Gitiles
Code Review Sign In
gerrit-public.fairphone.software / platform / external / llvm / 10368b64f83a38438bac7d6ff67031b48468b68d / . / docs / LangRef.html
blob: 09903f1016be76acc2decca4a81db68375ee86be [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>
Gordon Henriksen13fe5e32007-12-10 03:18:06 +0000[diff] [blame]29 <li><a href="#gc">Garbage Collector Names</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]30 <li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
31 <li><a href="#datalayout">Data Layout</a></li>
32 </ol>
33 </li>
34 <li><a href="#typesystem">Type System</a>
35 <ol>
Chris Lattner488772f2008-01-04 04:32:38 +0000[diff] [blame]36 <li><a href="#t_classifications">Type Classifications</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]37 <li><a href="#t_primitive">Primitive Types</a>
38 <ol>
Chris Lattner488772f2008-01-04 04:32:38 +0000[diff] [blame]39 <li><a href="#t_floating">Floating Point Types</a></li>
40 <li><a href="#t_void">Void Type</a></li>
41 <li><a href="#t_label">Label Type</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]42 </ol>
43 </li>
44 <li><a href="#t_derived">Derived Types</a>
45 <ol>
Chris Lattner251ab812007-12-18 06:18:21 +0000[diff] [blame]46 <li><a href="#t_integer">Integer Type</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]47 <li><a href="#t_array">Array Type</a></li>
48 <li><a href="#t_function">Function Type</a></li>
49 <li><a href="#t_pointer">Pointer Type</a></li>
50 <li><a href="#t_struct">Structure Type</a></li>
51 <li><a href="#t_pstruct">Packed Structure Type</a></li>
52 <li><a href="#t_vector">Vector Type</a></li>
53 <li><a href="#t_opaque">Opaque Type</a></li>
54 </ol>
55 </li>
56 </ol>
57 </li>
58 <li><a href="#constants">Constants</a>
59 <ol>
60 <li><a href="#simpleconstants">Simple Constants</a>
61 <li><a href="#aggregateconstants">Aggregate Constants</a>
62 <li><a href="#globalconstants">Global Variable and Function Addresses</a>
63 <li><a href="#undefvalues">Undefined Values</a>
64 <li><a href="#constantexprs">Constant Expressions</a>
65 </ol>
66 </li>
67 <li><a href="#othervalues">Other Values</a>
68 <ol>
69 <li><a href="#inlineasm">Inline Assembler Expressions</a>
70 </ol>
71 </li>
72 <li><a href="#instref">Instruction Reference</a>
73 <ol>
74 <li><a href="#terminators">Terminator Instructions</a>
75 <ol>
76 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
77 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
78 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
79 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
80 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
81 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
82 </ol>
83 </li>
84 <li><a href="#binaryops">Binary Operations</a>
85 <ol>
86 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
87 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
88 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
89 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
90 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
91 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
92 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
93 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
94 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
95 </ol>
96 </li>
97 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
98 <ol>
99 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
100 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
101 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
102 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
103 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
104 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
105 </ol>
106 </li>
107 <li><a href="#vectorops">Vector Operations</a>
108 <ol>
109 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
110 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
111 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
112 </ol>
113 </li>
114 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
115 <ol>
116 <li><a href="#i_malloc">'<tt>malloc</tt>' Instruction</a></li>
117 <li><a href="#i_free">'<tt>free</tt>' Instruction</a></li>
118 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
119 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
120 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
121 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
122 </ol>
123 </li>
124 <li><a href="#convertops">Conversion Operations</a>
125 <ol>
126 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
127 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
128 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
129 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
130 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
131 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
132 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
133 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
134 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
135 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
136 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
137 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
138 </ol>
139 <li><a href="#otherops">Other Operations</a>
140 <ol>
141 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
142 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
143 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
144 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
145 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
146 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
Devang Patela3cc5372008-03-10 20:49:15 +0000[diff] [blame]147 <li><a href="#i_getresult">'<tt>getresult</tt>' Instruction</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]148 </ol>
149 </li>
150 </ol>
151 </li>
152 <li><a href="#intrinsics">Intrinsic Functions</a>
153 <ol>
154 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
155 <ol>
156 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
157 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
158 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
159 </ol>
160 </li>
161 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
162 <ol>
163 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
164 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
165 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
166 </ol>
167 </li>
168 <li><a href="#int_codegen">Code Generator Intrinsics</a>
169 <ol>
170 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
171 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
172 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
173 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
174 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
175 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
176 <li><a href="#int_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
177 </ol>
178 </li>
179 <li><a href="#int_libc">Standard C Library Intrinsics</a>
180 <ol>
181 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
182 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
183 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
184 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
185 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
Dan Gohman361079c2007-10-15 20:30:11 +0000[diff] [blame]186 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
187 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
188 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]189 </ol>
190 </li>
191 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
192 <ol>
193 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
194 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
195 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
196 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
197 <li><a href="#int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic </a></li>
198 <li><a href="#int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic </a></li>
199 </ol>
200 </li>
201 <li><a href="#int_debugger">Debugger intrinsics</a></li>
202 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
Duncan Sands7407a9f2007-09-11 14:10:23 +0000[diff] [blame]203 <li><a href="#int_trampoline">Trampoline Intrinsic</a>
Duncan Sands38947cd2007-07-27 12:58:54 +0000[diff] [blame]204 <ol>
205 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
Duncan Sands38947cd2007-07-27 12:58:54 +0000[diff] [blame]206 </ol>
207 </li>
Andrew Lenharth785610d2008-02-16 01:24:58 +0000[diff] [blame]208 <li><a href="#int_atomics">Atomic intrinsics</a>
209 <ol>
Andrew Lenharthe44f3902008-02-21 06:45:13 +0000[diff] [blame]210 <li><a href="#int_memory_barrier"><tt>llvm.memory_barrier</tt></a></li>
211 <li><a href="#int_atomic_lcs"><tt>llvm.atomic.lcs</tt></a></li>
212 <li><a href="#int_atomic_las"><tt>llvm.atomic.las</tt></a></li>
213 <li><a href="#int_atomic_swap"><tt>llvm.atomic.swap</tt></a></li>
Andrew Lenharth785610d2008-02-16 01:24:58 +0000[diff] [blame]214 </ol>
215 </li>
Reid Spencerb043f672007-07-20 19:59:11 +0000[diff] [blame]216 <li><a href="#int_general">General intrinsics</a>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]217 <ol>
Reid Spencerb043f672007-07-20 19:59:11 +0000[diff] [blame]218 <li><a href="#int_var_annotation">
Tanya Lattner51369f32007-09-22 00:01:26 +0000[diff] [blame]219 <tt>llvm.var.annotation</tt>' Intrinsic</a></li>
Tanya Lattnerb306a9e2007-09-21 22:59:12 +0000[diff] [blame]220 <li><a href="#int_annotation">
Tanya Lattner51369f32007-09-22 00:01:26 +0000[diff] [blame]221 <tt>llvm.annotation.*</tt>' Intrinsic</a></li>
Anton Korobeynikove6e764f2008-01-15 22:31:34 +0000[diff] [blame]222 <li><a href="#int_trap">
223 <tt>llvm.trap</tt>' Intrinsic</a></li>
Tanya Lattnerb306a9e2007-09-21 22:59:12 +0000[diff] [blame]224 </ol>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]225 </li>
226 </ol>
227 </li>
228</ol>
229
230<div class="doc_author">
231 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
232 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
233</div>
234
235<!-- *********************************************************************** -->
236<div class="doc_section"> <a name="abstract">Abstract </a></div>
237<!-- *********************************************************************** -->
238
239<div class="doc_text">
240<p>This document is a reference manual for the LLVM assembly language.
241LLVM is an SSA based representation that provides type safety,
242low-level operations, flexibility, and the capability of representing
243'all' high-level languages cleanly. It is the common code
244representation used throughout all phases of the LLVM compilation
245strategy.</p>
246</div>
247
248<!-- *********************************************************************** -->
249<div class="doc_section"> <a name="introduction">Introduction</a> </div>
250<!-- *********************************************************************** -->
251
252<div class="doc_text">
253
254<p>The LLVM code representation is designed to be used in three
255different forms: as an in-memory compiler IR, as an on-disk bitcode
256representation (suitable for fast loading by a Just-In-Time compiler),
257and as a human readable assembly language representation. This allows
258LLVM to provide a powerful intermediate representation for efficient
259compiler transformations and analysis, while providing a natural means
260to debug and visualize the transformations. The three different forms
261of LLVM are all equivalent. This document describes the human readable
262representation and notation.</p>
263
264<p>The LLVM representation aims to be light-weight and low-level
265while being expressive, typed, and extensible at the same time. It
266aims to be a "universal IR" of sorts, by being at a low enough level
267that high-level ideas may be cleanly mapped to it (similar to how
268microprocessors are "universal IR's", allowing many source languages to
269be mapped to them). By providing type information, LLVM can be used as
270the target of optimizations: for example, through pointer analysis, it
271can be proven that a C automatic variable is never accessed outside of
272the current function... allowing it to be promoted to a simple SSA
273value instead of a memory location.</p>
274
275</div>
276
277<!-- _______________________________________________________________________ -->
278<div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
279
280<div class="doc_text">
281
282<p>It is important to note that this document describes 'well formed'
283LLVM assembly language. There is a difference between what the parser
284accepts and what is considered 'well formed'. For example, the
285following instruction is syntactically okay, but not well formed:</p>
286
287<div class="doc_code">
288<pre>
289%x = <a href="#i_add">add</a> i32 1, %x
290</pre>
291</div>
292
293<p>...because the definition of <tt>%x</tt> does not dominate all of
294its uses. The LLVM infrastructure provides a verification pass that may
295be used to verify that an LLVM module is well formed. This pass is
296automatically run by the parser after parsing input assembly and by
297the optimizer before it outputs bitcode. The violations pointed out
298by the verifier pass indicate bugs in transformation passes or input to
299the parser.</p>
300</div>
301
Chris Lattnera83fdc02007-10-03 17:34:29 +0000[diff] [blame]302<!-- Describe the typesetting conventions here. -->
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]303
304<!-- *********************************************************************** -->
305<div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
306<!-- *********************************************************************** -->
307
308<div class="doc_text">
309
Reid Spencerc8245b02007-08-07 14:34:28 +0000[diff] [blame]310 <p>LLVM identifiers come in two basic types: global and local. Global
311 identifiers (functions, global variables) begin with the @ character. Local
312 identifiers (register names, types) begin with the % character. Additionally,
313 there are three different formats for identifiers, for different purposes:
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]314
315<ol>
Reid Spencerc8245b02007-08-07 14:34:28 +0000[diff] [blame]316 <li>Named values are represented as a string of characters with their prefix.
317 For example, %foo, @DivisionByZero, %a.really.long.identifier. The actual
318 regular expression used is '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]319 Identifiers which require other characters in their names can be surrounded
Reid Spencerc8245b02007-08-07 14:34:28 +0000[diff] [blame]320 with quotes. In this way, anything except a <tt>&quot;</tt> character can
321 be used in a named value.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]322
Reid Spencerc8245b02007-08-07 14:34:28 +0000[diff] [blame]323 <li>Unnamed values are represented as an unsigned numeric value with their
324 prefix. For example, %12, @2, %44.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]325
326 <li>Constants, which are described in a <a href="#constants">section about
327 constants</a>, below.</li>
328</ol>
329
Reid Spencerc8245b02007-08-07 14:34:28 +0000[diff] [blame]330<p>LLVM requires that values start with a prefix for two reasons: Compilers
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]331don't need to worry about name clashes with reserved words, and the set of
332reserved words may be expanded in the future without penalty. Additionally,
333unnamed identifiers allow a compiler to quickly come up with a temporary
334variable without having to avoid symbol table conflicts.</p>
335
336<p>Reserved words in LLVM are very similar to reserved words in other
337languages. There are keywords for different opcodes
338('<tt><a href="#i_add">add</a></tt>',
339 '<tt><a href="#i_bitcast">bitcast</a></tt>',
340 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
341href="#t_void">void</a></tt>', '<tt><a href="#t_primitive">i32</a></tt>', etc...),
342and others. These reserved words cannot conflict with variable names, because
Reid Spencerc8245b02007-08-07 14:34:28 +0000[diff] [blame]343none of them start with a prefix character ('%' or '@').</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]344
345<p>Here is an example of LLVM code to multiply the integer variable
346'<tt>%X</tt>' by 8:</p>
347
348<p>The easy way:</p>
349
350<div class="doc_code">
351<pre>
352%result = <a href="#i_mul">mul</a> i32 %X, 8
353</pre>
354</div>
355
356<p>After strength reduction:</p>
357
358<div class="doc_code">
359<pre>
360%result = <a href="#i_shl">shl</a> i32 %X, i8 3
361</pre>
362</div>
363
364<p>And the hard way:</p>
365
366<div class="doc_code">
367<pre>
368<a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
369<a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
370%result = <a href="#i_add">add</a> i32 %1, %1
371</pre>
372</div>
373
374<p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
375important lexical features of LLVM:</p>
376
377<ol>
378
379 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
380 line.</li>
381
382 <li>Unnamed temporaries are created when the result of a computation is not
383 assigned to a named value.</li>
384
385 <li>Unnamed temporaries are numbered sequentially</li>
386
387</ol>
388
389<p>...and it also shows a convention that we follow in this document. When
390demonstrating instructions, we will follow an instruction with a comment that
391defines the type and name of value produced. Comments are shown in italic
392text.</p>
393
394</div>
395
396<!-- *********************************************************************** -->
397<div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
398<!-- *********************************************************************** -->
399
400<!-- ======================================================================= -->
401<div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
402</div>
403
404<div class="doc_text">
405
406<p>LLVM programs are composed of "Module"s, each of which is a
407translation unit of the input programs. Each module consists of
408functions, global variables, and symbol table entries. Modules may be
409combined together with the LLVM linker, which merges function (and
410global variable) definitions, resolves forward declarations, and merges
411symbol table entries. Here is an example of the "hello world" module:</p>
412
413<div class="doc_code">
414<pre><i>; Declare the string constant as a global constant...</i>
415<a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a
416 href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
417
418<i>; External declaration of the puts function</i>
419<a href="#functionstructure">declare</a> i32 @puts(i8 *) <i>; i32(i8 *)* </i>
420
421<i>; Definition of main function</i>
422define i32 @main() { <i>; i32()* </i>
423 <i>; Convert [13x i8 ]* to i8 *...</i>
424 %cast210 = <a
425 href="#i_getelementptr">getelementptr</a> [13 x i8 ]* @.LC0, i64 0, i64 0 <i>; i8 *</i>
426
427 <i>; Call puts function to write out the string to stdout...</i>
428 <a
429 href="#i_call">call</a> i32 @puts(i8 * %cast210) <i>; i32</i>
430 <a
431 href="#i_ret">ret</a> i32 0<br>}<br>
432</pre>
433</div>
434
435<p>This example is made up of a <a href="#globalvars">global variable</a>
436named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
437function, and a <a href="#functionstructure">function definition</a>
438for "<tt>main</tt>".</p>
439
440<p>In general, a module is made up of a list of global values,
441where both functions and global variables are global values. Global values are
442represented by a pointer to a memory location (in this case, a pointer to an
443array of char, and a pointer to a function), and have one of the following <a
444href="#linkage">linkage types</a>.</p>
445
446</div>
447
448<!-- ======================================================================= -->
449<div class="doc_subsection">
450 <a name="linkage">Linkage Types</a>
451</div>
452
453<div class="doc_text">
454
455<p>
456All Global Variables and Functions have one of the following types of linkage:
457</p>
458
459<dl>
460
461 <dt><tt><b><a name="linkage_internal">internal</a></b></tt> </dt>
462
463 <dd>Global values with internal linkage are only directly accessible by
464 objects in the current module. In particular, linking code into a module with
465 an internal global value may cause the internal to be renamed as necessary to
466 avoid collisions. Because the symbol is internal to the module, all
467 references can be updated. This corresponds to the notion of the
468 '<tt>static</tt>' keyword in C.
469 </dd>
470
471 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
472
473 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
474 the same name when linkage occurs. This is typically used to implement
475 inline functions, templates, or other code which must be generated in each
476 translation unit that uses it. Unreferenced <tt>linkonce</tt> globals are
477 allowed to be discarded.
478 </dd>
479
480 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
481
482 <dd>"<tt>weak</tt>" linkage is exactly the same as <tt>linkonce</tt> linkage,
483 except that unreferenced <tt>weak</tt> globals may not be discarded. This is
484 used for globals that may be emitted in multiple translation units, but that
485 are not guaranteed to be emitted into every translation unit that uses them.
486 One example of this are common globals in C, such as "<tt>int X;</tt>" at
487 global scope.
488 </dd>
489
490 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
491
492 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
493 pointer to array type. When two global variables with appending linkage are
494 linked together, the two global arrays are appended together. This is the
495 LLVM, typesafe, equivalent of having the system linker append together
496 "sections" with identical names when .o files are linked.
497 </dd>
498
499 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt>: </dt>
500 <dd>The semantics of this linkage follow the ELF model: the symbol is weak
501 until linked, if not linked, the symbol becomes null instead of being an
502 undefined reference.
503 </dd>
504
505 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
506
507 <dd>If none of the above identifiers are used, the global is externally
508 visible, meaning that it participates in linkage and can be used to resolve
509 external symbol references.
510 </dd>
511</dl>
512
513 <p>
514 The next two types of linkage are targeted for Microsoft Windows platform
515 only. They are designed to support importing (exporting) symbols from (to)
516 DLLs.
517 </p>
518
519 <dl>
520 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt>: </dt>
521
522 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
523 or variable via a global pointer to a pointer that is set up by the DLL
524 exporting the symbol. On Microsoft Windows targets, the pointer name is
525 formed by combining <code>_imp__</code> and the function or variable name.
526 </dd>
527
528 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt>: </dt>
529
530 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
531 pointer to a pointer in a DLL, so that it can be referenced with the
532 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
533 name is formed by combining <code>_imp__</code> and the function or variable
534 name.
535 </dd>
536
537</dl>
538
539<p><a name="linkage_external"></a>For example, since the "<tt>.LC0</tt>"
540variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
541variable and was linked with this one, one of the two would be renamed,
542preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
543external (i.e., lacking any linkage declarations), they are accessible
544outside of the current module.</p>
545<p>It is illegal for a function <i>declaration</i>
546to have any linkage type other than "externally visible", <tt>dllimport</tt>,
547or <tt>extern_weak</tt>.</p>
548<p>Aliases can have only <tt>external</tt>, <tt>internal</tt> and <tt>weak</tt>
549linkages.
550</div>
551
552<!-- ======================================================================= -->
553<div class="doc_subsection">
554 <a name="callingconv">Calling Conventions</a>
555</div>
556
557<div class="doc_text">
558
559<p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
560and <a href="#i_invoke">invokes</a> can all have an optional calling convention
561specified for the call. The calling convention of any pair of dynamic
562caller/callee must match, or the behavior of the program is undefined. The
563following calling conventions are supported by LLVM, and more may be added in
564the future:</p>
565
566<dl>
567 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
568
569 <dd>This calling convention (the default if no other calling convention is
570 specified) matches the target C calling conventions. This calling convention
571 supports varargs function calls and tolerates some mismatch in the declared
572 prototype and implemented declaration of the function (as does normal C).
573 </dd>
574
575 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
576
577 <dd>This calling convention attempts to make calls as fast as possible
578 (e.g. by passing things in registers). This calling convention allows the
579 target to use whatever tricks it wants to produce fast code for the target,
580 without having to conform to an externally specified ABI. Implementations of
581 this convention should allow arbitrary tail call optimization to be supported.
582 This calling convention does not support varargs and requires the prototype of
583 all callees to exactly match the prototype of the function definition.
584 </dd>
585
586 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
587
588 <dd>This calling convention attempts to make code in the caller as efficient
589 as possible under the assumption that the call is not commonly executed. As
590 such, these calls often preserve all registers so that the call does not break
591 any live ranges in the caller side. This calling convention does not support
592 varargs and requires the prototype of all callees to exactly match the
593 prototype of the function definition.
594 </dd>
595
596 <dt><b>"<tt>cc &lt;<em>n</em>&gt;</tt>" - Numbered convention</b>:</dt>
597
598 <dd>Any calling convention may be specified by number, allowing
599 target-specific calling conventions to be used. Target specific calling
600 conventions start at 64.
601 </dd>
602</dl>
603
604<p>More calling conventions can be added/defined on an as-needed basis, to
605support pascal conventions or any other well-known target-independent
606convention.</p>
607
608</div>
609
610<!-- ======================================================================= -->
611<div class="doc_subsection">
612 <a name="visibility">Visibility Styles</a>
613</div>
614
615<div class="doc_text">
616
617<p>
618All Global Variables and Functions have one of the following visibility styles:
619</p>
620
621<dl>
622 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
623
624 <dd>On ELF, default visibility means that the declaration is visible to other
625 modules and, in shared libraries, means that the declared entity may be
626 overridden. On Darwin, default visibility means that the declaration is
627 visible to other modules. Default visibility corresponds to "external
628 linkage" in the language.
629 </dd>
630
631 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
632
633 <dd>Two declarations of an object with hidden visibility refer to the same
634 object if they are in the same shared object. Usually, hidden visibility
635 indicates that the symbol will not be placed into the dynamic symbol table,
636 so no other module (executable or shared library) can reference it
637 directly.
638 </dd>
639
640 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
641
642 <dd>On ELF, protected visibility indicates that the symbol will be placed in
643 the dynamic symbol table, but that references within the defining module will
644 bind to the local symbol. That is, the symbol cannot be overridden by another
645 module.
646 </dd>
647</dl>
648
649</div>
650
651<!-- ======================================================================= -->
652<div class="doc_subsection">
653 <a name="globalvars">Global Variables</a>
654</div>
655
656<div class="doc_text">
657
658<p>Global variables define regions of memory allocated at compilation time
659instead of run-time. Global variables may optionally be initialized, may have
660an explicit section to be placed in, and may have an optional explicit alignment
661specified. A variable may be defined as "thread_local", which means that it
662will not be shared by threads (each thread will have a separated copy of the
663variable). A variable may be defined as a global "constant," which indicates
664that the contents of the variable will <b>never</b> be modified (enabling better
665optimization, allowing the global data to be placed in the read-only section of
666an executable, etc). Note that variables that need runtime initialization
667cannot be marked "constant" as there is a store to the variable.</p>
668
669<p>
670LLVM explicitly allows <em>declarations</em> of global variables to be marked
671constant, even if the final definition of the global is not. This capability
672can be used to enable slightly better optimization of the program, but requires
673the language definition to guarantee that optimizations based on the
674'constantness' are valid for the translation units that do not include the
675definition.
676</p>
677
678<p>As SSA values, global variables define pointer values that are in
679scope (i.e. they dominate) all basic blocks in the program. Global
680variables always define a pointer to their "content" type because they
681describe a region of memory, and all memory objects in LLVM are
682accessed through pointers.</p>
683
Christopher Lambdd0049d2007-12-11 09:31:00 +0000[diff] [blame]684<p>A global variable may be declared to reside in a target-specifc numbered
685address space. For targets that support them, address spaces may affect how
686optimizations are performed and/or what target instructions are used to access
Christopher Lamb20a39e92007-12-12 08:44:39 +0000[diff] [blame]687the variable. The default address space is zero. The address space qualifier
688must precede any other attributes.</p>
Christopher Lambdd0049d2007-12-11 09:31:00 +0000[diff] [blame]689
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]690<p>LLVM allows an explicit section to be specified for globals. If the target
691supports it, it will emit globals to the section specified.</p>
692
693<p>An explicit alignment may be specified for a global. If not present, or if
694the alignment is set to zero, the alignment of the global is set by the target
695to whatever it feels convenient. If an explicit alignment is specified, the
696global is forced to have at least that much alignment. All alignments must be
697a power of 2.</p>
698
Christopher Lambdd0049d2007-12-11 09:31:00 +0000[diff] [blame]699<p>For example, the following defines a global in a numbered address space with
700an initializer, section, and alignment:</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]701
702<div class="doc_code">
703<pre>
Christopher Lambdd0049d2007-12-11 09:31:00 +0000[diff] [blame]704@G = constant float 1.0 addrspace(5), section "foo", align 4
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]705</pre>
706</div>
707
708</div>
709
710
711<!-- ======================================================================= -->
712<div class="doc_subsection">
713 <a name="functionstructure">Functions</a>
714</div>
715
716<div class="doc_text">
717
718<p>LLVM function definitions consist of the "<tt>define</tt>" keyord,
719an optional <a href="#linkage">linkage type</a>, an optional
720<a href="#visibility">visibility style</a>, an optional
721<a href="#callingconv">calling convention</a>, a return type, an optional
722<a href="#paramattrs">parameter attribute</a> for the return type, a function
723name, a (possibly empty) argument list (each with optional
724<a href="#paramattrs">parameter attributes</a>), an optional section, an
Gordon Henriksend606f5b2007-12-10 03:30:21 +0000[diff] [blame]725optional alignment, an optional <a href="#gc">garbage collector name</a>, an
Gordon Henriksen13fe5e32007-12-10 03:18:06 +0000[diff] [blame]726opening curly brace, a list of basic blocks, and a closing curly brace.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]727
728LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
729optional <a href="#linkage">linkage type</a>, an optional
730<a href="#visibility">visibility style</a>, an optional
731<a href="#callingconv">calling convention</a>, a return type, an optional
732<a href="#paramattrs">parameter attribute</a> for the return type, a function
Gordon Henriksen13fe5e32007-12-10 03:18:06 +0000[diff] [blame]733name, a possibly empty list of arguments, an optional alignment, and an optional
Gordon Henriksend606f5b2007-12-10 03:30:21 +0000[diff] [blame]734<a href="#gc">garbage collector name</a>.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]735
736<p>A function definition contains a list of basic blocks, forming the CFG for
737the function. Each basic block may optionally start with a label (giving the
738basic block a symbol table entry), contains a list of instructions, and ends
739with a <a href="#terminators">terminator</a> instruction (such as a branch or
740function return).</p>
741
742<p>The first basic block in a function is special in two ways: it is immediately
743executed on entrance to the function, and it is not allowed to have predecessor
744basic blocks (i.e. there can not be any branches to the entry block of a
745function). Because the block can have no predecessors, it also cannot have any
746<a href="#i_phi">PHI nodes</a>.</p>
747
748<p>LLVM allows an explicit section to be specified for functions. If the target
749supports it, it will emit functions to the section specified.</p>
750
751<p>An explicit alignment may be specified for a function. If not present, or if
752the alignment is set to zero, the alignment of the function is set by the target
753to whatever it feels convenient. If an explicit alignment is specified, the
754function is forced to have at least that much alignment. All alignments must be
755a power of 2.</p>
756
757</div>
758
759
760<!-- ======================================================================= -->
761<div class="doc_subsection">
762 <a name="aliasstructure">Aliases</a>
763</div>
764<div class="doc_text">
765 <p>Aliases act as "second name" for the aliasee value (which can be either
Anton Korobeynikov96822812008-03-22 08:36:14 +0000[diff] [blame]766 function, global variable, another alias or bitcast of global value). Aliases
767 may have an optional <a href="#linkage">linkage type</a>, and an
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]768 optional <a href="#visibility">visibility style</a>.</p>
769
770 <h5>Syntax:</h5>
771
772<div class="doc_code">
773<pre>
774@&lt;Name&gt; = [Linkage] [Visibility] alias &lt;AliaseeTy&gt; @&lt;Aliasee&gt;
775</pre>
776</div>
777
778</div>
779
780
781
782<!-- ======================================================================= -->
783<div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
784<div class="doc_text">
785 <p>The return type and each parameter of a function type may have a set of
786 <i>parameter attributes</i> associated with them. Parameter attributes are
787 used to communicate additional information about the result or parameters of
Duncan Sandsf5588dc2007-11-27 13:23:08 +0000[diff] [blame]788 a function. Parameter attributes are considered to be part of the function,
789 not of the function type, so functions with different parameter attributes
790 can have the same function type.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]791
792 <p>Parameter attributes are simple keywords that follow the type specified. If
793 multiple parameter attributes are needed, they are space separated. For
794 example:</p>
795
796<div class="doc_code">
797<pre>
Duncan Sandsf5588dc2007-11-27 13:23:08 +0000[diff] [blame]798declare i32 @printf(i8* noalias , ...) nounwind
799declare i32 @atoi(i8*) nounwind readonly
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]800</pre>
801</div>
802
Duncan Sandsf5588dc2007-11-27 13:23:08 +0000[diff] [blame]803 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
804 <tt>readonly</tt>) come immediately after the argument list.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]805
806 <p>Currently, only the following parameter attributes are defined:</p>
807 <dl>
Reid Spencerf234bed2007-07-19 23:13:04 +0000[diff] [blame]808 <dt><tt>zeroext</tt></dt>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]809 <dd>This indicates that the parameter should be zero extended just before
810 a call to this function.</dd>
Chris Lattner275e6be2008-01-11 06:20:47 +0000[diff] [blame]811
Reid Spencerf234bed2007-07-19 23:13:04 +0000[diff] [blame]812 <dt><tt>signext</tt></dt>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]813 <dd>This indicates that the parameter should be sign extended just before
814 a call to this function.</dd>
Chris Lattner275e6be2008-01-11 06:20:47 +0000[diff] [blame]815
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]816 <dt><tt>inreg</tt></dt>
817 <dd>This indicates that the parameter should be placed in register (if
818 possible) during assembling function call. Support for this attribute is
819 target-specific</dd>
Chris Lattner275e6be2008-01-11 06:20:47 +0000[diff] [blame]820
821 <dt><tt>byval</tt></dt>
Chris Lattner04c86182008-01-15 04:34:22 +0000[diff] [blame]822 <dd>This indicates that the pointer parameter should really be passed by
823 value to the function. The attribute implies that a hidden copy of the
824 pointee is made between the caller and the callee, so the callee is unable
825 to modify the value in the callee. This attribute is only valid on llvm
826 pointer arguments. It is generally used to pass structs and arrays by
827 value, but is also valid on scalars (even though this is silly).</dd>
Chris Lattner275e6be2008-01-11 06:20:47 +0000[diff] [blame]828
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]829 <dt><tt>sret</tt></dt>
Duncan Sands616cc032008-02-18 04:19:38 +0000[diff] [blame]830 <dd>This indicates that the pointer parameter specifies the address of a
831 structure that is the return value of the function in the source program.
Duncan Sandsef0e9e42008-03-17 12:17:41 +0000[diff] [blame]832 Loads and stores to the structure are assumed not to trap.
Duncan Sands616cc032008-02-18 04:19:38 +0000[diff] [blame]833 May only be applied to the first parameter.</dd>
Chris Lattner275e6be2008-01-11 06:20:47 +0000[diff] [blame]834
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]835 <dt><tt>noalias</tt></dt>
Owen Andersonc4fc4cd2008-02-18 04:09:01 +0000[diff] [blame]836 <dd>This indicates that the parameter does not alias any global or any other
837 parameter. The caller is responsible for ensuring that this is the case,
838 usually by placing the value in a stack allocation.</dd>
Chris Lattner275e6be2008-01-11 06:20:47 +0000[diff] [blame]839
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]840 <dt><tt>noreturn</tt></dt>
841 <dd>This function attribute indicates that the function never returns. This
842 indicates to LLVM that every call to this function should be treated as if
843 an <tt>unreachable</tt> instruction immediately followed the call.</dd>
Chris Lattner275e6be2008-01-11 06:20:47 +0000[diff] [blame]844
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]845 <dt><tt>nounwind</tt></dt>
Duncan Sandsef0e9e42008-03-17 12:17:41 +0000[diff] [blame]846 <dd>This function attribute indicates that no exceptions unwind out of the
847 function. Usually this is because the function makes no use of exceptions,
848 but it may also be that the function catches any exceptions thrown when
849 executing it.</dd>
850
Duncan Sands4ee46812007-07-27 19:57:41 +0000[diff] [blame]851 <dt><tt>nest</tt></dt>
852 <dd>This indicates that the parameter can be excised using the
853 <a href="#int_trampoline">trampoline intrinsics</a>.</dd>
Duncan Sands13e13f82007-11-22 20:23:04 +0000[diff] [blame]854 <dt><tt>readonly</tt></dt>
Duncan Sandsd69c0e62007-11-14 21:14:02 +0000[diff] [blame]855 <dd>This function attribute indicates that the function has no side-effects
Duncan Sands13e13f82007-11-22 20:23:04 +0000[diff] [blame]856 except for producing a return value or throwing an exception. The value
857 returned must only depend on the function arguments and/or global variables.
858 It may use values obtained by dereferencing pointers.</dd>
859 <dt><tt>readnone</tt></dt>
860 <dd>A <tt>readnone</tt> function has the same restrictions as a <tt>readonly</tt>
Duncan Sandsd69c0e62007-11-14 21:14:02 +0000[diff] [blame]861 function, but in addition it is not allowed to dereference any pointer arguments
862 or global variables.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]863 </dl>
864
865</div>
866
867<!-- ======================================================================= -->
868<div class="doc_subsection">
Gordon Henriksen13fe5e32007-12-10 03:18:06 +0000[diff] [blame]869 <a name="gc">Garbage Collector Names</a>
870</div>
871
872<div class="doc_text">
873<p>Each function may specify a garbage collector name, which is simply a
874string.</p>
875
876<div class="doc_code"><pre
877>define void @f() gc "name" { ...</pre></div>
878
879<p>The compiler declares the supported values of <i>name</i>. Specifying a
880collector which will cause the compiler to alter its output in order to support
881the named garbage collection algorithm.</p>
882</div>
883
884<!-- ======================================================================= -->
885<div class="doc_subsection">
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]886 <a name="moduleasm">Module-Level Inline Assembly</a>
887</div>
888
889<div class="doc_text">
890<p>
891Modules may contain "module-level inline asm" blocks, which corresponds to the
892GCC "file scope inline asm" blocks. These blocks are internally concatenated by
893LLVM and treated as a single unit, but may be separated in the .ll file if
894desired. The syntax is very simple:
895</p>
896
897<div class="doc_code">
898<pre>
899module asm "inline asm code goes here"
900module asm "more can go here"
901</pre>
902</div>
903
904<p>The strings can contain any character by escaping non-printable characters.
905 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
906 for the number.
907</p>
908
909<p>
910 The inline asm code is simply printed to the machine code .s file when
911 assembly code is generated.
912</p>
913</div>
914
915<!-- ======================================================================= -->
916<div class="doc_subsection">
917 <a name="datalayout">Data Layout</a>
918</div>
919
920<div class="doc_text">
921<p>A module may specify a target specific data layout string that specifies how
922data is to be laid out in memory. The syntax for the data layout is simply:</p>
923<pre> target datalayout = "<i>layout specification</i>"</pre>
924<p>The <i>layout specification</i> consists of a list of specifications
925separated by the minus sign character ('-'). Each specification starts with a
926letter and may include other information after the letter to define some
927aspect of the data layout. The specifications accepted are as follows: </p>
928<dl>
929 <dt><tt>E</tt></dt>
930 <dd>Specifies that the target lays out data in big-endian form. That is, the
931 bits with the most significance have the lowest address location.</dd>
932 <dt><tt>e</tt></dt>
933 <dd>Specifies that hte target lays out data in little-endian form. That is,
934 the bits with the least significance have the lowest address location.</dd>
935 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
936 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
937 <i>preferred</i> alignments. All sizes are in bits. Specifying the <i>pref</i>
938 alignment is optional. If omitted, the preceding <tt>:</tt> should be omitted
939 too.</dd>
940 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
941 <dd>This specifies the alignment for an integer type of a given bit
942 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
943 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
944 <dd>This specifies the alignment for a vector type of a given bit
945 <i>size</i>.</dd>
946 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
947 <dd>This specifies the alignment for a floating point type of a given bit
948 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
949 (double).</dd>
950 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
951 <dd>This specifies the alignment for an aggregate type of a given bit
952 <i>size</i>.</dd>
953</dl>
954<p>When constructing the data layout for a given target, LLVM starts with a
955default set of specifications which are then (possibly) overriden by the
956specifications in the <tt>datalayout</tt> keyword. The default specifications
957are given in this list:</p>
958<ul>
959 <li><tt>E</tt> - big endian</li>
960 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
961 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
962 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
963 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
964 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
965 <li><tt>i64:32:64</tt> - i64 has abi alignment of 32-bits but preferred
966 alignment of 64-bits</li>
967 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
968 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
969 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
970 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
971 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
972</ul>
973<p>When llvm is determining the alignment for a given type, it uses the
974following rules:
975<ol>
976 <li>If the type sought is an exact match for one of the specifications, that
977 specification is used.</li>
978 <li>If no match is found, and the type sought is an integer type, then the
979 smallest integer type that is larger than the bitwidth of the sought type is
980 used. If none of the specifications are larger than the bitwidth then the the
981 largest integer type is used. For example, given the default specifications
982 above, the i7 type will use the alignment of i8 (next largest) while both
983 i65 and i256 will use the alignment of i64 (largest specified).</li>
984 <li>If no match is found, and the type sought is a vector type, then the
985 largest vector type that is smaller than the sought vector type will be used
986 as a fall back. This happens because <128 x double> can be implemented in
987 terms of 64 <2 x double>, for example.</li>
988</ol>
989</div>
990
991<!-- *********************************************************************** -->
992<div class="doc_section"> <a name="typesystem">Type System</a> </div>
993<!-- *********************************************************************** -->
994
995<div class="doc_text">
996
997<p>The LLVM type system is one of the most important features of the
998intermediate representation. Being typed enables a number of
999optimizations to be performed on the IR directly, without having to do
1000extra analyses on the side before the transformation. A strong type
1001system makes it easier to read the generated code and enables novel
1002analyses and transformations that are not feasible to perform on normal
1003three address code representations.</p>
1004
1005</div>
1006
1007<!-- ======================================================================= -->
Chris Lattner488772f2008-01-04 04:32:38 +0000[diff] [blame]1008<div class="doc_subsection"> <a name="t_classifications">Type
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1009Classifications</a> </div>
1010<div class="doc_text">
Chris Lattner488772f2008-01-04 04:32:38 +0000[diff] [blame]1011<p>The types fall into a few useful
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1012classifications:</p>
1013
1014<table border="1" cellspacing="0" cellpadding="4">
1015 <tbody>
1016 <tr><th>Classification</th><th>Types</th></tr>
1017 <tr>
Chris Lattner488772f2008-01-04 04:32:38 +0000[diff] [blame]1018 <td><a href="#t_integer">integer</a></td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1019 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1020 </tr>
1021 <tr>
Chris Lattner488772f2008-01-04 04:32:38 +0000[diff] [blame]1022 <td><a href="#t_floating">floating point</a></td>
1023 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1024 </tr>
1025 <tr>
1026 <td><a name="t_firstclass">first class</a></td>
Chris Lattner488772f2008-01-04 04:32:38 +0000[diff] [blame]1027 <td><a href="#t_integer">integer</a>,
1028 <a href="#t_floating">floating point</a>,
1029 <a href="#t_pointer">pointer</a>,
1030 <a href="#t_vector">vector</a>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1031 </td>
1032 </tr>
Chris Lattner488772f2008-01-04 04:32:38 +0000[diff] [blame]1033 <tr>
1034 <td><a href="#t_primitive">primitive</a></td>
1035 <td><a href="#t_label">label</a>,
1036 <a href="#t_void">void</a>,
1037 <a href="#t_integer">integer</a>,
1038 <a href="#t_floating">floating point</a>.</td>
1039 </tr>
1040 <tr>
1041 <td><a href="#t_derived">derived</a></td>
1042 <td><a href="#t_integer">integer</a>,
1043 <a href="#t_array">array</a>,
1044 <a href="#t_function">function</a>,
1045 <a href="#t_pointer">pointer</a>,
1046 <a href="#t_struct">structure</a>,
1047 <a href="#t_pstruct">packed structure</a>,
1048 <a href="#t_vector">vector</a>,
1049 <a href="#t_opaque">opaque</a>.
1050 </tr>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1051 </tbody>
1052</table>
1053
1054<p>The <a href="#t_firstclass">first class</a> types are perhaps the
1055most important. Values of these types are the only ones which can be
1056produced by instructions, passed as arguments, or used as operands to
1057instructions. This means that all structures and arrays must be
1058manipulated either by pointer or by component.</p>
1059</div>
1060
1061<!-- ======================================================================= -->
Chris Lattner488772f2008-01-04 04:32:38 +0000[diff] [blame]1062<div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
Chris Lattner86437612008-01-04 04:34:14 +0000[diff] [blame]1063
Chris Lattner488772f2008-01-04 04:32:38 +0000[diff] [blame]1064<div class="doc_text">
1065<p>The primitive types are the fundamental building blocks of the LLVM
1066system.</p>
1067
Chris Lattner86437612008-01-04 04:34:14 +0000[diff] [blame]1068</div>
1069
Chris Lattner488772f2008-01-04 04:32:38 +0000[diff] [blame]1070<!-- _______________________________________________________________________ -->
1071<div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1072
1073<div class="doc_text">
1074 <table>
1075 <tbody>
1076 <tr><th>Type</th><th>Description</th></tr>
1077 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1078 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1079 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1080 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1081 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1082 </tbody>
1083 </table>
1084</div>
1085
1086<!-- _______________________________________________________________________ -->
1087<div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1088
1089<div class="doc_text">
1090<h5>Overview:</h5>
1091<p>The void type does not represent any value and has no size.</p>
1092
1093<h5>Syntax:</h5>
1094
1095<pre>
1096 void
1097</pre>
1098</div>
1099
1100<!-- _______________________________________________________________________ -->
1101<div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1102
1103<div class="doc_text">
1104<h5>Overview:</h5>
1105<p>The label type represents code labels.</p>
1106
1107<h5>Syntax:</h5>
1108
1109<pre>
1110 label
1111</pre>
1112</div>
1113
1114
1115<!-- ======================================================================= -->
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1116<div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1117
1118<div class="doc_text">
1119
1120<p>The real power in LLVM comes from the derived types in the system.
1121This is what allows a programmer to represent arrays, functions,
1122pointers, and other useful types. Note that these derived types may be
1123recursive: For example, it is possible to have a two dimensional array.</p>
1124
1125</div>
1126
1127<!-- _______________________________________________________________________ -->
1128<div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1129
1130<div class="doc_text">
1131
1132<h5>Overview:</h5>
1133<p>The integer type is a very simple derived type that simply specifies an
1134arbitrary bit width for the integer type desired. Any bit width from 1 bit to
11352^23-1 (about 8 million) can be specified.</p>
1136
1137<h5>Syntax:</h5>
1138
1139<pre>
1140 iN
1141</pre>
1142
1143<p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1144value.</p>
1145
1146<h5>Examples:</h5>
1147<table class="layout">
Chris Lattner251ab812007-12-18 06:18:21 +0000[diff] [blame]1148 <tbody>
1149 <tr>
1150 <td><tt>i1</tt></td>
1151 <td>a single-bit integer.</td>
1152 </tr><tr>
1153 <td><tt>i32</tt></td>
1154 <td>a 32-bit integer.</td>
1155 </tr><tr>
1156 <td><tt>i1942652</tt></td>
1157 <td>a really big integer of over 1 million bits.</td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1158 </tr>
Chris Lattner251ab812007-12-18 06:18:21 +0000[diff] [blame]1159 </tbody>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1160</table>
1161</div>
1162
1163<!-- _______________________________________________________________________ -->
1164<div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1165
1166<div class="doc_text">
1167
1168<h5>Overview:</h5>
1169
1170<p>The array type is a very simple derived type that arranges elements
1171sequentially in memory. The array type requires a size (number of
1172elements) and an underlying data type.</p>
1173
1174<h5>Syntax:</h5>
1175
1176<pre>
1177 [&lt;# elements&gt; x &lt;elementtype&gt;]
1178</pre>
1179
1180<p>The number of elements is a constant integer value; elementtype may
1181be any type with a size.</p>
1182
1183<h5>Examples:</h5>
1184<table class="layout">
1185 <tr class="layout">
Chris Lattner7311d222007-12-19 05:04:11 +0000[diff] [blame]1186 <td class="left"><tt>[40 x i32]</tt></td>
1187 <td class="left">Array of 40 32-bit integer values.</td>
1188 </tr>
1189 <tr class="layout">
1190 <td class="left"><tt>[41 x i32]</tt></td>
1191 <td class="left">Array of 41 32-bit integer values.</td>
1192 </tr>
1193 <tr class="layout">
1194 <td class="left"><tt>[4 x i8]</tt></td>
1195 <td class="left">Array of 4 8-bit integer values.</td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1196 </tr>
1197</table>
1198<p>Here are some examples of multidimensional arrays:</p>
1199<table class="layout">
1200 <tr class="layout">
Chris Lattner7311d222007-12-19 05:04:11 +0000[diff] [blame]1201 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1202 <td class="left">3x4 array of 32-bit integer values.</td>
1203 </tr>
1204 <tr class="layout">
1205 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1206 <td class="left">12x10 array of single precision floating point values.</td>
1207 </tr>
1208 <tr class="layout">
1209 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1210 <td class="left">2x3x4 array of 16-bit integer values.</td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1211 </tr>
1212</table>
1213
1214<p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1215length array. Normally, accesses past the end of an array are undefined in
1216LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
1217As a special case, however, zero length arrays are recognized to be variable
1218length. This allows implementation of 'pascal style arrays' with the LLVM
1219type "{ i32, [0 x float]}", for example.</p>
1220
1221</div>
1222
1223<!-- _______________________________________________________________________ -->
1224<div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1225<div class="doc_text">
1226<h5>Overview:</h5>
1227<p>The function type can be thought of as a function signature. It
Devang Pateld4ba41d2008-03-24 05:35:41 +0000[diff] [blame]1228consists of a return type and a list of formal parameter types. The
Devang Pateld5404c02008-03-24 20:52:42 +0000[diff] [blame]1229return type of a function type is a scalar type or a void type or a struct type.
1230If the return type is a struct type then all struct elements must be of first
Devang Patel6dddba22008-03-24 18:10:52 +0000[diff] [blame]1231class types. Function types are usually used to build virtual function tables
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1232(which are structures of pointers to functions), for indirect function
1233calls, and when defining a function.</p>
Devang Patela3cc5372008-03-10 20:49:15 +0000[diff] [blame]1234
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1235<h5>Syntax:</h5>
Devang Patela3cc5372008-03-10 20:49:15 +0000[diff] [blame]1236<pre> &lt;returntype list&gt; (&lt;parameter list&gt;)<br></pre>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1237<p>...where '<tt>&lt;parameter list&gt;</tt>' is a comma-separated list of type
1238specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1239which indicates that the function takes a variable number of arguments.
1240Variable argument functions can access their arguments with the <a
Devang Patela3cc5372008-03-10 20:49:15 +0000[diff] [blame]1241 href="#int_varargs">variable argument handling intrinsic</a> functions.
1242'<tt>&lt;returntype list&gt;</tt>' is a comma-separated list of
1243<a href="#t_firstclass">first class</a> type specifiers.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1244<h5>Examples:</h5>
1245<table class="layout">
1246 <tr class="layout">
1247 <td class="left"><tt>i32 (i32)</tt></td>
1248 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1249 </td>
1250 </tr><tr class="layout">
Reid Spencerf234bed2007-07-19 23:13:04 +0000[diff] [blame]1251 <td class="left"><tt>float&nbsp;(i16&nbsp;signext,&nbsp;i32&nbsp;*)&nbsp;*
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1252 </tt></td>
1253 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1254 an <tt>i16</tt> that should be sign extended and a
1255 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1256 <tt>float</tt>.
1257 </td>
1258 </tr><tr class="layout">
1259 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1260 <td class="left">A vararg function that takes at least one
1261 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1262 which returns an integer. This is the signature for <tt>printf</tt> in
1263 LLVM.
1264 </td>
Devang Pateld4ba41d2008-03-24 05:35:41 +0000[diff] [blame]1265 </tr><tr class="layout">
1266 <td class="left"><tt>{i32, i32} (i32)</tt></td>
Devang Patel6dddba22008-03-24 18:10:52 +0000[diff] [blame]1267 <td class="left">A function taking an <tt>i32></tt>, returning two
1268 <tt> i32 </tt> values as an aggregate of type <tt>{ i32, i32 }</tt>
Devang Pateld4ba41d2008-03-24 05:35:41 +0000[diff] [blame]1269 </td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1270 </tr>
1271</table>
1272
1273</div>
1274<!-- _______________________________________________________________________ -->
1275<div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1276<div class="doc_text">
1277<h5>Overview:</h5>
1278<p>The structure type is used to represent a collection of data members
1279together in memory. The packing of the field types is defined to match
1280the ABI of the underlying processor. The elements of a structure may
1281be any type that has a size.</p>
1282<p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1283and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1284field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1285instruction.</p>
1286<h5>Syntax:</h5>
1287<pre> { &lt;type list&gt; }<br></pre>
1288<h5>Examples:</h5>
1289<table class="layout">
1290 <tr class="layout">
1291 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1292 <td class="left">A triple of three <tt>i32</tt> values</td>
1293 </tr><tr class="layout">
1294 <td class="left"><tt>{&nbsp;float,&nbsp;i32&nbsp;(i32)&nbsp;*&nbsp;}</tt></td>
1295 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1296 second element is a <a href="#t_pointer">pointer</a> to a
1297 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1298 an <tt>i32</tt>.</td>
1299 </tr>
1300</table>
1301</div>
1302
1303<!-- _______________________________________________________________________ -->
1304<div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1305</div>
1306<div class="doc_text">
1307<h5>Overview:</h5>
1308<p>The packed structure type is used to represent a collection of data members
1309together in memory. There is no padding between fields. Further, the alignment
1310of a packed structure is 1 byte. The elements of a packed structure may
1311be any type that has a size.</p>
1312<p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1313and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1314field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1315instruction.</p>
1316<h5>Syntax:</h5>
1317<pre> &lt; { &lt;type list&gt; } &gt; <br></pre>
1318<h5>Examples:</h5>
1319<table class="layout">
1320 <tr class="layout">
1321 <td class="left"><tt>&lt; { i32, i32, i32 } &gt;</tt></td>
1322 <td class="left">A triple of three <tt>i32</tt> values</td>
1323 </tr><tr class="layout">
Chris Lattner7311d222007-12-19 05:04:11 +0000[diff] [blame]1324 <td class="left"><tt>&lt; { float, i32 (i32)* } &gt;</tt></td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1325 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1326 second element is a <a href="#t_pointer">pointer</a> to a
1327 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1328 an <tt>i32</tt>.</td>
1329 </tr>
1330</table>
1331</div>
1332
1333<!-- _______________________________________________________________________ -->
1334<div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1335<div class="doc_text">
1336<h5>Overview:</h5>
1337<p>As in many languages, the pointer type represents a pointer or
Christopher Lambdd0049d2007-12-11 09:31:00 +0000[diff] [blame]1338reference to another object, which must live in memory. Pointer types may have
1339an optional address space attribute defining the target-specific numbered
1340address space where the pointed-to object resides. The default address space is
1341zero.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1342<h5>Syntax:</h5>
1343<pre> &lt;type&gt; *<br></pre>
1344<h5>Examples:</h5>
1345<table class="layout">
1346 <tr class="layout">
Chris Lattner7311d222007-12-19 05:04:11 +0000[diff] [blame]1347 <td class="left"><tt>[4x i32]*</tt></td>
1348 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1349 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1350 </tr>
1351 <tr class="layout">
1352 <td class="left"><tt>i32 (i32 *) *</tt></td>
1353 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1354 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
Chris Lattner7311d222007-12-19 05:04:11 +0000[diff] [blame]1355 <tt>i32</tt>.</td>
1356 </tr>
1357 <tr class="layout">
1358 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1359 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1360 that resides in address space #5.</td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1361 </tr>
1362</table>
1363</div>
1364
1365<!-- _______________________________________________________________________ -->
1366<div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1367<div class="doc_text">
1368
1369<h5>Overview:</h5>
1370
1371<p>A vector type is a simple derived type that represents a vector
1372of elements. Vector types are used when multiple primitive data
1373are operated in parallel using a single instruction (SIMD).
1374A vector type requires a size (number of
1375elements) and an underlying primitive data type. Vectors must have a power
1376of two length (1, 2, 4, 8, 16 ...). Vector types are
1377considered <a href="#t_firstclass">first class</a>.</p>
1378
1379<h5>Syntax:</h5>
1380
1381<pre>
1382 &lt; &lt;# elements&gt; x &lt;elementtype&gt; &gt;
1383</pre>
1384
1385<p>The number of elements is a constant integer value; elementtype may
1386be any integer or floating point type.</p>
1387
1388<h5>Examples:</h5>
1389
1390<table class="layout">
1391 <tr class="layout">
Chris Lattner7311d222007-12-19 05:04:11 +0000[diff] [blame]1392 <td class="left"><tt>&lt;4 x i32&gt;</tt></td>
1393 <td class="left">Vector of 4 32-bit integer values.</td>
1394 </tr>
1395 <tr class="layout">
1396 <td class="left"><tt>&lt;8 x float&gt;</tt></td>
1397 <td class="left">Vector of 8 32-bit floating-point values.</td>
1398 </tr>
1399 <tr class="layout">
1400 <td class="left"><tt>&lt;2 x i64&gt;</tt></td>
1401 <td class="left">Vector of 2 64-bit integer values.</td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1402 </tr>
1403</table>
1404</div>
1405
1406<!-- _______________________________________________________________________ -->
1407<div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1408<div class="doc_text">
1409
1410<h5>Overview:</h5>
1411
1412<p>Opaque types are used to represent unknown types in the system. This
Gordon Henriksenda0706e2007-10-14 00:34:53 +0000[diff] [blame]1413corresponds (for example) to the C notion of a forward declared structure type.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1414In LLVM, opaque types can eventually be resolved to any type (not just a
1415structure type).</p>
1416
1417<h5>Syntax:</h5>
1418
1419<pre>
1420 opaque
1421</pre>
1422
1423<h5>Examples:</h5>
1424
1425<table class="layout">
1426 <tr class="layout">
Chris Lattner7311d222007-12-19 05:04:11 +0000[diff] [blame]1427 <td class="left"><tt>opaque</tt></td>
1428 <td class="left">An opaque type.</td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1429 </tr>
1430</table>
1431</div>
1432
1433
1434<!-- *********************************************************************** -->
1435<div class="doc_section"> <a name="constants">Constants</a> </div>
1436<!-- *********************************************************************** -->
1437
1438<div class="doc_text">
1439
1440<p>LLVM has several different basic types of constants. This section describes
1441them all and their syntax.</p>
1442
1443</div>
1444
1445<!-- ======================================================================= -->
1446<div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1447
1448<div class="doc_text">
1449
1450<dl>
1451 <dt><b>Boolean constants</b></dt>
1452
1453 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1454 constants of the <tt><a href="#t_primitive">i1</a></tt> type.
1455 </dd>
1456
1457 <dt><b>Integer constants</b></dt>
1458
1459 <dd>Standard integers (such as '4') are constants of the <a
1460 href="#t_integer">integer</a> type. Negative numbers may be used with
1461 integer types.
1462 </dd>
1463
1464 <dt><b>Floating point constants</b></dt>
1465
1466 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1467 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
Chris Lattnerdc15b1d2008-04-01 18:45:27 +0000[diff] [blame]1468 notation (see below). The assembler requires the exact decimal value of
1469 a floating-point constant. For example, the assembler accepts 1.25 but
1470 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
1471 constants must have a <a href="#t_floating">floating point</a> type. </dd>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1472
1473 <dt><b>Null pointer constants</b></dt>
1474
1475 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1476 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1477
1478</dl>
1479
1480<p>The one non-intuitive notation for constants is the optional hexadecimal form
1481of floating point constants. For example, the form '<tt>double
14820x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
14834.5e+15</tt>'. The only time hexadecimal floating point constants are required
1484(and the only time that they are generated by the disassembler) is when a
1485floating point constant must be emitted but it cannot be represented as a
1486decimal floating point number. For example, NaN's, infinities, and other
1487special values are represented in their IEEE hexadecimal format so that
1488assembly and disassembly do not cause any bits to change in the constants.</p>
1489
1490</div>
1491
1492<!-- ======================================================================= -->
1493<div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1494</div>
1495
1496<div class="doc_text">
1497<p>Aggregate constants arise from aggregation of simple constants
1498and smaller aggregate constants.</p>
1499
1500<dl>
1501 <dt><b>Structure constants</b></dt>
1502
1503 <dd>Structure constants are represented with notation similar to structure
1504 type definitions (a comma separated list of elements, surrounded by braces
Chris Lattner6c8de962007-12-25 20:34:52 +0000[diff] [blame]1505 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
1506 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]1507 must have <a href="#t_struct">structure type</a>, and the number and
1508 types of elements must match those specified by the type.
1509 </dd>
1510
1511 <dt><b>Array constants</b></dt>
1512
1513 <dd>Array constants are represented with notation similar to array type
1514 definitions (a comma separated list of elements, surrounded by square brackets
1515 (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74 ]</tt>". Array
1516 constants must have <a href="#t_array">array type</a>, and the number and
1517 types of elements must match those specified by the type.
1518 </dd>
1519
1520 <dt><b>Vector constants</b></dt>
1521
1522 <dd>Vector constants are represented with notation similar to vector type
1523 definitions (a comma separated list of elements, surrounded by
1524 less-than/greater-than's (<tt>&lt;&gt;</tt>)). For example: "<tt>&lt; i32 42,
1525 i32 11, i32 74, i32 100 &gt;</tt>". Vector constants must have <a
1526 href="#t_vector">vector type</a>, and the number and types of elements must
1527 match those specified by the type.
1528 </dd>
1529
1530 <dt><b>Zero initialization</b></dt>
1531
1532 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1533 value to zero of <em>any</em> type, including scalar and aggregate types.
1534 This is often used to avoid having to print large zero initializers (e.g. for
1535 large arrays) and is always exactly equivalent to using explicit zero
1536 initializers.
1537 </dd>
1538</dl>
1539
1540</div>
1541
1542<!-- ======================================================================= -->
1543<div class="doc_subsection">
1544 <a name="globalconstants">Global Variable and Function Addresses</a>
1545</div>
1546
1547<div class="doc_text">
1548
1549<p>The addresses of <a href="#globalvars">global variables</a> and <a
1550href="#functionstructure">functions</a> are always implicitly valid (link-time)
1551constants. These constants are explicitly referenced when the <a
1552href="#identifiers">identifier for the global</a> is used and always have <a
1553href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1554file:</p>
1555
1556<div class="doc_code">
1557<pre>
1558@X = global i32 17
1559@Y = global i32 42
1560@Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1561</pre>
1562</div>
1563
1564</div>
1565
1566<!-- ======================================================================= -->
1567<div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1568<div class="doc_text">
1569 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1570 no specific value. Undefined values may be of any type and be used anywhere
1571 a constant is permitted.</p>
1572
1573 <p>Undefined values indicate to the compiler that the program is well defined
1574 no matter what value is used, giving the compiler more freedom to optimize.
1575 </p>
1576</div>
1577
1578<!-- ======================================================================= -->
1579<div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1580</div>
1581
1582<div class="doc_text">
1583
1584<p>Constant expressions are used to allow expressions involving other constants
1585to be used as constants. Constant expressions may be of any <a
1586href="#t_firstclass">first class</a> type and may involve any LLVM operation
1587that does not have side effects (e.g. load and call are not supported). The
1588following is the syntax for constant expressions:</p>
1589
1590<dl>
1591 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1592 <dd>Truncate a constant to another type. The bit size of CST must be larger
1593 than the bit size of TYPE. Both types must be integers.</dd>
1594
1595 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1596 <dd>Zero extend a constant to another type. The bit size of CST must be
1597 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1598
1599 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1600 <dd>Sign extend a constant to another type. The bit size of CST must be
1601 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1602
1603 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1604 <dd>Truncate a floating point constant to another floating point type. The
1605 size of CST must be larger than the size of TYPE. Both types must be
1606 floating point.</dd>
1607
1608 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1609 <dd>Floating point extend a constant to another type. The size of CST must be
1610 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1611
Reid Spencere6adee82007-07-31 14:40:14 +0000[diff] [blame]1612 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1613 <dd>Convert a floating point constant to the corresponding unsigned integer
Nate Begeman78246ca2007-11-17 03:58:34 +0000[diff] [blame]1614 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1615 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1616 of the same number of elements. If the value won't fit in the integer type,
1617 the results are undefined.</dd>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1618
1619 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
1620 <dd>Convert a floating point constant to the corresponding signed integer
Nate Begeman78246ca2007-11-17 03:58:34 +0000[diff] [blame]1621 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1622 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1623 of the same number of elements. If the value won't fit in the integer type,
1624 the results are undefined.</dd>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1625
1626 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
1627 <dd>Convert an unsigned integer constant to the corresponding floating point
Nate Begeman78246ca2007-11-17 03:58:34 +0000[diff] [blame]1628 constant. TYPE must be a scalar or vector floating point type. CST must be of
1629 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1630 of the same number of elements. If the value won't fit in the floating point
1631 type, the results are undefined.</dd>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1632
1633 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
1634 <dd>Convert a signed integer constant to the corresponding floating point
Nate Begeman78246ca2007-11-17 03:58:34 +0000[diff] [blame]1635 constant. TYPE must be a scalar or vector floating point type. CST must be of
1636 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1637 of the same number of elements. If the value won't fit in the floating point
1638 type, the results are undefined.</dd>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1639
1640 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
1641 <dd>Convert a pointer typed constant to the corresponding integer constant
1642 TYPE must be an integer type. CST must be of pointer type. The CST value is
1643 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1644
1645 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
1646 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1647 pointer type. CST must be of integer type. The CST value is zero extended,
1648 truncated, or unchanged to make it fit in a pointer size. This one is
1649 <i>really</i> dangerous!</dd>
1650
1651 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
1652 <dd>Convert a constant, CST, to another TYPE. The size of CST and TYPE must be
1653 identical (same number of bits). The conversion is done as if the CST value
1654 was stored to memory and read back as TYPE. In other words, no bits change
1655 with this operator, just the type. This can be used for conversion of
1656 vector types to any other type, as long as they have the same bit width. For
1657 pointers it is only valid to cast to another pointer type.
1658 </dd>
1659
1660 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1661
1662 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1663 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1664 instruction, the index list may have zero or more indexes, which are required
1665 to make sense for the type of "CSTPTR".</dd>
1666
1667 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1668
1669 <dd>Perform the <a href="#i_select">select operation</a> on
1670 constants.</dd>
1671
1672 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
1673 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
1674
1675 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
1676 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
1677
1678 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1679
1680 <dd>Perform the <a href="#i_extractelement">extractelement
1681 operation</a> on constants.
1682
1683 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1684
1685 <dd>Perform the <a href="#i_insertelement">insertelement
1686 operation</a> on constants.</dd>
1687
1688
1689 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
1690
1691 <dd>Perform the <a href="#i_shufflevector">shufflevector
1692 operation</a> on constants.</dd>
1693
1694 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1695
1696 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1697 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1698 binary</a> operations. The constraints on operands are the same as those for
1699 the corresponding instruction (e.g. no bitwise operations on floating point
1700 values are allowed).</dd>
1701</dl>
1702</div>
1703
1704<!-- *********************************************************************** -->
1705<div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1706<!-- *********************************************************************** -->
1707
1708<!-- ======================================================================= -->
1709<div class="doc_subsection">
1710<a name="inlineasm">Inline Assembler Expressions</a>
1711</div>
1712
1713<div class="doc_text">
1714
1715<p>
1716LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
1717Module-Level Inline Assembly</a>) through the use of a special value. This
1718value represents the inline assembler as a string (containing the instructions
1719to emit), a list of operand constraints (stored as a string), and a flag that
1720indicates whether or not the inline asm expression has side effects. An example
1721inline assembler expression is:
1722</p>
1723
1724<div class="doc_code">
1725<pre>
1726i32 (i32) asm "bswap $0", "=r,r"
1727</pre>
1728</div>
1729
1730<p>
1731Inline assembler expressions may <b>only</b> be used as the callee operand of
1732a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
1733</p>
1734
1735<div class="doc_code">
1736<pre>
1737%X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
1738</pre>
1739</div>
1740
1741<p>
1742Inline asms with side effects not visible in the constraint list must be marked
1743as having side effects. This is done through the use of the
1744'<tt>sideeffect</tt>' keyword, like so:
1745</p>
1746
1747<div class="doc_code">
1748<pre>
1749call void asm sideeffect "eieio", ""()
1750</pre>
1751</div>
1752
1753<p>TODO: The format of the asm and constraints string still need to be
1754documented here. Constraints on what can be done (e.g. duplication, moving, etc
1755need to be documented).
1756</p>
1757
1758</div>
1759
1760<!-- *********************************************************************** -->
1761<div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1762<!-- *********************************************************************** -->
1763
1764<div class="doc_text">
1765
1766<p>The LLVM instruction set consists of several different
1767classifications of instructions: <a href="#terminators">terminator
1768instructions</a>, <a href="#binaryops">binary instructions</a>,
1769<a href="#bitwiseops">bitwise binary instructions</a>, <a
1770 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1771instructions</a>.</p>
1772
1773</div>
1774
1775<!-- ======================================================================= -->
1776<div class="doc_subsection"> <a name="terminators">Terminator
1777Instructions</a> </div>
1778
1779<div class="doc_text">
1780
1781<p>As mentioned <a href="#functionstructure">previously</a>, every
1782basic block in a program ends with a "Terminator" instruction, which
1783indicates which block should be executed after the current block is
1784finished. These terminator instructions typically yield a '<tt>void</tt>'
1785value: they produce control flow, not values (the one exception being
1786the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1787<p>There are six different terminator instructions: the '<a
1788 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1789instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1790the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1791 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1792 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1793
1794</div>
1795
1796<!-- _______________________________________________________________________ -->
1797<div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1798Instruction</a> </div>
1799<div class="doc_text">
1800<h5>Syntax:</h5>
1801<pre> ret &lt;type&gt; &lt;value&gt; <i>; Return a value from a non-void function</i>
1802 ret void <i>; Return from void function</i>
Devang Patela3cc5372008-03-10 20:49:15 +0000[diff] [blame]1803 ret &lt;type&gt; &lt;value&gt;, &lt;type&gt; &lt;value&gt; <i>; Return two values from a non-void function </i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1804</pre>
1805<h5>Overview:</h5>
1806<p>The '<tt>ret</tt>' instruction is used to return control flow (and a
1807value) from a function back to the caller.</p>
1808<p>There are two forms of the '<tt>ret</tt>' instruction: one that
1809returns a value and then causes control flow, and one that just causes
1810control flow to occur.</p>
1811<h5>Arguments:</h5>
Devang Patela3cc5372008-03-10 20:49:15 +0000[diff] [blame]1812<p>The '<tt>ret</tt>' instruction may return one or multiple values. The
Devang Patelec8a5b02008-03-11 05:51:59 +0000[diff] [blame]1813type of each return value must be a '<a href="#t_firstclass">first class</a>'
1814 type. Note that a function is not <a href="#wellformed">well formed</a>
Devang Patela3cc5372008-03-10 20:49:15 +0000[diff] [blame]1815if there exists a '<tt>ret</tt>' instruction inside of the function that
Devang Patelec8a5b02008-03-11 05:51:59 +0000[diff] [blame]1816returns values that do not match the return type of the function.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1817<h5>Semantics:</h5>
1818<p>When the '<tt>ret</tt>' instruction is executed, control flow
1819returns back to the calling function's context. If the caller is a "<a
1820 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1821the instruction after the call. If the caller was an "<a
1822 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1823at the beginning of the "normal" destination block. If the instruction
1824returns a value, that value shall set the call or invoke instruction's
Devang Patela3cc5372008-03-10 20:49:15 +0000[diff] [blame]1825return value. If the instruction returns multiple values then these
Devang Patelec8a5b02008-03-11 05:51:59 +0000[diff] [blame]1826values can only be accessed through a '<a href="#i_getresult"><tt>getresult</tt>
1827</a>' instruction.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1828<h5>Example:</h5>
1829<pre> ret i32 5 <i>; Return an integer value of 5</i>
1830 ret void <i>; Return from a void function</i>
Devang Patela3cc5372008-03-10 20:49:15 +0000[diff] [blame]1831 ret i32 4, i8 2 <i>; Return two values 4 and 2 </i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1832</pre>
1833</div>
1834<!-- _______________________________________________________________________ -->
1835<div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1836<div class="doc_text">
1837<h5>Syntax:</h5>
1838<pre> br i1 &lt;cond&gt;, label &lt;iftrue&gt;, label &lt;iffalse&gt;<br> br label &lt;dest&gt; <i>; Unconditional branch</i>
1839</pre>
1840<h5>Overview:</h5>
1841<p>The '<tt>br</tt>' instruction is used to cause control flow to
1842transfer to a different basic block in the current function. There are
1843two forms of this instruction, corresponding to a conditional branch
1844and an unconditional branch.</p>
1845<h5>Arguments:</h5>
1846<p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1847single '<tt>i1</tt>' value and two '<tt>label</tt>' values. The
1848unconditional form of the '<tt>br</tt>' instruction takes a single
1849'<tt>label</tt>' value as a target.</p>
1850<h5>Semantics:</h5>
1851<p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
1852argument is evaluated. If the value is <tt>true</tt>, control flows
1853to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1854control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1855<h5>Example:</h5>
1856<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
1857 href="#i_ret">ret</a> i32 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> i32 0<br></pre>
1858</div>
1859<!-- _______________________________________________________________________ -->
1860<div class="doc_subsubsection">
1861 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
1862</div>
1863
1864<div class="doc_text">
1865<h5>Syntax:</h5>
1866
1867<pre>
1868 switch &lt;intty&gt; &lt;value&gt;, label &lt;defaultdest&gt; [ &lt;intty&gt; &lt;val&gt;, label &lt;dest&gt; ... ]
1869</pre>
1870
1871<h5>Overview:</h5>
1872
1873<p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
1874several different places. It is a generalization of the '<tt>br</tt>'
1875instruction, allowing a branch to occur to one of many possible
1876destinations.</p>
1877
1878
1879<h5>Arguments:</h5>
1880
1881<p>The '<tt>switch</tt>' instruction uses three parameters: an integer
1882comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
1883an array of pairs of comparison value constants and '<tt>label</tt>'s. The
1884table is not allowed to contain duplicate constant entries.</p>
1885
1886<h5>Semantics:</h5>
1887
1888<p>The <tt>switch</tt> instruction specifies a table of values and
1889destinations. When the '<tt>switch</tt>' instruction is executed, this
1890table is searched for the given value. If the value is found, control flow is
1891transfered to the corresponding destination; otherwise, control flow is
1892transfered to the default destination.</p>
1893
1894<h5>Implementation:</h5>
1895
1896<p>Depending on properties of the target machine and the particular
1897<tt>switch</tt> instruction, this instruction may be code generated in different
1898ways. For example, it could be generated as a series of chained conditional
1899branches or with a lookup table.</p>
1900
1901<h5>Example:</h5>
1902
1903<pre>
1904 <i>; Emulate a conditional br instruction</i>
1905 %Val = <a href="#i_zext">zext</a> i1 %value to i32
1906 switch i32 %Val, label %truedest [i32 0, label %falsedest ]
1907
1908 <i>; Emulate an unconditional br instruction</i>
1909 switch i32 0, label %dest [ ]
1910
1911 <i>; Implement a jump table:</i>
1912 switch i32 %val, label %otherwise [ i32 0, label %onzero
1913 i32 1, label %onone
1914 i32 2, label %ontwo ]
1915</pre>
1916</div>
1917
1918<!-- _______________________________________________________________________ -->
1919<div class="doc_subsubsection">
1920 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
1921</div>
1922
1923<div class="doc_text">
1924
1925<h5>Syntax:</h5>
1926
1927<pre>
Nick Lewyckya1c11a12008-03-16 07:18:12 +0000[diff] [blame]1928 &lt;result&gt; = invoke [<a href="#callingconv">cconv</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]1929 to label &lt;normal label&gt; unwind label &lt;exception label&gt;
1930</pre>
1931
1932<h5>Overview:</h5>
1933
1934<p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
1935function, with the possibility of control flow transfer to either the
1936'<tt>normal</tt>' label or the
1937'<tt>exception</tt>' label. If the callee function returns with the
1938"<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
1939"normal" label. If the callee (or any indirect callees) returns with the "<a
1940href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
Devang Patela3cc5372008-03-10 20:49:15 +0000[diff] [blame]1941continued at the dynamically nearest "exception" label. If the callee function
Devang Patelec8a5b02008-03-11 05:51:59 +0000[diff] [blame]1942returns multiple values then individual return values are only accessible through
1943a '<tt><a href="#i_getresult">getresult</a></tt>' instruction.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1944
1945<h5>Arguments:</h5>
1946
1947<p>This instruction requires several arguments:</p>
1948
1949<ol>
1950 <li>
1951 The optional "cconv" marker indicates which <a href="#callingconv">calling
1952 convention</a> the call should use. If none is specified, the call defaults
1953 to using C calling conventions.
1954 </li>
1955 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
1956 function value being invoked. In most cases, this is a direct function
1957 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
1958 an arbitrary pointer to function value.
1959 </li>
1960
1961 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
1962 function to be invoked. </li>
1963
1964 <li>'<tt>function args</tt>': argument list whose types match the function
1965 signature argument types. If the function signature indicates the function
1966 accepts a variable number of arguments, the extra arguments can be
1967 specified. </li>
1968
1969 <li>'<tt>normal label</tt>': the label reached when the called function
1970 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
1971
1972 <li>'<tt>exception label</tt>': the label reached when a callee returns with
1973 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
1974
1975</ol>
1976
1977<h5>Semantics:</h5>
1978
1979<p>This instruction is designed to operate as a standard '<tt><a
1980href="#i_call">call</a></tt>' instruction in most regards. The primary
1981difference is that it establishes an association with a label, which is used by
1982the runtime library to unwind the stack.</p>
1983
1984<p>This instruction is used in languages with destructors to ensure that proper
1985cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
1986exception. Additionally, this is important for implementation of
1987'<tt>catch</tt>' clauses in high-level languages that support them.</p>
1988
1989<h5>Example:</h5>
1990<pre>
Nick Lewyckya1c11a12008-03-16 07:18:12 +0000[diff] [blame]1991 %retval = invoke i32 @Test(i32 15) to label %Continue
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1992 unwind label %TestCleanup <i>; {i32}:retval set</i>
Nick Lewyckya1c11a12008-03-16 07:18:12 +0000[diff] [blame]1993 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]1994 unwind label %TestCleanup <i>; {i32}:retval set</i>
1995</pre>
1996</div>
1997
1998
1999<!-- _______________________________________________________________________ -->
2000
2001<div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
2002Instruction</a> </div>
2003
2004<div class="doc_text">
2005
2006<h5>Syntax:</h5>
2007<pre>
2008 unwind
2009</pre>
2010
2011<h5>Overview:</h5>
2012
2013<p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
2014at the first callee in the dynamic call stack which used an <a
2015href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
2016primarily used to implement exception handling.</p>
2017
2018<h5>Semantics:</h5>
2019
2020<p>The '<tt>unwind</tt>' intrinsic causes execution of the current function to
2021immediately halt. The dynamic call stack is then searched for the first <a
2022href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
2023execution continues at the "exceptional" destination block specified by the
2024<tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
2025dynamic call chain, undefined behavior results.</p>
2026</div>
2027
2028<!-- _______________________________________________________________________ -->
2029
2030<div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
2031Instruction</a> </div>
2032
2033<div class="doc_text">
2034
2035<h5>Syntax:</h5>
2036<pre>
2037 unreachable
2038</pre>
2039
2040<h5>Overview:</h5>
2041
2042<p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
2043instruction is used to inform the optimizer that a particular portion of the
2044code is not reachable. This can be used to indicate that the code after a
2045no-return function cannot be reached, and other facts.</p>
2046
2047<h5>Semantics:</h5>
2048
2049<p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
2050</div>
2051
2052
2053
2054<!-- ======================================================================= -->
2055<div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
2056<div class="doc_text">
2057<p>Binary operators are used to do most of the computation in a
Chris Lattnerab596d92008-04-01 18:47:32 +0000[diff] [blame]2058program. They require two operands of the same type, execute an operation on them, and
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2059produce a single value. The operands might represent
2060multiple 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]2061The result value has the same type as its operands.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2062<p>There are several different binary operators:</p>
2063</div>
2064<!-- _______________________________________________________________________ -->
2065<div class="doc_subsubsection"> <a name="i_add">'<tt>add</tt>'
2066Instruction</a> </div>
2067<div class="doc_text">
2068<h5>Syntax:</h5>
2069<pre> &lt;result&gt; = add &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2070</pre>
2071<h5>Overview:</h5>
2072<p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
2073<h5>Arguments:</h5>
2074<p>The two arguments to the '<tt>add</tt>' instruction must be either <a
2075 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a> values.
2076 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2077Both arguments must have identical types.</p>
2078<h5>Semantics:</h5>
2079<p>The value produced is the integer or floating point sum of the two
2080operands.</p>
Chris Lattner9aba1e22008-01-28 00:36:27 +0000[diff] [blame]2081<p>If an integer sum has unsigned overflow, the result returned is the
2082mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2083the result.</p>
2084<p>Because LLVM integers use a two's complement representation, this
2085instruction is appropriate for both signed and unsigned integers.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2086<h5>Example:</h5>
2087<pre> &lt;result&gt; = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
2088</pre>
2089</div>
2090<!-- _______________________________________________________________________ -->
2091<div class="doc_subsubsection"> <a name="i_sub">'<tt>sub</tt>'
2092Instruction</a> </div>
2093<div class="doc_text">
2094<h5>Syntax:</h5>
2095<pre> &lt;result&gt; = sub &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2096</pre>
2097<h5>Overview:</h5>
2098<p>The '<tt>sub</tt>' instruction returns the difference of its two
2099operands.</p>
2100<p>Note that the '<tt>sub</tt>' instruction is used to represent the '<tt>neg</tt>'
2101instruction present in most other intermediate representations.</p>
2102<h5>Arguments:</h5>
2103<p>The two arguments to the '<tt>sub</tt>' instruction must be either <a
2104 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
2105values.
2106This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2107Both arguments must have identical types.</p>
2108<h5>Semantics:</h5>
2109<p>The value produced is the integer or floating point difference of
2110the two operands.</p>
Chris Lattner9aba1e22008-01-28 00:36:27 +0000[diff] [blame]2111<p>If an integer difference has unsigned overflow, the result returned is the
2112mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2113the result.</p>
2114<p>Because LLVM integers use a two's complement representation, this
2115instruction is appropriate for both signed and unsigned integers.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2116<h5>Example:</h5>
2117<pre>
2118 &lt;result&gt; = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
2119 &lt;result&gt; = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
2120</pre>
2121</div>
2122<!-- _______________________________________________________________________ -->
2123<div class="doc_subsubsection"> <a name="i_mul">'<tt>mul</tt>'
2124Instruction</a> </div>
2125<div class="doc_text">
2126<h5>Syntax:</h5>
2127<pre> &lt;result&gt; = mul &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2128</pre>
2129<h5>Overview:</h5>
2130<p>The '<tt>mul</tt>' instruction returns the product of its two
2131operands.</p>
2132<h5>Arguments:</h5>
2133<p>The two arguments to the '<tt>mul</tt>' instruction must be either <a
2134 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
2135values.
2136This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2137Both arguments must have identical types.</p>
2138<h5>Semantics:</h5>
2139<p>The value produced is the integer or floating point product of the
2140two operands.</p>
Chris Lattner9aba1e22008-01-28 00:36:27 +0000[diff] [blame]2141<p>If the result of an integer multiplication has unsigned overflow,
2142the result returned is the mathematical result modulo
21432<sup>n</sup>, where n is the bit width of the result.</p>
2144<p>Because LLVM integers use a two's complement representation, and the
2145result is the same width as the operands, this instruction returns the
2146correct result for both signed and unsigned integers. If a full product
2147(e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands
2148should be sign-extended or zero-extended as appropriate to the
2149width of the full product.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2150<h5>Example:</h5>
2151<pre> &lt;result&gt; = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
2152</pre>
2153</div>
2154<!-- _______________________________________________________________________ -->
2155<div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
2156</a></div>
2157<div class="doc_text">
2158<h5>Syntax:</h5>
2159<pre> &lt;result&gt; = udiv &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2160</pre>
2161<h5>Overview:</h5>
2162<p>The '<tt>udiv</tt>' instruction returns the quotient of its two
2163operands.</p>
2164<h5>Arguments:</h5>
2165<p>The two arguments to the '<tt>udiv</tt>' instruction must be
2166<a href="#t_integer">integer</a> values. Both arguments must have identical
2167types. This instruction can also take <a href="#t_vector">vector</a> versions
2168of the values in which case the elements must be integers.</p>
2169<h5>Semantics:</h5>
Chris Lattner9aba1e22008-01-28 00:36:27 +0000[diff] [blame]2170<p>The value produced is the unsigned integer quotient of the two operands.</p>
2171<p>Note that unsigned integer division and signed integer division are distinct
2172operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
2173<p>Division by zero leads to undefined behavior.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2174<h5>Example:</h5>
2175<pre> &lt;result&gt; = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2176</pre>
2177</div>
2178<!-- _______________________________________________________________________ -->
2179<div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
2180</a> </div>
2181<div class="doc_text">
2182<h5>Syntax:</h5>
2183<pre> &lt;result&gt; = sdiv &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2184</pre>
2185<h5>Overview:</h5>
2186<p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
2187operands.</p>
2188<h5>Arguments:</h5>
2189<p>The two arguments to the '<tt>sdiv</tt>' instruction must be
2190<a href="#t_integer">integer</a> values. Both arguments must have identical
2191types. This instruction can also take <a href="#t_vector">vector</a> versions
2192of the values in which case the elements must be integers.</p>
2193<h5>Semantics:</h5>
Chris Lattnerdc15b1d2008-04-01 18:45:27 +0000[diff] [blame]2194<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]2195<p>Note that signed integer division and unsigned integer division are distinct
2196operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
2197<p>Division by zero leads to undefined behavior. Overflow also leads to
2198undefined behavior; this is a rare case, but can occur, for example,
2199by doing a 32-bit division of -2147483648 by -1.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2200<h5>Example:</h5>
2201<pre> &lt;result&gt; = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2202</pre>
2203</div>
2204<!-- _______________________________________________________________________ -->
2205<div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
2206Instruction</a> </div>
2207<div class="doc_text">
2208<h5>Syntax:</h5>
2209<pre> &lt;result&gt; = fdiv &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2210</pre>
2211<h5>Overview:</h5>
2212<p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
2213operands.</p>
2214<h5>Arguments:</h5>
2215<p>The two arguments to the '<tt>fdiv</tt>' instruction must be
2216<a href="#t_floating">floating point</a> values. Both arguments must have
2217identical types. This instruction can also take <a href="#t_vector">vector</a>
2218versions of floating point values.</p>
2219<h5>Semantics:</h5>
2220<p>The value produced is the floating point quotient of the two operands.</p>
2221<h5>Example:</h5>
2222<pre> &lt;result&gt; = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
2223</pre>
2224</div>
2225<!-- _______________________________________________________________________ -->
2226<div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
2227</div>
2228<div class="doc_text">
2229<h5>Syntax:</h5>
2230<pre> &lt;result&gt; = urem &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2231</pre>
2232<h5>Overview:</h5>
2233<p>The '<tt>urem</tt>' instruction returns the remainder from the
2234unsigned division of its two arguments.</p>
2235<h5>Arguments:</h5>
2236<p>The two arguments to the '<tt>urem</tt>' instruction must be
2237<a href="#t_integer">integer</a> values. Both arguments must have identical
Dan Gohman3e3fd8c2007-11-05 23:35:22 +0000[diff] [blame]2238types. This instruction can also take <a href="#t_vector">vector</a> versions
2239of the values in which case the elements must be integers.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2240<h5>Semantics:</h5>
2241<p>This instruction returns the unsigned integer <i>remainder</i> of a division.
Chris Lattnerdc15b1d2008-04-01 18:45:27 +0000[diff] [blame]2242This instruction always performs an unsigned division to get the remainder.</p>
Chris Lattner9aba1e22008-01-28 00:36:27 +0000[diff] [blame]2243<p>Note that unsigned integer remainder and signed integer remainder are
2244distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
2245<p>Taking the remainder of a division by zero leads to undefined behavior.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2246<h5>Example:</h5>
2247<pre> &lt;result&gt; = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2248</pre>
2249
2250</div>
2251<!-- _______________________________________________________________________ -->
2252<div class="doc_subsubsection"> <a name="i_srem">'<tt>srem</tt>'
2253Instruction</a> </div>
2254<div class="doc_text">
2255<h5>Syntax:</h5>
2256<pre> &lt;result&gt; = srem &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2257</pre>
2258<h5>Overview:</h5>
2259<p>The '<tt>srem</tt>' instruction returns the remainder from the
Dan Gohman3e3fd8c2007-11-05 23:35:22 +0000[diff] [blame]2260signed division of its two operands. This instruction can also take
2261<a href="#t_vector">vector</a> versions of the values in which case
2262the elements must be integers.</p>
Chris Lattner08497ce2008-01-04 04:33:49 +0000[diff] [blame]2263
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2264<h5>Arguments:</h5>
2265<p>The two arguments to the '<tt>srem</tt>' instruction must be
2266<a href="#t_integer">integer</a> values. Both arguments must have identical
2267types.</p>
2268<h5>Semantics:</h5>
2269<p>This instruction returns the <i>remainder</i> of a division (where the result
2270has the same sign as the dividend, <tt>var1</tt>), not the <i>modulo</i>
2271operator (where the result has the same sign as the divisor, <tt>var2</tt>) of
2272a value. For more information about the difference, see <a
2273 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
2274Math Forum</a>. For a table of how this is implemented in various languages,
2275please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
2276Wikipedia: modulo operation</a>.</p>
Chris Lattner9aba1e22008-01-28 00:36:27 +0000[diff] [blame]2277<p>Note that signed integer remainder and unsigned integer remainder are
2278distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
2279<p>Taking the remainder of a division by zero leads to undefined behavior.
2280Overflow also leads to undefined behavior; this is a rare case, but can occur,
2281for example, by taking the remainder of a 32-bit division of -2147483648 by -1.
2282(The remainder doesn't actually overflow, but this rule lets srem be
2283implemented using instructions that return both the result of the division
2284and the remainder.)</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2285<h5>Example:</h5>
2286<pre> &lt;result&gt; = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2287</pre>
2288
2289</div>
2290<!-- _______________________________________________________________________ -->
2291<div class="doc_subsubsection"> <a name="i_frem">'<tt>frem</tt>'
2292Instruction</a> </div>
2293<div class="doc_text">
2294<h5>Syntax:</h5>
2295<pre> &lt;result&gt; = frem &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2296</pre>
2297<h5>Overview:</h5>
2298<p>The '<tt>frem</tt>' instruction returns the remainder from the
2299division of its two operands.</p>
2300<h5>Arguments:</h5>
2301<p>The two arguments to the '<tt>frem</tt>' instruction must be
2302<a href="#t_floating">floating point</a> values. Both arguments must have
Dan Gohman3e3fd8c2007-11-05 23:35:22 +0000[diff] [blame]2303identical types. This instruction can also take <a href="#t_vector">vector</a>
2304versions of floating point values.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2305<h5>Semantics:</h5>
Chris Lattnerdc15b1d2008-04-01 18:45:27 +0000[diff] [blame]2306<p>This instruction returns the <i>remainder</i> of a division.
2307The remainder has the same sign as the dividend.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2308<h5>Example:</h5>
2309<pre> &lt;result&gt; = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
2310</pre>
2311</div>
2312
2313<!-- ======================================================================= -->
2314<div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
2315Operations</a> </div>
2316<div class="doc_text">
2317<p>Bitwise binary operators are used to do various forms of
2318bit-twiddling in a program. They are generally very efficient
2319instructions and can commonly be strength reduced from other
Chris Lattnerdc15b1d2008-04-01 18:45:27 +0000[diff] [blame]2320instructions. They require two operands of the same type, execute an operation on them,
2321and 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]2322</div>
2323
2324<!-- _______________________________________________________________________ -->
2325<div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
2326Instruction</a> </div>
2327<div class="doc_text">
2328<h5>Syntax:</h5>
2329<pre> &lt;result&gt; = shl &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2330</pre>
Chris Lattnerd939d9f2007-10-03 21:01:14 +0000[diff] [blame]2331
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2332<h5>Overview:</h5>
Chris Lattnerd939d9f2007-10-03 21:01:14 +0000[diff] [blame]2333
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2334<p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2335the left a specified number of bits.</p>
Chris Lattnerd939d9f2007-10-03 21:01:14 +0000[diff] [blame]2336
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2337<h5>Arguments:</h5>
Chris Lattnerd939d9f2007-10-03 21:01:14 +0000[diff] [blame]2338
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2339<p>Both arguments to the '<tt>shl</tt>' instruction must be the same <a
2340 href="#t_integer">integer</a> type.</p>
Chris Lattnerd939d9f2007-10-03 21:01:14 +0000[diff] [blame]2341
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2342<h5>Semantics:</h5>
Chris Lattnerd939d9f2007-10-03 21:01:14 +0000[diff] [blame]2343
Chris Lattnerdc15b1d2008-04-01 18:45:27 +0000[diff] [blame]2344<p>The value produced is <tt>var1</tt> * 2<sup><tt>var2</tt></sup> mod 2<sup>n</sup>,
2345where n is the width of the result. If <tt>var2</tt> is (statically or dynamically) negative or
2346equal to or larger than the number of bits in <tt>var1</tt>, the result is undefined.</p>
Chris Lattnerd939d9f2007-10-03 21:01:14 +0000[diff] [blame]2347
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2348<h5>Example:</h5><pre>
2349 &lt;result&gt; = shl i32 4, %var <i>; yields {i32}: 4 &lt;&lt; %var</i>
2350 &lt;result&gt; = shl i32 4, 2 <i>; yields {i32}: 16</i>
2351 &lt;result&gt; = shl i32 1, 10 <i>; yields {i32}: 1024</i>
Chris Lattnerd939d9f2007-10-03 21:01:14 +0000[diff] [blame]2352 &lt;result&gt; = shl i32 1, 32 <i>; undefined</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2353</pre>
2354</div>
2355<!-- _______________________________________________________________________ -->
2356<div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
2357Instruction</a> </div>
2358<div class="doc_text">
2359<h5>Syntax:</h5>
2360<pre> &lt;result&gt; = lshr &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2361</pre>
2362
2363<h5>Overview:</h5>
2364<p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
2365operand shifted to the right a specified number of bits with zero fill.</p>
2366
2367<h5>Arguments:</h5>
2368<p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
2369<a href="#t_integer">integer</a> type.</p>
2370
2371<h5>Semantics:</h5>
Chris Lattnerd939d9f2007-10-03 21:01:14 +0000[diff] [blame]2372
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2373<p>This instruction always performs a logical shift right operation. The most
2374significant bits of the result will be filled with zero bits after the
Chris Lattnerd939d9f2007-10-03 21:01:14 +0000[diff] [blame]2375shift. If <tt>var2</tt> is (statically or dynamically) equal to or larger than
2376the number of bits in <tt>var1</tt>, the result is undefined.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2377
2378<h5>Example:</h5>
2379<pre>
2380 &lt;result&gt; = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
2381 &lt;result&gt; = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
2382 &lt;result&gt; = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
2383 &lt;result&gt; = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
Chris Lattnerd939d9f2007-10-03 21:01:14 +0000[diff] [blame]2384 &lt;result&gt; = lshr i32 1, 32 <i>; undefined</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2385</pre>
2386</div>
2387
2388<!-- _______________________________________________________________________ -->
2389<div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
2390Instruction</a> </div>
2391<div class="doc_text">
2392
2393<h5>Syntax:</h5>
2394<pre> &lt;result&gt; = ashr &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2395</pre>
2396
2397<h5>Overview:</h5>
2398<p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
2399operand shifted to the right a specified number of bits with sign extension.</p>
2400
2401<h5>Arguments:</h5>
2402<p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
2403<a href="#t_integer">integer</a> type.</p>
2404
2405<h5>Semantics:</h5>
2406<p>This instruction always performs an arithmetic shift right operation,
2407The most significant bits of the result will be filled with the sign bit
Chris Lattnerd939d9f2007-10-03 21:01:14 +0000[diff] [blame]2408of <tt>var1</tt>. If <tt>var2</tt> is (statically or dynamically) equal to or
2409larger than the number of bits in <tt>var1</tt>, the result is undefined.
2410</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2411
2412<h5>Example:</h5>
2413<pre>
2414 &lt;result&gt; = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
2415 &lt;result&gt; = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
2416 &lt;result&gt; = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
2417 &lt;result&gt; = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
Chris Lattnerd939d9f2007-10-03 21:01:14 +0000[diff] [blame]2418 &lt;result&gt; = ashr i32 1, 32 <i>; undefined</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2419</pre>
2420</div>
2421
2422<!-- _______________________________________________________________________ -->
2423<div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
2424Instruction</a> </div>
2425<div class="doc_text">
2426<h5>Syntax:</h5>
2427<pre> &lt;result&gt; = and &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2428</pre>
2429<h5>Overview:</h5>
2430<p>The '<tt>and</tt>' instruction returns the bitwise logical and of
2431its two operands.</p>
2432<h5>Arguments:</h5>
2433<p>The two arguments to the '<tt>and</tt>' instruction must be <a
2434 href="#t_integer">integer</a> values. Both arguments must have
2435identical types.</p>
2436<h5>Semantics:</h5>
2437<p>The truth table used for the '<tt>and</tt>' instruction is:</p>
2438<p> </p>
2439<div style="align: center">
2440<table border="1" cellspacing="0" cellpadding="4">
2441 <tbody>
2442 <tr>
2443 <td>In0</td>
2444 <td>In1</td>
2445 <td>Out</td>
2446 </tr>
2447 <tr>
2448 <td>0</td>
2449 <td>0</td>
2450 <td>0</td>
2451 </tr>
2452 <tr>
2453 <td>0</td>
2454 <td>1</td>
2455 <td>0</td>
2456 </tr>
2457 <tr>
2458 <td>1</td>
2459 <td>0</td>
2460 <td>0</td>
2461 </tr>
2462 <tr>
2463 <td>1</td>
2464 <td>1</td>
2465 <td>1</td>
2466 </tr>
2467 </tbody>
2468</table>
2469</div>
2470<h5>Example:</h5>
2471<pre> &lt;result&gt; = and i32 4, %var <i>; yields {i32}:result = 4 &amp; %var</i>
2472 &lt;result&gt; = and i32 15, 40 <i>; yields {i32}:result = 8</i>
2473 &lt;result&gt; = and i32 4, 8 <i>; yields {i32}:result = 0</i>
2474</pre>
2475</div>
2476<!-- _______________________________________________________________________ -->
2477<div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
2478<div class="doc_text">
2479<h5>Syntax:</h5>
2480<pre> &lt;result&gt; = or &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2481</pre>
2482<h5>Overview:</h5>
2483<p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
2484or of its two operands.</p>
2485<h5>Arguments:</h5>
2486<p>The two arguments to the '<tt>or</tt>' instruction must be <a
2487 href="#t_integer">integer</a> values. Both arguments must have
2488identical types.</p>
2489<h5>Semantics:</h5>
2490<p>The truth table used for the '<tt>or</tt>' instruction is:</p>
2491<p> </p>
2492<div style="align: center">
2493<table border="1" cellspacing="0" cellpadding="4">
2494 <tbody>
2495 <tr>
2496 <td>In0</td>
2497 <td>In1</td>
2498 <td>Out</td>
2499 </tr>
2500 <tr>
2501 <td>0</td>
2502 <td>0</td>
2503 <td>0</td>
2504 </tr>
2505 <tr>
2506 <td>0</td>
2507 <td>1</td>
2508 <td>1</td>
2509 </tr>
2510 <tr>
2511 <td>1</td>
2512 <td>0</td>
2513 <td>1</td>
2514 </tr>
2515 <tr>
2516 <td>1</td>
2517 <td>1</td>
2518 <td>1</td>
2519 </tr>
2520 </tbody>
2521</table>
2522</div>
2523<h5>Example:</h5>
2524<pre> &lt;result&gt; = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
2525 &lt;result&gt; = or i32 15, 40 <i>; yields {i32}:result = 47</i>
2526 &lt;result&gt; = or i32 4, 8 <i>; yields {i32}:result = 12</i>
2527</pre>
2528</div>
2529<!-- _______________________________________________________________________ -->
2530<div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
2531Instruction</a> </div>
2532<div class="doc_text">
2533<h5>Syntax:</h5>
2534<pre> &lt;result&gt; = xor &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2535</pre>
2536<h5>Overview:</h5>
2537<p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
2538or of its two operands. The <tt>xor</tt> is used to implement the
2539"one's complement" operation, which is the "~" operator in C.</p>
2540<h5>Arguments:</h5>
2541<p>The two arguments to the '<tt>xor</tt>' instruction must be <a
2542 href="#t_integer">integer</a> values. Both arguments must have
2543identical types.</p>
2544<h5>Semantics:</h5>
2545<p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
2546<p> </p>
2547<div style="align: center">
2548<table border="1" cellspacing="0" cellpadding="4">
2549 <tbody>
2550 <tr>
2551 <td>In0</td>
2552 <td>In1</td>
2553 <td>Out</td>
2554 </tr>
2555 <tr>
2556 <td>0</td>
2557 <td>0</td>
2558 <td>0</td>
2559 </tr>
2560 <tr>
2561 <td>0</td>
2562 <td>1</td>
2563 <td>1</td>
2564 </tr>
2565 <tr>
2566 <td>1</td>
2567 <td>0</td>
2568 <td>1</td>
2569 </tr>
2570 <tr>
2571 <td>1</td>
2572 <td>1</td>
2573 <td>0</td>
2574 </tr>
2575 </tbody>
2576</table>
2577</div>
2578<p> </p>
2579<h5>Example:</h5>
2580<pre> &lt;result&gt; = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
2581 &lt;result&gt; = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
2582 &lt;result&gt; = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
2583 &lt;result&gt; = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
2584</pre>
2585</div>
2586
2587<!-- ======================================================================= -->
2588<div class="doc_subsection">
2589 <a name="vectorops">Vector Operations</a>
2590</div>
2591
2592<div class="doc_text">
2593
2594<p>LLVM supports several instructions to represent vector operations in a
2595target-independent manner. These instructions cover the element-access and
2596vector-specific operations needed to process vectors effectively. While LLVM
2597does directly support these vector operations, many sophisticated algorithms
2598will want to use target-specific intrinsics to take full advantage of a specific
2599target.</p>
2600
2601</div>
2602
2603<!-- _______________________________________________________________________ -->
2604<div class="doc_subsubsection">
2605 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2606</div>
2607
2608<div class="doc_text">
2609
2610<h5>Syntax:</h5>
2611
2612<pre>
2613 &lt;result&gt; = extractelement &lt;n x &lt;ty&gt;&gt; &lt;val&gt;, i32 &lt;idx&gt; <i>; yields &lt;ty&gt;</i>
2614</pre>
2615
2616<h5>Overview:</h5>
2617
2618<p>
2619The '<tt>extractelement</tt>' instruction extracts a single scalar
2620element from a vector at a specified index.
2621</p>
2622
2623
2624<h5>Arguments:</h5>
2625
2626<p>
2627The first operand of an '<tt>extractelement</tt>' instruction is a
2628value of <a href="#t_vector">vector</a> type. The second operand is
2629an index indicating the position from which to extract the element.
2630The index may be a variable.</p>
2631
2632<h5>Semantics:</h5>
2633
2634<p>
2635The result is a scalar of the same type as the element type of
2636<tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2637<tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2638results are undefined.
2639</p>
2640
2641<h5>Example:</h5>
2642
2643<pre>
2644 %result = extractelement &lt;4 x i32&gt; %vec, i32 0 <i>; yields i32</i>
2645</pre>
2646</div>
2647
2648
2649<!-- _______________________________________________________________________ -->
2650<div class="doc_subsubsection">
2651 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
2652</div>
2653
2654<div class="doc_text">
2655
2656<h5>Syntax:</h5>
2657
2658<pre>
2659 &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>
2660</pre>
2661
2662<h5>Overview:</h5>
2663
2664<p>
2665The '<tt>insertelement</tt>' instruction inserts a scalar
2666element into a vector at a specified index.
2667</p>
2668
2669
2670<h5>Arguments:</h5>
2671
2672<p>
2673The first operand of an '<tt>insertelement</tt>' instruction is a
2674value of <a href="#t_vector">vector</a> type. The second operand is a
2675scalar value whose type must equal the element type of the first
2676operand. The third operand is an index indicating the position at
2677which to insert the value. The index may be a variable.</p>
2678
2679<h5>Semantics:</h5>
2680
2681<p>
2682The result is a vector of the same type as <tt>val</tt>. Its
2683element values are those of <tt>val</tt> except at position
2684<tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
2685exceeds the length of <tt>val</tt>, the results are undefined.
2686</p>
2687
2688<h5>Example:</h5>
2689
2690<pre>
2691 %result = insertelement &lt;4 x i32&gt; %vec, i32 1, i32 0 <i>; yields &lt;4 x i32&gt;</i>
2692</pre>
2693</div>
2694
2695<!-- _______________________________________________________________________ -->
2696<div class="doc_subsubsection">
2697 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
2698</div>
2699
2700<div class="doc_text">
2701
2702<h5>Syntax:</h5>
2703
2704<pre>
2705 &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>
2706</pre>
2707
2708<h5>Overview:</h5>
2709
2710<p>
2711The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
2712from two input vectors, returning a vector of the same type.
2713</p>
2714
2715<h5>Arguments:</h5>
2716
2717<p>
2718The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
2719with types that match each other and types that match the result of the
2720instruction. The third argument is a shuffle mask, which has the same number
2721of elements as the other vector type, but whose element type is always 'i32'.
2722</p>
2723
2724<p>
2725The shuffle mask operand is required to be a constant vector with either
2726constant integer or undef values.
2727</p>
2728
2729<h5>Semantics:</h5>
2730
2731<p>
2732The elements of the two input vectors are numbered from left to right across
2733both of the vectors. The shuffle mask operand specifies, for each element of
2734the result vector, which element of the two input registers the result element
2735gets. The element selector may be undef (meaning "don't care") and the second
2736operand may be undef if performing a shuffle from only one vector.
2737</p>
2738
2739<h5>Example:</h5>
2740
2741<pre>
2742 %result = shufflevector &lt;4 x i32&gt; %v1, &lt;4 x i32&gt; %v2,
2743 &lt;4 x i32&gt; &lt;i32 0, i32 4, i32 1, i32 5&gt; <i>; yields &lt;4 x i32&gt;</i>
2744 %result = shufflevector &lt;4 x i32&gt; %v1, &lt;4 x i32&gt; undef,
2745 &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.
2746</pre>
2747</div>
2748
2749
2750<!-- ======================================================================= -->
2751<div class="doc_subsection">
2752 <a name="memoryops">Memory Access and Addressing Operations</a>
2753</div>
2754
2755<div class="doc_text">
2756
2757<p>A key design point of an SSA-based representation is how it
2758represents memory. In LLVM, no memory locations are in SSA form, which
2759makes things very simple. This section describes how to read, write,
2760allocate, and free memory in LLVM.</p>
2761
2762</div>
2763
2764<!-- _______________________________________________________________________ -->
2765<div class="doc_subsubsection">
2766 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
2767</div>
2768
2769<div class="doc_text">
2770
2771<h5>Syntax:</h5>
2772
2773<pre>
2774 &lt;result&gt; = malloc &lt;type&gt;[, i32 &lt;NumElements&gt;][, align &lt;alignment&gt;] <i>; yields {type*}:result</i>
2775</pre>
2776
2777<h5>Overview:</h5>
2778
2779<p>The '<tt>malloc</tt>' instruction allocates memory from the system
Christopher Lambcfe00962007-12-17 01:00:21 +0000[diff] [blame]2780heap and returns a pointer to it. The object is always allocated in the generic
2781address space (address space zero).</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2782
2783<h5>Arguments:</h5>
2784
2785<p>The '<tt>malloc</tt>' instruction allocates
2786<tt>sizeof(&lt;type&gt;)*NumElements</tt>
2787bytes of memory from the operating system and returns a pointer of the
2788appropriate type to the program. If "NumElements" is specified, it is the
Gabor Greif5082cf42008-02-09 22:24:34 +0000[diff] [blame]2789number of elements allocated, otherwise "NumElements" is defaulted to be one.
Chris Lattner10368b62008-04-02 00:38:26 +0000[diff] [blame^]2790If a constant alignment is specified, the value result of the allocation is guaranteed to
Gabor Greif5082cf42008-02-09 22:24:34 +0000[diff] [blame]2791be aligned to at least that boundary. If not specified, or if zero, the target can
2792choose to align the allocation on any convenient boundary.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2793
2794<p>'<tt>type</tt>' must be a sized type.</p>
2795
2796<h5>Semantics:</h5>
2797
2798<p>Memory is allocated using the system "<tt>malloc</tt>" function, and
Chris Lattner10368b62008-04-02 00:38:26 +0000[diff] [blame^]2799a pointer is returned. Allocating zero bytes is undefined. The result is null
2800if there is insufficient memory available.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2801
2802<h5>Example:</h5>
2803
2804<pre>
2805 %array = malloc [4 x i8 ] <i>; yields {[%4 x i8]*}:array</i>
2806
2807 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
2808 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
2809 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
2810 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
2811 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
2812</pre>
2813</div>
2814
2815<!-- _______________________________________________________________________ -->
2816<div class="doc_subsubsection">
2817 <a name="i_free">'<tt>free</tt>' Instruction</a>
2818</div>
2819
2820<div class="doc_text">
2821
2822<h5>Syntax:</h5>
2823
2824<pre>
2825 free &lt;type&gt; &lt;value&gt; <i>; yields {void}</i>
2826</pre>
2827
2828<h5>Overview:</h5>
2829
2830<p>The '<tt>free</tt>' instruction returns memory back to the unused
2831memory heap to be reallocated in the future.</p>
2832
2833<h5>Arguments:</h5>
2834
2835<p>'<tt>value</tt>' shall be a pointer value that points to a value
2836that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
2837instruction.</p>
2838
2839<h5>Semantics:</h5>
2840
2841<p>Access to the memory pointed to by the pointer is no longer defined
Chris Lattner10368b62008-04-02 00:38:26 +0000[diff] [blame^]2842after this instruction executes. If the pointer is null, the result is
2843undefined.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2844
2845<h5>Example:</h5>
2846
2847<pre>
2848 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
2849 free [4 x i8]* %array
2850</pre>
2851</div>
2852
2853<!-- _______________________________________________________________________ -->
2854<div class="doc_subsubsection">
2855 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
2856</div>
2857
2858<div class="doc_text">
2859
2860<h5>Syntax:</h5>
2861
2862<pre>
2863 &lt;result&gt; = alloca &lt;type&gt;[, i32 &lt;NumElements&gt;][, align &lt;alignment&gt;] <i>; yields {type*}:result</i>
2864</pre>
2865
2866<h5>Overview:</h5>
2867
2868<p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
2869currently executing function, to be automatically released when this function
Christopher Lambcfe00962007-12-17 01:00:21 +0000[diff] [blame]2870returns to its caller. The object is always allocated in the generic address
2871space (address space zero).</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2872
2873<h5>Arguments:</h5>
2874
2875<p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(&lt;type&gt;)*NumElements</tt>
2876bytes of memory on the runtime stack, returning a pointer of the
Gabor Greif5082cf42008-02-09 22:24:34 +0000[diff] [blame]2877appropriate type to the program. If "NumElements" is specified, it is the
2878number of elements allocated, otherwise "NumElements" is defaulted to be one.
Chris Lattner10368b62008-04-02 00:38:26 +0000[diff] [blame^]2879If a constant alignment is specified, the value result of the allocation is guaranteed
Gabor Greif5082cf42008-02-09 22:24:34 +0000[diff] [blame]2880to be aligned to at least that boundary. If not specified, or if zero, the target
2881can choose to align the allocation on any convenient boundary.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2882
2883<p>'<tt>type</tt>' may be any sized type.</p>
2884
2885<h5>Semantics:</h5>
2886
2887<p>Memory is allocated; a pointer is returned. '<tt>alloca</tt>'d
2888memory is automatically released when the function returns. The '<tt>alloca</tt>'
2889instruction is commonly used to represent automatic variables that must
2890have an address available. When the function returns (either with the <tt><a
2891 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^]2892instructions), the memory is reclaimed. Allocating zero bytes
2893is legal, but the result is undefined.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2894
2895<h5>Example:</h5>
2896
2897<pre>
2898 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
2899 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
2900 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
2901 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
2902</pre>
2903</div>
2904
2905<!-- _______________________________________________________________________ -->
2906<div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
2907Instruction</a> </div>
2908<div class="doc_text">
2909<h5>Syntax:</h5>
2910<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>
2911<h5>Overview:</h5>
2912<p>The '<tt>load</tt>' instruction is used to read from memory.</p>
2913<h5>Arguments:</h5>
2914<p>The argument to the '<tt>load</tt>' instruction specifies the memory
2915address from which to load. The pointer must point to a <a
2916 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
2917marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
2918the number or order of execution of this <tt>load</tt> with other
2919volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
2920instructions. </p>
Chris Lattner21003c82008-01-06 21:04:43 +0000[diff] [blame]2921<p>
Chris Lattner10368b62008-04-02 00:38:26 +0000[diff] [blame^]2922The optional constant "align" argument specifies the alignment of the operation
Chris Lattner21003c82008-01-06 21:04:43 +0000[diff] [blame]2923(that is, the alignment of the memory address). A value of 0 or an
2924omitted "align" argument means that the operation has the preferential
2925alignment for the target. It is the responsibility of the code emitter
2926to ensure that the alignment information is correct. Overestimating
2927the alignment results in an undefined behavior. Underestimating the
2928alignment may produce less efficient code. An alignment of 1 is always
2929safe.
2930</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2931<h5>Semantics:</h5>
2932<p>The location of memory pointed to is loaded.</p>
2933<h5>Examples:</h5>
2934<pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
2935 <a
2936 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
2937 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
2938</pre>
2939</div>
2940<!-- _______________________________________________________________________ -->
2941<div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
2942Instruction</a> </div>
2943<div class="doc_text">
2944<h5>Syntax:</h5>
2945<pre> store &lt;ty&gt; &lt;value&gt;, &lt;ty&gt;* &lt;pointer&gt;[, align &lt;alignment&gt;] <i>; yields {void}</i>
2946 volatile store &lt;ty&gt; &lt;value&gt;, &lt;ty&gt;* &lt;pointer&gt;[, align &lt;alignment&gt;] <i>; yields {void}</i>
2947</pre>
2948<h5>Overview:</h5>
2949<p>The '<tt>store</tt>' instruction is used to write to memory.</p>
2950<h5>Arguments:</h5>
2951<p>There are two arguments to the '<tt>store</tt>' instruction: a value
2952to 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^]2953operand must be a pointer to the <a href="#t_firstclass">first class</a> type
2954of the '<tt>&lt;value&gt;</tt>'
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2955operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
2956optimizer is not allowed to modify the number or order of execution of
2957this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
2958 href="#i_store">store</a></tt> instructions.</p>
Chris Lattner21003c82008-01-06 21:04:43 +0000[diff] [blame]2959<p>
Chris Lattner10368b62008-04-02 00:38:26 +0000[diff] [blame^]2960The optional constant "align" argument specifies the alignment of the operation
Chris Lattner21003c82008-01-06 21:04:43 +0000[diff] [blame]2961(that is, the alignment of the memory address). A value of 0 or an
2962omitted "align" argument means that the operation has the preferential
2963alignment for the target. It is the responsibility of the code emitter
2964to ensure that the alignment information is correct. Overestimating
2965the alignment results in an undefined behavior. Underestimating the
2966alignment may produce less efficient code. An alignment of 1 is always
2967safe.
2968</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]2969<h5>Semantics:</h5>
2970<p>The contents of memory are updated to contain '<tt>&lt;value&gt;</tt>'
2971at the location specified by the '<tt>&lt;pointer&gt;</tt>' operand.</p>
2972<h5>Example:</h5>
2973<pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
Bill Wendling63ffa142007-10-22 05:10:05 +0000[diff] [blame]2974 store i32 3, i32* %ptr <i>; yields {void}</i>
2975 %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]2976</pre>
2977</div>
2978
2979<!-- _______________________________________________________________________ -->
2980<div class="doc_subsubsection">
2981 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
2982</div>
2983
2984<div class="doc_text">
2985<h5>Syntax:</h5>
2986<pre>
2987 &lt;result&gt; = getelementptr &lt;ty&gt;* &lt;ptrval&gt;{, &lt;ty&gt; &lt;idx&gt;}*
2988</pre>
2989
2990<h5>Overview:</h5>
2991
2992<p>
2993The '<tt>getelementptr</tt>' instruction is used to get the address of a
2994subelement of an aggregate data structure.</p>
2995
2996<h5>Arguments:</h5>
2997
2998<p>This instruction takes a list of integer operands that indicate what
2999elements of the aggregate object to index to. The actual types of the arguments
3000provided depend on the type of the first pointer argument. The
3001'<tt>getelementptr</tt>' instruction is used to index down through the type
3002levels of a structure or to a specific index in an array. When indexing into a
3003structure, only <tt>i32</tt> integer constants are allowed. When indexing
3004into an array or pointer, only integers of 32 or 64 bits are allowed, and will
3005be sign extended to 64-bit values.</p>
3006
3007<p>For example, let's consider a C code fragment and how it gets
3008compiled to LLVM:</p>
3009
3010<div class="doc_code">
3011<pre>
3012struct RT {
3013 char A;
3014 int B[10][20];
3015 char C;
3016};
3017struct ST {
3018 int X;
3019 double Y;
3020 struct RT Z;
3021};
3022
3023int *foo(struct ST *s) {
3024 return &amp;s[1].Z.B[5][13];
3025}
3026</pre>
3027</div>
3028
3029<p>The LLVM code generated by the GCC frontend is:</p>
3030
3031<div class="doc_code">
3032<pre>
3033%RT = type { i8 , [10 x [20 x i32]], i8 }
3034%ST = type { i32, double, %RT }
3035
3036define i32* %foo(%ST* %s) {
3037entry:
3038 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
3039 ret i32* %reg
3040}
3041</pre>
3042</div>
3043
3044<h5>Semantics:</h5>
3045
3046<p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
3047on the pointer type that is being indexed into. <a href="#t_pointer">Pointer</a>
3048and <a href="#t_array">array</a> types can use a 32-bit or 64-bit
3049<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^]3050to 64-bits. <a href="#t_struct">Structure</a> and <a href="#t_pstruct">packed
3051structure</a> types require <tt>i32</tt> <b>constants</b>.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3052
3053<p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
3054type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
3055}</tt>' type, a structure. The second index indexes into the third element of
3056the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
3057i8 }</tt>' type, another structure. The third index indexes into the second
3058element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
3059array. The two dimensions of the array are subscripted into, yielding an
3060'<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
3061to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
3062
3063<p>Note that it is perfectly legal to index partially through a
3064structure, returning a pointer to an inner element. Because of this,
3065the LLVM code for the given testcase is equivalent to:</p>
3066
3067<pre>
3068 define i32* %foo(%ST* %s) {
3069 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
3070 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
3071 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
3072 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
3073 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
3074 ret i32* %t5
3075 }
3076</pre>
3077
3078<p>Note that it is undefined to access an array out of bounds: array and
3079pointer indexes must always be within the defined bounds of the array type.
3080The one exception for this rules is zero length arrays. These arrays are
3081defined to be accessible as variable length arrays, which requires access
3082beyond the zero'th element.</p>
3083
3084<p>The getelementptr instruction is often confusing. For some more insight
3085into how it works, see <a href="GetElementPtr.html">the getelementptr
3086FAQ</a>.</p>
3087
3088<h5>Example:</h5>
3089
3090<pre>
3091 <i>; yields [12 x i8]*:aptr</i>
3092 %aptr = getelementptr {i32, [12 x i8]}* %sptr, i64 0, i32 1
3093</pre>
3094</div>
3095
3096<!-- ======================================================================= -->
3097<div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
3098</div>
3099<div class="doc_text">
3100<p>The instructions in this category are the conversion instructions (casting)
3101which all take a single operand and a type. They perform various bit conversions
3102on the operand.</p>
3103</div>
3104
3105<!-- _______________________________________________________________________ -->
3106<div class="doc_subsubsection">
3107 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
3108</div>
3109<div class="doc_text">
3110
3111<h5>Syntax:</h5>
3112<pre>
3113 &lt;result&gt; = trunc &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3114</pre>
3115
3116<h5>Overview:</h5>
3117<p>
3118The '<tt>trunc</tt>' instruction truncates its operand to the type <tt>ty2</tt>.
3119</p>
3120
3121<h5>Arguments:</h5>
3122<p>
3123The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
3124be an <a href="#t_integer">integer</a> type, and a type that specifies the size
3125and type of the result, which must be an <a href="#t_integer">integer</a>
3126type. The bit size of <tt>value</tt> must be larger than the bit size of
3127<tt>ty2</tt>. Equal sized types are not allowed.</p>
3128
3129<h5>Semantics:</h5>
3130<p>
3131The '<tt>trunc</tt>' instruction truncates the high order bits in <tt>value</tt>
3132and converts the remaining bits to <tt>ty2</tt>. Since the source size must be
3133larger than the destination size, <tt>trunc</tt> cannot be a <i>no-op cast</i>.
3134It will always truncate bits.</p>
3135
3136<h5>Example:</h5>
3137<pre>
3138 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
3139 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
3140 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
3141</pre>
3142</div>
3143
3144<!-- _______________________________________________________________________ -->
3145<div class="doc_subsubsection">
3146 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
3147</div>
3148<div class="doc_text">
3149
3150<h5>Syntax:</h5>
3151<pre>
3152 &lt;result&gt; = zext &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3153</pre>
3154
3155<h5>Overview:</h5>
3156<p>The '<tt>zext</tt>' instruction zero extends its operand to type
3157<tt>ty2</tt>.</p>
3158
3159
3160<h5>Arguments:</h5>
3161<p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
3162<a href="#t_integer">integer</a> type, and a type to cast it to, which must
3163also be of <a href="#t_integer">integer</a> type. The bit size of the
3164<tt>value</tt> must be smaller than the bit size of the destination type,
3165<tt>ty2</tt>.</p>
3166
3167<h5>Semantics:</h5>
3168<p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
3169bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
3170
3171<p>When zero extending from i1, the result will always be either 0 or 1.</p>
3172
3173<h5>Example:</h5>
3174<pre>
3175 %X = zext i32 257 to i64 <i>; yields i64:257</i>
3176 %Y = zext i1 true to i32 <i>; yields i32:1</i>
3177</pre>
3178</div>
3179
3180<!-- _______________________________________________________________________ -->
3181<div class="doc_subsubsection">
3182 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
3183</div>
3184<div class="doc_text">
3185
3186<h5>Syntax:</h5>
3187<pre>
3188 &lt;result&gt; = sext &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3189</pre>
3190
3191<h5>Overview:</h5>
3192<p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
3193
3194<h5>Arguments:</h5>
3195<p>
3196The '<tt>sext</tt>' instruction takes a value to cast, which must be of
3197<a href="#t_integer">integer</a> type, and a type to cast it to, which must
3198also be of <a href="#t_integer">integer</a> type. The bit size of the
3199<tt>value</tt> must be smaller than the bit size of the destination type,
3200<tt>ty2</tt>.</p>
3201
3202<h5>Semantics:</h5>
3203<p>
3204The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
3205bit (highest order bit) of the <tt>value</tt> until it reaches the bit size of
3206the type <tt>ty2</tt>.</p>
3207
3208<p>When sign extending from i1, the extension always results in -1 or 0.</p>
3209
3210<h5>Example:</h5>
3211<pre>
3212 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
3213 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
3214</pre>
3215</div>
3216
3217<!-- _______________________________________________________________________ -->
3218<div class="doc_subsubsection">
3219 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
3220</div>
3221
3222<div class="doc_text">
3223
3224<h5>Syntax:</h5>
3225
3226<pre>
3227 &lt;result&gt; = fptrunc &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3228</pre>
3229
3230<h5>Overview:</h5>
3231<p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
3232<tt>ty2</tt>.</p>
3233
3234
3235<h5>Arguments:</h5>
3236<p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
3237 point</a> value to cast and a <a href="#t_floating">floating point</a> type to
3238cast it to. The size of <tt>value</tt> must be larger than the size of
3239<tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
3240<i>no-op cast</i>.</p>
3241
3242<h5>Semantics:</h5>
3243<p> The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
3244<a href="#t_floating">floating point</a> type to a smaller
3245<a href="#t_floating">floating point</a> type. If the value cannot fit within
3246the destination type, <tt>ty2</tt>, then the results are undefined.</p>
3247
3248<h5>Example:</h5>
3249<pre>
3250 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
3251 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
3252</pre>
3253</div>
3254
3255<!-- _______________________________________________________________________ -->
3256<div class="doc_subsubsection">
3257 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
3258</div>
3259<div class="doc_text">
3260
3261<h5>Syntax:</h5>
3262<pre>
3263 &lt;result&gt; = fpext &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3264</pre>
3265
3266<h5>Overview:</h5>
3267<p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
3268floating point value.</p>
3269
3270<h5>Arguments:</h5>
3271<p>The '<tt>fpext</tt>' instruction takes a
3272<a href="#t_floating">floating point</a> <tt>value</tt> to cast,
3273and a <a href="#t_floating">floating point</a> type to cast it to. The source
3274type must be smaller than the destination type.</p>
3275
3276<h5>Semantics:</h5>
3277<p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
3278<a href="#t_floating">floating point</a> type to a larger
3279<a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
3280used to make a <i>no-op cast</i> because it always changes bits. Use
3281<tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
3282
3283<h5>Example:</h5>
3284<pre>
3285 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
3286 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
3287</pre>
3288</div>
3289
3290<!-- _______________________________________________________________________ -->
3291<div class="doc_subsubsection">
3292 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
3293</div>
3294<div class="doc_text">
3295
3296<h5>Syntax:</h5>
3297<pre>
Reid Spencere6adee82007-07-31 14:40:14 +0000[diff] [blame]3298 &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]3299</pre>
3300
3301<h5>Overview:</h5>
Reid Spencere6adee82007-07-31 14:40:14 +0000[diff] [blame]3302<p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3303unsigned integer equivalent of type <tt>ty2</tt>.
3304</p>
3305
3306<h5>Arguments:</h5>
Reid Spencere6adee82007-07-31 14:40:14 +0000[diff] [blame]3307<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]3308scalar or vector <a href="#t_floating">floating point</a> value, and a type
3309to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3310type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3311vector integer type with the same number of elements as <tt>ty</tt></p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3312
3313<h5>Semantics:</h5>
Reid Spencere6adee82007-07-31 14:40:14 +0000[diff] [blame]3314<p> The '<tt>fptoui</tt>' instruction converts its
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3315<a href="#t_floating">floating point</a> operand into the nearest (rounding
3316towards zero) unsigned integer value. If the value cannot fit in <tt>ty2</tt>,
3317the results are undefined.</p>
3318
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3319<h5>Example:</h5>
3320<pre>
Reid Spencere6adee82007-07-31 14:40:14 +0000[diff] [blame]3321 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
Chris Lattner681f1e82007-09-22 03:17:52 +0000[diff] [blame]3322 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
Reid Spencere6adee82007-07-31 14:40:14 +0000[diff] [blame]3323 %X = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3324</pre>
3325</div>
3326
3327<!-- _______________________________________________________________________ -->
3328<div class="doc_subsubsection">
3329 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
3330</div>
3331<div class="doc_text">
3332
3333<h5>Syntax:</h5>
3334<pre>
3335 &lt;result&gt; = fptosi &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3336</pre>
3337
3338<h5>Overview:</h5>
3339<p>The '<tt>fptosi</tt>' instruction converts
3340<a href="#t_floating">floating point</a> <tt>value</tt> to type <tt>ty2</tt>.
3341</p>
3342
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3343<h5>Arguments:</h5>
3344<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]3345scalar or vector <a href="#t_floating">floating point</a> value, and a type
3346to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3347type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3348vector integer type with the same number of elements as <tt>ty</tt></p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3349
3350<h5>Semantics:</h5>
3351<p>The '<tt>fptosi</tt>' instruction converts its
3352<a href="#t_floating">floating point</a> operand into the nearest (rounding
3353towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
3354the results are undefined.</p>
3355
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3356<h5>Example:</h5>
3357<pre>
3358 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
Chris Lattner681f1e82007-09-22 03:17:52 +0000[diff] [blame]3359 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3360 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
3361</pre>
3362</div>
3363
3364<!-- _______________________________________________________________________ -->
3365<div class="doc_subsubsection">
3366 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
3367</div>
3368<div class="doc_text">
3369
3370<h5>Syntax:</h5>
3371<pre>
3372 &lt;result&gt; = uitofp &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3373</pre>
3374
3375<h5>Overview:</h5>
3376<p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
3377integer and converts that value to the <tt>ty2</tt> type.</p>
3378
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3379<h5>Arguments:</h5>
Nate Begeman78246ca2007-11-17 03:58:34 +0000[diff] [blame]3380<p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
3381scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3382to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3383type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3384floating point type with the same number of elements as <tt>ty</tt></p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3385
3386<h5>Semantics:</h5>
3387<p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
3388integer quantity and converts it to the corresponding floating point value. If
3389the value cannot fit in the floating point value, the results are undefined.</p>
3390
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3391<h5>Example:</h5>
3392<pre>
3393 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
3394 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
3395</pre>
3396</div>
3397
3398<!-- _______________________________________________________________________ -->
3399<div class="doc_subsubsection">
3400 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
3401</div>
3402<div class="doc_text">
3403
3404<h5>Syntax:</h5>
3405<pre>
3406 &lt;result&gt; = sitofp &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3407</pre>
3408
3409<h5>Overview:</h5>
3410<p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed
3411integer and converts that value to the <tt>ty2</tt> type.</p>
3412
3413<h5>Arguments:</h5>
Nate Begeman78246ca2007-11-17 03:58:34 +0000[diff] [blame]3414<p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
3415scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3416to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3417type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3418floating point type with the same number of elements as <tt>ty</tt></p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3419
3420<h5>Semantics:</h5>
3421<p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed
3422integer quantity and converts it to the corresponding floating point value. If
3423the value cannot fit in the floating point value, the results are undefined.</p>
3424
3425<h5>Example:</h5>
3426<pre>
3427 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
3428 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
3429</pre>
3430</div>
3431
3432<!-- _______________________________________________________________________ -->
3433<div class="doc_subsubsection">
3434 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
3435</div>
3436<div class="doc_text">
3437
3438<h5>Syntax:</h5>
3439<pre>
3440 &lt;result&gt; = ptrtoint &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3441</pre>
3442
3443<h5>Overview:</h5>
3444<p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
3445the integer type <tt>ty2</tt>.</p>
3446
3447<h5>Arguments:</h5>
3448<p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
3449must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
3450<tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.
3451
3452<h5>Semantics:</h5>
3453<p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
3454<tt>ty2</tt> by interpreting the pointer value as an integer and either
3455truncating or zero extending that value to the size of the integer type. If
3456<tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
3457<tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
3458are the same size, then nothing is done (<i>no-op cast</i>) other than a type
3459change.</p>
3460
3461<h5>Example:</h5>
3462<pre>
3463 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
3464 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
3465</pre>
3466</div>
3467
3468<!-- _______________________________________________________________________ -->
3469<div class="doc_subsubsection">
3470 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
3471</div>
3472<div class="doc_text">
3473
3474<h5>Syntax:</h5>
3475<pre>
3476 &lt;result&gt; = inttoptr &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3477</pre>
3478
3479<h5>Overview:</h5>
3480<p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to
3481a pointer type, <tt>ty2</tt>.</p>
3482
3483<h5>Arguments:</h5>
3484<p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
3485value to cast, and a type to cast it to, which must be a
3486<a href="#t_pointer">pointer</a> type.
3487
3488<h5>Semantics:</h5>
3489<p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
3490<tt>ty2</tt> by applying either a zero extension or a truncation depending on
3491the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
3492size of a pointer then a truncation is done. If <tt>value</tt> is smaller than
3493the size of a pointer then a zero extension is done. If they are the same size,
3494nothing is done (<i>no-op cast</i>).</p>
3495
3496<h5>Example:</h5>
3497<pre>
3498 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
3499 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
3500 %Y = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
3501</pre>
3502</div>
3503
3504<!-- _______________________________________________________________________ -->
3505<div class="doc_subsubsection">
3506 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
3507</div>
3508<div class="doc_text">
3509
3510<h5>Syntax:</h5>
3511<pre>
3512 &lt;result&gt; = bitcast &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3513</pre>
3514
3515<h5>Overview:</h5>
3516<p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3517<tt>ty2</tt> without changing any bits.</p>
3518
3519<h5>Arguments:</h5>
3520<p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be
3521a first class value, and a type to cast it to, which must also be a <a
3522 href="#t_firstclass">first class</a> type. The bit sizes of <tt>value</tt>
3523and the destination type, <tt>ty2</tt>, must be identical. If the source
3524type is a pointer, the destination type must also be a pointer.</p>
3525
3526<h5>Semantics:</h5>
3527<p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3528<tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
3529this conversion. The conversion is done as if the <tt>value</tt> had been
3530stored to memory and read back as type <tt>ty2</tt>. Pointer types may only be
3531converted to other pointer types with this instruction. To convert pointers to
3532other types, use the <a href="#i_inttoptr">inttoptr</a> or
3533<a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
3534
3535<h5>Example:</h5>
3536<pre>
3537 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
3538 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
3539 %Z = bitcast <2xint> %V to i64; <i>; yields i64: %V</i>
3540</pre>
3541</div>
3542
3543<!-- ======================================================================= -->
3544<div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
3545<div class="doc_text">
3546<p>The instructions in this category are the "miscellaneous"
3547instructions, which defy better classification.</p>
3548</div>
3549
3550<!-- _______________________________________________________________________ -->
3551<div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
3552</div>
3553<div class="doc_text">
3554<h5>Syntax:</h5>
3555<pre> &lt;result&gt; = icmp &lt;cond&gt; &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {i1}:result</i>
3556</pre>
3557<h5>Overview:</h5>
3558<p>The '<tt>icmp</tt>' instruction returns a boolean value based on comparison
Chris Lattner10368b62008-04-02 00:38:26 +0000[diff] [blame^]3559of its two integer or pointer operands.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3560<h5>Arguments:</h5>
3561<p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
3562the condition code indicating the kind of comparison to perform. It is not
3563a value, just a keyword. The possible condition code are:
3564<ol>
3565 <li><tt>eq</tt>: equal</li>
3566 <li><tt>ne</tt>: not equal </li>
3567 <li><tt>ugt</tt>: unsigned greater than</li>
3568 <li><tt>uge</tt>: unsigned greater or equal</li>
3569 <li><tt>ult</tt>: unsigned less than</li>
3570 <li><tt>ule</tt>: unsigned less or equal</li>
3571 <li><tt>sgt</tt>: signed greater than</li>
3572 <li><tt>sge</tt>: signed greater or equal</li>
3573 <li><tt>slt</tt>: signed less than</li>
3574 <li><tt>sle</tt>: signed less or equal</li>
3575</ol>
3576<p>The remaining two arguments must be <a href="#t_integer">integer</a> or
3577<a href="#t_pointer">pointer</a> typed. They must also be identical types.</p>
3578<h5>Semantics:</h5>
3579<p>The '<tt>icmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3580the condition code given as <tt>cond</tt>. The comparison performed always
3581yields a <a href="#t_primitive">i1</a> result, as follows:
3582<ol>
3583 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
3584 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3585 </li>
3586 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
3587 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3588 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
3589 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3590 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
3591 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3592 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
3593 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3594 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
3595 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3596 <li><tt>sgt</tt>: interprets the operands as signed values and yields
3597 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3598 <li><tt>sge</tt>: interprets the operands as signed values and yields
3599 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3600 <li><tt>slt</tt>: interprets the operands as signed values and yields
3601 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3602 <li><tt>sle</tt>: interprets the operands as signed values and yields
3603 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3604</ol>
3605<p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
3606values are compared as if they were integers.</p>
3607
3608<h5>Example:</h5>
3609<pre> &lt;result&gt; = icmp eq i32 4, 5 <i>; yields: result=false</i>
3610 &lt;result&gt; = icmp ne float* %X, %X <i>; yields: result=false</i>
3611 &lt;result&gt; = icmp ult i16 4, 5 <i>; yields: result=true</i>
3612 &lt;result&gt; = icmp sgt i16 4, 5 <i>; yields: result=false</i>
3613 &lt;result&gt; = icmp ule i16 -4, 5 <i>; yields: result=false</i>
3614 &lt;result&gt; = icmp sge i16 4, 5 <i>; yields: result=false</i>
3615</pre>
3616</div>
3617
3618<!-- _______________________________________________________________________ -->
3619<div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
3620</div>
3621<div class="doc_text">
3622<h5>Syntax:</h5>
3623<pre> &lt;result&gt; = fcmp &lt;cond&gt; &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {i1}:result</i>
3624</pre>
3625<h5>Overview:</h5>
3626<p>The '<tt>fcmp</tt>' instruction returns a boolean value based on comparison
3627of its floating point operands.</p>
3628<h5>Arguments:</h5>
3629<p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
3630the condition code indicating the kind of comparison to perform. It is not
3631a value, just a keyword. The possible condition code are:
3632<ol>
3633 <li><tt>false</tt>: no comparison, always returns false</li>
3634 <li><tt>oeq</tt>: ordered and equal</li>
3635 <li><tt>ogt</tt>: ordered and greater than </li>
3636 <li><tt>oge</tt>: ordered and greater than or equal</li>
3637 <li><tt>olt</tt>: ordered and less than </li>
3638 <li><tt>ole</tt>: ordered and less than or equal</li>
3639 <li><tt>one</tt>: ordered and not equal</li>
3640 <li><tt>ord</tt>: ordered (no nans)</li>
3641 <li><tt>ueq</tt>: unordered or equal</li>
3642 <li><tt>ugt</tt>: unordered or greater than </li>
3643 <li><tt>uge</tt>: unordered or greater than or equal</li>
3644 <li><tt>ult</tt>: unordered or less than </li>
3645 <li><tt>ule</tt>: unordered or less than or equal</li>
3646 <li><tt>une</tt>: unordered or not equal</li>
3647 <li><tt>uno</tt>: unordered (either nans)</li>
3648 <li><tt>true</tt>: no comparison, always returns true</li>
3649</ol>
3650<p><i>Ordered</i> means that neither operand is a QNAN while
3651<i>unordered</i> means that either operand may be a QNAN.</p>
3652<p>The <tt>val1</tt> and <tt>val2</tt> arguments must be
3653<a href="#t_floating">floating point</a> typed. They must have identical
3654types.</p>
3655<h5>Semantics:</h5>
3656<p>The '<tt>fcmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3657the condition code given as <tt>cond</tt>. The comparison performed always
3658yields a <a href="#t_primitive">i1</a> result, as follows:
3659<ol>
3660 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
3661 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3662 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3663 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3664 <tt>var1</tt> is greather than <tt>var2</tt>.</li>
3665 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3666 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3667 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3668 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3669 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3670 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3671 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3672 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3673 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
3674 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
3675 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3676 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
3677 <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3678 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
3679 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3680 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
3681 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3682 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
3683 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3684 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
3685 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3686 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
3687 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
3688</ol>
3689
3690<h5>Example:</h5>
3691<pre> &lt;result&gt; = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
3692 &lt;result&gt; = icmp one float 4.0, 5.0 <i>; yields: result=true</i>
3693 &lt;result&gt; = icmp olt float 4.0, 5.0 <i>; yields: result=true</i>
3694 &lt;result&gt; = icmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
3695</pre>
3696</div>
3697
3698<!-- _______________________________________________________________________ -->
3699<div class="doc_subsubsection"> <a name="i_phi">'<tt>phi</tt>'
3700Instruction</a> </div>
3701<div class="doc_text">
3702<h5>Syntax:</h5>
3703<pre> &lt;result&gt; = phi &lt;ty&gt; [ &lt;val0&gt;, &lt;label0&gt;], ...<br></pre>
3704<h5>Overview:</h5>
3705<p>The '<tt>phi</tt>' instruction is used to implement the &#966; node in
3706the SSA graph representing the function.</p>
3707<h5>Arguments:</h5>
3708<p>The type of the incoming values is specified with the first type
3709field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
3710as arguments, with one pair for each predecessor basic block of the
3711current block. Only values of <a href="#t_firstclass">first class</a>
3712type may be used as the value arguments to the PHI node. Only labels
3713may be used as the label arguments.</p>
3714<p>There must be no non-phi instructions between the start of a basic
3715block and the PHI instructions: i.e. PHI instructions must be first in
3716a basic block.</p>
3717<h5>Semantics:</h5>
3718<p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
3719specified by the pair corresponding to the predecessor basic block that executed
3720just prior to the current block.</p>
3721<h5>Example:</h5>
3722<pre>Loop: ; Infinite loop that counts from 0 on up...<br> %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]<br> %nextindvar = add i32 %indvar, 1<br> br label %Loop<br></pre>
3723</div>
3724
3725<!-- _______________________________________________________________________ -->
3726<div class="doc_subsubsection">
3727 <a name="i_select">'<tt>select</tt>' Instruction</a>
3728</div>
3729
3730<div class="doc_text">
3731
3732<h5>Syntax:</h5>
3733
3734<pre>
3735 &lt;result&gt; = select i1 &lt;cond&gt;, &lt;ty&gt; &lt;val1&gt;, &lt;ty&gt; &lt;val2&gt; <i>; yields ty</i>
3736</pre>
3737
3738<h5>Overview:</h5>
3739
3740<p>
3741The '<tt>select</tt>' instruction is used to choose one value based on a
3742condition, without branching.
3743</p>
3744
3745
3746<h5>Arguments:</h5>
3747
3748<p>
3749The '<tt>select</tt>' instruction requires a boolean value indicating the condition, and two values of the same <a href="#t_firstclass">first class</a> type.
3750</p>
3751
3752<h5>Semantics:</h5>
3753
3754<p>
3755If the boolean condition evaluates to true, the instruction returns the first
3756value argument; otherwise, it returns the second value argument.
3757</p>
3758
3759<h5>Example:</h5>
3760
3761<pre>
3762 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
3763</pre>
3764</div>
3765
3766
3767<!-- _______________________________________________________________________ -->
3768<div class="doc_subsubsection">
3769 <a name="i_call">'<tt>call</tt>' Instruction</a>
3770</div>
3771
3772<div class="doc_text">
3773
3774<h5>Syntax:</h5>
3775<pre>
Nick Lewycky93082fc2007-09-08 13:57:50 +0000[diff] [blame]3776 &lt;result&gt; = [tail] call [<a href="#callingconv">cconv</a>] &lt;ty&gt; [&lt;fnty&gt;*] &lt;fnptrval&gt;(&lt;param list&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3777</pre>
3778
3779<h5>Overview:</h5>
3780
3781<p>The '<tt>call</tt>' instruction represents a simple function call.</p>
3782
3783<h5>Arguments:</h5>
3784
3785<p>This instruction requires several arguments:</p>
3786
3787<ol>
3788 <li>
3789 <p>The optional "tail" marker indicates whether the callee function accesses
3790 any allocas or varargs in the caller. If the "tail" marker is present, the
3791 function call is eligible for tail call optimization. Note that calls may
3792 be marked "tail" even if they do not occur before a <a
3793 href="#i_ret"><tt>ret</tt></a> instruction.
3794 </li>
3795 <li>
3796 <p>The optional "cconv" marker indicates which <a href="#callingconv">calling
3797 convention</a> the call should use. If none is specified, the call defaults
3798 to using C calling conventions.
3799 </li>
3800 <li>
Nick Lewycky93082fc2007-09-08 13:57:50 +0000[diff] [blame]3801 <p>'<tt>ty</tt>': the type of the call instruction itself which is also
3802 the type of the return value. Functions that return no value are marked
3803 <tt><a href="#t_void">void</a></tt>.</p>
3804 </li>
3805 <li>
3806 <p>'<tt>fnty</tt>': shall be the signature of the pointer to function
3807 value being invoked. The argument types must match the types implied by
3808 this signature. This type can be omitted if the function is not varargs
3809 and if the function type does not return a pointer to a function.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3810 </li>
3811 <li>
3812 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
3813 be invoked. In most cases, this is a direct function invocation, but
3814 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
3815 to function value.</p>
3816 </li>
3817 <li>
3818 <p>'<tt>function args</tt>': argument list whose types match the
3819 function signature argument types. All arguments must be of
3820 <a href="#t_firstclass">first class</a> type. If the function signature
3821 indicates the function accepts a variable number of arguments, the extra
3822 arguments can be specified.</p>
3823 </li>
3824</ol>
3825
3826<h5>Semantics:</h5>
3827
3828<p>The '<tt>call</tt>' instruction is used to cause control flow to
3829transfer to a specified function, with its incoming arguments bound to
3830the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
3831instruction in the called function, control flow continues with the
3832instruction after the function call, and the return value of the
Chris Lattner5e893ef2008-03-21 17:24:17 +0000[diff] [blame]3833function is bound to the result argument. If the callee returns multiple
3834values then the return values of the function are only accessible through
3835the '<tt><a href="#i_getresult">getresult</a></tt>' instruction.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3836
3837<h5>Example:</h5>
3838
3839<pre>
Nick Lewycky93082fc2007-09-08 13:57:50 +0000[diff] [blame]3840 %retval = call i32 @test(i32 %argc)
Chris Lattner5e893ef2008-03-21 17:24:17 +0000[diff] [blame]3841 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42) <i>; yields i32</i>
3842 %X = tail call i32 @foo() <i>; yields i32</i>
3843 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
3844 call void %foo(i8 97 signext)
Devang Patela3cc5372008-03-10 20:49:15 +0000[diff] [blame]3845
3846 %struct.A = type { i32, i8 }
Chris Lattner5e893ef2008-03-21 17:24:17 +0000[diff] [blame]3847 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
3848 %gr = getresult %struct.A %r, 0 <i>; yields i32</i>
3849 %gr1 = getresult %struct.A %r, 1 <i>; yields i8</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3850</pre>
3851
3852</div>
3853
3854<!-- _______________________________________________________________________ -->
3855<div class="doc_subsubsection">
3856 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
3857</div>
3858
3859<div class="doc_text">
3860
3861<h5>Syntax:</h5>
3862
3863<pre>
3864 &lt;resultval&gt; = va_arg &lt;va_list*&gt; &lt;arglist&gt;, &lt;argty&gt;
3865</pre>
3866
3867<h5>Overview:</h5>
3868
3869<p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
3870the "variable argument" area of a function call. It is used to implement the
3871<tt>va_arg</tt> macro in C.</p>
3872
3873<h5>Arguments:</h5>
3874
3875<p>This instruction takes a <tt>va_list*</tt> value and the type of
3876the argument. It returns a value of the specified argument type and
3877increments the <tt>va_list</tt> to point to the next argument. The
3878actual type of <tt>va_list</tt> is target specific.</p>
3879
3880<h5>Semantics:</h5>
3881
3882<p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
3883type from the specified <tt>va_list</tt> and causes the
3884<tt>va_list</tt> to point to the next argument. For more information,
3885see the variable argument handling <a href="#int_varargs">Intrinsic
3886Functions</a>.</p>
3887
3888<p>It is legal for this instruction to be called in a function which does not
3889take a variable number of arguments, for example, the <tt>vfprintf</tt>
3890function.</p>
3891
3892<p><tt>va_arg</tt> is an LLVM instruction instead of an <a
3893href="#intrinsics">intrinsic function</a> because it takes a type as an
3894argument.</p>
3895
3896<h5>Example:</h5>
3897
3898<p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
3899
3900</div>
3901
Devang Patela3cc5372008-03-10 20:49:15 +0000[diff] [blame]3902<!-- _______________________________________________________________________ -->
3903<div class="doc_subsubsection">
3904 <a name="i_getresult">'<tt>getresult</tt>' Instruction</a>
3905</div>
3906
3907<div class="doc_text">
3908
3909<h5>Syntax:</h5>
3910<pre>
Chris Lattneree9da3f2008-03-21 17:20:51 +0000[diff] [blame]3911 &lt;resultval&gt; = getresult &lt;type&gt; &lt;retval&gt;, &lt;index&gt;
Devang Patela3cc5372008-03-10 20:49:15 +0000[diff] [blame]3912</pre>
Chris Lattneree9da3f2008-03-21 17:20:51 +0000[diff] [blame]3913
Devang Patela3cc5372008-03-10 20:49:15 +0000[diff] [blame]3914<h5>Overview:</h5>
3915
3916<p> The '<tt>getresult</tt>' instruction is used to extract individual values
Chris Lattneree9da3f2008-03-21 17:20:51 +0000[diff] [blame]3917from a '<tt><a href="#i_call">call</a></tt>'
3918or '<tt><a href="#i_invoke">invoke</a></tt>' instruction that returns multiple
3919results.</p>
Devang Patela3cc5372008-03-10 20:49:15 +0000[diff] [blame]3920
3921<h5>Arguments:</h5>
3922
Chris Lattneree9da3f2008-03-21 17:20:51 +0000[diff] [blame]3923<p>The '<tt>getresult</tt>' instruction takes a call or invoke value as its
3924first argument. The value must have <a href="#t_struct">structure type</a>.
Chris Lattner10368b62008-04-02 00:38:26 +0000[diff] [blame^]3925The second argument is a constant unsigned index value which must be in range for
Chris Lattneree9da3f2008-03-21 17:20:51 +0000[diff] [blame]3926the number of values returned by the call.</p>
Devang Patela3cc5372008-03-10 20:49:15 +0000[diff] [blame]3927
3928<h5>Semantics:</h5>
3929
Chris Lattneree9da3f2008-03-21 17:20:51 +0000[diff] [blame]3930<p>The '<tt>getresult</tt>' instruction extracts the element identified by
3931'<tt>index</tt>' from the aggregate value.</p>
Devang Patela3cc5372008-03-10 20:49:15 +0000[diff] [blame]3932
3933<h5>Example:</h5>
3934
3935<pre>
3936 %struct.A = type { i32, i8 }
3937
3938 %r = call %struct.A @foo()
Chris Lattneree9da3f2008-03-21 17:20:51 +0000[diff] [blame]3939 %gr = getresult %struct.A %r, 0 <i>; yields i32:%gr</i>
3940 %gr1 = getresult %struct.A %r, 1 <i>; yields i8:%gr1</i>
Devang Patela3cc5372008-03-10 20:49:15 +0000[diff] [blame]3941 add i32 %gr, 42
3942 add i8 %gr1, 41
3943</pre>
3944
3945</div>
3946
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3947<!-- *********************************************************************** -->
3948<div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
3949<!-- *********************************************************************** -->
3950
3951<div class="doc_text">
3952
3953<p>LLVM supports the notion of an "intrinsic function". These functions have
3954well known names and semantics and are required to follow certain restrictions.
3955Overall, these intrinsics represent an extension mechanism for the LLVM
3956language that does not require changing all of the transformations in LLVM when
3957adding to the language (or the bitcode reader/writer, the parser, etc...).</p>
3958
3959<p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
3960prefix is reserved in LLVM for intrinsic names; thus, function names may not
3961begin with this prefix. Intrinsic functions must always be external functions:
3962you cannot define the body of intrinsic functions. Intrinsic functions may
3963only be used in call or invoke instructions: it is illegal to take the address
3964of an intrinsic function. Additionally, because intrinsic functions are part
3965of the LLVM language, it is required if any are added that they be documented
3966here.</p>
3967
Chandler Carrutha228e392007-08-04 01:51:18 +0000[diff] [blame]3968<p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
3969a family of functions that perform the same operation but on different data
3970types. Because LLVM can represent over 8 million different integer types,
3971overloading is used commonly to allow an intrinsic function to operate on any
3972integer type. One or more of the argument types or the result type can be
3973overloaded to accept any integer type. Argument types may also be defined as
3974exactly matching a previous argument's type or the result type. This allows an
3975intrinsic function which accepts multiple arguments, but needs all of them to
3976be of the same type, to only be overloaded with respect to a single argument or
3977the result.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3978
Chandler Carrutha228e392007-08-04 01:51:18 +0000[diff] [blame]3979<p>Overloaded intrinsics will have the names of its overloaded argument types
3980encoded into its function name, each preceded by a period. Only those types
3981which are overloaded result in a name suffix. Arguments whose type is matched
3982against another type do not. For example, the <tt>llvm.ctpop</tt> function can
3983take an integer of any width and returns an integer of exactly the same integer
3984width. This leads to a family of functions such as
3985<tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29 %val)</tt>.
3986Only one type, the return type, is overloaded, and only one type suffix is
3987required. Because the argument's type is matched against the return type, it
3988does not require its own name suffix.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]3989
3990<p>To learn how to add an intrinsic function, please see the
3991<a href="ExtendingLLVM.html">Extending LLVM Guide</a>.
3992</p>
3993
3994</div>
3995
3996<!-- ======================================================================= -->
3997<div class="doc_subsection">
3998 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
3999</div>
4000
4001<div class="doc_text">
4002
4003<p>Variable argument support is defined in LLVM with the <a
4004 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
4005intrinsic functions. These functions are related to the similarly
4006named macros defined in the <tt>&lt;stdarg.h&gt;</tt> header file.</p>
4007
4008<p>All of these functions operate on arguments that use a
4009target-specific value type "<tt>va_list</tt>". The LLVM assembly
4010language reference manual does not define what this type is, so all
4011transformations should be prepared to handle these functions regardless of
4012the type used.</p>
4013
4014<p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
4015instruction and the variable argument handling intrinsic functions are
4016used.</p>
4017
4018<div class="doc_code">
4019<pre>
4020define i32 @test(i32 %X, ...) {
4021 ; Initialize variable argument processing
4022 %ap = alloca i8*
4023 %ap2 = bitcast i8** %ap to i8*
4024 call void @llvm.va_start(i8* %ap2)
4025
4026 ; Read a single integer argument
4027 %tmp = va_arg i8** %ap, i32
4028
4029 ; Demonstrate usage of llvm.va_copy and llvm.va_end
4030 %aq = alloca i8*
4031 %aq2 = bitcast i8** %aq to i8*
4032 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
4033 call void @llvm.va_end(i8* %aq2)
4034
4035 ; Stop processing of arguments.
4036 call void @llvm.va_end(i8* %ap2)
4037 ret i32 %tmp
4038}
4039
4040declare void @llvm.va_start(i8*)
4041declare void @llvm.va_copy(i8*, i8*)
4042declare void @llvm.va_end(i8*)
4043</pre>
4044</div>
4045
4046</div>
4047
4048<!-- _______________________________________________________________________ -->
4049<div class="doc_subsubsection">
4050 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
4051</div>
4052
4053
4054<div class="doc_text">
4055<h5>Syntax:</h5>
4056<pre> declare void %llvm.va_start(i8* &lt;arglist&gt;)<br></pre>
4057<h5>Overview:</h5>
4058<P>The '<tt>llvm.va_start</tt>' intrinsic initializes
4059<tt>*&lt;arglist&gt;</tt> for subsequent use by <tt><a
4060href="#i_va_arg">va_arg</a></tt>.</p>
4061
4062<h5>Arguments:</h5>
4063
4064<P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
4065
4066<h5>Semantics:</h5>
4067
4068<P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
4069macro available in C. In a target-dependent way, it initializes the
4070<tt>va_list</tt> element to which the argument points, so that the next call to
4071<tt>va_arg</tt> will produce the first variable argument passed to the function.
4072Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
4073last argument of the function as the compiler can figure that out.</p>
4074
4075</div>
4076
4077<!-- _______________________________________________________________________ -->
4078<div class="doc_subsubsection">
4079 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
4080</div>
4081
4082<div class="doc_text">
4083<h5>Syntax:</h5>
4084<pre> declare void @llvm.va_end(i8* &lt;arglist&gt;)<br></pre>
4085<h5>Overview:</h5>
4086
4087<p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*&lt;arglist&gt;</tt>,
4088which has been initialized previously with <tt><a href="#int_va_start">llvm.va_start</a></tt>
4089or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
4090
4091<h5>Arguments:</h5>
4092
4093<p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
4094
4095<h5>Semantics:</h5>
4096
4097<p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
4098macro available in C. In a target-dependent way, it destroys the
4099<tt>va_list</tt> element to which the argument points. Calls to <a
4100href="#int_va_start"><tt>llvm.va_start</tt></a> and <a href="#int_va_copy">
4101<tt>llvm.va_copy</tt></a> must be matched exactly with calls to
4102<tt>llvm.va_end</tt>.</p>
4103
4104</div>
4105
4106<!-- _______________________________________________________________________ -->
4107<div class="doc_subsubsection">
4108 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
4109</div>
4110
4111<div class="doc_text">
4112
4113<h5>Syntax:</h5>
4114
4115<pre>
4116 declare void @llvm.va_copy(i8* &lt;destarglist&gt;, i8* &lt;srcarglist&gt;)
4117</pre>
4118
4119<h5>Overview:</h5>
4120
4121<p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
4122from the source argument list to the destination argument list.</p>
4123
4124<h5>Arguments:</h5>
4125
4126<p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
4127The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
4128
4129
4130<h5>Semantics:</h5>
4131
4132<p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
4133macro available in C. In a target-dependent way, it copies the source
4134<tt>va_list</tt> element into the destination <tt>va_list</tt> element. This
4135intrinsic is necessary because the <tt><a href="#int_va_start">
4136llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
4137example, memory allocation.</p>
4138
4139</div>
4140
4141<!-- ======================================================================= -->
4142<div class="doc_subsection">
4143 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
4144</div>
4145
4146<div class="doc_text">
4147
4148<p>
4149LLVM support for <a href="GarbageCollection.html">Accurate Garbage
4150Collection</a> requires the implementation and generation of these intrinsics.
4151These intrinsics allow identification of <a href="#int_gcroot">GC roots on the
4152stack</a>, as well as garbage collector implementations that require <a
4153href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a> barriers.
4154Front-ends for type-safe garbage collected languages should generate these
4155intrinsics to make use of the LLVM garbage collectors. For more details, see <a
4156href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
4157</p>
Christopher Lambcfe00962007-12-17 01:00:21 +0000[diff] [blame]4158
4159<p>The garbage collection intrinsics only operate on objects in the generic
4160 address space (address space zero).</p>
4161
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4162</div>
4163
4164<!-- _______________________________________________________________________ -->
4165<div class="doc_subsubsection">
4166 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
4167</div>
4168
4169<div class="doc_text">
4170
4171<h5>Syntax:</h5>
4172
4173<pre>
Chris Lattner38bd5dd2007-09-21 17:30:40 +0000[diff] [blame]4174 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4175</pre>
4176
4177<h5>Overview:</h5>
4178
4179<p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
4180the code generator, and allows some metadata to be associated with it.</p>
4181
4182<h5>Arguments:</h5>
4183
4184<p>The first argument specifies the address of a stack object that contains the
4185root pointer. The second pointer (which must be either a constant or a global
4186value address) contains the meta-data to be associated with the root.</p>
4187
4188<h5>Semantics:</h5>
4189
4190<p>At runtime, a call to this intrinsics stores a null pointer into the "ptrloc"
4191location. At compile-time, the code generator generates information to allow
Gordon Henriksen40393542007-12-25 02:31:26 +0000[diff] [blame]4192the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
4193intrinsic may only be used in a function which <a href="#gc">specifies a GC
4194algorithm</a>.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4195
4196</div>
4197
4198
4199<!-- _______________________________________________________________________ -->
4200<div class="doc_subsubsection">
4201 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
4202</div>
4203
4204<div class="doc_text">
4205
4206<h5>Syntax:</h5>
4207
4208<pre>
Chris Lattner38bd5dd2007-09-21 17:30:40 +0000[diff] [blame]4209 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4210</pre>
4211
4212<h5>Overview:</h5>
4213
4214<p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
4215locations, allowing garbage collector implementations that require read
4216barriers.</p>
4217
4218<h5>Arguments:</h5>
4219
4220<p>The second argument is the address to read from, which should be an address
4221allocated from the garbage collector. The first object is a pointer to the
4222start of the referenced object, if needed by the language runtime (otherwise
4223null).</p>
4224
4225<h5>Semantics:</h5>
4226
4227<p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
4228instruction, but may be replaced with substantially more complex code by the
Gordon Henriksen40393542007-12-25 02:31:26 +0000[diff] [blame]4229garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
4230may only be used in a function which <a href="#gc">specifies a GC
4231algorithm</a>.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4232
4233</div>
4234
4235
4236<!-- _______________________________________________________________________ -->
4237<div class="doc_subsubsection">
4238 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
4239</div>
4240
4241<div class="doc_text">
4242
4243<h5>Syntax:</h5>
4244
4245<pre>
Chris Lattner38bd5dd2007-09-21 17:30:40 +0000[diff] [blame]4246 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4247</pre>
4248
4249<h5>Overview:</h5>
4250
4251<p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
4252locations, allowing garbage collector implementations that require write
4253barriers (such as generational or reference counting collectors).</p>
4254
4255<h5>Arguments:</h5>
4256
4257<p>The first argument is the reference to store, the second is the start of the
4258object to store it to, and the third is the address of the field of Obj to
4259store to. If the runtime does not require a pointer to the object, Obj may be
4260null.</p>
4261
4262<h5>Semantics:</h5>
4263
4264<p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
4265instruction, but may be replaced with substantially more complex code by the
Gordon Henriksen40393542007-12-25 02:31:26 +0000[diff] [blame]4266garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
4267may only be used in a function which <a href="#gc">specifies a GC
4268algorithm</a>.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4269
4270</div>
4271
4272
4273
4274<!-- ======================================================================= -->
4275<div class="doc_subsection">
4276 <a name="int_codegen">Code Generator Intrinsics</a>
4277</div>
4278
4279<div class="doc_text">
4280<p>
4281These intrinsics are provided by LLVM to expose special features that may only
4282be implemented with code generator support.
4283</p>
4284
4285</div>
4286
4287<!-- _______________________________________________________________________ -->
4288<div class="doc_subsubsection">
4289 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
4290</div>
4291
4292<div class="doc_text">
4293
4294<h5>Syntax:</h5>
4295<pre>
4296 declare i8 *@llvm.returnaddress(i32 &lt;level&gt;)
4297</pre>
4298
4299<h5>Overview:</h5>
4300
4301<p>
4302The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
4303target-specific value indicating the return address of the current function
4304or one of its callers.
4305</p>
4306
4307<h5>Arguments:</h5>
4308
4309<p>
4310The argument to this intrinsic indicates which function to return the address
4311for. Zero indicates the calling function, one indicates its caller, etc. The
4312argument is <b>required</b> to be a constant integer value.
4313</p>
4314
4315<h5>Semantics:</h5>
4316
4317<p>
4318The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
4319the return address of the specified call frame, or zero if it cannot be
4320identified. The value returned by this intrinsic is likely to be incorrect or 0
4321for arguments other than zero, so it should only be used for debugging purposes.
4322</p>
4323
4324<p>
4325Note that calling this intrinsic does not prevent function inlining or other
4326aggressive transformations, so the value returned may not be that of the obvious
4327source-language caller.
4328</p>
4329</div>
4330
4331
4332<!-- _______________________________________________________________________ -->
4333<div class="doc_subsubsection">
4334 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
4335</div>
4336
4337<div class="doc_text">
4338
4339<h5>Syntax:</h5>
4340<pre>
Chris Lattner38bd5dd2007-09-21 17:30:40 +0000[diff] [blame]4341 declare i8 *@llvm.frameaddress(i32 &lt;level&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4342</pre>
4343
4344<h5>Overview:</h5>
4345
4346<p>
4347The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
4348target-specific frame pointer value for the specified stack frame.
4349</p>
4350
4351<h5>Arguments:</h5>
4352
4353<p>
4354The argument to this intrinsic indicates which function to return the frame
4355pointer for. Zero indicates the calling function, one indicates its caller,
4356etc. The argument is <b>required</b> to be a constant integer value.
4357</p>
4358
4359<h5>Semantics:</h5>
4360
4361<p>
4362The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
4363the frame address of the specified call frame, or zero if it cannot be
4364identified. The value returned by this intrinsic is likely to be incorrect or 0
4365for arguments other than zero, so it should only be used for debugging purposes.
4366</p>
4367
4368<p>
4369Note that calling this intrinsic does not prevent function inlining or other
4370aggressive transformations, so the value returned may not be that of the obvious
4371source-language caller.
4372</p>
4373</div>
4374
4375<!-- _______________________________________________________________________ -->
4376<div class="doc_subsubsection">
4377 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
4378</div>
4379
4380<div class="doc_text">
4381
4382<h5>Syntax:</h5>
4383<pre>
Chris Lattner38bd5dd2007-09-21 17:30:40 +0000[diff] [blame]4384 declare i8 *@llvm.stacksave()
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4385</pre>
4386
4387<h5>Overview:</h5>
4388
4389<p>
4390The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
4391the function stack, for use with <a href="#int_stackrestore">
4392<tt>llvm.stackrestore</tt></a>. This is useful for implementing language
4393features like scoped automatic variable sized arrays in C99.
4394</p>
4395
4396<h5>Semantics:</h5>
4397
4398<p>
4399This intrinsic returns a opaque pointer value that can be passed to <a
4400href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
4401<tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
4402<tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
4403state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
4404practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
4405that were allocated after the <tt>llvm.stacksave</tt> was executed.
4406</p>
4407
4408</div>
4409
4410<!-- _______________________________________________________________________ -->
4411<div class="doc_subsubsection">
4412 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
4413</div>
4414
4415<div class="doc_text">
4416
4417<h5>Syntax:</h5>
4418<pre>
4419 declare void @llvm.stackrestore(i8 * %ptr)
4420</pre>
4421
4422<h5>Overview:</h5>
4423
4424<p>
4425The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
4426the function stack to the state it was in when the corresponding <a
4427href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
4428useful for implementing language features like scoped automatic variable sized
4429arrays in C99.
4430</p>
4431
4432<h5>Semantics:</h5>
4433
4434<p>
4435See the description for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.
4436</p>
4437
4438</div>
4439
4440
4441<!-- _______________________________________________________________________ -->
4442<div class="doc_subsubsection">
4443 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
4444</div>
4445
4446<div class="doc_text">
4447
4448<h5>Syntax:</h5>
4449<pre>
Chris Lattner38bd5dd2007-09-21 17:30:40 +0000[diff] [blame]4450 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]4451</pre>
4452
4453<h5>Overview:</h5>
4454
4455
4456<p>
4457The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
4458a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
4459no
4460effect on the behavior of the program but can change its performance
4461characteristics.
4462</p>
4463
4464<h5>Arguments:</h5>
4465
4466<p>
4467<tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
4468determining if the fetch should be for a read (0) or write (1), and
4469<tt>locality</tt> is a temporal locality specifier ranging from (0) - no
4470locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
4471<tt>locality</tt> arguments must be constant integers.
4472</p>
4473
4474<h5>Semantics:</h5>
4475
4476<p>
4477This intrinsic does not modify the behavior of the program. In particular,
4478prefetches cannot trap and do not produce a value. On targets that support this
4479intrinsic, the prefetch can provide hints to the processor cache for better
4480performance.
4481</p>
4482
4483</div>
4484
4485<!-- _______________________________________________________________________ -->
4486<div class="doc_subsubsection">
4487 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
4488</div>
4489
4490<div class="doc_text">
4491
4492<h5>Syntax:</h5>
4493<pre>
Chris Lattner38bd5dd2007-09-21 17:30:40 +0000[diff] [blame]4494 declare void @llvm.pcmarker(i32 &lt;id&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4495</pre>
4496
4497<h5>Overview:</h5>
4498
4499
4500<p>
4501The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
4502(PC) in a region of
4503code to simulators and other tools. The method is target specific, but it is
4504expected that the marker will use exported symbols to transmit the PC of the marker.
4505The marker makes no guarantees that it will remain with any specific instruction
4506after optimizations. It is possible that the presence of a marker will inhibit
4507optimizations. The intended use is to be inserted after optimizations to allow
4508correlations of simulation runs.
4509</p>
4510
4511<h5>Arguments:</h5>
4512
4513<p>
4514<tt>id</tt> is a numerical id identifying the marker.
4515</p>
4516
4517<h5>Semantics:</h5>
4518
4519<p>
4520This intrinsic does not modify the behavior of the program. Backends that do not
4521support this intrinisic may ignore it.
4522</p>
4523
4524</div>
4525
4526<!-- _______________________________________________________________________ -->
4527<div class="doc_subsubsection">
4528 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
4529</div>
4530
4531<div class="doc_text">
4532
4533<h5>Syntax:</h5>
4534<pre>
4535 declare i64 @llvm.readcyclecounter( )
4536</pre>
4537
4538<h5>Overview:</h5>
4539
4540
4541<p>
4542The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
4543counter register (or similar low latency, high accuracy clocks) on those targets
4544that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
4545As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
4546should only be used for small timings.
4547</p>
4548
4549<h5>Semantics:</h5>
4550
4551<p>
4552When directly supported, reading the cycle counter should not modify any memory.
4553Implementations are allowed to either return a application specific value or a
4554system wide value. On backends without support, this is lowered to a constant 0.
4555</p>
4556
4557</div>
4558
4559<!-- ======================================================================= -->
4560<div class="doc_subsection">
4561 <a name="int_libc">Standard C Library Intrinsics</a>
4562</div>
4563
4564<div class="doc_text">
4565<p>
4566LLVM provides intrinsics for a few important standard C library functions.
4567These intrinsics allow source-language front-ends to pass information about the
4568alignment of the pointer arguments to the code generator, providing opportunity
4569for more efficient code generation.
4570</p>
4571
4572</div>
4573
4574<!-- _______________________________________________________________________ -->
4575<div class="doc_subsubsection">
4576 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
4577</div>
4578
4579<div class="doc_text">
4580
4581<h5>Syntax:</h5>
4582<pre>
4583 declare void @llvm.memcpy.i32(i8 * &lt;dest&gt;, i8 * &lt;src&gt;,
4584 i32 &lt;len&gt;, i32 &lt;align&gt;)
4585 declare void @llvm.memcpy.i64(i8 * &lt;dest&gt;, i8 * &lt;src&gt;,
4586 i64 &lt;len&gt;, i32 &lt;align&gt;)
4587</pre>
4588
4589<h5>Overview:</h5>
4590
4591<p>
4592The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4593location to the destination location.
4594</p>
4595
4596<p>
4597Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
4598intrinsics do not return a value, and takes an extra alignment argument.
4599</p>
4600
4601<h5>Arguments:</h5>
4602
4603<p>
4604The first argument is a pointer to the destination, the second is a pointer to
4605the source. The third argument is an integer argument
4606specifying the number of bytes to copy, and the fourth argument is the alignment
4607of the source and destination locations.
4608</p>
4609
4610<p>
4611If the call to this intrinisic has an alignment value that is not 0 or 1, then
4612the caller guarantees that both the source and destination pointers are aligned
4613to that boundary.
4614</p>
4615
4616<h5>Semantics:</h5>
4617
4618<p>
4619The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4620location to the destination location, which are not allowed to overlap. It
4621copies "len" bytes of memory over. If the argument is known to be aligned to
4622some boundary, this can be specified as the fourth argument, otherwise it should
4623be set to 0 or 1.
4624</p>
4625</div>
4626
4627
4628<!-- _______________________________________________________________________ -->
4629<div class="doc_subsubsection">
4630 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
4631</div>
4632
4633<div class="doc_text">
4634
4635<h5>Syntax:</h5>
4636<pre>
4637 declare void @llvm.memmove.i32(i8 * &lt;dest&gt;, i8 * &lt;src&gt;,
4638 i32 &lt;len&gt;, i32 &lt;align&gt;)
4639 declare void @llvm.memmove.i64(i8 * &lt;dest&gt;, i8 * &lt;src&gt;,
4640 i64 &lt;len&gt;, i32 &lt;align&gt;)
4641</pre>
4642
4643<h5>Overview:</h5>
4644
4645<p>
4646The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
4647location to the destination location. It is similar to the
Chris Lattnerdba16ea2008-01-06 19:51:52 +0000[diff] [blame]4648'<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to overlap.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4649</p>
4650
4651<p>
4652Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
4653intrinsics do not return a value, and takes an extra alignment argument.
4654</p>
4655
4656<h5>Arguments:</h5>
4657
4658<p>
4659The first argument is a pointer to the destination, the second is a pointer to
4660the source. The third argument is an integer argument
4661specifying the number of bytes to copy, and the fourth argument is the alignment
4662of the source and destination locations.
4663</p>
4664
4665<p>
4666If the call to this intrinisic has an alignment value that is not 0 or 1, then
4667the caller guarantees that the source and destination pointers are aligned to
4668that boundary.
4669</p>
4670
4671<h5>Semantics:</h5>
4672
4673<p>
4674The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
4675location to the destination location, which may overlap. It
4676copies "len" bytes of memory over. If the argument is known to be aligned to
4677some boundary, this can be specified as the fourth argument, otherwise it should
4678be set to 0 or 1.
4679</p>
4680</div>
4681
4682
4683<!-- _______________________________________________________________________ -->
4684<div class="doc_subsubsection">
4685 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
4686</div>
4687
4688<div class="doc_text">
4689
4690<h5>Syntax:</h5>
4691<pre>
4692 declare void @llvm.memset.i32(i8 * &lt;dest&gt;, i8 &lt;val&gt;,
4693 i32 &lt;len&gt;, i32 &lt;align&gt;)
4694 declare void @llvm.memset.i64(i8 * &lt;dest&gt;, i8 &lt;val&gt;,
4695 i64 &lt;len&gt;, i32 &lt;align&gt;)
4696</pre>
4697
4698<h5>Overview:</h5>
4699
4700<p>
4701The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
4702byte value.
4703</p>
4704
4705<p>
4706Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
4707does not return a value, and takes an extra alignment argument.
4708</p>
4709
4710<h5>Arguments:</h5>
4711
4712<p>
4713The first argument is a pointer to the destination to fill, the second is the
4714byte value to fill it with, the third argument is an integer
4715argument specifying the number of bytes to fill, and the fourth argument is the
4716known alignment of destination location.
4717</p>
4718
4719<p>
4720If the call to this intrinisic has an alignment value that is not 0 or 1, then
4721the caller guarantees that the destination pointer is aligned to that boundary.
4722</p>
4723
4724<h5>Semantics:</h5>
4725
4726<p>
4727The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
4728the
4729destination location. If the argument is known to be aligned to some boundary,
4730this can be specified as the fourth argument, otherwise it should be set to 0 or
47311.
4732</p>
4733</div>
4734
4735
4736<!-- _______________________________________________________________________ -->
4737<div class="doc_subsubsection">
4738 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
4739</div>
4740
4741<div class="doc_text">
4742
4743<h5>Syntax:</h5>
Dale Johannesenf9adbb62007-10-02 17:47:38 +0000[diff] [blame]4744<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]4745floating point or vector of floating point type. Not all targets support all
4746types however.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4747<pre>
Dale Johannesenf9adbb62007-10-02 17:47:38 +0000[diff] [blame]4748 declare float @llvm.sqrt.f32(float %Val)
4749 declare double @llvm.sqrt.f64(double %Val)
4750 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
4751 declare fp128 @llvm.sqrt.f128(fp128 %Val)
4752 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4753</pre>
4754
4755<h5>Overview:</h5>
4756
4757<p>
4758The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
Dan Gohman361079c2007-10-15 20:30:11 +0000[diff] [blame]4759returning the same value as the libm '<tt>sqrt</tt>' functions would. Unlike
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4760<tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
Chris Lattnerf914a002008-01-29 07:00:44 +0000[diff] [blame]4761negative numbers other than -0.0 (which allows for better optimization, because
4762there is no need to worry about errno being set). <tt>llvm.sqrt(-0.0)</tt> is
4763defined to return -0.0 like IEEE sqrt.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4764</p>
4765
4766<h5>Arguments:</h5>
4767
4768<p>
4769The argument and return value are floating point numbers of the same type.
4770</p>
4771
4772<h5>Semantics:</h5>
4773
4774<p>
4775This function returns the sqrt of the specified operand if it is a nonnegative
4776floating point number.
4777</p>
4778</div>
4779
4780<!-- _______________________________________________________________________ -->
4781<div class="doc_subsubsection">
4782 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
4783</div>
4784
4785<div class="doc_text">
4786
4787<h5>Syntax:</h5>
Dale Johannesenf9adbb62007-10-02 17:47:38 +0000[diff] [blame]4788<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]4789floating point or vector of floating point type. Not all targets support all
4790types however.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4791<pre>
Dale Johannesenf9adbb62007-10-02 17:47:38 +0000[diff] [blame]4792 declare float @llvm.powi.f32(float %Val, i32 %power)
4793 declare double @llvm.powi.f64(double %Val, i32 %power)
4794 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
4795 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
4796 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4797</pre>
4798
4799<h5>Overview:</h5>
4800
4801<p>
4802The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
4803specified (positive or negative) power. The order of evaluation of
Dan Gohman361079c2007-10-15 20:30:11 +0000[diff] [blame]4804multiplications is not defined. When a vector of floating point type is
4805used, the second argument remains a scalar integer value.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4806</p>
4807
4808<h5>Arguments:</h5>
4809
4810<p>
4811The second argument is an integer power, and the first is a value to raise to
4812that power.
4813</p>
4814
4815<h5>Semantics:</h5>
4816
4817<p>
4818This function returns the first value raised to the second power with an
4819unspecified sequence of rounding operations.</p>
4820</div>
4821
Dan Gohman361079c2007-10-15 20:30:11 +0000[diff] [blame]4822<!-- _______________________________________________________________________ -->
4823<div class="doc_subsubsection">
4824 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
4825</div>
4826
4827<div class="doc_text">
4828
4829<h5>Syntax:</h5>
4830<p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
4831floating point or vector of floating point type. Not all targets support all
4832types however.
4833<pre>
4834 declare float @llvm.sin.f32(float %Val)
4835 declare double @llvm.sin.f64(double %Val)
4836 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
4837 declare fp128 @llvm.sin.f128(fp128 %Val)
4838 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
4839</pre>
4840
4841<h5>Overview:</h5>
4842
4843<p>
4844The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.
4845</p>
4846
4847<h5>Arguments:</h5>
4848
4849<p>
4850The argument and return value are floating point numbers of the same type.
4851</p>
4852
4853<h5>Semantics:</h5>
4854
4855<p>
4856This function returns the sine of the specified operand, returning the
4857same values as the libm <tt>sin</tt> functions would, and handles error
Dan Gohmaneaba92e2007-10-17 18:05:13 +0000[diff] [blame]4858conditions in the same way.</p>
Dan Gohman361079c2007-10-15 20:30:11 +0000[diff] [blame]4859</div>
4860
4861<!-- _______________________________________________________________________ -->
4862<div class="doc_subsubsection">
4863 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
4864</div>
4865
4866<div class="doc_text">
4867
4868<h5>Syntax:</h5>
4869<p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
4870floating point or vector of floating point type. Not all targets support all
4871types however.
4872<pre>
4873 declare float @llvm.cos.f32(float %Val)
4874 declare double @llvm.cos.f64(double %Val)
4875 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
4876 declare fp128 @llvm.cos.f128(fp128 %Val)
4877 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
4878</pre>
4879
4880<h5>Overview:</h5>
4881
4882<p>
4883The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.
4884</p>
4885
4886<h5>Arguments:</h5>
4887
4888<p>
4889The argument and return value are floating point numbers of the same type.
4890</p>
4891
4892<h5>Semantics:</h5>
4893
4894<p>
4895This function returns the cosine of the specified operand, returning the
4896same values as the libm <tt>cos</tt> functions would, and handles error
Dan Gohmaneaba92e2007-10-17 18:05:13 +0000[diff] [blame]4897conditions in the same way.</p>
Dan Gohman361079c2007-10-15 20:30:11 +0000[diff] [blame]4898</div>
4899
4900<!-- _______________________________________________________________________ -->
4901<div class="doc_subsubsection">
4902 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
4903</div>
4904
4905<div class="doc_text">
4906
4907<h5>Syntax:</h5>
4908<p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
4909floating point or vector of floating point type. Not all targets support all
4910types however.
4911<pre>
4912 declare float @llvm.pow.f32(float %Val, float %Power)
4913 declare double @llvm.pow.f64(double %Val, double %Power)
4914 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
4915 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
4916 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
4917</pre>
4918
4919<h5>Overview:</h5>
4920
4921<p>
4922The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
4923specified (positive or negative) power.
4924</p>
4925
4926<h5>Arguments:</h5>
4927
4928<p>
4929The second argument is a floating point power, and the first is a value to
4930raise to that power.
4931</p>
4932
4933<h5>Semantics:</h5>
4934
4935<p>
4936This function returns the first value raised to the second power,
4937returning the
4938same values as the libm <tt>pow</tt> functions would, and handles error
Dan Gohmaneaba92e2007-10-17 18:05:13 +0000[diff] [blame]4939conditions in the same way.</p>
Dan Gohman361079c2007-10-15 20:30:11 +0000[diff] [blame]4940</div>
4941
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4942
4943<!-- ======================================================================= -->
4944<div class="doc_subsection">
4945 <a name="int_manip">Bit Manipulation Intrinsics</a>
4946</div>
4947
4948<div class="doc_text">
4949<p>
4950LLVM provides intrinsics for a few important bit manipulation operations.
4951These allow efficient code generation for some algorithms.
4952</p>
4953
4954</div>
4955
4956<!-- _______________________________________________________________________ -->
4957<div class="doc_subsubsection">
4958 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
4959</div>
4960
4961<div class="doc_text">
4962
4963<h5>Syntax:</h5>
4964<p>This is an overloaded intrinsic function. You can use bswap on any integer
Chandler Carrutha228e392007-08-04 01:51:18 +0000[diff] [blame]4965type that is an even number of bytes (i.e. BitWidth % 16 == 0).
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4966<pre>
Chandler Carrutha228e392007-08-04 01:51:18 +0000[diff] [blame]4967 declare i16 @llvm.bswap.i16(i16 &lt;id&gt;)
4968 declare i32 @llvm.bswap.i32(i32 &lt;id&gt;)
4969 declare i64 @llvm.bswap.i64(i64 &lt;id&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4970</pre>
4971
4972<h5>Overview:</h5>
4973
4974<p>
4975The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
4976values with an even number of bytes (positive multiple of 16 bits). These are
4977useful for performing operations on data that is not in the target's native
4978byte order.
4979</p>
4980
4981<h5>Semantics:</h5>
4982
4983<p>
Chandler Carrutha228e392007-08-04 01:51:18 +0000[diff] [blame]4984The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4985and low byte of the input i16 swapped. Similarly, the <tt>llvm.bswap.i32</tt>
4986intrinsic returns an i32 value that has the four bytes of the input i32
4987swapped, 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]4988i32 will have its bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i48</tt>,
4989<tt>llvm.bswap.i64</tt> and other intrinsics extend this concept to
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]4990additional even-byte lengths (6 bytes, 8 bytes and more, respectively).
4991</p>
4992
4993</div>
4994
4995<!-- _______________________________________________________________________ -->
4996<div class="doc_subsubsection">
4997 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
4998</div>
4999
5000<div class="doc_text">
5001
5002<h5>Syntax:</h5>
5003<p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
5004width. Not all targets support all bit widths however.
5005<pre>
Chandler Carrutha228e392007-08-04 01:51:18 +0000[diff] [blame]5006 declare i8 @llvm.ctpop.i8 (i8 &lt;src&gt;)
5007 declare i16 @llvm.ctpop.i16(i16 &lt;src&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5008 declare i32 @llvm.ctpop.i32(i32 &lt;src&gt;)
Chandler Carrutha228e392007-08-04 01:51:18 +0000[diff] [blame]5009 declare i64 @llvm.ctpop.i64(i64 &lt;src&gt;)
5010 declare i256 @llvm.ctpop.i256(i256 &lt;src&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5011</pre>
5012
5013<h5>Overview:</h5>
5014
5015<p>
5016The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
5017value.
5018</p>
5019
5020<h5>Arguments:</h5>
5021
5022<p>
5023The only argument is the value to be counted. The argument may be of any
5024integer type. The return type must match the argument type.
5025</p>
5026
5027<h5>Semantics:</h5>
5028
5029<p>
5030The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
5031</p>
5032</div>
5033
5034<!-- _______________________________________________________________________ -->
5035<div class="doc_subsubsection">
5036 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
5037</div>
5038
5039<div class="doc_text">
5040
5041<h5>Syntax:</h5>
5042<p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
5043integer bit width. Not all targets support all bit widths however.
5044<pre>
Chandler Carrutha228e392007-08-04 01:51:18 +0000[diff] [blame]5045 declare i8 @llvm.ctlz.i8 (i8 &lt;src&gt;)
5046 declare i16 @llvm.ctlz.i16(i16 &lt;src&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5047 declare i32 @llvm.ctlz.i32(i32 &lt;src&gt;)
Chandler Carrutha228e392007-08-04 01:51:18 +0000[diff] [blame]5048 declare i64 @llvm.ctlz.i64(i64 &lt;src&gt;)
5049 declare i256 @llvm.ctlz.i256(i256 &lt;src&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5050</pre>
5051
5052<h5>Overview:</h5>
5053
5054<p>
5055The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
5056leading zeros in a variable.
5057</p>
5058
5059<h5>Arguments:</h5>
5060
5061<p>
5062The only argument is the value to be counted. The argument may be of any
5063integer type. The return type must match the argument type.
5064</p>
5065
5066<h5>Semantics:</h5>
5067
5068<p>
5069The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
5070in a variable. If the src == 0 then the result is the size in bits of the type
5071of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.
5072</p>
5073</div>
5074
5075
5076
5077<!-- _______________________________________________________________________ -->
5078<div class="doc_subsubsection">
5079 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
5080</div>
5081
5082<div class="doc_text">
5083
5084<h5>Syntax:</h5>
5085<p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
5086integer bit width. Not all targets support all bit widths however.
5087<pre>
Chandler Carrutha228e392007-08-04 01:51:18 +0000[diff] [blame]5088 declare i8 @llvm.cttz.i8 (i8 &lt;src&gt;)
5089 declare i16 @llvm.cttz.i16(i16 &lt;src&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5090 declare i32 @llvm.cttz.i32(i32 &lt;src&gt;)
Chandler Carrutha228e392007-08-04 01:51:18 +0000[diff] [blame]5091 declare i64 @llvm.cttz.i64(i64 &lt;src&gt;)
5092 declare i256 @llvm.cttz.i256(i256 &lt;src&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5093</pre>
5094
5095<h5>Overview:</h5>
5096
5097<p>
5098The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
5099trailing zeros.
5100</p>
5101
5102<h5>Arguments:</h5>
5103
5104<p>
5105The only argument is the value to be counted. The argument may be of any
5106integer type. The return type must match the argument type.
5107</p>
5108
5109<h5>Semantics:</h5>
5110
5111<p>
5112The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
5113in a variable. If the src == 0 then the result is the size in bits of the type
5114of src. For example, <tt>llvm.cttz(2) = 1</tt>.
5115</p>
5116</div>
5117
5118<!-- _______________________________________________________________________ -->
5119<div class="doc_subsubsection">
5120 <a name="int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic</a>
5121</div>
5122
5123<div class="doc_text">
5124
5125<h5>Syntax:</h5>
5126<p>This is an overloaded intrinsic. You can use <tt>llvm.part.select</tt>
5127on any integer bit width.
5128<pre>
Chandler Carrutha228e392007-08-04 01:51:18 +0000[diff] [blame]5129 declare i17 @llvm.part.select.i17 (i17 %val, i32 %loBit, i32 %hiBit)
5130 declare i29 @llvm.part.select.i29 (i29 %val, i32 %loBit, i32 %hiBit)
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5131</pre>
5132
5133<h5>Overview:</h5>
5134<p>The '<tt>llvm.part.select</tt>' family of intrinsic functions selects a
5135range of bits from an integer value and returns them in the same bit width as
5136the original value.</p>
5137
5138<h5>Arguments:</h5>
5139<p>The first argument, <tt>%val</tt> and the result may be integer types of
5140any bit width but they must have the same bit width. The second and third
5141arguments must be <tt>i32</tt> type since they specify only a bit index.</p>
5142
5143<h5>Semantics:</h5>
5144<p>The operation of the '<tt>llvm.part.select</tt>' intrinsic has two modes
5145of operation: forwards and reverse. If <tt>%loBit</tt> is greater than
5146<tt>%hiBits</tt> then the intrinsic operates in reverse mode. Otherwise it
5147operates in forward mode.</p>
5148<p>In forward mode, this intrinsic is the equivalent of shifting <tt>%val</tt>
5149right by <tt>%loBit</tt> bits and then ANDing it with a mask with
5150only the <tt>%hiBit - %loBit</tt> bits set, as follows:</p>
5151<ol>
5152 <li>The <tt>%val</tt> is shifted right (LSHR) by the number of bits specified
5153 by <tt>%loBits</tt>. This normalizes the value to the low order bits.</li>
5154 <li>The <tt>%loBits</tt> value is subtracted from the <tt>%hiBits</tt> value
5155 to determine the number of bits to retain.</li>
5156 <li>A mask of the retained bits is created by shifting a -1 value.</li>
5157 <li>The mask is ANDed with <tt>%val</tt> to produce the result.
5158</ol>
5159<p>In reverse mode, a similar computation is made except that the bits are
5160returned in the reverse order. So, for example, if <tt>X</tt> has the value
5161<tt>i16 0x0ACF (101011001111)</tt> and we apply
5162<tt>part.select(i16 X, 8, 3)</tt> to it, we get back the value
5163<tt>i16 0x0026 (000000100110)</tt>.</p>
5164</div>
5165
5166<div class="doc_subsubsection">
5167 <a name="int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic</a>
5168</div>
5169
5170<div class="doc_text">
5171
5172<h5>Syntax:</h5>
5173<p>This is an overloaded intrinsic. You can use <tt>llvm.part.set</tt>
5174on any integer bit width.
5175<pre>
Chandler Carrutha228e392007-08-04 01:51:18 +0000[diff] [blame]5176 declare i17 @llvm.part.set.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
5177 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]5178</pre>
5179
5180<h5>Overview:</h5>
5181<p>The '<tt>llvm.part.set</tt>' family of intrinsic functions replaces a range
5182of bits in an integer value with another integer value. It returns the integer
5183with the replaced bits.</p>
5184
5185<h5>Arguments:</h5>
5186<p>The first argument, <tt>%val</tt> and the result may be integer types of
5187any bit width but they must have the same bit width. <tt>%val</tt> is the value
5188whose bits will be replaced. The second argument, <tt>%repl</tt> may be an
5189integer of any bit width. The third and fourth arguments must be <tt>i32</tt>
5190type since they specify only a bit index.</p>
5191
5192<h5>Semantics:</h5>
5193<p>The operation of the '<tt>llvm.part.set</tt>' intrinsic has two modes
5194of operation: forwards and reverse. If <tt>%lo</tt> is greater than
5195<tt>%hi</tt> then the intrinsic operates in reverse mode. Otherwise it
5196operates in forward mode.</p>
5197<p>For both modes, the <tt>%repl</tt> value is prepared for use by either
5198truncating it down to the size of the replacement area or zero extending it
5199up to that size.</p>
5200<p>In forward mode, the bits between <tt>%lo</tt> and <tt>%hi</tt> (inclusive)
5201are replaced with corresponding bits from <tt>%repl</tt>. That is the 0th bit
5202in <tt>%repl</tt> replaces the <tt>%lo</tt>th bit in <tt>%val</tt> and etc. up
5203to the <tt>%hi</tt>th bit.
5204<p>In reverse mode, a similar computation is made except that the bits are
5205reversed. That is, the <tt>0</tt>th bit in <tt>%repl</tt> replaces the
5206<tt>%hi</tt> bit in <tt>%val</tt> and etc. down to the <tt>%lo</tt>th bit.
5207<h5>Examples:</h5>
5208<pre>
5209 llvm.part.set(0xFFFF, 0, 4, 7) -&gt; 0xFF0F
5210 llvm.part.set(0xFFFF, 0, 7, 4) -&gt; 0xFF0F
5211 llvm.part.set(0xFFFF, 1, 7, 4) -&gt; 0xFF8F
5212 llvm.part.set(0xFFFF, F, 8, 3) -&gt; 0xFFE7
5213 llvm.part.set(0xFFFF, 0, 3, 8) -&gt; 0xFE07
5214</pre>
5215</div>
5216
5217<!-- ======================================================================= -->
5218<div class="doc_subsection">
5219 <a name="int_debugger">Debugger Intrinsics</a>
5220</div>
5221
5222<div class="doc_text">
5223<p>
5224The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
5225are described in the <a
5226href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
5227Debugging</a> document.
5228</p>
5229</div>
5230
5231
5232<!-- ======================================================================= -->
5233<div class="doc_subsection">
5234 <a name="int_eh">Exception Handling Intrinsics</a>
5235</div>
5236
5237<div class="doc_text">
5238<p> The LLVM exception handling intrinsics (which all start with
5239<tt>llvm.eh.</tt> prefix), are described in the <a
5240href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
5241Handling</a> document. </p>
5242</div>
5243
5244<!-- ======================================================================= -->
5245<div class="doc_subsection">
Duncan Sands7407a9f2007-09-11 14:10:23 +0000[diff] [blame]5246 <a name="int_trampoline">Trampoline Intrinsic</a>
Duncan Sands38947cd2007-07-27 12:58:54 +0000[diff] [blame]5247</div>
5248
5249<div class="doc_text">
5250<p>
Duncan Sands7407a9f2007-09-11 14:10:23 +0000[diff] [blame]5251 This intrinsic makes it possible to excise one parameter, marked with
Duncan Sands38947cd2007-07-27 12:58:54 +0000[diff] [blame]5252 the <tt>nest</tt> attribute, from a function. The result is a callable
5253 function pointer lacking the nest parameter - the caller does not need
5254 to provide a value for it. Instead, the value to use is stored in
5255 advance in a "trampoline", a block of memory usually allocated
5256 on the stack, which also contains code to splice the nest value into the
5257 argument list. This is used to implement the GCC nested function address
5258 extension.
5259</p>
5260<p>
5261 For example, if the function is
5262 <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]5263 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]5264<pre>
Duncan Sands7407a9f2007-09-11 14:10:23 +0000[diff] [blame]5265 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
5266 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
5267 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
5268 %fp = bitcast i8* %p to i32 (i32, i32)*
Duncan Sands38947cd2007-07-27 12:58:54 +0000[diff] [blame]5269</pre>
Bill Wendlinge262b4c2007-09-22 09:23:55 +0000[diff] [blame]5270 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
5271 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
Duncan Sands38947cd2007-07-27 12:58:54 +0000[diff] [blame]5272</div>
5273
5274<!-- _______________________________________________________________________ -->
5275<div class="doc_subsubsection">
5276 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
5277</div>
5278<div class="doc_text">
5279<h5>Syntax:</h5>
5280<pre>
Duncan Sands7407a9f2007-09-11 14:10:23 +0000[diff] [blame]5281declare 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]5282</pre>
5283<h5>Overview:</h5>
5284<p>
Duncan Sands7407a9f2007-09-11 14:10:23 +0000[diff] [blame]5285 This fills the memory pointed to by <tt>tramp</tt> with code
5286 and returns a function pointer suitable for executing it.
Duncan Sands38947cd2007-07-27 12:58:54 +0000[diff] [blame]5287</p>
5288<h5>Arguments:</h5>
5289<p>
5290 The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
5291 pointers. The <tt>tramp</tt> argument must point to a sufficiently large
5292 and sufficiently aligned block of memory; this memory is written to by the
Duncan Sands35012212007-08-22 23:39:54 +0000[diff] [blame]5293 intrinsic. Note that the size and the alignment are target-specific - LLVM
5294 currently provides no portable way of determining them, so a front-end that
5295 generates this intrinsic needs to have some target-specific knowledge.
5296 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]5297</p>
5298<h5>Semantics:</h5>
5299<p>
5300 The block of memory pointed to by <tt>tramp</tt> is filled with target
Duncan Sands7407a9f2007-09-11 14:10:23 +0000[diff] [blame]5301 dependent code, turning it into a function. A pointer to this function is
5302 returned, but needs to be bitcast to an
Duncan Sands38947cd2007-07-27 12:58:54 +0000[diff] [blame]5303 <a href="#int_trampoline">appropriate function pointer type</a>
Duncan Sands7407a9f2007-09-11 14:10:23 +0000[diff] [blame]5304 before being called. The new function's signature is the same as that of
5305 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
5306 removed. At most one such <tt>nest</tt> argument is allowed, and it must be
5307 of pointer type. Calling the new function is equivalent to calling
5308 <tt>func</tt> with the same argument list, but with <tt>nval</tt> used for the
5309 missing <tt>nest</tt> argument. If, after calling
5310 <tt>llvm.init.trampoline</tt>, the memory pointed to by <tt>tramp</tt> is
5311 modified, then the effect of any later call to the returned function pointer is
5312 undefined.
Duncan Sands38947cd2007-07-27 12:58:54 +0000[diff] [blame]5313</p>
5314</div>
5315
5316<!-- ======================================================================= -->
5317<div class="doc_subsection">
Andrew Lenharth785610d2008-02-16 01:24:58 +0000[diff] [blame]5318 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
5319</div>
5320
5321<div class="doc_text">
5322<p>
5323 These intrinsic functions expand the "universal IR" of LLVM to represent
5324 hardware constructs for atomic operations and memory synchronization. This
5325 provides an interface to the hardware, not an interface to the programmer. It
5326 is aimed at a low enough level to allow any programming models or APIs which
5327 need atomic behaviors to map cleanly onto it. It is also modeled primarily on
5328 hardware behavior. Just as hardware provides a "universal IR" for source
5329 languages, it also provides a starting point for developing a "universal"
5330 atomic operation and synchronization IR.
5331</p>
5332<p>
5333 These do <em>not</em> form an API such as high-level threading libraries,
5334 software transaction memory systems, atomic primitives, and intrinsic
5335 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
5336 application libraries. The hardware interface provided by LLVM should allow
5337 a clean implementation of all of these APIs and parallel programming models.
5338 No one model or paradigm should be selected above others unless the hardware
5339 itself ubiquitously does so.
5340
5341</p>
5342</div>
5343
5344<!-- _______________________________________________________________________ -->
5345<div class="doc_subsubsection">
5346 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
5347</div>
5348<div class="doc_text">
5349<h5>Syntax:</h5>
5350<pre>
5351declare void @llvm.memory.barrier( i1 &lt;ll&gt;, i1 &lt;ls&gt;, i1 &lt;sl&gt;, i1 &lt;ss&gt;,
5352i1 &lt;device&gt; )
5353
5354</pre>
5355<h5>Overview:</h5>
5356<p>
5357 The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
5358 specific pairs of memory access types.
5359</p>
5360<h5>Arguments:</h5>
5361<p>
5362 The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
5363 The first four arguments enables a specific barrier as listed below. The fith
5364 argument specifies that the barrier applies to io or device or uncached memory.
5365
5366</p>
5367 <ul>
5368 <li><tt>ll</tt>: load-load barrier</li>
5369 <li><tt>ls</tt>: load-store barrier</li>
5370 <li><tt>sl</tt>: store-load barrier</li>
5371 <li><tt>ss</tt>: store-store barrier</li>
5372 <li><tt>device</tt>: barrier applies to device and uncached memory also.
5373 </ul>
5374<h5>Semantics:</h5>
5375<p>
5376 This intrinsic causes the system to enforce some ordering constraints upon
5377 the loads and stores of the program. This barrier does not indicate
5378 <em>when</em> any events will occur, it only enforces an <em>order</em> in
5379 which they occur. For any of the specified pairs of load and store operations
5380 (f.ex. load-load, or store-load), all of the first operations preceding the
5381 barrier will complete before any of the second operations succeeding the
5382 barrier begin. Specifically the semantics for each pairing is as follows:
5383</p>
5384 <ul>
5385 <li><tt>ll</tt>: All loads before the barrier must complete before any load
5386 after the barrier begins.</li>
5387
5388 <li><tt>ls</tt>: All loads before the barrier must complete before any
5389 store after the barrier begins.</li>
5390 <li><tt>ss</tt>: All stores before the barrier must complete before any
5391 store after the barrier begins.</li>
5392 <li><tt>sl</tt>: All stores before the barrier must complete before any
5393 load after the barrier begins.</li>
5394 </ul>
5395<p>
5396 These semantics are applied with a logical "and" behavior when more than one
5397 is enabled in a single memory barrier intrinsic.
5398</p>
5399<p>
5400 Backends may implement stronger barriers than those requested when they do not
5401 support as fine grained a barrier as requested. Some architectures do not
5402 need all types of barriers and on such architectures, these become noops.
5403</p>
5404<h5>Example:</h5>
5405<pre>
5406%ptr = malloc i32
5407 store i32 4, %ptr
5408
5409%result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
5410 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
5411 <i>; guarantee the above finishes</i>
5412 store i32 8, %ptr <i>; before this begins</i>
5413</pre>
5414</div>
5415
Andrew Lenharthe44f3902008-02-21 06:45:13 +0000[diff] [blame]5416<!-- _______________________________________________________________________ -->
5417<div class="doc_subsubsection">
5418 <a name="int_atomic_lcs">'<tt>llvm.atomic.lcs.*</tt>' Intrinsic</a>
5419</div>
5420<div class="doc_text">
5421<h5>Syntax:</h5>
5422<p>
5423 This is an overloaded intrinsic. You can use <tt>llvm.atomic.lcs</tt> on any
5424 integer bit width. Not all targets support all bit widths however.</p>
5425
5426<pre>
5427declare i8 @llvm.atomic.lcs.i8( i8* &lt;ptr&gt;, i8 &lt;cmp&gt;, i8 &lt;val&gt; )
5428declare i16 @llvm.atomic.lcs.i16( i16* &lt;ptr&gt;, i16 &lt;cmp&gt;, i16 &lt;val&gt; )
5429declare i32 @llvm.atomic.lcs.i32( i32* &lt;ptr&gt;, i32 &lt;cmp&gt;, i32 &lt;val&gt; )
5430declare i64 @llvm.atomic.lcs.i64( i64* &lt;ptr&gt;, i64 &lt;cmp&gt;, i64 &lt;val&gt; )
5431
5432</pre>
5433<h5>Overview:</h5>
5434<p>
5435 This loads a value in memory and compares it to a given value. If they are
5436 equal, it stores a new value into the memory.
5437</p>
5438<h5>Arguments:</h5>
5439<p>
5440 The <tt>llvm.atomic.lcs</tt> intrinsic takes three arguments. The result as
5441 well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
5442 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
5443 this integer type. While any bit width integer may be used, targets may only
5444 lower representations they support in hardware.
5445
5446</p>
5447<h5>Semantics:</h5>
5448<p>
5449 This entire intrinsic must be executed atomically. It first loads the value
5450 in memory pointed to by <tt>ptr</tt> and compares it with the value
5451 <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the memory. The
5452 loaded value is yielded in all cases. This provides the equivalent of an
5453 atomic compare-and-swap operation within the SSA framework.
5454</p>
5455<h5>Examples:</h5>
5456
5457<pre>
5458%ptr = malloc i32
5459 store i32 4, %ptr
5460
5461%val1 = add i32 4, 4
5462%result1 = call i32 @llvm.atomic.lcs.i32( i32* %ptr, i32 4, %val1 )
5463 <i>; yields {i32}:result1 = 4</i>
5464%stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
5465%memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
5466
5467%val2 = add i32 1, 1
5468%result2 = call i32 @llvm.atomic.lcs.i32( i32* %ptr, i32 5, %val2 )
5469 <i>; yields {i32}:result2 = 8</i>
5470%stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
5471
5472%memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
5473</pre>
5474</div>
5475
5476<!-- _______________________________________________________________________ -->
5477<div class="doc_subsubsection">
5478 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
5479</div>
5480<div class="doc_text">
5481<h5>Syntax:</h5>
5482
5483<p>
5484 This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
5485 integer bit width. Not all targets support all bit widths however.</p>
5486<pre>
5487declare i8 @llvm.atomic.swap.i8( i8* &lt;ptr&gt;, i8 &lt;val&gt; )
5488declare i16 @llvm.atomic.swap.i16( i16* &lt;ptr&gt;, i16 &lt;val&gt; )
5489declare i32 @llvm.atomic.swap.i32( i32* &lt;ptr&gt;, i32 &lt;val&gt; )
5490declare i64 @llvm.atomic.swap.i64( i64* &lt;ptr&gt;, i64 &lt;val&gt; )
5491
5492</pre>
5493<h5>Overview:</h5>
5494<p>
5495 This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
5496 the value from memory. It then stores the value in <tt>val</tt> in the memory
5497 at <tt>ptr</tt>.
5498</p>
5499<h5>Arguments:</h5>
5500
5501<p>
5502 The <tt>llvm.atomic.ls</tt> intrinsic takes two arguments. Both the
5503 <tt>val</tt> argument and the result must be integers of the same bit width.
5504 The first argument, <tt>ptr</tt>, must be a pointer to a value of this
5505 integer type. The targets may only lower integer representations they
5506 support.
5507</p>
5508<h5>Semantics:</h5>
5509<p>
5510 This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
5511 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
5512 equivalent of an atomic swap operation within the SSA framework.
5513
5514</p>
5515<h5>Examples:</h5>
5516<pre>
5517%ptr = malloc i32
5518 store i32 4, %ptr
5519
5520%val1 = add i32 4, 4
5521%result1 = call i32 @llvm.atomic.swap.i32( i32* %ptr, i32 %val1 )
5522 <i>; yields {i32}:result1 = 4</i>
5523%stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
5524%memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
5525
5526%val2 = add i32 1, 1
5527%result2 = call i32 @llvm.atomic.swap.i32( i32* %ptr, i32 %val2 )
5528 <i>; yields {i32}:result2 = 8</i>
5529
5530%stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
5531%memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
5532</pre>
5533</div>
5534
5535<!-- _______________________________________________________________________ -->
5536<div class="doc_subsubsection">
5537 <a name="int_atomic_las">'<tt>llvm.atomic.las.*</tt>' Intrinsic</a>
5538
5539</div>
5540<div class="doc_text">
5541<h5>Syntax:</h5>
5542<p>
5543 This is an overloaded intrinsic. You can use <tt>llvm.atomic.las</tt> on any
5544 integer bit width. Not all targets support all bit widths however.</p>
5545<pre>
5546declare i8 @llvm.atomic.las.i8.( i8* &lt;ptr&gt;, i8 &lt;delta&gt; )
5547declare i16 @llvm.atomic.las.i16.( i16* &lt;ptr&gt;, i16 &lt;delta&gt; )
5548declare i32 @llvm.atomic.las.i32.( i32* &lt;ptr&gt;, i32 &lt;delta&gt; )
5549declare i64 @llvm.atomic.las.i64.( i64* &lt;ptr&gt;, i64 &lt;delta&gt; )
5550
5551</pre>
5552<h5>Overview:</h5>
5553<p>
5554 This intrinsic adds <tt>delta</tt> to the value stored in memory at
5555 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
5556</p>
5557<h5>Arguments:</h5>
5558<p>
5559
5560 The intrinsic takes two arguments, the first a pointer to an integer value
5561 and the second an integer value. The result is also an integer value. These
5562 integer types can have any bit width, but they must all have the same bit
5563 width. The targets may only lower integer representations they support.
5564</p>
5565<h5>Semantics:</h5>
5566<p>
5567 This intrinsic does a series of operations atomically. It first loads the
5568 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
5569 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
5570</p>
5571
5572<h5>Examples:</h5>
5573<pre>
5574%ptr = malloc i32
5575 store i32 4, %ptr
5576%result1 = call i32 @llvm.atomic.las.i32( i32* %ptr, i32 4 )
5577 <i>; yields {i32}:result1 = 4</i>
5578%result2 = call i32 @llvm.atomic.las.i32( i32* %ptr, i32 2 )
5579 <i>; yields {i32}:result2 = 8</i>
5580%result3 = call i32 @llvm.atomic.las.i32( i32* %ptr, i32 5 )
5581 <i>; yields {i32}:result3 = 10</i>
5582%memval = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
5583</pre>
5584</div>
5585
Andrew Lenharth785610d2008-02-16 01:24:58 +0000[diff] [blame]5586
5587<!-- ======================================================================= -->
5588<div class="doc_subsection">
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5589 <a name="int_general">General Intrinsics</a>
5590</div>
5591
5592<div class="doc_text">
5593<p> This class of intrinsics is designed to be generic and has
5594no specific purpose. </p>
5595</div>
5596
5597<!-- _______________________________________________________________________ -->
5598<div class="doc_subsubsection">
5599 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
5600</div>
5601
5602<div class="doc_text">
5603
5604<h5>Syntax:</h5>
5605<pre>
5606 declare void @llvm.var.annotation(i8* &lt;val&gt;, i8* &lt;str&gt;, i8* &lt;str&gt;, i32 &lt;int&gt; )
5607</pre>
5608
5609<h5>Overview:</h5>
5610
5611<p>
5612The '<tt>llvm.var.annotation</tt>' intrinsic
5613</p>
5614
5615<h5>Arguments:</h5>
5616
5617<p>
5618The first argument is a pointer to a value, the second is a pointer to a
5619global string, the third is a pointer to a global string which is the source
5620file name, and the last argument is the line number.
5621</p>
5622
5623<h5>Semantics:</h5>
5624
5625<p>
Anton Korobeynikove6e764f2008-01-15 22:31:34 +0000[diff] [blame]5626This intrinsic allows annotation of local variables with arbitrary strings.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5627This can be useful for special purpose optimizations that want to look for these
Anton Korobeynikove6e764f2008-01-15 22:31:34 +0000[diff] [blame]5628annotations. These have no other defined use, they are ignored by code
5629generation and optimization.
5630</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5631</div>
5632
Tanya Lattnerb306a9e2007-09-21 22:59:12 +0000[diff] [blame]5633<!-- _______________________________________________________________________ -->
5634<div class="doc_subsubsection">
Tanya Lattnerc9869b12007-09-21 23:57:59 +0000[diff] [blame]5635 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
Tanya Lattnerb306a9e2007-09-21 22:59:12 +0000[diff] [blame]5636</div>
5637
5638<div class="doc_text">
5639
5640<h5>Syntax:</h5>
Tanya Lattnere545be72007-09-21 23:56:27 +0000[diff] [blame]5641<p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
5642any integer bit width.
5643</p>
Tanya Lattnerb306a9e2007-09-21 22:59:12 +0000[diff] [blame]5644<pre>
Tanya Lattner09161fe2007-09-22 00:03:01 +0000[diff] [blame]5645 declare i8 @llvm.annotation.i8(i8 &lt;val&gt;, i8* &lt;str&gt;, i8* &lt;str&gt;, i32 &lt;int&gt; )
5646 declare i16 @llvm.annotation.i16(i16 &lt;val&gt;, i8* &lt;str&gt;, i8* &lt;str&gt;, i32 &lt;int&gt; )
5647 declare i32 @llvm.annotation.i32(i32 &lt;val&gt;, i8* &lt;str&gt;, i8* &lt;str&gt;, i32 &lt;int&gt; )
5648 declare i64 @llvm.annotation.i64(i64 &lt;val&gt;, i8* &lt;str&gt;, i8* &lt;str&gt;, i32 &lt;int&gt; )
5649 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]5650</pre>
5651
5652<h5>Overview:</h5>
Tanya Lattnere545be72007-09-21 23:56:27 +0000[diff] [blame]5653
5654<p>
5655The '<tt>llvm.annotation</tt>' intrinsic.
Tanya Lattnerb306a9e2007-09-21 22:59:12 +0000[diff] [blame]5656</p>
5657
5658<h5>Arguments:</h5>
5659
5660<p>
5661The first argument is an integer value (result of some expression),
5662the second is a pointer to a global string, the third is a pointer to a global
5663string which is the source file name, and the last argument is the line number.
Tanya Lattnere545be72007-09-21 23:56:27 +0000[diff] [blame]5664It returns the value of the first argument.
Tanya Lattnerb306a9e2007-09-21 22:59:12 +0000[diff] [blame]5665</p>
5666
5667<h5>Semantics:</h5>
5668
5669<p>
5670This intrinsic allows annotations to be put on arbitrary expressions
5671with arbitrary strings. This can be useful for special purpose optimizations
5672that want to look for these annotations. These have no other defined use, they
5673are ignored by code generation and optimization.
5674</div>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5675
Anton Korobeynikove6e764f2008-01-15 22:31:34 +0000[diff] [blame]5676<!-- _______________________________________________________________________ -->
5677<div class="doc_subsubsection">
5678 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
5679</div>
5680
5681<div class="doc_text">
5682
5683<h5>Syntax:</h5>
5684<pre>
5685 declare void @llvm.trap()
5686</pre>
5687
5688<h5>Overview:</h5>
5689
5690<p>
5691The '<tt>llvm.trap</tt>' intrinsic
5692</p>
5693
5694<h5>Arguments:</h5>
5695
5696<p>
5697None
5698</p>
5699
5700<h5>Semantics:</h5>
5701
5702<p>
5703This intrinsics is lowered to the target dependent trap instruction. If the
5704target does not have a trap instruction, this intrinsic will be lowered to the
5705call of the abort() function.
5706</p>
5707</div>
5708
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5709<!-- *********************************************************************** -->
5710<hr>
5711<address>
5712 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
5713 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
5714 <a href="http://validator.w3.org/check/referer"><img
Chris Lattner08497ce2008-01-04 04:33:49 +0000[diff] [blame]5715 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5716
5717 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
5718 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
5719 Last modified: $Date$
5720</address>
Chris Lattner08497ce2008-01-04 04:33:49 +0000[diff] [blame]5721
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000[diff] [blame]5722</body>
5723</html>