_primitive">primitive</a></td>
1133 <td><a href="#t_label">label</a>,
1134 <a href="#t_void">void</a>,
Chris Lattner488772f2008-01-04 04:32:38 +0000[diff] [blame]1135 <a href="#t_floating">floating point</a>.</td>
1136 </tr>
1137 <tr>
1138 <td><a href="#t_derived">derived</a></td>
1139 <td><a href="#t_integer">integer</a>,
1140 <a href="#t_array">array</a>,
1141 <a href="#t_function">function</a>,
1142 <a href="#t_pointer">pointer</a>,
1143 <a href="#t_struct">structure</a>,
1144 <a href="#t_pstruct">packed structure</a>,
1145 <a href="#t_vector">vector</a>,
1146 <a href="#t_opaque">opaque</a>.
1147 </tr>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1148 </tbody>
1149</table>
1150
1151<p>The <a href="#t_firstclass">first class</a> types are perhaps the
1152most important. Values of these types are the only ones which can be
1153produced by instructions, passed as arguments, or used as operands to
Dan Gohman4f29e422008-05-23 21:53:15 +0000[diff] [blame]1154instructions.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1155</div>
1156
1157<!-- ======================================================================= -->
Chris Lattner488772f2008-01-04 04:32:38 +0000[diff] [blame]1158<div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
Chris Lattner86437612008-01-04 04:34:14 +0000[diff] [blame]1159
Chris Lattner488772f2008-01-04 04:32:38 +0000[diff] [blame]1160<div class="doc_text">
1161<p>The primitive types are the fundamental building blocks of the LLVM
1162system.</p>
1163
Chris Lattner86437612008-01-04 04:34:14 +0000[diff] [blame]1164</div>
1165
Chris Lattner488772f2008-01-04 04:32:38 +0000[diff] [blame]1166<!-- _______________________________________________________________________ -->
1167<div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1168
1169<div class="doc_text">
1170 <table>
1171 <tbody>
1172 <tr><th>Type</th><th>Description</th></tr>
1173 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1174 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1175 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1176 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1177 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1178 </tbody>
1179 </table>
1180</div>
1181
1182<!-- _______________________________________________________________________ -->
1183<div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1184
1185<div class="doc_text">
1186<h5>Overview:</h5>
1187<p>The void type does not represent any value and has no size.</p>
1188
1189<h5>Syntax:</h5>
1190
1191<pre>
1192 void
1193</pre>
1194</div>
1195
1196<!-- _______________________________________________________________________ -->
1197<div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1198
1199<div class="doc_text">
1200<h5>Overview:</h5>
1201<p>The label type represents code labels.</p>
1202
1203<h5>Syntax:</h5>
1204
1205<pre>
1206 label
1207</pre>
1208</div>
1209
1210
1211<!-- ======================================================================= -->
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1212<div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1213
1214<div class="doc_text">
1215
1216<p>The real power in LLVM comes from the derived types in the system.
1217This is what allows a programmer to represent arrays, functions,
1218pointers, and other useful types. Note that these derived types may be
1219recursive: For example, it is possible to have a two dimensional array.</p>
1220
1221</div>
1222
1223<!-- _______________________________________________________________________ -->
1224<div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1225
1226<div class="doc_text">
1227
1228<h5>Overview:</h5>
1229<p>The integer type is a very simple derived type that simply specifies an
1230arbitrary bit width for the integer type desired. Any bit width from 1 bit to
12312^23-1 (about 8 million) can be specified.</p>
1232
1233<h5>Syntax:</h5>
1234
1235<pre>
1236 iN
1237</pre>
1238
1239<p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1240value.</p>
1241
1242<h5>Examples:</h5>
1243<table class="layout">
Chris Lattner251ab812007-12-18 06:18:21 +0000[diff] [blame]1244 <tbody>
1245 <tr>
1246 <td><tt>i1</tt></td>
1247 <td>a single-bit integer.</td>
1248 </tr><tr>
1249 <td><tt>i32</tt></td>
1250 <td>a 32-bit integer.</td>
1251 </tr><tr>
1252 <td><tt>i1942652</tt></td>
1253 <td>a really big integer of over 1 million bits.</td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1254 </tr>
Chris Lattner251ab812007-12-18 06:18:21 +0000[diff] [blame]1255 </tbody>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1256</table>
1257</div>
1258
1259<!-- _______________________________________________________________________ -->
1260<div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1261
1262<div class="doc_text">
1263
1264<h5>Overview:</h5>
1265
1266<p>The array type is a very simple derived type that arranges elements
1267sequentially in memory. The array type requires a size (number of
1268elements) and an underlying data type.</p>
1269
1270<h5>Syntax:</h5>
1271
1272<pre>
1273 [&lt;# elements&gt; x &lt;elementtype&gt;]
1274</pre>
1275
1276<p>The number of elements is a constant integer value; elementtype may
1277be any type with a size.</p>
1278
1279<h5>Examples:</h5>
1280<table class="layout">
1281 <tr class="layout">
Chris Lattner7311d222007-12-19 05:04:11 +0000[diff] [blame]1282 <td class="left"><tt>[40 x i32]</tt></td>
1283 <td class="left">Array of 40 32-bit integer values.</td>
1284 </tr>
1285 <tr class="layout">
1286 <td class="left"><tt>[41 x i32]</tt></td>
1287 <td class="left">Array of 41 32-bit integer values.</td>
1288 </tr>
1289 <tr class="layout">
1290 <td class="left"><tt>[4 x i8]</tt></td>
1291 <td class="left">Array of 4 8-bit integer values.</td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1292 </tr>
1293</table>
1294<p>Here are some examples of multidimensional arrays:</p>
1295<table class="layout">
1296 <tr class="layout">
Chris Lattner7311d222007-12-19 05:04:11 +0000[diff] [blame]1297 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1298 <td class="left">3x4 array of 32-bit integer values.</td>
1299 </tr>
1300 <tr class="layout">
1301 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1302 <td class="left">12x10 array of single precision floating point values.</td>
1303 </tr>
1304 <tr class="layout">
1305 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1306 <td class="left">2x3x4 array of 16-bit integer values.</td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1307 </tr>
1308</table>
1309
1310<p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1311length array. Normally, accesses past the end of an array are undefined in
1312LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
1313As a special case, however, zero length arrays are recognized to be variable
1314length. This allows implementation of 'pascal style arrays' with the LLVM
1315type "{ i32, [0 x float]}", for example.</p>
1316
1317</div>
1318
1319<!-- _______________________________________________________________________ -->
1320<div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1321<div class="doc_text">
Chris Lattner43030e72008-04-23 04:59:35 +0000[diff] [blame]1322
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1323<h5>Overview:</h5>
Chris Lattner43030e72008-04-23 04:59:35 +0000[diff] [blame]1324
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1325<p>The function type can be thought of as a function signature. It
Devang Pateld4ba41d2008-03-24 05:35:41 +0000[diff] [blame]1326consists of a return type and a list of formal parameter types. The
Chris Lattner43030e72008-04-23 04:59:35 +0000[diff] [blame]1327return 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]1328If 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]1329class types, and the struct must have at least one element.</p>
Devang Patela3cc5372008-03-10 20:49:15 +0000[diff] [blame]1330
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1331<h5>Syntax:</h5>
Chris Lattner43030e72008-04-23 04:59:35 +0000[diff] [blame]1332
1333<pre>
1334 &lt;returntype list&gt; (&lt;parameter list&gt;)
1335</pre>
1336
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1337<p>...where '<tt>&lt;parameter list&gt;</tt>' is a comma-separated list of type
1338specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1339which indicates that the function takes a variable number of arguments.
1340Variable argument functions can access their arguments with the <a
Devang Patela3cc5372008-03-10 20:49:15 +0000[diff] [blame]1341 href="#int_varargs">variable argument handling intrinsic</a> functions.
1342'<tt>&lt;returntype list&gt;</tt>' is a comma-separated list of
1343<a href="#t_firstclass">first class</a> type specifiers.</p>
Chris Lattner43030e72008-04-23 04:59:35 +0000[diff] [blame]1344
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1345<h5>Examples:</h5>
1346<table class="layout">
1347 <tr class="layout">
1348 <td class="left"><tt>i32 (i32)</tt></td>
1349 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1350 </td>
1351 </tr><tr class="layout">
Reid Spencerf234bed2007-07-19 23:13:04 +0000[diff] [blame]1352 <td class="left"><tt>float&nbsp;(i16&nbsp;signext,&nbsp;i32&nbsp;*)&nbsp;*
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1353 </tt></td>
1354 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1355 an <tt>i16</tt> that should be sign extended and a
1356 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1357 <tt>float</tt>.
1358 </td>
1359 </tr><tr class="layout">
1360 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1361 <td class="left">A vararg function that takes at least one
1362 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1363 which returns an integer. This is the signature for <tt>printf</tt> in
1364 LLVM.
1365 </td>
Devang Pateld4ba41d2008-03-24 05:35:41 +0000[diff] [blame]1366 </tr><tr class="layout">
1367 <td class="left"><tt>{i32, i32} (i32)</tt></td>
Devang Patel6dddba22008-03-24 18:10:52 +0000[diff] [blame]1368 <td class="left">A function taking an <tt>i32></tt>, returning two
1369 <tt> i32 </tt> values as an aggregate of type <tt>{ i32, i32 }</tt>
Devang Pateld4ba41d2008-03-24 05:35:41 +0000[diff] [blame]1370 </td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1371 </tr>
1372</table>
1373
1374</div>
1375<!-- _______________________________________________________________________ -->
1376<div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1377<div class="doc_text">
1378<h5>Overview:</h5>
1379<p>The structure type is used to represent a collection of data members
1380together in memory. The packing of the field types is defined to match
1381the ABI of the underlying processor. The elements of a structure may
1382be any type that has a size.</p>
1383<p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1384and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1385field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1386instruction.</p>
1387<h5>Syntax:</h5>
1388<pre> { &lt;type list&gt; }<br></pre>
1389<h5>Examples:</h5>
1390<table class="layout">
1391 <tr class="layout">
1392 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1393 <td class="left">A triple of three <tt>i32</tt> values</td>
1394 </tr><tr class="layout">
1395 <td class="left"><tt>{&nbsp;float,&nbsp;i32&nbsp;(i32)&nbsp;*&nbsp;}</tt></td>
1396 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1397 second element is a <a href="#t_pointer">pointer</a> to a
1398 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1399 an <tt>i32</tt>.</td>
1400 </tr>
1401</table>
1402</div>
1403
1404<!-- _______________________________________________________________________ -->
1405<div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1406</div>
1407<div class="doc_text">
1408<h5>Overview:</h5>
1409<p>The packed structure type is used to represent a collection of data members
1410together in memory. There is no padding between fields. Further, the alignment
1411of a packed structure is 1 byte. The elements of a packed structure may
1412be any type that has a size.</p>
1413<p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1414and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1415field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1416instruction.</p>
1417<h5>Syntax:</h5>
1418<pre> &lt; { &lt;type list&gt; } &gt; <br></pre>
1419<h5>Examples:</h5>
1420<table class="layout">
1421 <tr class="layout">
1422 <td class="left"><tt>&lt; { i32, i32, i32 } &gt;</tt></td>
1423 <td class="left">A triple of three <tt>i32</tt> values</td>
1424 </tr><tr class="layout">
Bill Wendling74d3eac2008-09-07 10:26:33 +0000[diff] [blame]1425 <td class="left">
1426<tt>&lt;&nbsp;{&nbsp;float,&nbsp;i32&nbsp;(i32)*&nbsp;}&nbsp;&gt;</tt></td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1427 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1428 second element is a <a href="#t_pointer">pointer</a> to a
1429 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1430 an <tt>i32</tt>.</td>
1431 </tr>
1432</table>
1433</div>
1434
1435<!-- _______________________________________________________________________ -->
1436<div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1437<div class="doc_text">
1438<h5>Overview:</h5>
1439<p>As in many languages, the pointer type represents a pointer or
Christopher Lambdd0049d2007-12-11 09:31:00 +0000[diff] [blame]1440reference to another object, which must live in memory. Pointer types may have
1441an optional address space attribute defining the target-specific numbered
1442address space where the pointed-to object resides. The default address space is
1443zero.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1444<h5>Syntax:</h5>
1445<pre> &lt;type&gt; *<br></pre>
1446<h5>Examples:</h5>
1447<table class="layout">
1448 <tr class="layout">
Chris Lattner7311d222007-12-19 05:04:11 +0000[diff] [blame]1449 <td class="left"><tt>[4x i32]*</tt></td>
1450 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1451 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1452 </tr>
1453 <tr class="layout">
1454 <td class="left"><tt>i32 (i32 *) *</tt></td>
1455 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1456 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
Chris Lattner7311d222007-12-19 05:04:11 +0000[diff] [blame]1457 <tt>i32</tt>.</td>
1458 </tr>
1459 <tr class="layout">
1460 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1461 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1462 that resides in address space #5.</td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1463 </tr>
1464</table>
1465</div>
1466
1467<!-- _______________________________________________________________________ -->
1468<div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1469<div class="doc_text">
1470
1471<h5>Overview:</h5>
1472
1473<p>A vector type is a simple derived type that represents a vector
1474of elements. Vector types are used when multiple primitive data
1475are operated in parallel using a single instruction (SIMD).
1476A vector type requires a size (number of
1477elements) and an underlying primitive data type. Vectors must have a power
1478of two length (1, 2, 4, 8, 16 ...). Vector types are
1479considered <a href="#t_firstclass">first class</a>.</p>
1480
1481<h5>Syntax:</h5>
1482
1483<pre>
1484 &lt; &lt;# elements&gt; x &lt;elementtype&gt; &gt;
1485</pre>
1486
1487<p>The number of elements is a constant integer value; elementtype may
1488be any integer or floating point type.</p>
1489
1490<h5>Examples:</h5>
1491
1492<table class="layout">
1493 <tr class="layout">
Chris Lattner7311d222007-12-19 05:04:11 +0000[diff] [blame]1494 <td class="left"><tt>&lt;4 x i32&gt;</tt></td>
1495 <td class="left">Vector of 4 32-bit integer values.</td>
1496 </tr>
1497 <tr class="layout">
1498 <td class="left"><tt>&lt;8 x float&gt;</tt></td>
1499 <td class="left">Vector of 8 32-bit floating-point values.</td>
1500 </tr>
1501 <tr class="layout">
1502 <td class="left"><tt>&lt;2 x i64&gt;</tt></td>
1503 <td class="left">Vector of 2 64-bit integer values.</td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1504 </tr>
1505</table>
1506</div>
1507
1508<!-- _______________________________________________________________________ -->
1509<div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1510<div class="doc_text">
1511
1512<h5>Overview:</h5>
1513
1514<p>Opaque types are used to represent unknown types in the system. This
Gordon Henriksenda0706e2007-10-14 00:34:53 +0000[diff] [blame]1515corresponds (for example) to the C notion of a forward declared structure type.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1516In LLVM, opaque types can eventually be resolved to any type (not just a
1517structure type).</p>
1518
1519<h5>Syntax:</h5>
1520
1521<pre>
1522 opaque
1523</pre>
1524
1525<h5>Examples:</h5>
1526
1527<table class="layout">
1528 <tr class="layout">
Chris Lattner7311d222007-12-19 05:04:11 +0000[diff] [blame]1529 <td class="left"><tt>opaque</tt></td>
1530 <td class="left">An opaque type.</td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1531 </tr>
1532</table>
1533</div>
1534
1535
1536<!-- *********************************************************************** -->
1537<div class="doc_section"> <a name="constants">Constants</a> </div>
1538<!-- *********************************************************************** -->
1539
1540<div class="doc_text">
1541
1542<p>LLVM has several different basic types of constants. This section describes
1543them all and their syntax.</p>
1544
1545</div>
1546
1547<!-- ======================================================================= -->
1548<div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1549
1550<div class="doc_text">
1551
1552<dl>
1553 <dt><b>Boolean constants</b></dt>
1554
1555 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1556 constants of the <tt><a href="#t_primitive">i1</a></tt> type.
1557 </dd>
1558
1559 <dt><b>Integer constants</b></dt>
1560
1561 <dd>Standard integers (such as '4') are constants of the <a
1562 href="#t_integer">integer</a> type. Negative numbers may be used with
1563 integer types.
1564 </dd>
1565
1566 <dt><b>Floating point constants</b></dt>
1567
1568 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1569 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
Chris Lattnerdc15b1d2008-04-01 18:45:27 +0000[diff] [blame]1570 notation (see below). The assembler requires the exact decimal value of
1571 a floating-point constant. For example, the assembler accepts 1.25 but
1572 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
1573 constants must have a <a href="#t_floating">floating point</a> type. </dd>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1574
1575 <dt><b>Null pointer constants</b></dt>
1576
1577 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1578 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1579
1580</dl>
1581
1582<p>The one non-intuitive notation for constants is the optional hexadecimal form
1583of floating point constants. For example, the form '<tt>double
15840x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
15854.5e+15</tt>'. The only time hexadecimal floating point constants are required
1586(and the only time that they are generated by the disassembler) is when a
1587floating point constant must be emitted but it cannot be represented as a
1588decimal floating point number. For example, NaN's, infinities, and other
1589special values are represented in their IEEE hexadecimal format so that
1590assembly and disassembly do not cause any bits to change in the constants.</p>
1591
1592</div>
1593
1594<!-- ======================================================================= -->
1595<div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1596</div>
1597
1598<div class="doc_text">
1599<p>Aggregate constants arise from aggregation of simple constants
1600and smaller aggregate constants.</p>
1601
1602<dl>
1603 <dt><b>Structure constants</b></dt>
1604
1605 <dd>Structure constants are represented with notation similar to structure
1606 type definitions (a comma separated list of elements, surrounded by braces
Chris Lattner6c8de962007-12-25 20:34:52 +0000[diff] [blame]1607 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
1608 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]1609 must have <a href="#t_struct">structure type</a>, and the number and
1610 types of elements must match those specified by the type.
1611 </dd>
1612
1613 <dt><b>Array constants</b></dt>
1614
1615 <dd>Array constants are represented with notation similar to array type
1616 definitions (a comma separated list of elements, surrounded by square brackets
1617 (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74 ]</tt>". Array
1618 constants must have <a href="#t_array">array type</a>, and the number and
1619 types of elements must match those specified by the type.
1620 </dd>
1621
1622 <dt><b>Vector constants</b></dt>
1623
1624 <dd>Vector constants are represented with notation similar to vector type
1625 definitions (a comma separated list of elements, surrounded by
1626 less-than/greater-than's (<tt>&lt;&gt;</tt>)). For example: "<tt>&lt; i32 42,
1627 i32 11, i32 74, i32 100 &gt;</tt>". Vector constants must have <a
1628 href="#t_vector">vector type</a>, and the number and types of elements must
1629 match those specified by the type.
1630 </dd>
1631
1632 <dt><b>Zero initialization</b></dt>
1633
1634 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1635 value to zero of <em>any</em> type, including scalar and aggregate types.
1636 This is often used to avoid having to print large zero initializers (e.g. for
1637 large arrays) and is always exactly equivalent to using explicit zero
1638 initializers.
1639 </dd>
1640</dl>
1641
1642</div>
1643
1644<!-- ======================================================================= -->
1645<div class="doc_subsection">
1646 <a name="globalconstants">Global Variable and Function Addresses</a>
1647</div>
1648
1649<div class="doc_text">
1650
1651<p>The addresses of <a href="#globalvars">global variables</a> and <a
1652href="#functionstructure">functions</a> are always implicitly valid (link-time)
1653constants. These constants are explicitly referenced when the <a
1654href="#identifiers">identifier for the global</a> is used and always have <a
1655href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1656file:</p>
1657
1658<div class="doc_code">
1659<pre>
1660@X = global i32 17
1661@Y = global i32 42
1662@Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1663</pre>
1664</div>
1665
1666</div>
1667
1668<!-- ======================================================================= -->
1669<div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1670<div class="doc_text">
1671 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1672 no specific value. Undefined values may be of any type and be used anywhere
1673 a constant is permitted.</p>
1674
1675 <p>Undefined values indicate to the compiler that the program is well defined
1676 no matter what value is used, giving the compiler more freedom to optimize.
1677 </p>
1678</div>
1679
1680<!-- ======================================================================= -->
1681<div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1682</div>
1683
1684<div class="doc_text">
1685
1686<p>Constant expressions are used to allow expressions involving other constants
1687to be used as constants. Constant expressions may be of any <a
1688href="#t_firstclass">first class</a> type and may involve any LLVM operation
1689that does not have side effects (e.g. load and call are not supported). The
1690following is the syntax for constant expressions:</p>
1691
1692<dl>
1693 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1694 <dd>Truncate a constant to another type. The bit size of CST must be larger
1695 than the bit size of TYPE. Both types must be integers.</dd>
1696
1697 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1698 <dd>Zero extend a constant to another type. The bit size of CST must be
1699 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1700
1701 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1702 <dd>Sign extend a constant to another type. The bit size of CST must be
1703 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1704
1705 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1706 <dd>Truncate a floating point constant to another floating point type. The
1707 size of CST must be larger than the size of TYPE. Both types must be
1708 floating point.</dd>
1709
1710 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1711 <dd>Floating point extend a constant to another type. The size of CST must be
1712 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1713
Reid Spencere6adee82007-07-31 14:40:14 +0000[diff] [blame]1714 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1715 <dd>Convert a floating point constant to the corresponding unsigned integer
Nate Begeman78246ca2007-11-17 03:58:34 +0000[diff] [blame]1716 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1717 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1718 of the same number of elements. If the value won't fit in the integer type,
1719 the results are undefined.</dd>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1720
1721 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
1722 <dd>Convert a floating point constant to the corresponding signed integer
Nate Begeman78246ca2007-11-17 03:58:34 +0000[diff] [blame]1723 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1724 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1725 of the same number of elements. If the value won't fit in the integer type,
1726 the results are undefined.</dd>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1727
1728 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
1729 <dd>Convert an unsigned integer constant to the corresponding floating point
Nate Begeman78246ca2007-11-17 03:58:34 +0000[diff] [blame]1730 constant. TYPE must be a scalar or vector floating point type. CST must be of
1731 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1732 of the same number of elements. If the value won't fit in the floating point
1733 type, the results are undefined.</dd>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1734
1735 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
1736 <dd>Convert a signed integer constant to the corresponding floating point
Nate Begeman78246ca2007-11-17 03:58:34 +0000[diff] [blame]1737 constant. TYPE must be a scalar or vector floating point type. CST must be of
1738 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1739 of the same number of elements. If the value won't fit in the floating point
1740 type, the results are undefined.</dd>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1741
1742 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
1743 <dd>Convert a pointer typed constant to the corresponding integer constant
1744 TYPE must be an integer type. CST must be of pointer type. The CST value is
1745 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1746
1747 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
1748 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1749 pointer type. CST must be of integer type. The CST value is zero extended,
1750 truncated, or unchanged to make it fit in a pointer size. This one is
1751 <i>really</i> dangerous!</dd>
1752
1753 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
1754 <dd>Convert a constant, CST, to another TYPE. The size of CST and TYPE must be
1755 identical (same number of bits). The conversion is done as if the CST value
1756 was stored to memory and read back as TYPE. In other words, no bits change
1757 with this operator, just the type. This can be used for conversion of
1758 vector types to any other type, as long as they have the same bit width. For
Dan Gohman7305fa02008-09-08 16:45:59 +0000[diff] [blame]1759 pointers it is only valid to cast to another pointer type. It is not valid
1760 to bitcast to or from an aggregate type.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1761 </dd>
1762
1763 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1764
1765 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1766 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1767 instruction, the index list may have zero or more indexes, which are required
1768 to make sense for the type of "CSTPTR".</dd>
1769
1770 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1771
1772 <dd>Perform the <a href="#i_select">select operation</a> on
1773 constants.</dd>
1774
1775 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
1776 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
1777
1778 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
1779 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
1780
Nate Begeman646fa482008-05-12 19:01:56 +0000[diff] [blame]1781 <dt><b><tt>vicmp COND ( VAL1, VAL2 )</tt></b></dt>
1782 <dd>Performs the <a href="#i_vicmp">vicmp operation</a> on constants.</dd>
1783
1784 <dt><b><tt>vfcmp COND ( VAL1, VAL2 )</tt></b></dt>
1785 <dd>Performs the <a href="#i_vfcmp">vfcmp operation</a> on constants.</dd>
1786
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1787 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1788
1789 <dd>Perform the <a href="#i_extractelement">extractelement
1790 operation</a> on constants.
1791
1792 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1793
1794 <dd>Perform the <a href="#i_insertelement">insertelement
1795 operation</a> on constants.</dd>
1796
1797
1798 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
1799
1800 <dd>Perform the <a href="#i_shufflevector">shufflevector
1801 operation</a> on constants.</dd>
1802
1803 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1804
1805 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1806 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1807 binary</a> operations. The constraints on operands are the same as those for
1808 the corresponding instruction (e.g. no bitwise operations on floating point
1809 values are allowed).</dd>
1810</dl>
1811</div>
1812
1813<!-- *********************************************************************** -->
1814<div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1815<!-- *********************************************************************** -->
1816
1817<!-- ======================================================================= -->
1818<div class="doc_subsection">
1819<a name="inlineasm">Inline Assembler Expressions</a>
1820</div>
1821
1822<div class="doc_text">
1823
1824<p>
1825LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
1826Module-Level Inline Assembly</a>) through the use of a special value. This
1827value represents the inline assembler as a string (containing the instructions
1828to emit), a list of operand constraints (stored as a string), and a flag that
1829indicates whether or not the inline asm expression has side effects. An example
1830inline assembler expression is:
1831</p>
1832
1833<div class="doc_code">
1834<pre>
1835i32 (i32) asm "bswap $0", "=r,r"
1836</pre>
1837</div>
1838
1839<p>
1840Inline assembler expressions may <b>only</b> be used as the callee operand of
1841a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
1842</p>
1843
1844<div class="doc_code">
1845<pre>
1846%X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
1847</pre>
1848</div>
1849
1850<p>
1851Inline asms with side effects not visible in the constraint list must be marked
1852as having side effects. This is done through the use of the
1853'<tt>sideeffect</tt>' keyword, like so:
1854</p>
1855
1856<div class="doc_code">
1857<pre>
1858call void asm sideeffect "eieio", ""()
1859</pre>
1860</div>
1861
1862<p>TODO: The format of the asm and constraints string still need to be
1863documented here. Constraints on what can be done (e.g. duplication, moving, etc
Chris Lattner1cdd64e2008-10-04 18:36:02 +0000[diff] [blame]1864need to be documented). This is probably best done by reference to another
1865document that covers inline asm from a holistic perspective.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1866</p>
1867
1868</div>
1869
1870<!-- *********************************************************************** -->
1871<div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1872<!-- *********************************************************************** -->
1873
1874<div class="doc_text">
1875
1876<p>The LLVM instruction set consists of several different
1877classifications of instructions: <a href="#terminators">terminator
1878instructions</a>, <a href="#binaryops">binary instructions</a>,
1879<a href="#bitwiseops">bitwise binary instructions</a>, <a
1880 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1881instructions</a>.</p>
1882
1883</div>
1884
1885<!-- ======================================================================= -->
1886<div class="doc_subsection"> <a name="terminators">Terminator
1887Instructions</a> </div>
1888
1889<div class="doc_text">
1890
1891<p>As mentioned <a href="#functionstructure">previously</a>, every
1892basic block in a program ends with a "Terminator" instruction, which
1893indicates which block should be executed after the current block is
1894finished. These terminator instructions typically yield a '<tt>void</tt>'
1895value: they produce control flow, not values (the one exception being
1896the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1897<p>There are six different terminator instructions: the '<a
1898 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1899instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1900the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1901 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1902 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1903
1904</div>
1905
1906<!-- _______________________________________________________________________ -->
1907<div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1908Instruction</a> </div>
1909<div class="doc_text">
1910<h5>Syntax:</h5>
Dan Gohman3e700032008-10-04 19:00:07 +0000[diff] [blame]1911<pre>
1912 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]1913 ret void <i>; Return from void function</i>
1914</pre>
Chris Lattner43030e72008-04-23 04:59:35 +0000[diff] [blame]1915
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1916<h5>Overview:</h5>
Chris Lattner43030e72008-04-23 04:59:35 +0000[diff] [blame]1917
Dan Gohman3e700032008-10-04 19:00:07 +0000[diff] [blame]1918<p>The '<tt>ret</tt>' instruction is used to return control flow (and
1919optionally a value) from a function back to the caller.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1920<p>There are two forms of the '<tt>ret</tt>' instruction: one that
Dan Gohman3e700032008-10-04 19:00:07 +0000[diff] [blame]1921returns a value and then causes control flow, and one that just causes
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1922control flow to occur.</p>
Chris Lattner43030e72008-04-23 04:59:35 +0000[diff] [blame]1923
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1924<h5>Arguments:</h5>
Chris Lattner43030e72008-04-23 04:59:35 +0000[diff] [blame]1925
Dan Gohman3e700032008-10-04 19:00:07 +0000[diff] [blame]1926<p>The '<tt>ret</tt>' instruction optionally accepts a single argument,
1927the return value. The type of the return value must be a
1928'<a href="#t_firstclass">first class</a>' type.</p>
1929
1930<p>A function is not <a href="#wellformed">well formed</a> if
1931it it has a non-void return type and contains a '<tt>ret</tt>'
1932instruction with no return value or a return value with a type that
1933does not match its type, or if it has a void return type and contains
1934a '<tt>ret</tt>' instruction with a return value.</p>
Chris Lattner43030e72008-04-23 04:59:35 +0000[diff] [blame]1935
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1936<h5>Semantics:</h5>
Chris Lattner43030e72008-04-23 04:59:35 +0000[diff] [blame]1937
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1938<p>When the '<tt>ret</tt>' instruction is executed, control flow
1939returns back to the calling function's context. If the caller is a "<a
1940 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1941the instruction after the call. If the caller was an "<a
1942 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1943at the beginning of the "normal" destination block. If the instruction
1944returns a value, that value shall set the call or invoke instruction's
Dan Gohman3e700032008-10-04 19:00:07 +0000[diff] [blame]1945return value.
Chris Lattner43030e72008-04-23 04:59:35 +0000[diff] [blame]1946
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1947<h5>Example:</h5>
Chris Lattner43030e72008-04-23 04:59:35 +0000[diff] [blame]1948
1949<pre>
1950 ret i32 5 <i>; Return an integer value of 5</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1951 ret void <i>; Return from a void function</i>
Dan Gohman3e700032008-10-04 19:00:07 +0000[diff] [blame]1952 ret { i32, i8 } { i32 4, i8 2 } <i>; Return an aggregate of values 4 and 2</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1953</pre>
1954</div>
1955<!-- _______________________________________________________________________ -->
1956<div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1957<div class="doc_text">
1958<h5>Syntax:</h5>
1959<pre> br i1 &lt;cond&gt;, label &lt;iftrue&gt;, label &lt;iffalse&gt;<br> br label &lt;dest&gt; <i>; Unconditional branch</i>
1960</pre>
1961<h5>Overview:</h5>
1962<p>The '<tt>br</tt>' instruction is used to cause control flow to
1963transfer to a different basic block in the current function. There are
1964two forms of this instruction, corresponding to a conditional branch
1965and an unconditional branch.</p>
1966<h5>Arguments:</h5>
1967<p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1968single '<tt>i1</tt>' value and two '<tt>label</tt>' values. The
1969unconditional form of the '<tt>br</tt>' instruction takes a single
1970'<tt>label</tt>' value as a target.</p>
1971<h5>Semantics:</h5>
1972<p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
1973argument is evaluated. If the value is <tt>true</tt>, control flows
1974to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1975control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1976<h5>Example:</h5>
1977<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
1978 href="#i_ret">ret</a> i32 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> i32 0<br></pre>
1979</div>
1980<!-- _______________________________________________________________________ -->
1981<div class="doc_subsubsection">
1982 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
1983</div>
1984
1985<div class="doc_text">
1986<h5>Syntax:</h5>
1987
1988<pre>
1989 switch &lt;intty&gt; &lt;value&gt;, label &lt;defaultdest&gt; [ &lt;intty&gt; &lt;val&gt;, label &lt;dest&gt; ... ]
1990</pre>
1991
1992<h5>Overview:</h5>
1993
1994<p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
1995several different places. It is a generalization of the '<tt>br</tt>'
1996instruction, allowing a branch to occur to one of many possible
1997destinations.</p>
1998
1999
2000<h5>Arguments:</h5>
2001
2002<p>The '<tt>switch</tt>' instruction uses three parameters: an integer
2003comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
2004an array of pairs of comparison value constants and '<tt>label</tt>'s. The
2005table is not allowed to contain duplicate constant entries.</p>
2006
2007<h5>Semantics:</h5>
2008
2009<p>The <tt>switch</tt> instruction specifies a table of values and
2010destinations. When the '<tt>switch</tt>' instruction is executed, this
2011table is searched for the given value. If the value is found, control flow is
2012transfered to the corresponding destination; otherwise, control flow is
2013transfered to the default destination.</p>
2014
2015<h5>Implementation:</h5>
2016
2017<p>Depending on properties of the target machine and the particular
2018<tt>switch</tt> instruction, this instruction may be code generated in different
2019ways. For example, it could be generated as a series of chained conditional
2020branches or with a lookup table.</p>
2021
2022<h5>Example:</h5>
2023
2024<pre>
2025 <i>; Emulate a conditional br instruction</i>
2026 %Val = <a href="#i_zext">zext</a> i1 %value to i32
2027 switch i32 %Val, label %truedest [i32 0, label %falsedest ]
2028
2029 <i>; Emulate an unconditional br instruction</i>
2030 switch i32 0, label %dest [ ]
2031
2032 <i>; Implement a jump table:</i>
2033 switch i32 %val, label %otherwise [ i32 0, label %onzero
2034 i32 1, label %onone
2035 i32 2, label %ontwo ]
2036</pre>
2037</div>
2038
2039<!-- _______________________________________________________________________ -->
2040<div class="doc_subsubsection">
2041 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
2042</div>
2043
2044<div class="doc_text">
2045
2046<h5>Syntax:</h5>
2047
2048<pre>
Devang Patelac2fc272008-10-06 18:50:38 +0000[diff] [blame]2049 &lt;result&gt; = invoke [<a href="#callingconv">cconv</a>] [<a href="#pa\
2050ramattrs">RetAttrs</a>] &lt;ptr to function ty&gt; &lt;function ptr val&gt;(&lt;function args&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2051 to label &lt;normal label&gt; unwind label &lt;exception label&gt;
2052</pre>
2053
2054<h5>Overview:</h5>
2055
2056<p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
2057function, with the possibility of control flow transfer to either the
2058'<tt>normal</tt>' label or the
2059'<tt>exception</tt>' label. If the callee function returns with the
2060"<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
2061"normal" label. If the callee (or any indirect callees) returns with the "<a
2062href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
Dan Gohman3e700032008-10-04 19:00:07 +0000[diff] [blame]2063continued at the dynamically nearest "exception" label.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2064
2065<h5>Arguments:</h5>
2066
2067<p>This instruction requires several arguments:</p>
2068
2069<ol>
2070 <li>
2071 The optional "cconv" marker indicates which <a href="#callingconv">calling
2072 convention</a> the call should use. If none is specified, the call defaults
2073 to using C calling conventions.
2074 </li>
Devang Patelac2fc272008-10-06 18:50:38 +0000[diff] [blame]2075
2076 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
2077 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>',
2078 and '<tt>inreg</tt>' attributes are valid here.</li>
2079
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2080 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
2081 function value being invoked. In most cases, this is a direct function
2082 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
2083 an arbitrary pointer to function value.
2084 </li>
2085
2086 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
2087 function to be invoked. </li>
2088
2089 <li>'<tt>function args</tt>': argument list whose types match the function
2090 signature argument types. If the function signature indicates the function
2091 accepts a variable number of arguments, the extra arguments can be
2092 specified. </li>
2093
2094 <li>'<tt>normal label</tt>': the label reached when the called function
2095 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
2096
2097 <li>'<tt>exception label</tt>': the label reached when a callee returns with
2098 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
2099
Devang Patelac2fc272008-10-06 18:50:38 +0000[diff] [blame]2100 <li>The optional <a href="fnattrs">function attributes</a> list. Only
2101 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
2102 '<tt>readnone</tt>' attributes are valid here.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2103</ol>
2104
2105<h5>Semantics:</h5>
2106
2107<p>This instruction is designed to operate as a standard '<tt><a
2108href="#i_call">call</a></tt>' instruction in most regards. The primary
2109difference is that it establishes an association with a label, which is used by
2110the runtime library to unwind the stack.</p>
2111
2112<p>This instruction is used in languages with destructors to ensure that proper
2113cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
2114exception. Additionally, this is important for implementation of
2115'<tt>catch</tt>' clauses in high-level languages that support them.</p>
2116
2117<h5>Example:</h5>
2118<pre>
Nick Lewyckya1c11a12008-03-16 07:18:12 +0000[diff] [blame]2119 %retval = invoke i32 @Test(i32 15) to label %Continue
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2120 unwind label %TestCleanup <i>; {i32}:retval set</i>
Nick Lewyckya1c11a12008-03-16 07:18:12 +0000[diff] [blame]2121 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2122 unwind label %TestCleanup <i>; {i32}:retval set</i>
2123</pre>
2124</div>
2125
2126
2127<!-- _______________________________________________________________________ -->
2128
2129<div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
2130Instruction</a> </div>
2131
2132<div class="doc_text">
2133
2134<h5>Syntax:</h5>
2135<pre>
2136 unwind
2137</pre>
2138
2139<h5>Overview:</h5>
2140
2141<p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
2142at the first callee in the dynamic call stack which used an <a
2143href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
2144primarily used to implement exception handling.</p>
2145
2146<h5>Semantics:</h5>
2147
Chris Lattner8b094fc2008-04-19 21:01:16 +0000[diff] [blame]2148<p>The '<tt>unwind</tt>' instruction causes execution of the current function to
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2149immediately halt. The dynamic call stack is then searched for the first <a
2150href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
2151execution continues at the "exceptional" destination block specified by the
2152<tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
2153dynamic call chain, undefined behavior results.</p>
2154</div>
2155
2156<!-- _______________________________________________________________________ -->
2157
2158<div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
2159Instruction</a> </div>
2160
2161<div class="doc_text">
2162
2163<h5>Syntax:</h5>
2164<pre>
2165 unreachable
2166</pre>
2167
2168<h5>Overview:</h5>
2169
2170<p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
2171instruction is used to inform the optimizer that a particular portion of the
2172code is not reachable. This can be used to indicate that the code after a
2173no-return function cannot be reached, and other facts.</p>
2174
2175<h5>Semantics:</h5>
2176
2177<p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
2178</div>
2179
2180
2181
2182<!-- ======================================================================= -->
2183<div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
2184<div class="doc_text">
2185<p>Binary operators are used to do most of the computation in a
Chris Lattnerab596d92008-04-01 18:47:32 +0000[diff] [blame]2186program. They require two operands of the same type, execute an operation on them, and
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2187produce a single value. The operands might represent
2188multiple 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]2189The result value has the same type as its operands.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2190<p>There are several different binary operators:</p>
2191</div>
2192<!-- _______________________________________________________________________ -->
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2193<div class="doc_subsubsection">
2194 <a name="i_add">'<tt>add</tt>' Instruction</a>
2195</div>
2196
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2197<div class="doc_text">
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2198
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2199<h5>Syntax:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2200
2201<pre>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2202 &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]2203</pre>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2204
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2205<h5>Overview:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2206
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2207<p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2208
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2209<h5>Arguments:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2210
2211<p>The two arguments to the '<tt>add</tt>' instruction must be <a
2212 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>, or
2213 <a href="#t_vector">vector</a> values. Both arguments must have identical
2214 types.</p>
2215
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2216<h5>Semantics:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2217
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2218<p>The value produced is the integer or floating point sum of the two
2219operands.</p>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2220
Chris Lattner9aba1e22008-01-28 00:36:27 +0000[diff] [blame]2221<p>If an integer sum has unsigned overflow, the result returned is the
2222mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2223the result.</p>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2224
Chris Lattner9aba1e22008-01-28 00:36:27 +0000[diff] [blame]2225<p>Because LLVM integers use a two's complement representation, this
2226instruction is appropriate for both signed and unsigned integers.</p>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2227
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2228<h5>Example:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2229
2230<pre>
2231 &lt;result&gt; = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2232</pre>
2233</div>
2234<!-- _______________________________________________________________________ -->
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2235<div class="doc_subsubsection">
2236 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
2237</div>
2238
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2239<div class="doc_text">
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2240
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2241<h5>Syntax:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2242
2243<pre>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2244 &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]2245</pre>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2246
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2247<h5>Overview:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2248
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2249<p>The '<tt>sub</tt>' instruction returns the difference of its two
2250operands.</p>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2251
2252<p>Note that the '<tt>sub</tt>' instruction is used to represent the
2253'<tt>neg</tt>' instruction present in most other intermediate
2254representations.</p>
2255
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2256<h5>Arguments:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2257
2258<p>The two arguments to the '<tt>sub</tt>' instruction must be <a
2259 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>,
2260 or <a href="#t_vector">vector</a> values. Both arguments must have identical
2261 types.</p>
2262
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2263<h5>Semantics:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2264
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2265<p>The value produced is the integer or floating point difference of
2266the two operands.</p>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2267
Chris Lattner9aba1e22008-01-28 00:36:27 +0000[diff] [blame]2268<p>If an integer difference has unsigned overflow, the result returned is the
2269mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2270the result.</p>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2271
Chris Lattner9aba1e22008-01-28 00:36:27 +0000[diff] [blame]2272<p>Because LLVM integers use a two's complement representation, this
2273instruction is appropriate for both signed and unsigned integers.</p>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2274
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2275<h5>Example:</h5>
2276<pre>
2277 &lt;result&gt; = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
2278 &lt;result&gt; = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
2279</pre>
2280</div>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2281
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2282<!-- _______________________________________________________________________ -->
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2283<div class="doc_subsubsection">
2284 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
2285</div>
2286
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2287<div class="doc_text">
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2288
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2289<h5>Syntax:</h5>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2290<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]2291</pre>
2292<h5>Overview:</h5>
2293<p>The '<tt>mul</tt>' instruction returns the product of its two
2294operands.</p>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2295
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2296<h5>Arguments:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2297
2298<p>The two arguments to the '<tt>mul</tt>' instruction must be <a
2299href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>,
2300or <a href="#t_vector">vector</a> values. Both arguments must have identical
2301types.</p>
2302
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2303<h5>Semantics:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2304
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2305<p>The value produced is the integer or floating point product of the
2306two operands.</p>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2307
Chris Lattner9aba1e22008-01-28 00:36:27 +0000[diff] [blame]2308<p>If the result of an integer multiplication has unsigned overflow,
2309the result returned is the mathematical result modulo
23102<sup>n</sup>, where n is the bit width of the result.</p>
2311<p>Because LLVM integers use a two's complement representation, and the
2312result is the same width as the operands, this instruction returns the
2313correct result for both signed and unsigned integers. If a full product
2314(e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands
2315should be sign-extended or zero-extended as appropriate to the
2316width of the full product.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2317<h5>Example:</h5>
2318<pre> &lt;result&gt; = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
2319</pre>
2320</div>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2321
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2322<!-- _______________________________________________________________________ -->
2323<div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
2324</a></div>
2325<div class="doc_text">
2326<h5>Syntax:</h5>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2327<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]2328</pre>
2329<h5>Overview:</h5>
2330<p>The '<tt>udiv</tt>' instruction returns the quotient of its two
2331operands.</p>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2332
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2333<h5>Arguments:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2334
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2335<p>The two arguments to the '<tt>udiv</tt>' instruction must be
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2336<a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2337values. Both arguments must have identical types.</p>
2338
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2339<h5>Semantics:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2340
Chris Lattner9aba1e22008-01-28 00:36:27 +0000[diff] [blame]2341<p>The value produced is the unsigned integer quotient of the two operands.</p>
2342<p>Note that unsigned integer division and signed integer division are distinct
2343operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
2344<p>Division by zero leads to undefined behavior.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2345<h5>Example:</h5>
2346<pre> &lt;result&gt; = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2347</pre>
2348</div>
2349<!-- _______________________________________________________________________ -->
2350<div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
2351</a> </div>
2352<div class="doc_text">
2353<h5>Syntax:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2354<pre>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2355 &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]2356</pre>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2357
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2358<h5>Overview:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2359
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2360<p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
2361operands.</p>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2362
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2363<h5>Arguments:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2364
2365<p>The two arguments to the '<tt>sdiv</tt>' instruction must be
2366<a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2367values. Both arguments must have identical types.</p>
2368
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2369<h5>Semantics:</h5>
Chris Lattnerdc15b1d2008-04-01 18:45:27 +0000[diff] [blame]2370<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]2371<p>Note that signed integer division and unsigned integer division are distinct
2372operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
2373<p>Division by zero leads to undefined behavior. Overflow also leads to
2374undefined behavior; this is a rare case, but can occur, for example,
2375by doing a 32-bit division of -2147483648 by -1.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2376<h5>Example:</h5>
2377<pre> &lt;result&gt; = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2378</pre>
2379</div>
2380<!-- _______________________________________________________________________ -->
2381<div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
2382Instruction</a> </div>
2383<div class="doc_text">
2384<h5>Syntax:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2385<pre>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2386 &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]2387</pre>
2388<h5>Overview:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2389
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2390<p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
2391operands.</p>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2392
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2393<h5>Arguments:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2394
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2395<p>The two arguments to the '<tt>fdiv</tt>' instruction must be
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2396<a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a>
2397of floating point values. Both arguments must have identical types.</p>
2398
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2399<h5>Semantics:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2400
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2401<p>The value produced is the floating point quotient of the two operands.</p>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2402
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2403<h5>Example:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2404
2405<pre>
2406 &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]2407</pre>
2408</div>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2409
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2410<!-- _______________________________________________________________________ -->
2411<div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
2412</div>
2413<div class="doc_text">
2414<h5>Syntax:</h5>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2415<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]2416</pre>
2417<h5>Overview:</h5>
2418<p>The '<tt>urem</tt>' instruction returns the remainder from the
2419unsigned division of its two arguments.</p>
2420<h5>Arguments:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2421<p>The two arguments to the '<tt>urem</tt>' instruction must be
2422<a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2423values. Both arguments must have identical types.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2424<h5>Semantics:</h5>
2425<p>This instruction returns the unsigned integer <i>remainder</i> of a division.
Chris Lattnerdc15b1d2008-04-01 18:45:27 +0000[diff] [blame]2426This instruction always performs an unsigned division to get the remainder.</p>
Chris Lattner9aba1e22008-01-28 00:36:27 +0000[diff] [blame]2427<p>Note that unsigned integer remainder and signed integer remainder are
2428distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
2429<p>Taking the remainder of a division by zero leads to undefined behavior.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2430<h5>Example:</h5>
2431<pre> &lt;result&gt; = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2432</pre>
2433
2434</div>
2435<!-- _______________________________________________________________________ -->
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2436<div class="doc_subsubsection">
2437 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
2438</div>
2439
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2440<div class="doc_text">
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2441
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2442<h5>Syntax:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2443
2444<pre>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2445 &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]2446</pre>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2447
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2448<h5>Overview:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2449
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2450<p>The '<tt>srem</tt>' instruction returns the remainder from the
Dan Gohman3e3fd8c2007-11-05 23:35:22 +0000[diff] [blame]2451signed division of its two operands. This instruction can also take
2452<a href="#t_vector">vector</a> versions of the values in which case
2453the elements must be integers.</p>
Chris Lattner08497ce2008-01-04 04:33:49 +0000[diff] [blame]2454
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2455<h5>Arguments:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2456
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2457<p>The two arguments to the '<tt>srem</tt>' instruction must be
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2458<a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2459values. Both arguments must have identical types.</p>
2460
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2461<h5>Semantics:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2462
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2463<p>This instruction returns the <i>remainder</i> of a division (where the result
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2464has the same sign as the dividend, <tt>op1</tt>), not the <i>modulo</i>
2465operator (where the result has the same sign as the divisor, <tt>op2</tt>) of
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2466a value. For more information about the difference, see <a
2467 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
2468Math Forum</a>. For a table of how this is implemented in various languages,
2469please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
2470Wikipedia: modulo operation</a>.</p>
Chris Lattner9aba1e22008-01-28 00:36:27 +0000[diff] [blame]2471<p>Note that signed integer remainder and unsigned integer remainder are
2472distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
2473<p>Taking the remainder of a division by zero leads to undefined behavior.
2474Overflow also leads to undefined behavior; this is a rare case, but can occur,
2475for example, by taking the remainder of a 32-bit division of -2147483648 by -1.
2476(The remainder doesn't actually overflow, but this rule lets srem be
2477implemented using instructions that return both the result of the division
2478and the remainder.)</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2479<h5>Example:</h5>
2480<pre> &lt;result&gt; = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2481</pre>
2482
2483</div>
2484<!-- _______________________________________________________________________ -->
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2485<div class="doc_subsubsection">
2486 <a name="i_frem">'<tt>frem</tt>' Instruction</a> </div>
2487
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2488<div class="doc_text">
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2489
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2490<h5>Syntax:</h5>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2491<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]2492</pre>
2493<h5>Overview:</h5>
2494<p>The '<tt>frem</tt>' instruction returns the remainder from the
2495division of its two operands.</p>
2496<h5>Arguments:</h5>
2497<p>The two arguments to the '<tt>frem</tt>' instruction must be
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2498<a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a>
2499of floating point values. Both arguments must have identical types.</p>
2500
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2501<h5>Semantics:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2502
Chris Lattnerdc15b1d2008-04-01 18:45:27 +0000[diff] [blame]2503<p>This instruction returns the <i>remainder</i> of a division.
2504The remainder has the same sign as the dividend.</p>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2505
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2506<h5>Example:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2507
2508<pre>
2509 &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]2510</pre>
2511</div>
2512
2513<!-- ======================================================================= -->
2514<div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
2515Operations</a> </div>
2516<div class="doc_text">
2517<p>Bitwise binary operators are used to do various forms of
2518bit-twiddling in a program. They are generally very efficient
2519instructions and can commonly be strength reduced from other
Chris Lattnerdc15b1d2008-04-01 18:45:27 +0000[diff] [blame]2520instructions. They require two operands of the same type, execute an operation on them,
2521and 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]2522</div>
2523
2524<!-- _______________________________________________________________________ -->
2525<div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
2526Instruction</a> </div>
2527<div class="doc_text">
2528<h5>Syntax:</h5>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2529<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]2530</pre>
Chris Lattnerd939d9f2007-10-03 21:01:14 +0000[diff] [blame]2531
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2532<h5>Overview:</h5>
Chris Lattnerd939d9f2007-10-03 21:01:14 +0000[diff] [blame]2533
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2534<p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2535the left a specified number of bits.</p>
Chris Lattnerd939d9f2007-10-03 21:01:14 +0000[diff] [blame]2536
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2537<h5>Arguments:</h5>
Chris Lattnerd939d9f2007-10-03 21:01:14 +0000[diff] [blame]2538
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2539<p>Both arguments to the '<tt>shl</tt>' instruction must be the same <a
Nate Begemanbb1ce942008-07-29 15:49:41 +0000[diff] [blame]2540 href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2541type. '<tt>op2</tt>' is treated as an unsigned value.</p>
Chris Lattnerd939d9f2007-10-03 21:01:14 +0000[diff] [blame]2542
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2543<h5>Semantics:</h5>
Chris Lattnerd939d9f2007-10-03 21:01:14 +0000[diff] [blame]2544
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2545<p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod 2<sup>n</sup>,
2546where n is the width of the result. If <tt>op2</tt> is (statically or dynamically) negative or
2547equal to or larger than the number of bits in <tt>op1</tt>, the result is undefined.</p>
Chris Lattnerd939d9f2007-10-03 21:01:14 +0000[diff] [blame]2548
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2549<h5>Example:</h5><pre>
2550 &lt;result&gt; = shl i32 4, %var <i>; yields {i32}: 4 &lt;&lt; %var</i>
2551 &lt;result&gt; = shl i32 4, 2 <i>; yields {i32}: 16</i>
2552 &lt;result&gt; = shl i32 1, 10 <i>; yields {i32}: 1024</i>
Chris Lattnerd939d9f2007-10-03 21:01:14 +0000[diff] [blame]2553 &lt;result&gt; = shl i32 1, 32 <i>; undefined</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2554</pre>
2555</div>
2556<!-- _______________________________________________________________________ -->
2557<div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
2558Instruction</a> </div>
2559<div class="doc_text">
2560<h5>Syntax:</h5>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2561<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]2562</pre>
2563
2564<h5>Overview:</h5>
2565<p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
2566operand shifted to the right a specified number of bits with zero fill.</p>
2567
2568<h5>Arguments:</h5>
2569<p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
Nate Begemanbb1ce942008-07-29 15:49:41 +0000[diff] [blame]2570<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]2571type. '<tt>op2</tt>' is treated as an unsigned value.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2572
2573<h5>Semantics:</h5>
Chris Lattnerd939d9f2007-10-03 21:01:14 +0000[diff] [blame]2574
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2575<p>This instruction always performs a logical shift right operation. The most
2576significant bits of the result will be filled with zero bits after the
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2577shift. If <tt>op2</tt> is (statically or dynamically) equal to or larger than
2578the number of bits in <tt>op1</tt>, the result is undefined.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2579
2580<h5>Example:</h5>
2581<pre>
2582 &lt;result&gt; = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
2583 &lt;result&gt; = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
2584 &lt;result&gt; = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
2585 &lt;result&gt; = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
Chris Lattnerd939d9f2007-10-03 21:01:14 +0000[diff] [blame]2586 &lt;result&gt; = lshr i32 1, 32 <i>; undefined</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2587</pre>
2588</div>
2589
2590<!-- _______________________________________________________________________ -->
2591<div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
2592Instruction</a> </div>
2593<div class="doc_text">
2594
2595<h5>Syntax:</h5>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2596<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]2597</pre>
2598
2599<h5>Overview:</h5>
2600<p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
2601operand shifted to the right a specified number of bits with sign extension.</p>
2602
2603<h5>Arguments:</h5>
2604<p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
Nate Begemanbb1ce942008-07-29 15:49:41 +0000[diff] [blame]2605<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]2606type. '<tt>op2</tt>' is treated as an unsigned value.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2607
2608<h5>Semantics:</h5>
2609<p>This instruction always performs an arithmetic shift right operation,
2610The most significant bits of the result will be filled with the sign bit
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2611of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
2612larger than the number of bits in <tt>op1</tt>, the result is undefined.
Chris Lattnerd939d9f2007-10-03 21:01:14 +0000[diff] [blame]2613</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2614
2615<h5>Example:</h5>
2616<pre>
2617 &lt;result&gt; = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
2618 &lt;result&gt; = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
2619 &lt;result&gt; = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
2620 &lt;result&gt; = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
Chris Lattnerd939d9f2007-10-03 21:01:14 +0000[diff] [blame]2621 &lt;result&gt; = ashr i32 1, 32 <i>; undefined</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2622</pre>
2623</div>
2624
2625<!-- _______________________________________________________________________ -->
2626<div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
2627Instruction</a> </div>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2628
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2629<div class="doc_text">
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2630
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2631<h5>Syntax:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2632
2633<pre>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2634 &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]2635</pre>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2636
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2637<h5>Overview:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2638
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2639<p>The '<tt>and</tt>' instruction returns the bitwise logical and of
2640its two operands.</p>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2641
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2642<h5>Arguments:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2643
2644<p>The two arguments to the '<tt>and</tt>' instruction must be
2645<a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2646values. Both arguments must have identical types.</p>
2647
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2648<h5>Semantics:</h5>
2649<p>The truth table used for the '<tt>and</tt>' instruction is:</p>
2650<p> </p>
Bill Wendling1e19eac2008-09-07 10:29:20 +0000[diff] [blame]2651<div>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2652<table border="1" cellspacing="0" cellpadding="4">
2653 <tbody>
2654 <tr>
2655 <td>In0</td>
2656 <td>In1</td>
2657 <td>Out</td>
2658 </tr>
2659 <tr>
2660 <td>0</td>
2661 <td>0</td>
2662 <td>0</td>
2663 </tr>
2664 <tr>
2665 <td>0</td>
2666 <td>1</td>
2667 <td>0</td>
2668 </tr>
2669 <tr>
2670 <td>1</td>
2671 <td>0</td>
2672 <td>0</td>
2673 </tr>
2674 <tr>
2675 <td>1</td>
2676 <td>1</td>
2677 <td>1</td>
2678 </tr>
2679 </tbody>
2680</table>
2681</div>
2682<h5>Example:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2683<pre>
2684 &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]2685 &lt;result&gt; = and i32 15, 40 <i>; yields {i32}:result = 8</i>
2686 &lt;result&gt; = and i32 4, 8 <i>; yields {i32}:result = 0</i>
2687</pre>
2688</div>
2689<!-- _______________________________________________________________________ -->
2690<div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
2691<div class="doc_text">
2692<h5>Syntax:</h5>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2693<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]2694</pre>
2695<h5>Overview:</h5>
2696<p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
2697or of its two operands.</p>
2698<h5>Arguments:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2699
2700<p>The two arguments to the '<tt>or</tt>' instruction must be
2701<a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2702values. Both arguments must have identical types.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2703<h5>Semantics:</h5>
2704<p>The truth table used for the '<tt>or</tt>' instruction is:</p>
2705<p> </p>
Bill Wendling1e19eac2008-09-07 10:29:20 +0000[diff] [blame]2706<div>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2707<table border="1" cellspacing="0" cellpadding="4">
2708 <tbody>
2709 <tr>
2710 <td>In0</td>
2711 <td>In1</td>
2712 <td>Out</td>
2713 </tr>
2714 <tr>
2715 <td>0</td>
2716 <td>0</td>
2717 <td>0</td>
2718 </tr>
2719 <tr>
2720 <td>0</td>
2721 <td>1</td>
2722 <td>1</td>
2723 </tr>
2724 <tr>
2725 <td>1</td>
2726 <td>0</td>
2727 <td>1</td>
2728 </tr>
2729 <tr>
2730 <td>1</td>
2731 <td>1</td>
2732 <td>1</td>
2733 </tr>
2734 </tbody>
2735</table>
2736</div>
2737<h5>Example:</h5>
2738<pre> &lt;result&gt; = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
2739 &lt;result&gt; = or i32 15, 40 <i>; yields {i32}:result = 47</i>
2740 &lt;result&gt; = or i32 4, 8 <i>; yields {i32}:result = 12</i>
2741</pre>
2742</div>
2743<!-- _______________________________________________________________________ -->
2744<div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
2745Instruction</a> </div>
2746<div class="doc_text">
2747<h5>Syntax:</h5>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]2748<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]2749</pre>
2750<h5>Overview:</h5>
2751<p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
2752or of its two operands. The <tt>xor</tt> is used to implement the
2753"one's complement" operation, which is the "~" operator in C.</p>
2754<h5>Arguments:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2755<p>The two arguments to the '<tt>xor</tt>' instruction must be
2756<a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2757values. Both arguments must have identical types.</p>
2758
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2759<h5>Semantics:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]2760
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2761<p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
2762<p> </p>
Bill Wendling1e19eac2008-09-07 10:29:20 +0000[diff] [blame]2763<div>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2764<table border="1" cellspacing="0" cellpadding="4">
2765 <tbody>
2766 <tr>
2767 <td>In0</td>
2768 <td>In1</td>
2769 <td>Out</td>
2770 </tr>
2771 <tr>
2772 <td>0</td>
2773 <td>0</td>
2774 <td>0</td>
2775 </tr>
2776 <tr>
2777 <td>0</td>
2778 <td>1</td>
2779 <td>1</td>
2780 </tr>
2781 <tr>
2782 <td>1</td>
2783 <td>0</td>
2784 <td>1</td>
2785 </tr>
2786 <tr>
2787 <td>1</td>
2788 <td>1</td>
2789 <td>0</td>
2790 </tr>
2791 </tbody>
2792</table>
2793</div>
2794<p> </p>
2795<h5>Example:</h5>
2796<pre> &lt;result&gt; = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
2797 &lt;result&gt; = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
2798 &lt;result&gt; = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
2799 &lt;result&gt; = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
2800</pre>
2801</div>
2802
2803<!-- ======================================================================= -->
2804<div class="doc_subsection">
2805 <a name="vectorops">Vector Operations</a>
2806</div>
2807
2808<div class="doc_text">
2809
2810<p>LLVM supports several instructions to represent vector operations in a
2811target-independent manner. These instructions cover the element-access and
2812vector-specific operations needed to process vectors effectively. While LLVM
2813does directly support these vector operations, many sophisticated algorithms
2814will want to use target-specific intrinsics to take full advantage of a specific
2815target.</p>
2816
2817</div>
2818
2819<!-- _______________________________________________________________________ -->
2820<div class="doc_subsubsection">
2821 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2822</div>
2823
2824<div class="doc_text">
2825
2826<h5>Syntax:</h5>
2827
2828<pre>
2829 &lt;result&gt; = extractelement &lt;n x &lt;ty&gt;&gt; &lt;val&gt;, i32 &lt;idx&gt; <i>; yields &lt;ty&gt;</i>
2830</pre>
2831
2832<h5>Overview:</h5>
2833
2834<p>
2835The '<tt>extractelement</tt>' instruction extracts a single scalar
2836element from a vector at a specified index.
2837</p>
2838
2839
2840<h5>Arguments:</h5>
2841
2842<p>
2843The first operand of an '<tt>extractelement</tt>' instruction is a
2844value of <a href="#t_vector">vector</a> type. The second operand is
2845an index indicating the position from which to extract the element.
2846The index may be a variable.</p>
2847
2848<h5>Semantics:</h5>
2849
2850<p>
2851The result is a scalar of the same type as the element type of
2852<tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2853<tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2854results are undefined.
2855</p>
2856
2857<h5>Example:</h5>
2858
2859<pre>
2860 %result = extractelement &lt;4 x i32&gt; %vec, i32 0 <i>; yields i32</i>
2861</pre>
2862</div>
2863
2864
2865<!-- _______________________________________________________________________ -->
2866<div class="doc_subsubsection">
2867 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
2868</div>
2869
2870<div class="doc_text">
2871
2872<h5>Syntax:</h5>
2873
2874<pre>
Dan Gohmanbcc3c502008-05-12 23:38:42 +0000[diff] [blame]2875 &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]2876</pre>
2877
2878<h5>Overview:</h5>
2879
2880<p>
2881The '<tt>insertelement</tt>' instruction inserts a scalar
2882element into a vector at a specified index.
2883</p>
2884
2885
2886<h5>Arguments:</h5>
2887
2888<p>
2889The first operand of an '<tt>insertelement</tt>' instruction is a
2890value of <a href="#t_vector">vector</a> type. The second operand is a
2891scalar value whose type must equal the element type of the first
2892operand. The third operand is an index indicating the position at
2893which to insert the value. The index may be a variable.</p>
2894
2895<h5>Semantics:</h5>
2896
2897<p>
2898The result is a vector of the same type as <tt>val</tt>. Its
2899element values are those of <tt>val</tt> except at position
2900<tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
2901exceeds the length of <tt>val</tt>, the results are undefined.
2902</p>
2903
2904<h5>Example:</h5>
2905
2906<pre>
2907 %result = insertelement &lt;4 x i32&gt; %vec, i32 1, i32 0 <i>; yields &lt;4 x i32&gt;</i>
2908</pre>
2909</div>
2910
2911<!-- _______________________________________________________________________ -->
2912<div class="doc_subsubsection">
2913 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
2914</div>
2915
2916<div class="doc_text">
2917
2918<h5>Syntax:</h5>
2919
2920<pre>
2921 &lt;result&gt; = shufflevector &lt;n x &lt;ty&gt;&gt; &lt;v1&gt;, &lt;n x &lt;ty&gt;&gt; &lt;v2&gt;, &lt;n x i32&gt; &lt;mask&gt; <i>; yields &lt;n x &lt;ty&gt;&gt;</i>
2922</pre>
2923
2924<h5>Overview:</h5>
2925
2926<p>
2927The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
2928from two input vectors, returning a vector of the same type.
2929</p>
2930
2931<h5>Arguments:</h5>
2932
2933<p>
2934The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
2935with types that match each other and types that match the result of the
2936instruction. The third argument is a shuffle mask, which has the same number
2937of elements as the other vector type, but whose element type is always 'i32'.
2938</p>
2939
2940<p>
2941The shuffle mask operand is required to be a constant vector with either
2942constant integer or undef values.
2943</p>
2944
2945<h5>Semantics:</h5>
2946
2947<p>
2948The elements of the two input vectors are numbered from left to right across
2949both of the vectors. The shuffle mask operand specifies, for each element of
2950the result vector, which element of the two input registers the result element
2951gets. The element selector may be undef (meaning "don't care") and the second
2952operand may be undef if performing a shuffle from only one vector.
2953</p>
2954
2955<h5>Example:</h5>
2956
2957<pre>
2958 %result = shufflevector &lt;4 x i32&gt; %v1, &lt;4 x i32&gt; %v2,
2959 &lt;4 x i32&gt; &lt;i32 0, i32 4, i32 1, i32 5&gt; <i>; yields &lt;4 x i32&gt;</i>
2960 %result = shufflevector &lt;4 x i32&gt; %v1, &lt;4 x i32&gt; undef,
2961 &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.
2962</pre>
2963</div>
2964
2965
2966<!-- ======================================================================= -->
2967<div class="doc_subsection">
Dan Gohman74d6faf2008-05-12 23:51:09 +0000[diff] [blame]2968 <a name="aggregateops">Aggregate Operations</a>
2969</div>
2970
2971<div class="doc_text">
2972
2973<p>LLVM supports several instructions for working with aggregate values.
2974</p>
2975
2976</div>
2977
2978<!-- _______________________________________________________________________ -->
2979<div class="doc_subsubsection">
2980 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
2981</div>
2982
2983<div class="doc_text">
2984
2985<h5>Syntax:</h5>
2986
2987<pre>
2988 &lt;result&gt; = extractvalue &lt;aggregate type&gt; &lt;val&gt;, &lt;idx&gt;{, &lt;idx&gt;}*
2989</pre>
2990
2991<h5>Overview:</h5>
2992
2993<p>
Dan Gohman6c6dea02008-05-13 18:16:06 +0000[diff] [blame]2994The '<tt>extractvalue</tt>' instruction extracts the value of a struct field
2995or array element from an aggregate value.
Dan Gohman74d6faf2008-05-12 23:51:09 +0000[diff] [blame]2996</p>
2997
2998
2999<h5>Arguments:</h5>
3000
3001<p>
3002The first operand of an '<tt>extractvalue</tt>' instruction is a
3003value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a>
Dan Gohman6c6dea02008-05-13 18:16:06 +0000[diff] [blame]3004type. The operands are constant indices to specify which value to extract
Dan Gohmane5febe42008-05-31 00:58:22 +0000[diff] [blame]3005in a similar manner as indices in a
Dan Gohman74d6faf2008-05-12 23:51:09 +0000[diff] [blame]3006'<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
3007</p>
3008
3009<h5>Semantics:</h5>
3010
3011<p>
3012The result is the value at the position in the aggregate specified by
3013the index operands.
3014</p>
3015
3016<h5>Example:</h5>
3017
3018<pre>
Dan Gohmane5febe42008-05-31 00:58:22 +0000[diff] [blame]3019 %result = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
Dan Gohman74d6faf2008-05-12 23:51:09 +0000[diff] [blame]3020</pre>
3021</div>
3022
3023
3024<!-- _______________________________________________________________________ -->
3025<div class="doc_subsubsection">
3026 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
3027</div>
3028
3029<div class="doc_text">
3030
3031<h5>Syntax:</h5>
3032
3033<pre>
Dan Gohmane5febe42008-05-31 00:58:22 +0000[diff] [blame]3034 &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]3035</pre>
3036
3037<h5>Overview:</h5>
3038
3039<p>
3040The '<tt>insertvalue</tt>' instruction inserts a value
Dan Gohman6c6dea02008-05-13 18:16:06 +0000[diff] [blame]3041into a struct field or array element in an aggregate.
Dan Gohman74d6faf2008-05-12 23:51:09 +0000[diff] [blame]3042</p>
3043
3044
3045<h5>Arguments:</h5>
3046
3047<p>
3048The first operand of an '<tt>insertvalue</tt>' instruction is a
3049value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a> type.
3050The second operand is a first-class value to insert.
Dan Gohman4f29e422008-05-23 21:53:15 +0000[diff] [blame]3051The following operands are constant indices
Dan Gohmane5febe42008-05-31 00:58:22 +0000[diff] [blame]3052indicating the position at which to insert the value in a similar manner as
Dan Gohman6c6dea02008-05-13 18:16:06 +0000[diff] [blame]3053indices in a
Dan Gohman74d6faf2008-05-12 23:51:09 +0000[diff] [blame]3054'<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
3055The value to insert must have the same type as the value identified
Dan Gohman6c6dea02008-05-13 18:16:06 +0000[diff] [blame]3056by the indices.
Dan Gohman74d6faf2008-05-12 23:51:09 +0000[diff] [blame]3057
3058<h5>Semantics:</h5>
3059
3060<p>
3061The result is an aggregate of the same type as <tt>val</tt>. Its
3062value is that of <tt>val</tt> except that the value at the position
Dan Gohman6c6dea02008-05-13 18:16:06 +0000[diff] [blame]3063specified by the indices is that of <tt>elt</tt>.
Dan Gohman74d6faf2008-05-12 23:51:09 +0000[diff] [blame]3064</p>
3065
3066<h5>Example:</h5>
3067
3068<pre>
Dan Gohmanb1aab4e2008-06-23 15:26:37 +0000[diff] [blame]3069 %result = insertvalue {i32, float} %agg, i32 1, 0 <i>; yields {i32, float}</i>
Dan Gohman74d6faf2008-05-12 23:51:09 +0000[diff] [blame]3070</pre>
3071</div>
3072
3073
3074<!-- ======================================================================= -->
3075<div class="doc_subsection">
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3076 <a name="memoryops">Memory Access and Addressing Operations</a>
3077</div>
3078
3079<div class="doc_text">
3080
3081<p>A key design point of an SSA-based representation is how it
3082represents memory. In LLVM, no memory locations are in SSA form, which
3083makes things very simple. This section describes how to read, write,
3084allocate, and free memory in LLVM.</p>
3085
3086</div>
3087
3088<!-- _______________________________________________________________________ -->
3089<div class="doc_subsubsection">
3090 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
3091</div>
3092
3093<div class="doc_text">
3094
3095<h5>Syntax:</h5>
3096
3097<pre>
3098 &lt;result&gt; = malloc &lt;type&gt;[, i32 &lt;NumElements&gt;][, align &lt;alignment&gt;] <i>; yields {type*}:result</i>
3099</pre>
3100
3101<h5>Overview:</h5>
3102
3103<p>The '<tt>malloc</tt>' instruction allocates memory from the system
Christopher Lambcfe00962007-12-17 01:00:21 +0000[diff] [blame]3104heap and returns a pointer to it. The object is always allocated in the generic
3105address space (address space zero).</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3106
3107<h5>Arguments:</h5>
3108
3109<p>The '<tt>malloc</tt>' instruction allocates
3110<tt>sizeof(&lt;type&gt;)*NumElements</tt>
3111bytes of memory from the operating system and returns a pointer of the
3112appropriate type to the program. If "NumElements" is specified, it is the
Gabor Greif5082cf42008-02-09 22:24:34 +0000[diff] [blame]3113number of elements allocated, otherwise "NumElements" is defaulted to be one.
Chris Lattner10368b62008-04-02 00:38:26 +0000[diff] [blame]3114If a constant alignment is specified, the value result of the allocation is guaranteed to
Gabor Greif5082cf42008-02-09 22:24:34 +0000[diff] [blame]3115be aligned to at least that boundary. If not specified, or if zero, the target can
3116choose to align the allocation on any convenient boundary.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3117
3118<p>'<tt>type</tt>' must be a sized type.</p>
3119
3120<h5>Semantics:</h5>
3121
3122<p>Memory is allocated using the system "<tt>malloc</tt>" function, and
Chris Lattner8b094fc2008-04-19 21:01:16 +0000[diff] [blame]3123a pointer is returned. The result of a zero byte allocattion is undefined. The
3124result is null if there is insufficient memory available.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3125
3126<h5>Example:</h5>
3127
3128<pre>
3129 %array = malloc [4 x i8 ] <i>; yields {[%4 x i8]*}:array</i>
3130
3131 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
3132 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
3133 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
3134 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
3135 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
3136</pre>
3137</div>
3138
3139<!-- _______________________________________________________________________ -->
3140<div class="doc_subsubsection">
3141 <a name="i_free">'<tt>free</tt>' Instruction</a>
3142</div>
3143
3144<div class="doc_text">
3145
3146<h5>Syntax:</h5>
3147
3148<pre>
3149 free &lt;type&gt; &lt;value&gt; <i>; yields {void}</i>
3150</pre>
3151
3152<h5>Overview:</h5>
3153
3154<p>The '<tt>free</tt>' instruction returns memory back to the unused
3155memory heap to be reallocated in the future.</p>
3156
3157<h5>Arguments:</h5>
3158
3159<p>'<tt>value</tt>' shall be a pointer value that points to a value
3160that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
3161instruction.</p>
3162
3163<h5>Semantics:</h5>
3164
3165<p>Access to the memory pointed to by the pointer is no longer defined
Chris Lattner329d6532008-04-19 22:41:32 +0000[diff] [blame]3166after this instruction executes. If the pointer is null, the operation
3167is a noop.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3168
3169<h5>Example:</h5>
3170
3171<pre>
3172 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
3173 free [4 x i8]* %array
3174</pre>
3175</div>
3176
3177<!-- _______________________________________________________________________ -->
3178<div class="doc_subsubsection">
3179 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
3180</div>
3181
3182<div class="doc_text">
3183
3184<h5>Syntax:</h5>
3185
3186<pre>
3187 &lt;result&gt; = alloca &lt;type&gt;[, i32 &lt;NumElements&gt;][, align &lt;alignment&gt;] <i>; yields {type*}:result</i>
3188</pre>
3189
3190<h5>Overview:</h5>
3191
3192<p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
3193currently executing function, to be automatically released when this function
Christopher Lambcfe00962007-12-17 01:00:21 +0000[diff] [blame]3194returns to its caller. The object is always allocated in the generic address
3195space (address space zero).</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3196
3197<h5>Arguments:</h5>
3198
3199<p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(&lt;type&gt;)*NumElements</tt>
3200bytes of memory on the runtime stack, returning a pointer of the
Gabor Greif5082cf42008-02-09 22:24:34 +0000[diff] [blame]3201appropriate type to the program. If "NumElements" is specified, it is the
3202number of elements allocated, otherwise "NumElements" is defaulted to be one.
Chris Lattner10368b62008-04-02 00:38:26 +0000[diff] [blame]3203If a constant alignment is specified, the value result of the allocation is guaranteed
Gabor Greif5082cf42008-02-09 22:24:34 +0000[diff] [blame]3204to be aligned to at least that boundary. If not specified, or if zero, the target
3205can choose to align the allocation on any convenient boundary.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3206
3207<p>'<tt>type</tt>' may be any sized type.</p>
3208
3209<h5>Semantics:</h5>
3210
Chris Lattner8b094fc2008-04-19 21:01:16 +0000[diff] [blame]3211<p>Memory is allocated; a pointer is returned. The operation is undefiend if
3212there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3213memory is automatically released when the function returns. The '<tt>alloca</tt>'
3214instruction is commonly used to represent automatic variables that must
3215have an address available. When the function returns (either with the <tt><a
3216 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]3217instructions), the memory is reclaimed. Allocating zero bytes
3218is legal, but the result is undefined.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3219
3220<h5>Example:</h5>
3221
3222<pre>
3223 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
3224 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
3225 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
3226 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
3227</pre>
3228</div>
3229
3230<!-- _______________________________________________________________________ -->
3231<div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
3232Instruction</a> </div>
3233<div class="doc_text">
3234<h5>Syntax:</h5>
3235<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>
3236<h5>Overview:</h5>
3237<p>The '<tt>load</tt>' instruction is used to read from memory.</p>
3238<h5>Arguments:</h5>
3239<p>The argument to the '<tt>load</tt>' instruction specifies the memory
3240address from which to load. The pointer must point to a <a
3241 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
3242marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
3243the number or order of execution of this <tt>load</tt> with other
3244volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
3245instructions. </p>
Chris Lattner21003c82008-01-06 21:04:43 +0000[diff] [blame]3246<p>
Chris Lattner10368b62008-04-02 00:38:26 +0000[diff] [blame]3247The optional constant "align" argument specifies the alignment of the operation
Chris Lattner21003c82008-01-06 21:04:43 +0000[diff] [blame]3248(that is, the alignment of the memory address). A value of 0 or an
3249omitted "align" argument means that the operation has the preferential
3250alignment for the target. It is the responsibility of the code emitter
3251to ensure that the alignment information is correct. Overestimating
3252the alignment results in an undefined behavior. Underestimating the
3253alignment may produce less efficient code. An alignment of 1 is always
3254safe.
3255</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3256<h5>Semantics:</h5>
3257<p>The location of memory pointed to is loaded.</p>
3258<h5>Examples:</h5>
3259<pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3260 <a
3261 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
3262 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
3263</pre>
3264</div>
3265<!-- _______________________________________________________________________ -->
3266<div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
3267Instruction</a> </div>
3268<div class="doc_text">
3269<h5>Syntax:</h5>
3270<pre> store &lt;ty&gt; &lt;value&gt;, &lt;ty&gt;* &lt;pointer&gt;[, align &lt;alignment&gt;] <i>; yields {void}</i>
3271 volatile store &lt;ty&gt; &lt;value&gt;, &lt;ty&gt;* &lt;pointer&gt;[, align &lt;alignment&gt;] <i>; yields {void}</i>
3272</pre>
3273<h5>Overview:</h5>
3274<p>The '<tt>store</tt>' instruction is used to write to memory.</p>
3275<h5>Arguments:</h5>
3276<p>There are two arguments to the '<tt>store</tt>' instruction: a value
3277to 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]3278operand must be a pointer to the <a href="#t_firstclass">first class</a> type
3279of the '<tt>&lt;value&gt;</tt>'
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3280operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
3281optimizer is not allowed to modify the number or order of execution of
3282this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
3283 href="#i_store">store</a></tt> instructions.</p>
Chris Lattner21003c82008-01-06 21:04:43 +0000[diff] [blame]3284<p>
Chris Lattner10368b62008-04-02 00:38:26 +0000[diff] [blame]3285The optional constant "align" argument specifies the alignment of the operation
Chris Lattner21003c82008-01-06 21:04:43 +0000[diff] [blame]3286(that is, the alignment of the memory address). A value of 0 or an
3287omitted "align" argument means that the operation has the preferential
3288alignment for the target. It is the responsibility of the code emitter
3289to ensure that the alignment information is correct. Overestimating
3290the alignment results in an undefined behavior. Underestimating the
3291alignment may produce less efficient code. An alignment of 1 is always
3292safe.
3293</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3294<h5>Semantics:</h5>
3295<p>The contents of memory are updated to contain '<tt>&lt;value&gt;</tt>'
3296at the location specified by the '<tt>&lt;pointer&gt;</tt>' operand.</p>
3297<h5>Example:</h5>
3298<pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
Bill Wendling63ffa142007-10-22 05:10:05 +0000[diff] [blame]3299 store i32 3, i32* %ptr <i>; yields {void}</i>
3300 %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]3301</pre>
3302</div>
3303
3304<!-- _______________________________________________________________________ -->
3305<div class="doc_subsubsection">
3306 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
3307</div>
3308
3309<div class="doc_text">
3310<h5>Syntax:</h5>
3311<pre>
3312 &lt;result&gt; = getelementptr &lt;ty&gt;* &lt;ptrval&gt;{, &lt;ty&gt; &lt;idx&gt;}*
3313</pre>
3314
3315<h5>Overview:</h5>
3316
3317<p>
3318The '<tt>getelementptr</tt>' instruction is used to get the address of a
3319subelement of an aggregate data structure.</p>
3320
3321<h5>Arguments:</h5>
3322
3323<p>This instruction takes a list of integer operands that indicate what
3324elements of the aggregate object to index to. The actual types of the arguments
3325provided depend on the type of the first pointer argument. The
3326'<tt>getelementptr</tt>' instruction is used to index down through the type
3327levels of a structure or to a specific index in an array. When indexing into a
3328structure, only <tt>i32</tt> integer constants are allowed. When indexing
Chris Lattnera7d94ba2008-04-24 05:59:56 +0000[diff] [blame]3329into an array or pointer, only integers of 32 or 64 bits are allowed; 32-bit
3330values will be sign extended to 64-bits if required.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3331
3332<p>For example, let's consider a C code fragment and how it gets
3333compiled to LLVM:</p>
3334
3335<div class="doc_code">
3336<pre>
3337struct RT {
3338 char A;
3339 int B[10][20];
3340 char C;
3341};
3342struct ST {
3343 int X;
3344 double Y;
3345 struct RT Z;
3346};
3347
3348int *foo(struct ST *s) {
3349 return &amp;s[1].Z.B[5][13];
3350}
3351</pre>
3352</div>
3353
3354<p>The LLVM code generated by the GCC frontend is:</p>
3355
3356<div class="doc_code">
3357<pre>
3358%RT = type { i8 , [10 x [20 x i32]], i8 }
3359%ST = type { i32, double, %RT }
3360
3361define i32* %foo(%ST* %s) {
3362entry:
3363 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
3364 ret i32* %reg
3365}
3366</pre>
3367</div>
3368
3369<h5>Semantics:</h5>
3370
3371<p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
3372on the pointer type that is being indexed into. <a href="#t_pointer">Pointer</a>
3373and <a href="#t_array">array</a> types can use a 32-bit or 64-bit
3374<a href="#t_integer">integer</a> type but the value will always be sign extended
Chris Lattner10368b62008-04-02 00:38:26 +0000[diff] [blame]3375to 64-bits. <a href="#t_struct">Structure</a> and <a href="#t_pstruct">packed
3376structure</a> types require <tt>i32</tt> <b>constants</b>.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3377
3378<p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
3379type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
3380}</tt>' type, a structure. The second index indexes into the third element of
3381the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
3382i8 }</tt>' type, another structure. The third index indexes into the second
3383element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
3384array. The two dimensions of the array are subscripted into, yielding an
3385'<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
3386to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
3387
3388<p>Note that it is perfectly legal to index partially through a
3389structure, returning a pointer to an inner element. Because of this,
3390the LLVM code for the given testcase is equivalent to:</p>
3391
3392<pre>
3393 define i32* %foo(%ST* %s) {
3394 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
3395 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
3396 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
3397 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
3398 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
3399 ret i32* %t5
3400 }
3401</pre>
3402
3403<p>Note that it is undefined to access an array out of bounds: array and
3404pointer indexes must always be within the defined bounds of the array type.
Chris Lattnera7d94ba2008-04-24 05:59:56 +0000[diff] [blame]3405The one exception for this rule is zero length arrays. These arrays are
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3406defined to be accessible as variable length arrays, which requires access
3407beyond the zero'th element.</p>
3408
3409<p>The getelementptr instruction is often confusing. For some more insight
3410into how it works, see <a href="GetElementPtr.html">the getelementptr
3411FAQ</a>.</p>
3412
3413<h5>Example:</h5>
3414
3415<pre>
3416 <i>; yields [12 x i8]*:aptr</i>
3417 %aptr = getelementptr {i32, [12 x i8]}* %sptr, i64 0, i32 1
3418</pre>
3419</div>
3420
3421<!-- ======================================================================= -->
3422<div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
3423</div>
3424<div class="doc_text">
3425<p>The instructions in this category are the conversion instructions (casting)
3426which all take a single operand and a type. They perform various bit conversions
3427on the operand.</p>
3428</div>
3429
3430<!-- _______________________________________________________________________ -->
3431<div class="doc_subsubsection">
3432 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
3433</div>
3434<div class="doc_text">
3435
3436<h5>Syntax:</h5>
3437<pre>
3438 &lt;result&gt; = trunc &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3439</pre>
3440
3441<h5>Overview:</h5>
3442<p>
3443The '<tt>trunc</tt>' instruction truncates its operand to the type <tt>ty2</tt>.
3444</p>
3445
3446<h5>Arguments:</h5>
3447<p>
3448The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
3449be an <a href="#t_integer">integer</a> type, and a type that specifies the size
3450and type of the result, which must be an <a href="#t_integer">integer</a>
3451type. The bit size of <tt>value</tt> must be larger than the bit size of
3452<tt>ty2</tt>. Equal sized types are not allowed.</p>
3453
3454<h5>Semantics:</h5>
3455<p>
3456The '<tt>trunc</tt>' instruction truncates the high order bits in <tt>value</tt>
3457and converts the remaining bits to <tt>ty2</tt>. Since the source size must be
3458larger than the destination size, <tt>trunc</tt> cannot be a <i>no-op cast</i>.
3459It will always truncate bits.</p>
3460
3461<h5>Example:</h5>
3462<pre>
3463 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
3464 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
3465 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
3466</pre>
3467</div>
3468
3469<!-- _______________________________________________________________________ -->
3470<div class="doc_subsubsection">
3471 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
3472</div>
3473<div class="doc_text">
3474
3475<h5>Syntax:</h5>
3476<pre>
3477 &lt;result&gt; = zext &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3478</pre>
3479
3480<h5>Overview:</h5>
3481<p>The '<tt>zext</tt>' instruction zero extends its operand to type
3482<tt>ty2</tt>.</p>
3483
3484
3485<h5>Arguments:</h5>
3486<p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
3487<a href="#t_integer">integer</a> type, and a type to cast it to, which must
3488also be of <a href="#t_integer">integer</a> type. The bit size of the
3489<tt>value</tt> must be smaller than the bit size of the destination type,
3490<tt>ty2</tt>.</p>
3491
3492<h5>Semantics:</h5>
3493<p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
3494bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
3495
3496<p>When zero extending from i1, the result will always be either 0 or 1.</p>
3497
3498<h5>Example:</h5>
3499<pre>
3500 %X = zext i32 257 to i64 <i>; yields i64:257</i>
3501 %Y = zext i1 true to i32 <i>; yields i32:1</i>
3502</pre>
3503</div>
3504
3505<!-- _______________________________________________________________________ -->
3506<div class="doc_subsubsection">
3507 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
3508</div>
3509<div class="doc_text">
3510
3511<h5>Syntax:</h5>
3512<pre>
3513 &lt;result&gt; = sext &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3514</pre>
3515
3516<h5>Overview:</h5>
3517<p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
3518
3519<h5>Arguments:</h5>
3520<p>
3521The '<tt>sext</tt>' instruction takes a value to cast, which must be of
3522<a href="#t_integer">integer</a> type, and a type to cast it to, which must
3523also be of <a href="#t_integer">integer</a> type. The bit size of the
3524<tt>value</tt> must be smaller than the bit size of the destination type,
3525<tt>ty2</tt>.</p>
3526
3527<h5>Semantics:</h5>
3528<p>
3529The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
3530bit (highest order bit) of the <tt>value</tt> until it reaches the bit size of
3531the type <tt>ty2</tt>.</p>
3532
3533<p>When sign extending from i1, the extension always results in -1 or 0.</p>
3534
3535<h5>Example:</h5>
3536<pre>
3537 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
3538 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
3539</pre>
3540</div>
3541
3542<!-- _______________________________________________________________________ -->
3543<div class="doc_subsubsection">
3544 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
3545</div>
3546
3547<div class="doc_text">
3548
3549<h5>Syntax:</h5>
3550
3551<pre>
3552 &lt;result&gt; = fptrunc &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3553</pre>
3554
3555<h5>Overview:</h5>
3556<p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
3557<tt>ty2</tt>.</p>
3558
3559
3560<h5>Arguments:</h5>
3561<p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
3562 point</a> value to cast and a <a href="#t_floating">floating point</a> type to
3563cast it to. The size of <tt>value</tt> must be larger than the size of
3564<tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
3565<i>no-op cast</i>.</p>
3566
3567<h5>Semantics:</h5>
3568<p> The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
3569<a href="#t_floating">floating point</a> type to a smaller
3570<a href="#t_floating">floating point</a> type. If the value cannot fit within
3571the destination type, <tt>ty2</tt>, then the results are undefined.</p>
3572
3573<h5>Example:</h5>
3574<pre>
3575 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
3576 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
3577</pre>
3578</div>
3579
3580<!-- _______________________________________________________________________ -->
3581<div class="doc_subsubsection">
3582 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
3583</div>
3584<div class="doc_text">
3585
3586<h5>Syntax:</h5>
3587<pre>
3588 &lt;result&gt; = fpext &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3589</pre>
3590
3591<h5>Overview:</h5>
3592<p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
3593floating point value.</p>
3594
3595<h5>Arguments:</h5>
3596<p>The '<tt>fpext</tt>' instruction takes a
3597<a href="#t_floating">floating point</a> <tt>value</tt> to cast,
3598and a <a href="#t_floating">floating point</a> type to cast it to. The source
3599type must be smaller than the destination type.</p>
3600
3601<h5>Semantics:</h5>
3602<p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
3603<a href="#t_floating">floating point</a> type to a larger
3604<a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
3605used to make a <i>no-op cast</i> because it always changes bits. Use
3606<tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
3607
3608<h5>Example:</h5>
3609<pre>
3610 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
3611 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
3612</pre>
3613</div>
3614
3615<!-- _______________________________________________________________________ -->
3616<div class="doc_subsubsection">
3617 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
3618</div>
3619<div class="doc_text">
3620
3621<h5>Syntax:</h5>
3622<pre>
Reid Spencere6adee82007-07-31 14:40:14 +0000[diff] [blame]3623 &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]3624</pre>
3625
3626<h5>Overview:</h5>
Reid Spencere6adee82007-07-31 14:40:14 +0000[diff] [blame]3627<p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3628unsigned integer equivalent of type <tt>ty2</tt>.
3629</p>
3630
3631<h5>Arguments:</h5>
Reid Spencere6adee82007-07-31 14:40:14 +0000[diff] [blame]3632<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]3633scalar or vector <a href="#t_floating">floating point</a> value, and a type
3634to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3635type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3636vector integer type with the same number of elements as <tt>ty</tt></p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3637
3638<h5>Semantics:</h5>
Reid Spencere6adee82007-07-31 14:40:14 +0000[diff] [blame]3639<p> The '<tt>fptoui</tt>' instruction converts its
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3640<a href="#t_floating">floating point</a> operand into the nearest (rounding
3641towards zero) unsigned integer value. If the value cannot fit in <tt>ty2</tt>,
3642the results are undefined.</p>
3643
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3644<h5>Example:</h5>
3645<pre>
Reid Spencere6adee82007-07-31 14:40:14 +0000[diff] [blame]3646 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
Chris Lattner681f1e82007-09-22 03:17:52 +0000[diff] [blame]3647 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
Reid Spencere6adee82007-07-31 14:40:14 +0000[diff] [blame]3648 %X = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3649</pre>
3650</div>
3651
3652<!-- _______________________________________________________________________ -->
3653<div class="doc_subsubsection">
3654 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
3655</div>
3656<div class="doc_text">
3657
3658<h5>Syntax:</h5>
3659<pre>
3660 &lt;result&gt; = fptosi &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3661</pre>
3662
3663<h5>Overview:</h5>
3664<p>The '<tt>fptosi</tt>' instruction converts
3665<a href="#t_floating">floating point</a> <tt>value</tt> to type <tt>ty2</tt>.
3666</p>
3667
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3668<h5>Arguments:</h5>
3669<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]3670scalar or vector <a href="#t_floating">floating point</a> value, and a type
3671to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3672type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3673vector integer type with the same number of elements as <tt>ty</tt></p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3674
3675<h5>Semantics:</h5>
3676<p>The '<tt>fptosi</tt>' instruction converts its
3677<a href="#t_floating">floating point</a> operand into the nearest (rounding
3678towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
3679the results are undefined.</p>
3680
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3681<h5>Example:</h5>
3682<pre>
3683 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
Chris Lattner681f1e82007-09-22 03:17:52 +0000[diff] [blame]3684 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3685 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
3686</pre>
3687</div>
3688
3689<!-- _______________________________________________________________________ -->
3690<div class="doc_subsubsection">
3691 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
3692</div>
3693<div class="doc_text">
3694
3695<h5>Syntax:</h5>
3696<pre>
3697 &lt;result&gt; = uitofp &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3698</pre>
3699
3700<h5>Overview:</h5>
3701<p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
3702integer and converts that value to the <tt>ty2</tt> type.</p>
3703
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3704<h5>Arguments:</h5>
Nate Begeman78246ca2007-11-17 03:58:34 +0000[diff] [blame]3705<p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
3706scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3707to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3708type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3709floating point type with the same number of elements as <tt>ty</tt></p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3710
3711<h5>Semantics:</h5>
3712<p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
3713integer quantity and converts it to the corresponding floating point value. If
3714the value cannot fit in the floating point value, the results are undefined.</p>
3715
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3716<h5>Example:</h5>
3717<pre>
3718 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
3719 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
3720</pre>
3721</div>
3722
3723<!-- _______________________________________________________________________ -->
3724<div class="doc_subsubsection">
3725 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
3726</div>
3727<div class="doc_text">
3728
3729<h5>Syntax:</h5>
3730<pre>
3731 &lt;result&gt; = sitofp &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3732</pre>
3733
3734<h5>Overview:</h5>
3735<p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed
3736integer and converts that value to the <tt>ty2</tt> type.</p>
3737
3738<h5>Arguments:</h5>
Nate Begeman78246ca2007-11-17 03:58:34 +0000[diff] [blame]3739<p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
3740scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3741to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3742type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3743floating point type with the same number of elements as <tt>ty</tt></p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3744
3745<h5>Semantics:</h5>
3746<p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed
3747integer quantity and converts it to the corresponding floating point value. If
3748the value cannot fit in the floating point value, the results are undefined.</p>
3749
3750<h5>Example:</h5>
3751<pre>
3752 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
3753 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
3754</pre>
3755</div>
3756
3757<!-- _______________________________________________________________________ -->
3758<div class="doc_subsubsection">
3759 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
3760</div>
3761<div class="doc_text">
3762
3763<h5>Syntax:</h5>
3764<pre>
3765 &lt;result&gt; = ptrtoint &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3766</pre>
3767
3768<h5>Overview:</h5>
3769<p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
3770the integer type <tt>ty2</tt>.</p>
3771
3772<h5>Arguments:</h5>
3773<p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
3774must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
3775<tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.
3776
3777<h5>Semantics:</h5>
3778<p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
3779<tt>ty2</tt> by interpreting the pointer value as an integer and either
3780truncating or zero extending that value to the size of the integer type. If
3781<tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
3782<tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
3783are the same size, then nothing is done (<i>no-op cast</i>) other than a type
3784change.</p>
3785
3786<h5>Example:</h5>
3787<pre>
3788 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
3789 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
3790</pre>
3791</div>
3792
3793<!-- _______________________________________________________________________ -->
3794<div class="doc_subsubsection">
3795 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
3796</div>
3797<div class="doc_text">
3798
3799<h5>Syntax:</h5>
3800<pre>
3801 &lt;result&gt; = inttoptr &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3802</pre>
3803
3804<h5>Overview:</h5>
3805<p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to
3806a pointer type, <tt>ty2</tt>.</p>
3807
3808<h5>Arguments:</h5>
3809<p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
3810value to cast, and a type to cast it to, which must be a
3811<a href="#t_pointer">pointer</a> type.
3812
3813<h5>Semantics:</h5>
3814<p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
3815<tt>ty2</tt> by applying either a zero extension or a truncation depending on
3816the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
3817size of a pointer then a truncation is done. If <tt>value</tt> is smaller than
3818the size of a pointer then a zero extension is done. If they are the same size,
3819nothing is done (<i>no-op cast</i>).</p>
3820
3821<h5>Example:</h5>
3822<pre>
3823 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
3824 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
3825 %Y = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
3826</pre>
3827</div>
3828
3829<!-- _______________________________________________________________________ -->
3830<div class="doc_subsubsection">
3831 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
3832</div>
3833<div class="doc_text">
3834
3835<h5>Syntax:</h5>
3836<pre>
3837 &lt;result&gt; = bitcast &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3838</pre>
3839
3840<h5>Overview:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]3841
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3842<p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3843<tt>ty2</tt> without changing any bits.</p>
3844
3845<h5>Arguments:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]3846
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3847<p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be
Dan Gohman7305fa02008-09-08 16:45:59 +0000[diff] [blame]3848a non-aggregate first class value, and a type to cast it to, which must also be
3849a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes of
3850<tt>value</tt>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3851and the destination type, <tt>ty2</tt>, must be identical. If the source
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]3852type is a pointer, the destination type must also be a pointer. This
3853instruction supports bitwise conversion of vectors to integers and to vectors
3854of other types (as long as they have the same size).</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3855
3856<h5>Semantics:</h5>
3857<p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3858<tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
3859this conversion. The conversion is done as if the <tt>value</tt> had been
3860stored to memory and read back as type <tt>ty2</tt>. Pointer types may only be
3861converted to other pointer types with this instruction. To convert pointers to
3862other types, use the <a href="#i_inttoptr">inttoptr</a> or
3863<a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
3864
3865<h5>Example:</h5>
3866<pre>
3867 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
3868 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
3869 %Z = bitcast <2xint> %V to i64; <i>; yields i64: %V</i>
3870</pre>
3871</div>
3872
3873<!-- ======================================================================= -->
3874<div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
3875<div class="doc_text">
3876<p>The instructions in this category are the "miscellaneous"
3877instructions, which defy better classification.</p>
3878</div>
3879
3880<!-- _______________________________________________________________________ -->
3881<div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
3882</div>
3883<div class="doc_text">
3884<h5>Syntax:</h5>
Dan Gohmanb60ca3c2008-09-09 01:02:47 +0000[diff] [blame]3885<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]3886</pre>
3887<h5>Overview:</h5>
Dan Gohmanb60ca3c2008-09-09 01:02:47 +0000[diff] [blame]3888<p>The '<tt>icmp</tt>' instruction returns a boolean value or
3889a vector of boolean values based on comparison
3890of its two integer, integer vector, or pointer operands.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3891<h5>Arguments:</h5>
3892<p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
3893the condition code indicating the kind of comparison to perform. It is not
3894a value, just a keyword. The possible condition code are:
3895<ol>
3896 <li><tt>eq</tt>: equal</li>
3897 <li><tt>ne</tt>: not equal </li>
3898 <li><tt>ugt</tt>: unsigned greater than</li>
3899 <li><tt>uge</tt>: unsigned greater or equal</li>
3900 <li><tt>ult</tt>: unsigned less than</li>
3901 <li><tt>ule</tt>: unsigned less or equal</li>
3902 <li><tt>sgt</tt>: signed greater than</li>
3903 <li><tt>sge</tt>: signed greater or equal</li>
3904 <li><tt>slt</tt>: signed less than</li>
3905 <li><tt>sle</tt>: signed less or equal</li>
3906</ol>
3907<p>The remaining two arguments must be <a href="#t_integer">integer</a> or
Dan Gohmanb60ca3c2008-09-09 01:02:47 +0000[diff] [blame]3908<a href="#t_pointer">pointer</a>
3909or integer <a href="#t_vector">vector</a> typed.
3910They must also be identical types.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3911<h5>Semantics:</h5>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]3912<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]3913the condition code given as <tt>cond</tt>. The comparison performed always
Dan Gohmanb60ca3c2008-09-09 01:02:47 +0000[diff] [blame]3914yields either an <a href="#t_primitive"><tt>i1</tt></a> or vector of <tt>i1</tt> result, as follows:
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3915<ol>
3916 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
3917 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3918 </li>
3919 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
3920 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3921 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]3922 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3923 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]3924 <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]3925 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]3926 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3927 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]3928 <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]3929 <li><tt>sgt</tt>: interprets the operands as signed values and yields
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]3930 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3931 <li><tt>sge</tt>: interprets the operands as signed values and yields
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]3932 <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]3933 <li><tt>slt</tt>: interprets the operands as signed values and yields
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]3934 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3935 <li><tt>sle</tt>: interprets the operands as signed values and yields
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]3936 <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]3937</ol>
3938<p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
3939values are compared as if they were integers.</p>
Dan Gohmanb60ca3c2008-09-09 01:02:47 +0000[diff] [blame]3940<p>If the operands are integer vectors, then they are compared
3941element by element. The result is an <tt>i1</tt> vector with
3942the same number of elements as the values being compared.
3943Otherwise, the result is an <tt>i1</tt>.
3944</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3945
3946<h5>Example:</h5>
3947<pre> &lt;result&gt; = icmp eq i32 4, 5 <i>; yields: result=false</i>
3948 &lt;result&gt; = icmp ne float* %X, %X <i>; yields: result=false</i>
3949 &lt;result&gt; = icmp ult i16 4, 5 <i>; yields: result=true</i>
3950 &lt;result&gt; = icmp sgt i16 4, 5 <i>; yields: result=false</i>
3951 &lt;result&gt; = icmp ule i16 -4, 5 <i>; yields: result=false</i>
3952 &lt;result&gt; = icmp sge i16 4, 5 <i>; yields: result=false</i>
3953</pre>
3954</div>
3955
3956<!-- _______________________________________________________________________ -->
3957<div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
3958</div>
3959<div class="doc_text">
3960<h5>Syntax:</h5>
Dan Gohmanb60ca3c2008-09-09 01:02:47 +0000[diff] [blame]3961<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]3962</pre>
3963<h5>Overview:</h5>
Dan Gohmanb60ca3c2008-09-09 01:02:47 +0000[diff] [blame]3964<p>The '<tt>fcmp</tt>' instruction returns a boolean value
3965or vector of boolean values based on comparison
3966of its operands.
3967<p>
3968If the operands are floating point scalars, then the result
3969type is a boolean (<a href="#t_primitive"><tt>i1</tt></a>).
3970</p>
3971<p>If the operands are floating point vectors, then the result type
3972is a vector of boolean with the same number of elements as the
3973operands being compared.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3974<h5>Arguments:</h5>
3975<p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
3976the condition code indicating the kind of comparison to perform. It is not
3977a value, just a keyword. The possible condition code are:
3978<ol>
3979 <li><tt>false</tt>: no comparison, always returns false</li>
3980 <li><tt>oeq</tt>: ordered and equal</li>
3981 <li><tt>ogt</tt>: ordered and greater than </li>
3982 <li><tt>oge</tt>: ordered and greater than or equal</li>
3983 <li><tt>olt</tt>: ordered and less than </li>
3984 <li><tt>ole</tt>: ordered and less than or equal</li>
3985 <li><tt>one</tt>: ordered and not equal</li>
3986 <li><tt>ord</tt>: ordered (no nans)</li>
3987 <li><tt>ueq</tt>: unordered or equal</li>
3988 <li><tt>ugt</tt>: unordered or greater than </li>
3989 <li><tt>uge</tt>: unordered or greater than or equal</li>
3990 <li><tt>ult</tt>: unordered or less than </li>
3991 <li><tt>ule</tt>: unordered or less than or equal</li>
3992 <li><tt>une</tt>: unordered or not equal</li>
3993 <li><tt>uno</tt>: unordered (either nans)</li>
3994 <li><tt>true</tt>: no comparison, always returns true</li>
3995</ol>
3996<p><i>Ordered</i> means that neither operand is a QNAN while
3997<i>unordered</i> means that either operand may be a QNAN.</p>
Dan Gohmanb60ca3c2008-09-09 01:02:47 +0000[diff] [blame]3998<p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be
3999either a <a href="#t_floating">floating point</a> type
4000or a <a href="#t_vector">vector</a> of floating point type.
4001They must have identical types.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4002<h5>Semantics:</h5>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]4003<p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
Dan Gohmanb60ca3c2008-09-09 01:02:47 +0000[diff] [blame]4004according to the condition code given as <tt>cond</tt>.
4005If the operands are vectors, then the vectors are compared
4006element by element.
4007Each comparison performed
4008always yields an <a href="#t_primitive">i1</a> result, as follows:
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4009<ol>
4010 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
4011 <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]4012 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4013 <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]4014 <tt>op1</tt> is greather than <tt>op2</tt>.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4015 <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]4016 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4017 <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]4018 <tt>op1</tt> is less than <tt>op2</tt>.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4019 <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]4020 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4021 <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]4022 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4023 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
4024 <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]4025 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4026 <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]4027 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4028 <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]4029 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4030 <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]4031 <tt>op1</tt> is less than <tt>op2</tt>.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4032 <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]4033 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4034 <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]4035 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4036 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
4037 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
4038</ol>
4039
4040<h5>Example:</h5>
4041<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]4042 &lt;result&gt; = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
4043 &lt;result&gt; = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
4044 &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]4045</pre>
4046</div>
4047
4048<!-- _______________________________________________________________________ -->
Nate Begeman646fa482008-05-12 19:01:56 +0000[diff] [blame]4049<div class="doc_subsubsection">
4050 <a name="i_vicmp">'<tt>vicmp</tt>' Instruction</a>
4051</div>
4052<div class="doc_text">
4053<h5>Syntax:</h5>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]4054<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]4055</pre>
4056<h5>Overview:</h5>
4057<p>The '<tt>vicmp</tt>' instruction returns an integer vector value based on
4058element-wise comparison of its two integer vector operands.</p>
4059<h5>Arguments:</h5>
4060<p>The '<tt>vicmp</tt>' instruction takes three operands. The first operand is
4061the condition code indicating the kind of comparison to perform. It is not
4062a value, just a keyword. The possible condition code are:
4063<ol>
4064 <li><tt>eq</tt>: equal</li>
4065 <li><tt>ne</tt>: not equal </li>
4066 <li><tt>ugt</tt>: unsigned greater than</li>
4067 <li><tt>uge</tt>: unsigned greater or equal</li>
4068 <li><tt>ult</tt>: unsigned less than</li>
4069 <li><tt>ule</tt>: unsigned less or equal</li>
4070 <li><tt>sgt</tt>: signed greater than</li>
4071 <li><tt>sge</tt>: signed greater or equal</li>
4072 <li><tt>slt</tt>: signed less than</li>
4073 <li><tt>sle</tt>: signed less or equal</li>
4074</ol>
Dan Gohmanb60ca3c2008-09-09 01:02:47 +0000[diff] [blame]4075<p>The remaining two arguments must be <a href="#t_vector">vector</a> or
Nate Begeman646fa482008-05-12 19:01:56 +0000[diff] [blame]4076<a href="#t_integer">integer</a> typed. They must also be identical types.</p>
4077<h5>Semantics:</h5>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]4078<p>The '<tt>vicmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
Nate Begeman646fa482008-05-12 19:01:56 +0000[diff] [blame]4079according to the condition code given as <tt>cond</tt>. The comparison yields a
4080<a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, of
4081identical type as the values being compared. The most significant bit in each
4082element is 1 if the element-wise comparison evaluates to true, and is 0
4083otherwise. All other bits of the result are undefined. The condition codes
4084are evaluated identically to the <a href="#i_icmp">'<tt>icmp</tt>'
4085instruction</a>.
4086
4087<h5>Example:</h5>
4088<pre>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]4089 &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>
4090 &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]4091</pre>
4092</div>
4093
4094<!-- _______________________________________________________________________ -->
4095<div class="doc_subsubsection">
4096 <a name="i_vfcmp">'<tt>vfcmp</tt>' Instruction</a>
4097</div>
4098<div class="doc_text">
4099<h5>Syntax:</h5>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]4100<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]4101<h5>Overview:</h5>
4102<p>The '<tt>vfcmp</tt>' instruction returns an integer vector value based on
4103element-wise comparison of its two floating point vector operands. The output
4104elements have the same width as the input elements.</p>
4105<h5>Arguments:</h5>
4106<p>The '<tt>vfcmp</tt>' instruction takes three operands. The first operand is
4107the condition code indicating the kind of comparison to perform. It is not
4108a value, just a keyword. The possible condition code are:
4109<ol>
4110 <li><tt>false</tt>: no comparison, always returns false</li>
4111 <li><tt>oeq</tt>: ordered and equal</li>
4112 <li><tt>ogt</tt>: ordered and greater than </li>
4113 <li><tt>oge</tt>: ordered and greater than or equal</li>
4114 <li><tt>olt</tt>: ordered and less than </li>
4115 <li><tt>ole</tt>: ordered and less than or equal</li>
4116 <li><tt>one</tt>: ordered and not equal</li>
4117 <li><tt>ord</tt>: ordered (no nans)</li>
4118 <li><tt>ueq</tt>: unordered or equal</li>
4119 <li><tt>ugt</tt>: unordered or greater than </li>
4120 <li><tt>uge</tt>: unordered or greater than or equal</li>
4121 <li><tt>ult</tt>: unordered or less than </li>
4122 <li><tt>ule</tt>: unordered or less than or equal</li>
4123 <li><tt>une</tt>: unordered or not equal</li>
4124 <li><tt>uno</tt>: unordered (either nans)</li>
4125 <li><tt>true</tt>: no comparison, always returns true</li>
4126</ol>
4127<p>The remaining two arguments must be <a href="#t_vector">vector</a> of
4128<a href="#t_floating">floating point</a> typed. They must also be identical
4129types.</p>
4130<h5>Semantics:</h5>
Gabor Greifd9068fe2008-08-07 21:46:00 +0000[diff] [blame]4131<p>The '<tt>vfcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
Nate Begeman646fa482008-05-12 19:01:56 +0000[diff] [blame]4132according to the condition code given as <tt>cond</tt>. The comparison yields a
4133<a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, with
4134an identical number of elements as the values being compared, and each element
4135having identical with to the width of the floating point elements. The most
4136significant bit in each element is 1 if the element-wise comparison evaluates to
4137true, and is 0 otherwise. All other bits of the result are undefined. The
4138condition codes are evaluated identically to the
4139<a href="#i_fcmp">'<tt>fcmp</tt>' instruction</a>.
4140
4141<h5>Example:</h5>
4142<pre>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]4143 &lt;result&gt; = vfcmp oeq &lt;2 x float&gt; &lt; float 4, float 0 &gt;, &lt; float 5, float 0 &gt; <i>; yields: result=&lt;2 x i32&gt; &lt; i32 0, i32 -1 &gt;</i>
4144 &lt;result&gt; = vfcmp ult &lt;2 x double&gt; &lt; double 1, double 2 &gt;, &lt; double 2, double 2&gt; <i>; yields: result=&lt;2 x i64&gt; &lt; i64 -1, i64 0 &gt;</i>
Nate Begeman646fa482008-05-12 19:01:56 +0000[diff] [blame]4145</pre>
4146</div>
4147
4148<!-- _______________________________________________________________________ -->
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]4149<div class="doc_subsubsection">
4150 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
4151</div>
4152
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4153<div class="doc_text">
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]4154
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4155<h5>Syntax:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]4156
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4157<pre> &lt;result&gt; = phi &lt;ty&gt; [ &lt;val0&gt;, &lt;label0&gt;], ...<br></pre>
4158<h5>Overview:</h5>
4159<p>The '<tt>phi</tt>' instruction is used to implement the &#966; node in
4160the SSA graph representing the function.</p>
4161<h5>Arguments:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]4162
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4163<p>The type of the incoming values is specified with the first type
4164field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
4165as arguments, with one pair for each predecessor basic block of the
4166current block. Only values of <a href="#t_firstclass">first class</a>
4167type may be used as the value arguments to the PHI node. Only labels
4168may be used as the label arguments.</p>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]4169
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4170<p>There must be no non-phi instructions between the start of a basic
4171block and the PHI instructions: i.e. PHI instructions must be first in
4172a basic block.</p>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]4173
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4174<h5>Semantics:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]4175
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4176<p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
4177specified by the pair corresponding to the predecessor basic block that executed
4178just prior to the current block.</p>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]4179
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4180<h5>Example:</h5>
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]4181<pre>
4182Loop: ; Infinite loop that counts from 0 on up...
4183 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
4184 %nextindvar = add i32 %indvar, 1
4185 br label %Loop
4186</pre>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4187</div>
4188
4189<!-- _______________________________________________________________________ -->
4190<div class="doc_subsubsection">
4191 <a name="i_select">'<tt>select</tt>' Instruction</a>
4192</div>
4193
4194<div class="doc_text">
4195
4196<h5>Syntax:</h5>
4197
4198<pre>
Dan Gohmanb60ca3c2008-09-09 01:02:47 +0000[diff] [blame]4199 &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>
4200
4201 <i>selty</i> is either i1 or {&lt;N x i1&gt}
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4202</pre>
4203
4204<h5>Overview:</h5>
4205
4206<p>
4207The '<tt>select</tt>' instruction is used to choose one value based on a
4208condition, without branching.
4209</p>
4210
4211
4212<h5>Arguments:</h5>
4213
4214<p>
Dan Gohmanb60ca3c2008-09-09 01:02:47 +0000[diff] [blame]4215The '<tt>select</tt>' instruction requires an 'i1' value or
4216a vector of 'i1' values indicating the
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]4217condition, and two values of the same <a href="#t_firstclass">first class</a>
Dan Gohmanb60ca3c2008-09-09 01:02:47 +0000[diff] [blame]4218type. If the val1/val2 are vectors and
4219the condition is a scalar, then entire vectors are selected, not
Chris Lattner6704c212008-05-20 20:48:21 +0000[diff] [blame]4220individual elements.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4221</p>
4222
4223<h5>Semantics:</h5>
4224
4225<p>
Dan Gohmanb60ca3c2008-09-09 01:02:47 +0000[diff] [blame]4226If 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]4227value argument; otherwise, it returns the second value argument.
4228</p>
Dan Gohmanb60ca3c2008-09-09 01:02:47 +0000[diff] [blame]4229<p>
4230If the condition is a vector of i1, then the value arguments must
4231be vectors of the same size, and the selection is done element
4232by element.
4233</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4234
4235<h5>Example:</h5>
4236
4237<pre>
4238 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
4239</pre>
4240</div>
4241
4242
4243<!-- _______________________________________________________________________ -->
4244<div class="doc_subsubsection">
4245 <a name="i_call">'<tt>call</tt>' Instruction</a>
4246</div>
4247
4248<div class="doc_text">
4249
4250<h5>Syntax:</h5>
4251<pre>
Devang Patelac2fc272008-10-06 18:50:38 +0000[diff] [blame]4252 &lt;result&gt; = [tail] call [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">RetAttrs</a>] &lt;ty&gt; [&lt;fnty&gt;*] &lt;fnptrval&gt;(&lt;param list&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4253</pre>
4254
4255<h5>Overview:</h5>
4256
4257<p>The '<tt>call</tt>' instruction represents a simple function call.</p>
4258
4259<h5>Arguments:</h5>
4260
4261<p>This instruction requires several arguments:</p>
4262
4263<ol>
4264 <li>
4265 <p>The optional "tail" marker indicates whether the callee function accesses
4266 any allocas or varargs in the caller. If the "tail" marker is present, the
4267 function call is eligible for tail call optimization. Note that calls may
4268 be marked "tail" even if they do not occur before a <a
4269 href="#i_ret"><tt>ret</tt></a> instruction.
4270 </li>
4271 <li>
4272 <p>The optional "cconv" marker indicates which <a href="#callingconv">calling
4273 convention</a> the call should use. If none is specified, the call defaults
4274 to using C calling conventions.
4275 </li>
Devang Patelac2fc272008-10-06 18:50:38 +0000[diff] [blame]4276
4277 <li>
4278 <p>The optional <a href="#paramattrs">Parameter Attributes</a> list for
4279 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>',
4280 and '<tt>inreg</tt>' attributes are valid here.</p>
4281 </li>
4282
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4283 <li>
Nick Lewycky93082fc2007-09-08 13:57:50 +0000[diff] [blame]4284 <p>'<tt>ty</tt>': the type of the call instruction itself which is also
4285 the type of the return value. Functions that return no value are marked
4286 <tt><a href="#t_void">void</a></tt>.</p>
4287 </li>
4288 <li>
4289 <p>'<tt>fnty</tt>': shall be the signature of the pointer to function
4290 value being invoked. The argument types must match the types implied by
4291 this signature. This type can be omitted if the function is not varargs
4292 and if the function type does not return a pointer to a function.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4293 </li>
4294 <li>
4295 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
4296 be invoked. In most cases, this is a direct function invocation, but
4297 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
4298 to function value.</p>
4299 </li>
4300 <li>
4301 <p>'<tt>function args</tt>': argument list whose types match the
4302 function signature argument types. All arguments must be of
4303 <a href="#t_firstclass">first class</a> type. If the function signature
4304 indicates the function accepts a variable number of arguments, the extra
4305 arguments can be specified.</p>
4306 </li>
Devang Patelac2fc272008-10-06 18:50:38 +0000[diff] [blame]4307 <li>
4308 <p>The optional <a href="fnattrs">function attributes</a> list. Only
4309 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
4310 '<tt>readnone</tt>' attributes are valid here.</p>
4311 </li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4312</ol>
4313
4314<h5>Semantics:</h5>
4315
4316<p>The '<tt>call</tt>' instruction is used to cause control flow to
4317transfer to a specified function, with its incoming arguments bound to
4318the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
4319instruction in the called function, control flow continues with the
4320instruction after the function call, and the return value of the
Dan Gohman3e700032008-10-04 19:00:07 +0000[diff] [blame]4321function is bound to the result argument.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4322
4323<h5>Example:</h5>
4324
4325<pre>
Nick Lewycky93082fc2007-09-08 13:57:50 +0000[diff] [blame]4326 %retval = call i32 @test(i32 %argc)
Chris Lattner5e893ef2008-03-21 17:24:17 +0000[diff] [blame]4327 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42) <i>; yields i32</i>
4328 %X = tail call i32 @foo() <i>; yields i32</i>
4329 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
4330 call void %foo(i8 97 signext)
Devang Patela3cc5372008-03-10 20:49:15 +0000[diff] [blame]4331
4332 %struct.A = type { i32, i8 }
Devang Patelac2fc272008-10-06 18:50:38 +0000[diff] [blame]4333 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
Dan Gohman3e700032008-10-04 19:00:07 +0000[diff] [blame]4334 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
4335 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
Matthijs Kooijman2c4e05a2008-10-07 10:03:45 +0000[diff] [blame^]4336 %Z = call void @foo() noreturn <i>; indicates that %foo never returns nomrally</i>
4337 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4338</pre>
4339
4340</div>
4341
4342<!-- _______________________________________________________________________ -->
4343<div class="doc_subsubsection">
4344 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
4345</div>
4346
4347<div class="doc_text">
4348
4349<h5>Syntax:</h5>
4350
4351<pre>
4352 &lt;resultval&gt; = va_arg &lt;va_list*&gt; &lt;arglist&gt;, &lt;argty&gt;
4353</pre>
4354
4355<h5>Overview:</h5>
4356
4357<p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
4358the "variable argument" area of a function call. It is used to implement the
4359<tt>va_arg</tt> macro in C.</p>
4360
4361<h5>Arguments:</h5>
4362
4363<p>This instruction takes a <tt>va_list*</tt> value and the type of
4364the argument. It returns a value of the specified argument type and
4365increments the <tt>va_list</tt> to point to the next argument. The
4366actual type of <tt>va_list</tt> is target specific.</p>
4367
4368<h5>Semantics:</h5>
4369
4370<p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
4371type from the specified <tt>va_list</tt> and causes the
4372<tt>va_list</tt> to point to the next argument. For more information,
4373see the variable argument handling <a href="#int_varargs">Intrinsic
4374Functions</a>.</p>
4375
4376<p>It is legal for this instruction to be called in a function which does not
4377take a variable number of arguments, for example, the <tt>vfprintf</tt>
4378function.</p>
4379
4380<p><tt>va_arg</tt> is an LLVM instruction instead of an <a
4381href="#intrinsics">intrinsic function</a> because it takes a type as an
4382argument.</p>
4383
4384<h5>Example:</h5>
4385
4386<p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
4387
4388</div>
4389
4390<!-- *********************************************************************** -->
4391<div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
4392<!-- *********************************************************************** -->
4393
4394<div class="doc_text">
4395
4396<p>LLVM supports the notion of an "intrinsic function". These functions have
4397well known names and semantics and are required to follow certain restrictions.
4398Overall, these intrinsics represent an extension mechanism for the LLVM
4399language that does not require changing all of the transformations in LLVM when
4400adding to the language (or the bitcode reader/writer, the parser, etc...).</p>
4401
4402<p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
4403prefix is reserved in LLVM for intrinsic names; thus, function names may not
4404begin with this prefix. Intrinsic functions must always be external functions:
4405you cannot define the body of intrinsic functions. Intrinsic functions may
4406only be used in call or invoke instructions: it is illegal to take the address
4407of an intrinsic function. Additionally, because intrinsic functions are part
4408of the LLVM language, it is required if any are added that they be documented
4409here.</p>
4410
Chandler Carrutha228e392007-08-04 01:51:18 +0000[diff] [blame]4411<p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
4412a family of functions that perform the same operation but on different data
4413types. Because LLVM can represent over 8 million different integer types,
4414overloading is used commonly to allow an intrinsic function to operate on any
4415integer type. One or more of the argument types or the result type can be
4416overloaded to accept any integer type. Argument types may also be defined as
4417exactly matching a previous argument's type or the result type. This allows an
4418intrinsic function which accepts multiple arguments, but needs all of them to
4419be of the same type, to only be overloaded with respect to a single argument or
4420the result.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4421
Chandler Carrutha228e392007-08-04 01:51:18 +0000[diff] [blame]4422<p>Overloaded intrinsics will have the names of its overloaded argument types
4423encoded into its function name, each preceded by a period. Only those types
4424which are overloaded result in a name suffix. Arguments whose type is matched
4425against another type do not. For example, the <tt>llvm.ctpop</tt> function can
4426take an integer of any width and returns an integer of exactly the same integer
4427width. This leads to a family of functions such as
4428<tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29 %val)</tt>.
4429Only one type, the return type, is overloaded, and only one type suffix is
4430required. Because the argument's type is matched against the return type, it
4431does not require its own name suffix.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4432
4433<p>To learn how to add an intrinsic function, please see the
4434<a href="ExtendingLLVM.html">Extending LLVM Guide</a>.
4435</p>
4436
4437</div>
4438
4439<!-- ======================================================================= -->
4440<div class="doc_subsection">
4441 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
4442</div>
4443
4444<div class="doc_text">
4445
4446<p>Variable argument support is defined in LLVM with the <a
4447 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
4448intrinsic functions. These functions are related to the similarly
4449named macros defined in the <tt>&lt;stdarg.h&gt;</tt> header file.</p>
4450
4451<p>All of these functions operate on arguments that use a
4452target-specific value type "<tt>va_list</tt>". The LLVM assembly
4453language reference manual does not define what this type is, so all
4454transformations should be prepared to handle these functions regardless of
4455the type used.</p>
4456
4457<p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
4458instruction and the variable argument handling intrinsic functions are
4459used.</p>
4460
4461<div class="doc_code">
4462<pre>
4463define i32 @test(i32 %X, ...) {
4464 ; Initialize variable argument processing
4465 %ap = alloca i8*
4466 %ap2 = bitcast i8** %ap to i8*
4467 call void @llvm.va_start(i8* %ap2)
4468
4469 ; Read a single integer argument
4470 %tmp = va_arg i8** %ap, i32
4471
4472 ; Demonstrate usage of llvm.va_copy and llvm.va_end
4473 %aq = alloca i8*
4474 %aq2 = bitcast i8** %aq to i8*
4475 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
4476 call void @llvm.va_end(i8* %aq2)
4477
4478 ; Stop processing of arguments.
4479 call void @llvm.va_end(i8* %ap2)
4480 ret i32 %tmp
4481}
4482
4483declare void @llvm.va_start(i8*)
4484declare void @llvm.va_copy(i8*, i8*)
4485declare void @llvm.va_end(i8*)
4486</pre>
4487</div>
4488
4489</div>
4490
4491<!-- _______________________________________________________________________ -->
4492<div class="doc_subsubsection">
4493 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
4494</div>
4495
4496
4497<div class="doc_text">
4498<h5>Syntax:</h5>
4499<pre> declare void %llvm.va_start(i8* &lt;arglist&gt;)<br></pre>
4500<h5>Overview:</h5>
4501<P>The '<tt>llvm.va_start</tt>' intrinsic initializes
4502<tt>*&lt;arglist&gt;</tt> for subsequent use by <tt><a
4503href="#i_va_arg">va_arg</a></tt>.</p>
4504
4505<h5>Arguments:</h5>
4506
4507<P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
4508
4509<h5>Semantics:</h5>
4510
4511<P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
4512macro available in C. In a target-dependent way, it initializes the
4513<tt>va_list</tt> element to which the argument points, so that the next call to
4514<tt>va_arg</tt> will produce the first variable argument passed to the function.
4515Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
4516last argument of the function as the compiler can figure that out.</p>
4517
4518</div>
4519
4520<!-- _______________________________________________________________________ -->
4521<div class="doc_subsubsection">
4522 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
4523</div>
4524
4525<div class="doc_text">
4526<h5>Syntax:</h5>
4527<pre> declare void @llvm.va_end(i8* &lt;arglist&gt;)<br></pre>
4528<h5>Overview:</h5>
4529
4530<p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*&lt;arglist&gt;</tt>,
4531which has been initialized previously with <tt><a href="#int_va_start">llvm.va_start</a></tt>
4532or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
4533
4534<h5>Arguments:</h5>
4535
4536<p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
4537
4538<h5>Semantics:</h5>
4539
4540<p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
4541macro available in C. In a target-dependent way, it destroys the
4542<tt>va_list</tt> element to which the argument points. Calls to <a
4543href="#int_va_start"><tt>llvm.va_start</tt></a> and <a href="#int_va_copy">
4544<tt>llvm.va_copy</tt></a> must be matched exactly with calls to
4545<tt>llvm.va_end</tt>.</p>
4546
4547</div>
4548
4549<!-- _______________________________________________________________________ -->
4550<div class="doc_subsubsection">
4551 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
4552</div>
4553
4554<div class="doc_text">
4555
4556<h5>Syntax:</h5>
4557
4558<pre>
4559 declare void @llvm.va_copy(i8* &lt;destarglist&gt;, i8* &lt;srcarglist&gt;)
4560</pre>
4561
4562<h5>Overview:</h5>
4563
4564<p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
4565from the source argument list to the destination argument list.</p>
4566
4567<h5>Arguments:</h5>
4568
4569<p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
4570The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
4571
4572
4573<h5>Semantics:</h5>
4574
4575<p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
4576macro available in C. In a target-dependent way, it copies the source
4577<tt>va_list</tt> element into the destination <tt>va_list</tt> element. This
4578intrinsic is necessary because the <tt><a href="#int_va_start">
4579llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
4580example, memory allocation.</p>
4581
4582</div>
4583
4584<!-- ======================================================================= -->
4585<div class="doc_subsection">
4586 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
4587</div>
4588
4589<div class="doc_text">
4590
4591<p>
4592LLVM support for <a href="GarbageCollection.html">Accurate Garbage
Chris Lattner96451482008-08-05 18:29:16 +0000[diff] [blame]4593Collection</a> (GC) requires the implementation and generation of these
4594intrinsics.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4595These intrinsics allow identification of <a href="#int_gcroot">GC roots on the
4596stack</a>, as well as garbage collector implementations that require <a
4597href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a> barriers.
4598Front-ends for type-safe garbage collected languages should generate these
4599intrinsics to make use of the LLVM garbage collectors. For more details, see <a
4600href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
4601</p>
Christopher Lambcfe00962007-12-17 01:00:21 +0000[diff] [blame]4602
4603<p>The garbage collection intrinsics only operate on objects in the generic
4604 address space (address space zero).</p>
4605
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4606</div>
4607
4608<!-- _______________________________________________________________________ -->
4609<div class="doc_subsubsection">
4610 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
4611</div>
4612
4613<div class="doc_text">
4614
4615<h5>Syntax:</h5>
4616
4617<pre>
Chris Lattner38bd5dd2007-09-21 17:30:40 +0000[diff] [blame]4618 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4619</pre>
4620
4621<h5>Overview:</h5>
4622
4623<p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
4624the code generator, and allows some metadata to be associated with it.</p>
4625
4626<h5>Arguments:</h5>
4627
4628<p>The first argument specifies the address of a stack object that contains the
4629root pointer. The second pointer (which must be either a constant or a global
4630value address) contains the meta-data to be associated with the root.</p>
4631
4632<h5>Semantics:</h5>
4633
Chris Lattnera7d94ba2008-04-24 05:59:56 +0000[diff] [blame]4634<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]4635location. At compile-time, the code generator generates information to allow
Gordon Henriksen40393542007-12-25 02:31:26 +0000[diff] [blame]4636the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
4637intrinsic may only be used in a function which <a href="#gc">specifies a GC
4638algorithm</a>.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4639
4640</div>
4641
4642
4643<!-- _______________________________________________________________________ -->
4644<div class="doc_subsubsection">
4645 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
4646</div>
4647
4648<div class="doc_text">
4649
4650<h5>Syntax:</h5>
4651
4652<pre>
Chris Lattner38bd5dd2007-09-21 17:30:40 +0000[diff] [blame]4653 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4654</pre>
4655
4656<h5>Overview:</h5>
4657
4658<p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
4659locations, allowing garbage collector implementations that require read
4660barriers.</p>
4661
4662<h5>Arguments:</h5>
4663
4664<p>The second argument is the address to read from, which should be an address
4665allocated from the garbage collector. The first object is a pointer to the
4666start of the referenced object, if needed by the language runtime (otherwise
4667null).</p>
4668
4669<h5>Semantics:</h5>
4670
4671<p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
4672instruction, but may be replaced with substantially more complex code by the
Gordon Henriksen40393542007-12-25 02:31:26 +0000[diff] [blame]4673garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
4674may only be used in a function which <a href="#gc">specifies a GC
4675algorithm</a>.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4676
4677</div>
4678
4679
4680<!-- _______________________________________________________________________ -->
4681<div class="doc_subsubsection">
4682 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
4683</div>
4684
4685<div class="doc_text">
4686
4687<h5>Syntax:</h5>
4688
4689<pre>
Chris Lattner38bd5dd2007-09-21 17:30:40 +0000[diff] [blame]4690 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4691</pre>
4692
4693<h5>Overview:</h5>
4694
4695<p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
4696locations, allowing garbage collector implementations that require write
4697barriers (such as generational or reference counting collectors).</p>
4698
4699<h5>Arguments:</h5>
4700
4701<p>The first argument is the reference to store, the second is the start of the
4702object to store it to, and the third is the address of the field of Obj to
4703store to. If the runtime does not require a pointer to the object, Obj may be
4704null.</p>
4705
4706<h5>Semantics:</h5>
4707
4708<p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
4709instruction, but may be replaced with substantially more complex code by the
Gordon Henriksen40393542007-12-25 02:31:26 +0000[diff] [blame]4710garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
4711may only be used in a function which <a href="#gc">specifies a GC
4712algorithm</a>.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4713
4714</div>
4715
4716
4717
4718<!-- ======================================================================= -->
4719<div class="doc_subsection">
4720 <a name="int_codegen">Code Generator Intrinsics</a>
4721</div>
4722
4723<div class="doc_text">
4724<p>
4725These intrinsics are provided by LLVM to expose special features that may only
4726be implemented with code generator support.
4727</p>
4728
4729</div>
4730
4731<!-- _______________________________________________________________________ -->
4732<div class="doc_subsubsection">
4733 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
4734</div>
4735
4736<div class="doc_text">
4737
4738<h5>Syntax:</h5>
4739<pre>
4740 declare i8 *@llvm.returnaddress(i32 &lt;level&gt;)
4741</pre>
4742
4743<h5>Overview:</h5>
4744
4745<p>
4746The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
4747target-specific value indicating the return address of the current function
4748or one of its callers.
4749</p>
4750
4751<h5>Arguments:</h5>
4752
4753<p>
4754The argument to this intrinsic indicates which function to return the address
4755for. Zero indicates the calling function, one indicates its caller, etc. The
4756argument is <b>required</b> to be a constant integer value.
4757</p>
4758
4759<h5>Semantics:</h5>
4760
4761<p>
4762The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
4763the return address of the specified call frame, or zero if it cannot be
4764identified. The value returned by this intrinsic is likely to be incorrect or 0
4765for arguments other than zero, so it should only be used for debugging purposes.
4766</p>
4767
4768<p>
4769Note that calling this intrinsic does not prevent function inlining or other
4770aggressive transformations, so the value returned may not be that of the obvious
4771source-language caller.
4772</p>
4773</div>
4774
4775
4776<!-- _______________________________________________________________________ -->
4777<div class="doc_subsubsection">
4778 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
4779</div>
4780
4781<div class="doc_text">
4782
4783<h5>Syntax:</h5>
4784<pre>
Chris Lattner38bd5dd2007-09-21 17:30:40 +0000[diff] [blame]4785 declare i8 *@llvm.frameaddress(i32 &lt;level&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4786</pre>
4787
4788<h5>Overview:</h5>
4789
4790<p>
4791The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
4792target-specific frame pointer value for the specified stack frame.
4793</p>
4794
4795<h5>Arguments:</h5>
4796
4797<p>
4798The argument to this intrinsic indicates which function to return the frame
4799pointer for. Zero indicates the calling function, one indicates its caller,
4800etc. The argument is <b>required</b> to be a constant integer value.
4801</p>
4802
4803<h5>Semantics:</h5>
4804
4805<p>
4806The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
4807the frame address of the specified call frame, or zero if it cannot be
4808identified. The value returned by this intrinsic is likely to be incorrect or 0
4809for arguments other than zero, so it should only be used for debugging purposes.
4810</p>
4811
4812<p>
4813Note that calling this intrinsic does not prevent function inlining or other
4814aggressive transformations, so the value returned may not be that of the obvious
4815source-language caller.
4816</p>
4817</div>
4818
4819<!-- _______________________________________________________________________ -->
4820<div class="doc_subsubsection">
4821 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
4822</div>
4823
4824<div class="doc_text">
4825
4826<h5>Syntax:</h5>
4827<pre>
Chris Lattner38bd5dd2007-09-21 17:30:40 +0000[diff] [blame]4828 declare i8 *@llvm.stacksave()
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4829</pre>
4830
4831<h5>Overview:</h5>
4832
4833<p>
4834The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
4835the function stack, for use with <a href="#int_stackrestore">
4836<tt>llvm.stackrestore</tt></a>. This is useful for implementing language
4837features like scoped automatic variable sized arrays in C99.
4838</p>
4839
4840<h5>Semantics:</h5>
4841
4842<p>
4843This intrinsic returns a opaque pointer value that can be passed to <a
4844href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
4845<tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
4846<tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
4847state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
4848practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
4849that were allocated after the <tt>llvm.stacksave</tt> was executed.
4850</p>
4851
4852</div>
4853
4854<!-- _______________________________________________________________________ -->
4855<div class="doc_subsubsection">
4856 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
4857</div>
4858
4859<div class="doc_text">
4860
4861<h5>Syntax:</h5>
4862<pre>
4863 declare void @llvm.stackrestore(i8 * %ptr)
4864</pre>
4865
4866<h5>Overview:</h5>
4867
4868<p>
4869The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
4870the function stack to the state it was in when the corresponding <a
4871href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
4872useful for implementing language features like scoped automatic variable sized
4873arrays in C99.
4874</p>
4875
4876<h5>Semantics:</h5>
4877
4878<p>
4879See the description for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.
4880</p>
4881
4882</div>
4883
4884
4885<!-- _______________________________________________________________________ -->
4886<div class="doc_subsubsection">
4887 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
4888</div>
4889
4890<div class="doc_text">
4891
4892<h5>Syntax:</h5>
4893<pre>
Chris Lattner38bd5dd2007-09-21 17:30:40 +0000[diff] [blame]4894 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]4895</pre>
4896
4897<h5>Overview:</h5>
4898
4899
4900<p>
4901The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
4902a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
4903no
4904effect on the behavior of the program but can change its performance
4905characteristics.
4906</p>
4907
4908<h5>Arguments:</h5>
4909
4910<p>
4911<tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
4912determining if the fetch should be for a read (0) or write (1), and
4913<tt>locality</tt> is a temporal locality specifier ranging from (0) - no
4914locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
4915<tt>locality</tt> arguments must be constant integers.
4916</p>
4917
4918<h5>Semantics:</h5>
4919
4920<p>
4921This intrinsic does not modify the behavior of the program. In particular,
4922prefetches cannot trap and do not produce a value. On targets that support this
4923intrinsic, the prefetch can provide hints to the processor cache for better
4924performance.
4925</p>
4926
4927</div>
4928
4929<!-- _______________________________________________________________________ -->
4930<div class="doc_subsubsection">
4931 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
4932</div>
4933
4934<div class="doc_text">
4935
4936<h5>Syntax:</h5>
4937<pre>
Chris Lattner38bd5dd2007-09-21 17:30:40 +0000[diff] [blame]4938 declare void @llvm.pcmarker(i32 &lt;id&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4939</pre>
4940
4941<h5>Overview:</h5>
4942
4943
4944<p>
4945The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
Chris Lattner96451482008-08-05 18:29:16 +0000[diff] [blame]4946(PC) in a region of
4947code to simulators and other tools. The method is target specific, but it is
4948expected that the marker will use exported symbols to transmit the PC of the
4949marker.
4950The marker makes no guarantees that it will remain with any specific instruction
4951after optimizations. It is possible that the presence of a marker will inhibit
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4952optimizations. The intended use is to be inserted after optimizations to allow
4953correlations of simulation runs.
4954</p>
4955
4956<h5>Arguments:</h5>
4957
4958<p>
4959<tt>id</tt> is a numerical id identifying the marker.
4960</p>
4961
4962<h5>Semantics:</h5>
4963
4964<p>
4965This intrinsic does not modify the behavior of the program. Backends that do not
4966support this intrinisic may ignore it.
4967</p>
4968
4969</div>
4970
4971<!-- _______________________________________________________________________ -->
4972<div class="doc_subsubsection">
4973 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
4974</div>
4975
4976<div class="doc_text">
4977
4978<h5>Syntax:</h5>
4979<pre>
4980 declare i64 @llvm.readcyclecounter( )
4981</pre>
4982
4983<h5>Overview:</h5>
4984
4985
4986<p>
4987The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
4988counter register (or similar low latency, high accuracy clocks) on those targets
4989that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
4990As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
4991should only be used for small timings.
4992</p>
4993
4994<h5>Semantics:</h5>
4995
4996<p>
4997When directly supported, reading the cycle counter should not modify any memory.
4998Implementations are allowed to either return a application specific value or a
4999system wide value. On backends without support, this is lowered to a constant 0.
5000</p>
5001
5002</div>
5003
5004<!-- ======================================================================= -->
5005<div class="doc_subsection">
5006 <a name="int_libc">Standard C Library Intrinsics</a>
5007</div>
5008
5009<div class="doc_text">
5010<p>
5011LLVM provides intrinsics for a few important standard C library functions.
5012These intrinsics allow source-language front-ends to pass information about the
5013alignment of the pointer arguments to the code generator, providing opportunity
5014for more efficient code generation.
5015</p>
5016
5017</div>
5018
5019<!-- _______________________________________________________________________ -->
5020<div class="doc_subsubsection">
5021 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
5022</div>
5023
5024<div class="doc_text">
5025
5026<h5>Syntax:</h5>
5027<pre>
5028 declare void @llvm.memcpy.i32(i8 * &lt;dest&gt;, i8 * &lt;src&gt;,
5029 i32 &lt;len&gt;, i32 &lt;align&gt;)
5030 declare void @llvm.memcpy.i64(i8 * &lt;dest&gt;, i8 * &lt;src&gt;,
5031 i64 &lt;len&gt;, i32 &lt;align&gt;)
5032</pre>
5033
5034<h5>Overview:</h5>
5035
5036<p>
5037The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
5038location to the destination location.
5039</p>
5040
5041<p>
5042Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
5043intrinsics do not return a value, and takes an extra alignment argument.
5044</p>
5045
5046<h5>Arguments:</h5>
5047
5048<p>
5049The first argument is a pointer to the destination, the second is a pointer to
5050the source. The third argument is an integer argument
5051specifying the number of bytes to copy, and the fourth argument is the alignment
5052of the source and destination locations.
5053</p>
5054
5055<p>
5056If the call to this intrinisic has an alignment value that is not 0 or 1, then
5057the caller guarantees that both the source and destination pointers are aligned
5058to that boundary.
5059</p>
5060
5061<h5>Semantics:</h5>
5062
5063<p>
5064The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
5065location to the destination location, which are not allowed to overlap. It
5066copies "len" bytes of memory over. If the argument is known to be aligned to
5067some boundary, this can be specified as the fourth argument, otherwise it should
5068be set to 0 or 1.
5069</p>
5070</div>
5071
5072
5073<!-- _______________________________________________________________________ -->
5074<div class="doc_subsubsection">
5075 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
5076</div>
5077
5078<div class="doc_text">
5079
5080<h5>Syntax:</h5>
5081<pre>
5082 declare void @llvm.memmove.i32(i8 * &lt;dest&gt;, i8 * &lt;src&gt;,
5083 i32 &lt;len&gt;, i32 &lt;align&gt;)
5084 declare void @llvm.memmove.i64(i8 * &lt;dest&gt;, i8 * &lt;src&gt;,
5085 i64 &lt;len&gt;, i32 &lt;align&gt;)
5086</pre>
5087
5088<h5>Overview:</h5>
5089
5090<p>
5091The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
5092location to the destination location. It is similar to the
Chris Lattnerdba16ea2008-01-06 19:51:52 +0000[diff] [blame]5093'<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to overlap.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5094</p>
5095
5096<p>
5097Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
5098intrinsics do not return a value, and takes an extra alignment argument.
5099</p>
5100
5101<h5>Arguments:</h5>
5102
5103<p>
5104The first argument is a pointer to the destination, the second is a pointer to
5105the source. The third argument is an integer argument
5106specifying the number of bytes to copy, and the fourth argument is the alignment
5107of the source and destination locations.
5108</p>
5109
5110<p>
5111If the call to this intrinisic has an alignment value that is not 0 or 1, then
5112the caller guarantees that the source and destination pointers are aligned to
5113that boundary.
5114</p>
5115
5116<h5>Semantics:</h5>
5117
5118<p>
5119The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
5120location to the destination location, which may overlap. It
5121copies "len" bytes of memory over. If the argument is known to be aligned to
5122some boundary, this can be specified as the fourth argument, otherwise it should
5123be set to 0 or 1.
5124</p>
5125</div>
5126
5127
5128<!-- _______________________________________________________________________ -->
5129<div class="doc_subsubsection">
5130 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
5131</div>
5132
5133<div class="doc_text">
5134
5135<h5>Syntax:</h5>
5136<pre>
5137 declare void @llvm.memset.i32(i8 * &lt;dest&gt;, i8 &lt;val&gt;,
5138 i32 &lt;len&gt;, i32 &lt;align&gt;)
5139 declare void @llvm.memset.i64(i8 * &lt;dest&gt;, i8 &lt;val&gt;,
5140 i64 &lt;len&gt;, i32 &lt;align&gt;)
5141</pre>
5142
5143<h5>Overview:</h5>
5144
5145<p>
5146The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
5147byte value.
5148</p>
5149
5150<p>
5151Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
5152does not return a value, and takes an extra alignment argument.
5153</p>
5154
5155<h5>Arguments:</h5>
5156
5157<p>
5158The first argument is a pointer to the destination to fill, the second is the
5159byte value to fill it with, the third argument is an integer
5160argument specifying the number of bytes to fill, and the fourth argument is the
5161known alignment of destination location.
5162</p>
5163
5164<p>
5165If the call to this intrinisic has an alignment value that is not 0 or 1, then
5166the caller guarantees that the destination pointer is aligned to that boundary.
5167</p>
5168
5169<h5>Semantics:</h5>
5170
5171<p>
5172The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
5173the
5174destination location. If the argument is known to be aligned to some boundary,
5175this can be specified as the fourth argument, otherwise it should be set to 0 or
51761.
5177</p>
5178</div>
5179
5180
5181<!-- _______________________________________________________________________ -->
5182<div class="doc_subsubsection">
5183 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
5184</div>
5185
5186<div class="doc_text">
5187
5188<h5>Syntax:</h5>
Dale Johannesenf9adbb62007-10-02 17:47:38 +0000[diff] [blame]5189<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]5190floating point or vector of floating point type. Not all targets support all
5191types however.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5192<pre>
Dale Johannesenf9adbb62007-10-02 17:47:38 +0000[diff] [blame]5193 declare float @llvm.sqrt.f32(float %Val)
5194 declare double @llvm.sqrt.f64(double %Val)
5195 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
5196 declare fp128 @llvm.sqrt.f128(fp128 %Val)
5197 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5198</pre>
5199
5200<h5>Overview:</h5>
5201
5202<p>
5203The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
Dan Gohman361079c2007-10-15 20:30:11 +0000[diff] [blame]5204returning the same value as the libm '<tt>sqrt</tt>' functions would. Unlike
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5205<tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
Chris Lattnerf914a002008-01-29 07:00:44 +0000[diff] [blame]5206negative numbers other than -0.0 (which allows for better optimization, because
5207there is no need to worry about errno being set). <tt>llvm.sqrt(-0.0)</tt> is
5208defined to return -0.0 like IEEE sqrt.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5209</p>
5210
5211<h5>Arguments:</h5>
5212
5213<p>
5214The argument and return value are floating point numbers of the same type.
5215</p>
5216
5217<h5>Semantics:</h5>
5218
5219<p>
5220This function returns the sqrt of the specified operand if it is a nonnegative
5221floating point number.
5222</p>
5223</div>
5224
5225<!-- _______________________________________________________________________ -->
5226<div class="doc_subsubsection">
5227 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
5228</div>
5229
5230<div class="doc_text">
5231
5232<h5>Syntax:</h5>
Dale Johannesenf9adbb62007-10-02 17:47:38 +0000[diff] [blame]5233<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]5234floating point or vector of floating point type. Not all targets support all
5235types however.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5236<pre>
Dale Johannesenf9adbb62007-10-02 17:47:38 +0000[diff] [blame]5237 declare float @llvm.powi.f32(float %Val, i32 %power)
5238 declare double @llvm.powi.f64(double %Val, i32 %power)
5239 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
5240 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
5241 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5242</pre>
5243
5244<h5>Overview:</h5>
5245
5246<p>
5247The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
5248specified (positive or negative) power. The order of evaluation of
Dan Gohman361079c2007-10-15 20:30:11 +0000[diff] [blame]5249multiplications is not defined. When a vector of floating point type is
5250used, the second argument remains a scalar integer value.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5251</p>
5252
5253<h5>Arguments:</h5>
5254
5255<p>
5256The second argument is an integer power, and the first is a value to raise to
5257that power.
5258</p>
5259
5260<h5>Semantics:</h5>
5261
5262<p>
5263This function returns the first value raised to the second power with an
5264unspecified sequence of rounding operations.</p>
5265</div>
5266
Dan Gohman361079c2007-10-15 20:30:11 +0000[diff] [blame]5267<!-- _______________________________________________________________________ -->
5268<div class="doc_subsubsection">
5269 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
5270</div>
5271
5272<div class="doc_text">
5273
5274<h5>Syntax:</h5>
5275<p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
5276floating point or vector of floating point type. Not all targets support all
5277types however.
5278<pre>
5279 declare float @llvm.sin.f32(float %Val)
5280 declare double @llvm.sin.f64(double %Val)
5281 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
5282 declare fp128 @llvm.sin.f128(fp128 %Val)
5283 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
5284</pre>
5285
5286<h5>Overview:</h5>
5287
5288<p>
5289The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.
5290</p>
5291
5292<h5>Arguments:</h5>
5293
5294<p>
5295The argument and return value are floating point numbers of the same type.
5296</p>
5297
5298<h5>Semantics:</h5>
5299
5300<p>
5301This function returns the sine of the specified operand, returning the
5302same values as the libm <tt>sin</tt> functions would, and handles error
Dan Gohmaneaba92e2007-10-17 18:05:13 +0000[diff] [blame]5303conditions in the same way.</p>
Dan Gohman361079c2007-10-15 20:30:11 +0000[diff] [blame]5304</div>
5305
5306<!-- _______________________________________________________________________ -->
5307<div class="doc_subsubsection">
5308 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
5309</div>
5310
5311<div class="doc_text">
5312
5313<h5>Syntax:</h5>
5314<p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
5315floating point or vector of floating point type. Not all targets support all
5316types however.
5317<pre>
5318 declare float @llvm.cos.f32(float %Val)
5319 declare double @llvm.cos.f64(double %Val)
5320 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
5321 declare fp128 @llvm.cos.f128(fp128 %Val)
5322 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
5323</pre>
5324
5325<h5>Overview:</h5>
5326
5327<p>
5328The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.
5329</p>
5330
5331<h5>Arguments:</h5>
5332
5333<p>
5334The argument and return value are floating point numbers of the same type.
5335</p>
5336
5337<h5>Semantics:</h5>
5338
5339<p>
5340This function returns the cosine of the specified operand, returning the
5341same values as the libm <tt>cos</tt> functions would, and handles error
Dan Gohmaneaba92e2007-10-17 18:05:13 +0000[diff] [blame]5342conditions in the same way.</p>
Dan Gohman361079c2007-10-15 20:30:11 +0000[diff] [blame]5343</div>
5344
5345<!-- _______________________________________________________________________ -->
5346<div class="doc_subsubsection">
5347 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
5348</div>
5349
5350<div class="doc_text">
5351
5352<h5>Syntax:</h5>
5353<p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
5354floating point or vector of floating point type. Not all targets support all
5355types however.
5356<pre>
5357 declare float @llvm.pow.f32(float %Val, float %Power)
5358 declare double @llvm.pow.f64(double %Val, double %Power)
5359 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
5360 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
5361 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
5362</pre>
5363
5364<h5>Overview:</h5>
5365
5366<p>
5367The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
5368specified (positive or negative) power.
5369</p>
5370
5371<h5>Arguments:</h5>
5372
5373<p>
5374The second argument is a floating point power, and the first is a value to
5375raise to that power.
5376</p>
5377
5378<h5>Semantics:</h5>
5379
5380<p>
5381This function returns the first value raised to the second power,
5382returning the
5383same values as the libm <tt>pow</tt> functions would, and handles error
Dan Gohmaneaba92e2007-10-17 18:05:13 +0000[diff] [blame]5384conditions in the same way.</p>
Dan Gohman361079c2007-10-15 20:30:11 +0000[diff] [blame]5385</div>
5386
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5387
5388<!-- ======================================================================= -->
5389<div class="doc_subsection">
5390 <a name="int_manip">Bit Manipulation Intrinsics</a>
5391</div>
5392
5393<div class="doc_text">
5394<p>
5395LLVM provides intrinsics for a few important bit manipulation operations.
5396These allow efficient code generation for some algorithms.
5397</p>
5398
5399</div>
5400
5401<!-- _______________________________________________________________________ -->
5402<div class="doc_subsubsection">
5403 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
5404</div>
5405
5406<div class="doc_text">
5407
5408<h5>Syntax:</h5>
5409<p>This is an overloaded intrinsic function. You can use bswap on any integer
Chandler Carrutha228e392007-08-04 01:51:18 +0000[diff] [blame]5410type that is an even number of bytes (i.e. BitWidth % 16 == 0).
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5411<pre>
Chandler Carrutha228e392007-08-04 01:51:18 +0000[diff] [blame]5412 declare i16 @llvm.bswap.i16(i16 &lt;id&gt;)
5413 declare i32 @llvm.bswap.i32(i32 &lt;id&gt;)
5414 declare i64 @llvm.bswap.i64(i64 &lt;id&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5415</pre>
5416
5417<h5>Overview:</h5>
5418
5419<p>
5420The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
5421values with an even number of bytes (positive multiple of 16 bits). These are
5422useful for performing operations on data that is not in the target's native
5423byte order.
5424</p>
5425
5426<h5>Semantics:</h5>
5427
5428<p>
Chandler Carrutha228e392007-08-04 01:51:18 +0000[diff] [blame]5429The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5430and low byte of the input i16 swapped. Similarly, the <tt>llvm.bswap.i32</tt>
5431intrinsic returns an i32 value that has the four bytes of the input i32
5432swapped, 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]5433i32 will have its bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i48</tt>,
5434<tt>llvm.bswap.i64</tt> and other intrinsics extend this concept to
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5435additional even-byte lengths (6 bytes, 8 bytes and more, respectively).
5436</p>
5437
5438</div>
5439
5440<!-- _______________________________________________________________________ -->
5441<div class="doc_subsubsection">
5442 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
5443</div>
5444
5445<div class="doc_text">
5446
5447<h5>Syntax:</h5>
5448<p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
5449width. Not all targets support all bit widths however.
5450<pre>
Chandler Carrutha228e392007-08-04 01:51:18 +0000[diff] [blame]5451 declare i8 @llvm.ctpop.i8 (i8 &lt;src&gt;)
5452 declare i16 @llvm.ctpop.i16(i16 &lt;src&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5453 declare i32 @llvm.ctpop.i32(i32 &lt;src&gt;)
Chandler Carrutha228e392007-08-04 01:51:18 +0000[diff] [blame]5454 declare i64 @llvm.ctpop.i64(i64 &lt;src&gt;)
5455 declare i256 @llvm.ctpop.i256(i256 &lt;src&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5456</pre>
5457
5458<h5>Overview:</h5>
5459
5460<p>
5461The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
5462value.
5463</p>
5464
5465<h5>Arguments:</h5>
5466
5467<p>
5468The only argument is the value to be counted. The argument may be of any
5469integer type. The return type must match the argument type.
5470</p>
5471
5472<h5>Semantics:</h5>
5473
5474<p>
5475The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
5476</p>
5477</div>
5478
5479<!-- _______________________________________________________________________ -->
5480<div class="doc_subsubsection">
5481 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
5482</div>
5483
5484<div class="doc_text">
5485
5486<h5>Syntax:</h5>
5487<p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
5488integer bit width. Not all targets support all bit widths however.
5489<pre>
Chandler Carrutha228e392007-08-04 01:51:18 +0000[diff] [blame]5490 declare i8 @llvm.ctlz.i8 (i8 &lt;src&gt;)
5491 declare i16 @llvm.ctlz.i16(i16 &lt;src&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5492 declare i32 @llvm.ctlz.i32(i32 &lt;src&gt;)
Chandler Carrutha228e392007-08-04 01:51:18 +0000[diff] [blame]5493 declare i64 @llvm.ctlz.i64(i64 &lt;src&gt;)
5494 declare i256 @llvm.ctlz.i256(i256 &lt;src&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5495</pre>
5496
5497<h5>Overview:</h5>
5498
5499<p>
5500The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
5501leading zeros in a variable.
5502</p>
5503
5504<h5>Arguments:</h5>
5505
5506<p>
5507The only argument is the value to be counted. The argument may be of any
5508integer type. The return type must match the argument type.
5509</p>
5510
5511<h5>Semantics:</h5>
5512
5513<p>
5514The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
5515in a variable. If the src == 0 then the result is the size in bits of the type
5516of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.
5517</p>
5518</div>
5519
5520
5521
5522<!-- _______________________________________________________________________ -->
5523<div class="doc_subsubsection">
5524 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
5525</div>
5526
5527<div class="doc_text">
5528
5529<h5>Syntax:</h5>
5530<p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
5531integer bit width. Not all targets support all bit widths however.
5532<pre>
Chandler Carrutha228e392007-08-04 01:51:18 +0000[diff] [blame]5533 declare i8 @llvm.cttz.i8 (i8 &lt;src&gt;)
5534 declare i16 @llvm.cttz.i16(i16 &lt;src&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5535 declare i32 @llvm.cttz.i32(i32 &lt;src&gt;)
Chandler Carrutha228e392007-08-04 01:51:18 +0000[diff] [blame]5536 declare i64 @llvm.cttz.i64(i64 &lt;src&gt;)
5537 declare i256 @llvm.cttz.i256(i256 &lt;src&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5538</pre>
5539
5540<h5>Overview:</h5>
5541
5542<p>
5543The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
5544trailing zeros.
5545</p>
5546
5547<h5>Arguments:</h5>
5548
5549<p>
5550The only argument is the value to be counted. The argument may be of any
5551integer type. The return type must match the argument type.
5552</p>
5553
5554<h5>Semantics:</h5>
5555
5556<p>
5557The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
5558in a variable. If the src == 0 then the result is the size in bits of the type
5559of src. For example, <tt>llvm.cttz(2) = 1</tt>.
5560</p>
5561</div>
5562
5563<!-- _______________________________________________________________________ -->
5564<div class="doc_subsubsection">
5565 <a name="int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic</a>
5566</div>
5567
5568<div class="doc_text">
5569
5570<h5>Syntax:</h5>
5571<p>This is an overloaded intrinsic. You can use <tt>llvm.part.select</tt>
5572on any integer bit width.
5573<pre>
Chandler Carrutha228e392007-08-04 01:51:18 +0000[diff] [blame]5574 declare i17 @llvm.part.select.i17 (i17 %val, i32 %loBit, i32 %hiBit)
5575 declare i29 @llvm.part.select.i29 (i29 %val, i32 %loBit, i32 %hiBit)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5576</pre>
5577
5578<h5>Overview:</h5>
5579<p>The '<tt>llvm.part.select</tt>' family of intrinsic functions selects a
5580range of bits from an integer value and returns them in the same bit width as
5581the original value.</p>
5582
5583<h5>Arguments:</h5>
5584<p>The first argument, <tt>%val</tt> and the result may be integer types of
5585any bit width but they must have the same bit width. The second and third
5586arguments must be <tt>i32</tt> type since they specify only a bit index.</p>
5587
5588<h5>Semantics:</h5>
5589<p>The operation of the '<tt>llvm.part.select</tt>' intrinsic has two modes
5590of operation: forwards and reverse. If <tt>%loBit</tt> is greater than
5591<tt>%hiBits</tt> then the intrinsic operates in reverse mode. Otherwise it
5592operates in forward mode.</p>
5593<p>In forward mode, this intrinsic is the equivalent of shifting <tt>%val</tt>
5594right by <tt>%loBit</tt> bits and then ANDing it with a mask with
5595only the <tt>%hiBit - %loBit</tt> bits set, as follows:</p>
5596<ol>
5597 <li>The <tt>%val</tt> is shifted right (LSHR) by the number of bits specified
5598 by <tt>%loBits</tt>. This normalizes the value to the low order bits.</li>
5599 <li>The <tt>%loBits</tt> value is subtracted from the <tt>%hiBits</tt> value
5600 to determine the number of bits to retain.</li>
5601 <li>A mask of the retained bits is created by shifting a -1 value.</li>
5602 <li>The mask is ANDed with <tt>%val</tt> to produce the result.
5603</ol>
5604<p>In reverse mode, a similar computation is made except that the bits are
5605returned in the reverse order. So, for example, if <tt>X</tt> has the value
5606<tt>i16 0x0ACF (101011001111)</tt> and we apply
5607<tt>part.select(i16 X, 8, 3)</tt> to it, we get back the value
5608<tt>i16 0x0026 (000000100110)</tt>.</p>
5609</div>
5610
5611<div class="doc_subsubsection">
5612 <a name="int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic</a>
5613</div>
5614
5615<div class="doc_text">
5616
5617<h5>Syntax:</h5>
5618<p>This is an overloaded intrinsic. You can use <tt>llvm.part.set</tt>
5619on any integer bit width.
5620<pre>
Chandler Carrutha228e392007-08-04 01:51:18 +0000[diff] [blame]5621 declare i17 @llvm.part.set.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
5622 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]5623</pre>
5624
5625<h5>Overview:</h5>
5626<p>The '<tt>llvm.part.set</tt>' family of intrinsic functions replaces a range
5627of bits in an integer value with another integer value. It returns the integer
5628with the replaced bits.</p>
5629
5630<h5>Arguments:</h5>
5631<p>The first argument, <tt>%val</tt> and the result may be integer types of
5632any bit width but they must have the same bit width. <tt>%val</tt> is the value
5633whose bits will be replaced. The second argument, <tt>%repl</tt> may be an
5634integer of any bit width. The third and fourth arguments must be <tt>i32</tt>
5635type since they specify only a bit index.</p>
5636
5637<h5>Semantics:</h5>
5638<p>The operation of the '<tt>llvm.part.set</tt>' intrinsic has two modes
5639of operation: forwards and reverse. If <tt>%lo</tt> is greater than
5640<tt>%hi</tt> then the intrinsic operates in reverse mode. Otherwise it
5641operates in forward mode.</p>
5642<p>For both modes, the <tt>%repl</tt> value is prepared for use by either
5643truncating it down to the size of the replacement area or zero extending it
5644up to that size.</p>
5645<p>In forward mode, the bits between <tt>%lo</tt> and <tt>%hi</tt> (inclusive)
5646are replaced with corresponding bits from <tt>%repl</tt>. That is the 0th bit
5647in <tt>%repl</tt> replaces the <tt>%lo</tt>th bit in <tt>%val</tt> and etc. up
5648to the <tt>%hi</tt>th bit.
5649<p>In reverse mode, a similar computation is made except that the bits are
5650reversed. That is, the <tt>0</tt>th bit in <tt>%repl</tt> replaces the
5651<tt>%hi</tt> bit in <tt>%val</tt> and etc. down to the <tt>%lo</tt>th bit.
5652<h5>Examples:</h5>
5653<pre>
5654 llvm.part.set(0xFFFF, 0, 4, 7) -&gt; 0xFF0F
5655 llvm.part.set(0xFFFF, 0, 7, 4) -&gt; 0xFF0F
5656 llvm.part.set(0xFFFF, 1, 7, 4) -&gt; 0xFF8F
5657 llvm.part.set(0xFFFF, F, 8, 3) -&gt; 0xFFE7
5658 llvm.part.set(0xFFFF, 0, 3, 8) -&gt; 0xFE07
5659</pre>
5660</div>
5661
5662<!-- ======================================================================= -->
5663<div class="doc_subsection">
5664 <a name="int_debugger">Debugger Intrinsics</a>
5665</div>
5666
5667<div class="doc_text">
5668<p>
5669The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
5670are described in the <a
5671href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
5672Debugging</a> document.
5673</p>
5674</div>
5675
5676
5677<!-- ======================================================================= -->
5678<div class="doc_subsection">
5679 <a name="int_eh">Exception Handling Intrinsics</a>
5680</div>
5681
5682<div class="doc_text">
5683<p> The LLVM exception handling intrinsics (which all start with
5684<tt>llvm.eh.</tt> prefix), are described in the <a
5685href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
5686Handling</a> document. </p>
5687</div>
5688
5689<!-- ======================================================================= -->
5690<div class="doc_subsection">
Duncan Sands7407a9f2007-09-11 14:10:23 +0000[diff] [blame]5691 <a name="int_trampoline">Trampoline Intrinsic</a>
Duncan Sands38947cd2007-07-27 12:58:54 +0000[diff] [blame]5692</div>
5693
5694<div class="doc_text">
5695<p>
Duncan Sands7407a9f2007-09-11 14:10:23 +0000[diff] [blame]5696 This intrinsic makes it possible to excise one parameter, marked with
Duncan Sands38947cd2007-07-27 12:58:54 +0000[diff] [blame]5697 the <tt>nest</tt> attribute, from a function. The result is a callable
5698 function pointer lacking the nest parameter - the caller does not need
5699 to provide a value for it. Instead, the value to use is stored in
5700 advance in a "trampoline", a block of memory usually allocated
5701 on the stack, which also contains code to splice the nest value into the
5702 argument list. This is used to implement the GCC nested function address
5703 extension.
5704</p>
5705<p>
5706 For example, if the function is
5707 <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]5708 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]5709<pre>
Duncan Sands7407a9f2007-09-11 14:10:23 +0000[diff] [blame]5710 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
5711 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
5712 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
5713 %fp = bitcast i8* %p to i32 (i32, i32)*
Duncan Sands38947cd2007-07-27 12:58:54 +0000[diff] [blame]5714</pre>
Bill Wendlinge262b4c2007-09-22 09:23:55 +0000[diff] [blame]5715 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
5716 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
Duncan Sands38947cd2007-07-27 12:58:54 +0000[diff] [blame]5717</div>
5718
5719<!-- _______________________________________________________________________ -->
5720<div class="doc_subsubsection">
5721 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
5722</div>
5723<div class="doc_text">
5724<h5>Syntax:</h5>
5725<pre>
Duncan Sands7407a9f2007-09-11 14:10:23 +0000[diff] [blame]5726declare 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]5727</pre>
5728<h5>Overview:</h5>
5729<p>
Duncan Sands7407a9f2007-09-11 14:10:23 +0000[diff] [blame]5730 This fills the memory pointed to by <tt>tramp</tt> with code
5731 and returns a function pointer suitable for executing it.
Duncan Sands38947cd2007-07-27 12:58:54 +0000[diff] [blame]5732</p>
5733<h5>Arguments:</h5>
5734<p>
5735 The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
5736 pointers. The <tt>tramp</tt> argument must point to a sufficiently large
5737 and sufficiently aligned block of memory; this memory is written to by the
Duncan Sands35012212007-08-22 23:39:54 +0000[diff] [blame]5738 intrinsic. Note that the size and the alignment are target-specific - LLVM
5739 currently provides no portable way of determining them, so a front-end that
5740 generates this intrinsic needs to have some target-specific knowledge.
5741 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]5742</p>
5743<h5>Semantics:</h5>
5744<p>
5745 The block of memory pointed to by <tt>tramp</tt> is filled with target
Duncan Sands7407a9f2007-09-11 14:10:23 +0000[diff] [blame]5746 dependent code, turning it into a function. A pointer to this function is
5747 returned, but needs to be bitcast to an
Duncan Sands38947cd2007-07-27 12:58:54 +0000[diff] [blame]5748 <a href="#int_trampoline">appropriate function pointer type</a>
Duncan Sands7407a9f2007-09-11 14:10:23 +0000[diff] [blame]5749 before being called. The new function's signature is the same as that of
5750 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
5751 removed. At most one such <tt>nest</tt> argument is allowed, and it must be
5752 of pointer type. Calling the new function is equivalent to calling
5753 <tt>func</tt> with the same argument list, but with <tt>nval</tt> used for the
5754 missing <tt>nest</tt> argument. If, after calling
5755 <tt>llvm.init.trampoline</tt>, the memory pointed to by <tt>tramp</tt> is
5756 modified, then the effect of any later call to the returned function pointer is
5757 undefined.
Duncan Sands38947cd2007-07-27 12:58:54 +0000[diff] [blame]5758</p>
5759</div>
5760
5761<!-- ======================================================================= -->
5762<div class="doc_subsection">
Andrew Lenharth785610d2008-02-16 01:24:58 +0000[diff] [blame]5763 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
5764</div>
5765
5766<div class="doc_text">
5767<p>
5768 These intrinsic functions expand the "universal IR" of LLVM to represent
5769 hardware constructs for atomic operations and memory synchronization. This
5770 provides an interface to the hardware, not an interface to the programmer. It
Chris Lattner96451482008-08-05 18:29:16 +0000[diff] [blame]5771 is aimed at a low enough level to allow any programming models or APIs
5772 (Application Programming Interfaces) which
Andrew Lenharth785610d2008-02-16 01:24:58 +0000[diff] [blame]5773 need atomic behaviors to map cleanly onto it. It is also modeled primarily on
5774 hardware behavior. Just as hardware provides a "universal IR" for source
5775 languages, it also provides a starting point for developing a "universal"
5776 atomic operation and synchronization IR.
5777</p>
5778<p>
5779 These do <em>not</em> form an API such as high-level threading libraries,
5780 software transaction memory systems, atomic primitives, and intrinsic
5781 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
5782 application libraries. The hardware interface provided by LLVM should allow
5783 a clean implementation of all of these APIs and parallel programming models.
5784 No one model or paradigm should be selected above others unless the hardware
5785 itself ubiquitously does so.
5786
5787</p>
5788</div>
5789
5790<!-- _______________________________________________________________________ -->
5791<div class="doc_subsubsection">
5792 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
5793</div>
5794<div class="doc_text">
5795<h5>Syntax:</h5>
5796<pre>
5797declare void @llvm.memory.barrier( i1 &lt;ll&gt;, i1 &lt;ls&gt;, i1 &lt;sl&gt;, i1 &lt;ss&gt;,
5798i1 &lt;device&gt; )
5799
5800</pre>
5801<h5>Overview:</h5>
5802<p>
5803 The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
5804 specific pairs of memory access types.
5805</p>
5806<h5>Arguments:</h5>
5807<p>
5808 The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
5809 The first four arguments enables a specific barrier as listed below. The fith
5810 argument specifies that the barrier applies to io or device or uncached memory.
5811
5812</p>
5813 <ul>
5814 <li><tt>ll</tt>: load-load barrier</li>
5815 <li><tt>ls</tt>: load-store barrier</li>
5816 <li><tt>sl</tt>: store-load barrier</li>
5817 <li><tt>ss</tt>: store-store barrier</li>
5818 <li><tt>device</tt>: barrier applies to device and uncached memory also.
5819 </ul>
5820<h5>Semantics:</h5>
5821<p>
5822 This intrinsic causes the system to enforce some ordering constraints upon
5823 the loads and stores of the program. This barrier does not indicate
5824 <em>when</em> any events will occur, it only enforces an <em>order</em> in
5825 which they occur. For any of the specified pairs of load and store operations
5826 (f.ex. load-load, or store-load), all of the first operations preceding the
5827 barrier will complete before any of the second operations succeeding the
5828 barrier begin. Specifically the semantics for each pairing is as follows:
5829</p>
5830 <ul>
5831 <li><tt>ll</tt>: All loads before the barrier must complete before any load
5832 after the barrier begins.</li>
5833
5834 <li><tt>ls</tt>: All loads before the barrier must complete before any
5835 store after the barrier begins.</li>
5836 <li><tt>ss</tt>: All stores before the barrier must complete before any
5837 store after the barrier begins.</li>
5838 <li><tt>sl</tt>: All stores before the barrier must complete before any
5839 load after the barrier begins.</li>
5840 </ul>
5841<p>
5842 These semantics are applied with a logical "and" behavior when more than one
5843 is enabled in a single memory barrier intrinsic.
5844</p>
5845<p>
5846 Backends may implement stronger barriers than those requested when they do not
5847 support as fine grained a barrier as requested. Some architectures do not
5848 need all types of barriers and on such architectures, these become noops.
5849</p>
5850<h5>Example:</h5>
5851<pre>
5852%ptr = malloc i32
5853 store i32 4, %ptr
5854
5855%result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
5856 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
5857 <i>; guarantee the above finishes</i>
5858 store i32 8, %ptr <i>; before this begins</i>
5859</pre>
5860</div>
5861
Andrew Lenharthe44f3902008-02-21 06:45:13 +0000[diff] [blame]5862<!-- _______________________________________________________________________ -->
5863<div class="doc_subsubsection">
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]5864 <a name="int_atomic_cmp_swap">'<tt>llvm.atomic.cmp.swap.*</tt>' Intrinsic</a>
Andrew Lenharthe44f3902008-02-21 06:45:13 +0000[diff] [blame]5865</div>
5866<div class="doc_text">
5867<h5>Syntax:</h5>
5868<p>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]5869 This is an overloaded intrinsic. You can use <tt>llvm.atomic.cmp.swap</tt> on
5870 any integer bit width and for different address spaces. Not all targets
5871 support all bit widths however.</p>
Andrew Lenharthe44f3902008-02-21 06:45:13 +0000[diff] [blame]5872
5873<pre>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]5874declare i8 @llvm.atomic.cmp.swap.i8.p0i8( i8* &lt;ptr&gt;, i8 &lt;cmp&gt;, i8 &lt;val&gt; )
5875declare i16 @llvm.atomic.cmp.swap.i16.p0i16( i16* &lt;ptr&gt;, i16 &lt;cmp&gt;, i16 &lt;val&gt; )
5876declare i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* &lt;ptr&gt;, i32 &lt;cmp&gt;, i32 &lt;val&gt; )
5877declare 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]5878
5879</pre>
5880<h5>Overview:</h5>
5881<p>
5882 This loads a value in memory and compares it to a given value. If they are
5883 equal, it stores a new value into the memory.
5884</p>
5885<h5>Arguments:</h5>
5886<p>
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]5887 The <tt>llvm.atomic.cmp.swap</tt> intrinsic takes three arguments. The result as
Andrew Lenharthe44f3902008-02-21 06:45:13 +0000[diff] [blame]5888 well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
5889 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
5890 this integer type. While any bit width integer may be used, targets may only
5891 lower representations they support in hardware.
5892
5893</p>
5894<h5>Semantics:</h5>
5895<p>
5896 This entire intrinsic must be executed atomically. It first loads the value
5897 in memory pointed to by <tt>ptr</tt> and compares it with the value
5898 <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the memory. The
5899 loaded value is yielded in all cases. This provides the equivalent of an
5900 atomic compare-and-swap operation within the SSA framework.
5901</p>
5902<h5>Examples:</h5>
5903
5904<pre>
5905%ptr = malloc i32
5906 store i32 4, %ptr
5907
5908%val1 = add i32 4, 4
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]5909%result1 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 4, %val1 )
Andrew Lenharthe44f3902008-02-21 06:45:13 +0000[diff] [blame]5910 <i>; yields {i32}:result1 = 4</i>
5911%stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
5912%memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
5913
5914%val2 = add i32 1, 1
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]5915%result2 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 5, %val2 )
Andrew Lenharthe44f3902008-02-21 06:45:13 +0000[diff] [blame]5916 <i>; yields {i32}:result2 = 8</i>
5917%stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
5918
5919%memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
5920</pre>
5921</div>
5922
5923<!-- _______________________________________________________________________ -->
5924<div class="doc_subsubsection">
5925 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
5926</div>
5927<div class="doc_text">
5928<h5>Syntax:</h5>
5929
5930<p>
5931 This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
5932 integer bit width. Not all targets support all bit widths however.</p>
5933<pre>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]5934declare i8 @llvm.atomic.swap.i8.p0i8( i8* &lt;ptr&gt;, i8 &lt;val&gt; )
5935declare i16 @llvm.atomic.swap.i16.p0i16( i16* &lt;ptr&gt;, i16 &lt;val&gt; )
5936declare i32 @llvm.atomic.swap.i32.p0i32( i32* &lt;ptr&gt;, i32 &lt;val&gt; )
5937declare i64 @llvm.atomic.swap.i64.p0i64( i64* &lt;ptr&gt;, i64 &lt;val&gt; )
Andrew Lenharthe44f3902008-02-21 06:45:13 +0000[diff] [blame]5938
5939</pre>
5940<h5>Overview:</h5>
5941<p>
5942 This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
5943 the value from memory. It then stores the value in <tt>val</tt> in the memory
5944 at <tt>ptr</tt>.
5945</p>
5946<h5>Arguments:</h5>
5947
5948<p>
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]5949 The <tt>llvm.atomic.swap</tt> intrinsic takes two arguments. Both the
Andrew Lenharthe44f3902008-02-21 06:45:13 +0000[diff] [blame]5950 <tt>val</tt> argument and the result must be integers of the same bit width.
5951 The first argument, <tt>ptr</tt>, must be a pointer to a value of this
5952 integer type. The targets may only lower integer representations they
5953 support.
5954</p>
5955<h5>Semantics:</h5>
5956<p>
5957 This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
5958 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
5959 equivalent of an atomic swap operation within the SSA framework.
5960
5961</p>
5962<h5>Examples:</h5>
5963<pre>
5964%ptr = malloc i32
5965 store i32 4, %ptr
5966
5967%val1 = add i32 4, 4
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]5968%result1 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val1 )
Andrew Lenharthe44f3902008-02-21 06:45:13 +0000[diff] [blame]5969 <i>; yields {i32}:result1 = 4</i>
5970%stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
5971%memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
5972
5973%val2 = add i32 1, 1
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]5974%result2 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val2 )
Andrew Lenharthe44f3902008-02-21 06:45:13 +0000[diff] [blame]5975 <i>; yields {i32}:result2 = 8</i>
5976
5977%stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
5978%memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
5979</pre>
5980</div>
5981
5982<!-- _______________________________________________________________________ -->
5983<div class="doc_subsubsection">
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]5984 <a name="int_atomic_load_add">'<tt>llvm.atomic.load.add.*</tt>' Intrinsic</a>
Andrew Lenharthe44f3902008-02-21 06:45:13 +0000[diff] [blame]5985
5986</div>
5987<div class="doc_text">
5988<h5>Syntax:</h5>
5989<p>
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]5990 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]5991 integer bit width. Not all targets support all bit widths however.</p>
5992<pre>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]5993declare i8 @llvm.atomic.load.add.i8..p0i8( i8* &lt;ptr&gt;, i8 &lt;delta&gt; )
5994declare i16 @llvm.atomic.load.add.i16..p0i16( i16* &lt;ptr&gt;, i16 &lt;delta&gt; )
5995declare i32 @llvm.atomic.load.add.i32..p0i32( i32* &lt;ptr&gt;, i32 &lt;delta&gt; )
5996declare 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]5997
5998</pre>
5999<h5>Overview:</h5>
6000<p>
6001 This intrinsic adds <tt>delta</tt> to the value stored in memory at
6002 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
6003</p>
6004<h5>Arguments:</h5>
6005<p>
6006
6007 The intrinsic takes two arguments, the first a pointer to an integer value
6008 and the second an integer value. The result is also an integer value. These
6009 integer types can have any bit width, but they must all have the same bit
6010 width. The targets may only lower integer representations they support.
6011</p>
6012<h5>Semantics:</h5>
6013<p>
6014 This intrinsic does a series of operations atomically. It first loads the
6015 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
6016 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
6017</p>
6018
6019<h5>Examples:</h5>
6020<pre>
6021%ptr = malloc i32
6022 store i32 4, %ptr
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6023%result1 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 4 )
Andrew Lenharthe44f3902008-02-21 06:45:13 +0000[diff] [blame]6024 <i>; yields {i32}:result1 = 4</i>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6025%result2 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 2 )
Andrew Lenharthe44f3902008-02-21 06:45:13 +0000[diff] [blame]6026 <i>; yields {i32}:result2 = 8</i>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6027%result3 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 5 )
Andrew Lenharthe44f3902008-02-21 06:45:13 +0000[diff] [blame]6028 <i>; yields {i32}:result3 = 10</i>
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6029%memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
Andrew Lenharthe44f3902008-02-21 06:45:13 +0000[diff] [blame]6030</pre>
6031</div>
6032
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6033<!-- _______________________________________________________________________ -->
6034<div class="doc_subsubsection">
6035 <a name="int_atomic_load_sub">'<tt>llvm.atomic.load.sub.*</tt>' Intrinsic</a>
6036
6037</div>
6038<div class="doc_text">
6039<h5>Syntax:</h5>
6040<p>
6041 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]6042 any integer bit width and for different address spaces. Not all targets
6043 support all bit widths however.</p>
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6044<pre>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6045declare i8 @llvm.atomic.load.sub.i8.p0i32( i8* &lt;ptr&gt;, i8 &lt;delta&gt; )
6046declare i16 @llvm.atomic.load.sub.i16.p0i32( i16* &lt;ptr&gt;, i16 &lt;delta&gt; )
6047declare i32 @llvm.atomic.load.sub.i32.p0i32( i32* &lt;ptr&gt;, i32 &lt;delta&gt; )
6048declare 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]6049
6050</pre>
6051<h5>Overview:</h5>
6052<p>
6053 This intrinsic subtracts <tt>delta</tt> to the value stored in memory at
6054 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
6055</p>
6056<h5>Arguments:</h5>
6057<p>
6058
6059 The intrinsic takes two arguments, the first a pointer to an integer value
6060 and the second an integer value. The result is also an integer value. These
6061 integer types can have any bit width, but they must all have the same bit
6062 width. The targets may only lower integer representations they support.
6063</p>
6064<h5>Semantics:</h5>
6065<p>
6066 This intrinsic does a series of operations atomically. It first loads the
6067 value stored at <tt>ptr</tt>. It then subtracts <tt>delta</tt>, stores the
6068 result to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
6069</p>
6070
6071<h5>Examples:</h5>
6072<pre>
6073%ptr = malloc i32
6074 store i32 8, %ptr
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6075%result1 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 4 )
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6076 <i>; yields {i32}:result1 = 8</i>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6077%result2 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 2 )
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6078 <i>; yields {i32}:result2 = 4</i>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6079%result3 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 5 )
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6080 <i>; yields {i32}:result3 = 2</i>
6081%memval1 = load i32* %ptr <i>; yields {i32}:memval1 = -3</i>
6082</pre>
6083</div>
6084
6085<!-- _______________________________________________________________________ -->
6086<div class="doc_subsubsection">
6087 <a name="int_atomic_load_and">'<tt>llvm.atomic.load.and.*</tt>' Intrinsic</a><br>
6088 <a name="int_atomic_load_nand">'<tt>llvm.atomic.load.nand.*</tt>' Intrinsic</a><br>
6089 <a name="int_atomic_load_or">'<tt>llvm.atomic.load.or.*</tt>' Intrinsic</a><br>
6090 <a name="int_atomic_load_xor">'<tt>llvm.atomic.load.xor.*</tt>' Intrinsic</a><br>
6091
6092</div>
6093<div class="doc_text">
6094<h5>Syntax:</h5>
6095<p>
6096 These are overloaded intrinsics. You can use <tt>llvm.atomic.load_and</tt>,
6097 <tt>llvm.atomic.load_nand</tt>, <tt>llvm.atomic.load_or</tt>, and
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6098 <tt>llvm.atomic.load_xor</tt> on any integer bit width and for different
6099 address spaces. Not all targets support all bit widths however.</p>
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6100<pre>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6101declare i8 @llvm.atomic.load.and.i8.p0i8( i8* &lt;ptr&gt;, i8 &lt;delta&gt; )
6102declare i16 @llvm.atomic.load.and.i16.p0i16( i16* &lt;ptr&gt;, i16 &lt;delta&gt; )
6103declare i32 @llvm.atomic.load.and.i32.p0i32( i32* &lt;ptr&gt;, i32 &lt;delta&gt; )
6104declare 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]6105
6106</pre>
6107
6108<pre>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6109declare i8 @llvm.atomic.load.or.i8.p0i8( i8* &lt;ptr&gt;, i8 &lt;delta&gt; )
6110declare i16 @llvm.atomic.load.or.i16.p0i16( i16* &lt;ptr&gt;, i16 &lt;delta&gt; )
6111declare i32 @llvm.atomic.load.or.i32.p0i32( i32* &lt;ptr&gt;, i32 &lt;delta&gt; )
6112declare 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]6113
6114</pre>
6115
6116<pre>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6117declare i8 @llvm.atomic.load.nand.i8.p0i32( i8* &lt;ptr&gt;, i8 &lt;delta&gt; )
6118declare i16 @llvm.atomic.load.nand.i16.p0i32( i16* &lt;ptr&gt;, i16 &lt;delta&gt; )
6119declare i32 @llvm.atomic.load.nand.i32.p0i32( i32* &lt;ptr&gt;, i32 &lt;delta&gt; )
6120declare 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]6121
6122</pre>
6123
6124<pre>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6125declare i8 @llvm.atomic.load.xor.i8.p0i32( i8* &lt;ptr&gt;, i8 &lt;delta&gt; )
6126declare i16 @llvm.atomic.load.xor.i16.p0i32( i16* &lt;ptr&gt;, i16 &lt;delta&gt; )
6127declare i32 @llvm.atomic.load.xor.i32.p0i32( i32* &lt;ptr&gt;, i32 &lt;delta&gt; )
6128declare 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]6129
6130</pre>
6131<h5>Overview:</h5>
6132<p>
6133 These intrinsics bitwise the operation (and, nand, or, xor) <tt>delta</tt> to
6134 the value stored in memory at <tt>ptr</tt>. It yields the original value
6135 at <tt>ptr</tt>.
6136</p>
6137<h5>Arguments:</h5>
6138<p>
6139
6140 These intrinsics take two arguments, the first a pointer to an integer value
6141 and the second an integer value. The result is also an integer value. These
6142 integer types can have any bit width, but they must all have the same bit
6143 width. The targets may only lower integer representations they support.
6144</p>
6145<h5>Semantics:</h5>
6146<p>
6147 These intrinsics does a series of operations atomically. They first load the
6148 value stored at <tt>ptr</tt>. They then do the bitwise operation
6149 <tt>delta</tt>, store the result to <tt>ptr</tt>. They yield the original
6150 value stored at <tt>ptr</tt>.
6151</p>
6152
6153<h5>Examples:</h5>
6154<pre>
6155%ptr = malloc i32
6156 store i32 0x0F0F, %ptr
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6157%result0 = call i32 @llvm.atomic.load.nand.i32.p0i32( i32* %ptr, i32 0xFF )
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6158 <i>; yields {i32}:result0 = 0x0F0F</i>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6159%result1 = call i32 @llvm.atomic.load.and.i32.p0i32( i32* %ptr, i32 0xFF )
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6160 <i>; yields {i32}:result1 = 0xFFFFFFF0</i>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6161%result2 = call i32 @llvm.atomic.load.or.i32.p0i32( i32* %ptr, i32 0F )
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6162 <i>; yields {i32}:result2 = 0xF0</i>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6163%result3 = call i32 @llvm.atomic.load.xor.i32.p0i32( i32* %ptr, i32 0F )
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6164 <i>; yields {i32}:result3 = FF</i>
6165%memval1 = load i32* %ptr <i>; yields {i32}:memval1 = F0</i>
6166</pre>
6167</div>
6168
6169
6170<!-- _______________________________________________________________________ -->
6171<div class="doc_subsubsection">
6172 <a name="int_atomic_load_max">'<tt>llvm.atomic.load.max.*</tt>' Intrinsic</a><br>
6173 <a name="int_atomic_load_min">'<tt>llvm.atomic.load.min.*</tt>' Intrinsic</a><br>
6174 <a name="int_atomic_load_umax">'<tt>llvm.atomic.load.umax.*</tt>' Intrinsic</a><br>
6175 <a name="int_atomic_load_umin">'<tt>llvm.atomic.load.umin.*</tt>' Intrinsic</a><br>
6176
6177</div>
6178<div class="doc_text">
6179<h5>Syntax:</h5>
6180<p>
6181 These are overloaded intrinsics. You can use <tt>llvm.atomic.load_max</tt>,
6182 <tt>llvm.atomic.load_min</tt>, <tt>llvm.atomic.load_umax</tt>, and
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6183 <tt>llvm.atomic.load_umin</tt> on any integer bit width and for different
6184 address spaces. Not all targets
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6185 support all bit widths however.</p>
6186<pre>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6187declare i8 @llvm.atomic.load.max.i8.p0i8( i8* &lt;ptr&gt;, i8 &lt;delta&gt; )
6188declare i16 @llvm.atomic.load.max.i16.p0i16( i16* &lt;ptr&gt;, i16 &lt;delta&gt; )
6189declare i32 @llvm.atomic.load.max.i32.p0i32( i32* &lt;ptr&gt;, i32 &lt;delta&gt; )
6190declare 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]6191
6192</pre>
6193
6194<pre>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6195declare i8 @llvm.atomic.load.min.i8.p0i8( i8* &lt;ptr&gt;, i8 &lt;delta&gt; )
6196declare i16 @llvm.atomic.load.min.i16.p0i16( i16* &lt;ptr&gt;, i16 &lt;delta&gt; )
6197declare i32 @llvm.atomic.load.min.i32..p0i32( i32* &lt;ptr&gt;, i32 &lt;delta&gt; )
6198declare 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]6199
6200</pre>
6201
6202<pre>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6203declare i8 @llvm.atomic.load.umax.i8.p0i8( i8* &lt;ptr&gt;, i8 &lt;delta&gt; )
6204declare i16 @llvm.atomic.load.umax.i16.p0i16( i16* &lt;ptr&gt;, i16 &lt;delta&gt; )
6205declare i32 @llvm.atomic.load.umax.i32.p0i32( i32* &lt;ptr&gt;, i32 &lt;delta&gt; )
6206declare 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]6207
6208</pre>
6209
6210<pre>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6211declare i8 @llvm.atomic.load.umin.i8..p0i8( i8* &lt;ptr&gt;, i8 &lt;delta&gt; )
6212declare i16 @llvm.atomic.load.umin.i16.p0i16( i16* &lt;ptr&gt;, i16 &lt;delta&gt; )
6213declare i32 @llvm.atomic.load.umin.i32..p0i32( i32* &lt;ptr&gt;, i32 &lt;delta&gt; )
6214declare 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]6215
6216</pre>
6217<h5>Overview:</h5>
6218<p>
6219 These intrinsics takes the signed or unsigned minimum or maximum of
6220 <tt>delta</tt> and the value stored in memory at <tt>ptr</tt>. It yields the
6221 original value at <tt>ptr</tt>.
6222</p>
6223<h5>Arguments:</h5>
6224<p>
6225
6226 These intrinsics take two arguments, the first a pointer to an integer value
6227 and the second an integer value. The result is also an integer value. These
6228 integer types can have any bit width, but they must all have the same bit
6229 width. The targets may only lower integer representations they support.
6230</p>
6231<h5>Semantics:</h5>
6232<p>
6233 These intrinsics does a series of operations atomically. They first load the
6234 value stored at <tt>ptr</tt>. They then do the signed or unsigned min or max
6235 <tt>delta</tt> and the value, store the result to <tt>ptr</tt>. They yield
6236 the original value stored at <tt>ptr</tt>.
6237</p>
6238
6239<h5>Examples:</h5>
6240<pre>
6241%ptr = malloc i32
6242 store i32 7, %ptr
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6243%result0 = call i32 @llvm.atomic.load.min.i32.p0i32( i32* %ptr, i32 -2 )
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6244 <i>; yields {i32}:result0 = 7</i>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6245%result1 = call i32 @llvm.atomic.load.max.i32.p0i32( i32* %ptr, i32 8 )
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6246 <i>; yields {i32}:result1 = -2</i>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6247%result2 = call i32 @llvm.atomic.load.umin.i32.p0i32( i32* %ptr, i32 10 )
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6248 <i>; yields {i32}:result2 = 8</i>
Mon P Wangce3ac892008-07-30 04:36:53 +0000[diff] [blame]6249%result3 = call i32 @llvm.atomic.load.umax.i32.p0i32( i32* %ptr, i32 30 )
Mon P Wang6bde9ec2008-06-25 08:15:39 +0000[diff] [blame]6250 <i>; yields {i32}:result3 = 8</i>
6251%memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 30</i>
6252</pre>
6253</div>
Andrew Lenharth785610d2008-02-16 01:24:58 +0000[diff] [blame]6254
6255<!-- ======================================================================= -->
6256<div class="doc_subsection">
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]6257 <a name="int_general">General Intrinsics</a>
6258</div>
6259
6260<div class="doc_text">
6261<p> This class of intrinsics is designed to be generic and has
6262no specific purpose. </p>
6263</div>
6264
6265<!-- _______________________________________________________________________ -->
6266<div class="doc_subsubsection">
6267 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
6268</div>
6269
6270<div class="doc_text">
6271
6272<h5>Syntax:</h5>
6273<pre>
6274 declare void @llvm.var.annotation(i8* &lt;val&gt;, i8* &lt;str&gt;, i8* &lt;str&gt;, i32 &lt;int&gt; )
6275</pre>
6276
6277<h5>Overview:</h5>
6278
6279<p>
6280The '<tt>llvm.var.annotation</tt>' intrinsic
6281</p>
6282
6283<h5>Arguments:</h5>
6284
6285<p>
6286The first argument is a pointer to a value, the second is a pointer to a
6287global string, the third is a pointer to a global string which is the source
6288file name, and the last argument is the line number.
6289</p>
6290
6291<h5>Semantics:</h5>
6292
6293<p>
Anton Korobeynikove6e764f2008-01-15 22:31:34 +0000[diff] [blame]6294This intrinsic allows annotation of local variables with arbitrary strings.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]6295This can be useful for special purpose optimizations that want to look for these
Anton Korobeynikove6e764f2008-01-15 22:31:34 +0000[diff] [blame]6296annotations. These have no other defined use, they are ignored by code
6297generation and optimization.
6298</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]6299</div>
6300
Tanya Lattnerb306a9e2007-09-21 22:59:12 +0000[diff] [blame]6301<!-- _______________________________________________________________________ -->
6302<div class="doc_subsubsection">
Tanya Lattnerc9869b12007-09-21 23:57:59 +0000[diff] [blame]6303 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
Tanya Lattnerb306a9e2007-09-21 22:59:12 +0000[diff] [blame]6304</div>
6305
6306<div class="doc_text">
6307
6308<h5>Syntax:</h5>
Tanya Lattnere545be72007-09-21 23:56:27 +0000[diff] [blame]6309<p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
6310any integer bit width.
6311</p>
Tanya Lattnerb306a9e2007-09-21 22:59:12 +0000[diff] [blame]6312<pre>
Tanya Lattner09161fe2007-09-22 00:03:01 +0000[diff] [blame]6313 declare i8 @llvm.annotation.i8(i8 &lt;val&gt;, i8* &lt;str&gt;, i8* &lt;str&gt;, i32 &lt;int&gt; )
6314 declare i16 @llvm.annotation.i16(i16 &lt;val&gt;, i8* &lt;str&gt;, i8* &lt;str&gt;, i32 &lt;int&gt; )
6315 declare i32 @llvm.annotation.i32(i32 &lt;val&gt;, i8* &lt;str&gt;, i8* &lt;str&gt;, i32 &lt;int&gt; )
6316 declare i64 @llvm.annotation.i64(i64 &lt;val&gt;, i8* &lt;str&gt;, i8* &lt;str&gt;, i32 &lt;int&gt; )
6317 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]6318</pre>
6319
6320<h5>Overview:</h5>
Tanya Lattnere545be72007-09-21 23:56:27 +0000[diff] [blame]6321
6322<p>
6323The '<tt>llvm.annotation</tt>' intrinsic.
Tanya Lattnerb306a9e2007-09-21 22:59:12 +0000[diff] [blame]6324</p>
6325
6326<h5>Arguments:</h5>
6327
6328<p>
6329The first argument is an integer value (result of some expression),
6330the second is a pointer to a global string, the third is a pointer to a global
6331string which is the source file name, and the last argument is the line number.
Tanya Lattnere545be72007-09-21 23:56:27 +0000[diff] [blame]6332It returns the value of the first argument.
Tanya Lattnerb306a9e2007-09-21 22:59:12 +0000[diff] [blame]6333</p>
6334
6335<h5>Semantics:</h5>
6336
6337<p>
6338This intrinsic allows annotations to be put on arbitrary expressions
6339with arbitrary strings. This can be useful for special purpose optimizations
6340that want to look for these annotations. These have no other defined use, they
6341are ignored by code generation and optimization.
6342</div>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]6343
Anton Korobeynikove6e764f2008-01-15 22:31:34 +0000[diff] [blame]6344<!-- _______________________________________________________________________ -->
6345<div class="doc_subsubsection">
6346 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
6347</div>
6348
6349<div class="doc_text">
6350
6351<h5>Syntax:</h5>
6352<pre>
6353 declare void @llvm.trap()
6354</pre>
6355
6356<h5>Overview:</h5>
6357
6358<p>
6359The '<tt>llvm.trap</tt>' intrinsic
6360</p>
6361
6362<h5>Arguments:</h5>
6363
6364<p>
6365None
6366</p>
6367
6368<h5>Semantics:</h5>
6369
6370<p>
6371This intrinsics is lowered to the target dependent trap instruction. If the
6372target does not have a trap instruction, this intrinsic will be lowered to the
6373call of the abort() function.
6374</p>
6375</div>
6376
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]6377<!-- *********************************************************************** -->
6378<hr>
6379<address>
6380 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
6381 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
6382 <a href="http://validator.w3.org/check/referer"><img
Chris Lattner08497ce2008-01-04 04:33:49 +0000[diff] [blame]6383 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]6384
6385 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
6386 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
6387 Last modified: $Date$
6388</address>
Chris Lattner08497ce2008-01-04 04:33:49 +0000[diff] [blame]6389
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]6390</body>
6391</html>