blob: 694fed821e2911e84ac7c20e6639832a6785106d [file] [log] [blame]
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001<!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
3<html>
4<head>
5 <title>LLVM Assembly Language Reference Manual</title>
6 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
7 <meta name="author" content="Chris Lattner">
8 <meta name="description"
9 content="LLVM Assembly Language Reference Manual.">
10 <link rel="stylesheet" href="llvm.css" type="text/css">
11</head>
12
13<body>
14
15<div class="doc_title"> LLVM Language Reference Manual </div>
16<ol>
17 <li><a href="#abstract">Abstract</a></li>
18 <li><a href="#introduction">Introduction</a></li>
19 <li><a href="#identifiers">Identifiers</a></li>
20 <li><a href="#highlevel">High Level Structure</a>
21 <ol>
22 <li><a href="#modulestructure">Module Structure</a></li>
23 <li><a href="#linkage">Linkage Types</a></li>
24 <li><a href="#callingconv">Calling Conventions</a></li>
25 <li><a href="#globalvars">Global Variables</a></li>
26 <li><a href="#functionstructure">Functions</a></li>
27 <li><a href="#aliasstructure">Aliases</a>
28 <li><a href="#paramattrs">Parameter Attributes</a></li>
Gordon Henriksen13fe5e32007-12-10 03:18:06 +000029 <li><a href="#gc">Garbage Collector Names</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +000030 <li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
31 <li><a href="#datalayout">Data Layout</a></li>
32 </ol>
33 </li>
34 <li><a href="#typesystem">Type System</a>
35 <ol>
Chris Lattner488772f2008-01-04 04:32:38 +000036 <li><a href="#t_classifications">Type Classifications</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +000037 <li><a href="#t_primitive">Primitive Types</a>
38 <ol>
Chris Lattner488772f2008-01-04 04:32:38 +000039 <li><a href="#t_floating">Floating Point Types</a></li>
40 <li><a href="#t_void">Void Type</a></li>
41 <li><a href="#t_label">Label Type</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +000042 </ol>
43 </li>
44 <li><a href="#t_derived">Derived Types</a>
45 <ol>
Chris Lattner251ab812007-12-18 06:18:21 +000046 <li><a href="#t_integer">Integer Type</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +000047 <li><a href="#t_array">Array Type</a></li>
48 <li><a href="#t_function">Function Type</a></li>
49 <li><a href="#t_pointer">Pointer Type</a></li>
50 <li><a href="#t_struct">Structure Type</a></li>
51 <li><a href="#t_pstruct">Packed Structure Type</a></li>
52 <li><a href="#t_vector">Vector Type</a></li>
53 <li><a href="#t_opaque">Opaque Type</a></li>
54 </ol>
55 </li>
56 </ol>
57 </li>
58 <li><a href="#constants">Constants</a>
59 <ol>
60 <li><a href="#simpleconstants">Simple Constants</a>
61 <li><a href="#aggregateconstants">Aggregate Constants</a>
62 <li><a href="#globalconstants">Global Variable and Function Addresses</a>
63 <li><a href="#undefvalues">Undefined Values</a>
64 <li><a href="#constantexprs">Constant Expressions</a>
65 </ol>
66 </li>
67 <li><a href="#othervalues">Other Values</a>
68 <ol>
69 <li><a href="#inlineasm">Inline Assembler Expressions</a>
70 </ol>
71 </li>
72 <li><a href="#instref">Instruction Reference</a>
73 <ol>
74 <li><a href="#terminators">Terminator Instructions</a>
75 <ol>
76 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
77 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
78 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
79 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
80 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
81 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
82 </ol>
83 </li>
84 <li><a href="#binaryops">Binary Operations</a>
85 <ol>
86 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
87 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
88 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
89 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
90 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
91 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
92 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
93 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
94 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
95 </ol>
96 </li>
97 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
98 <ol>
99 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
100 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
101 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
102 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
103 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
104 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
105 </ol>
106 </li>
107 <li><a href="#vectorops">Vector Operations</a>
108 <ol>
109 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
110 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
111 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
112 </ol>
113 </li>
Dan Gohman74d6faf2008-05-12 23:51:09 +0000114 <li><a href="#aggregateops">Aggregate Operations</a>
115 <ol>
116 <li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li>
117 <li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li>
118 </ol>
119 </li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000120 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
121 <ol>
122 <li><a href="#i_malloc">'<tt>malloc</tt>' Instruction</a></li>
123 <li><a href="#i_free">'<tt>free</tt>' Instruction</a></li>
124 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
125 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
126 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
127 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
128 </ol>
129 </li>
130 <li><a href="#convertops">Conversion Operations</a>
131 <ol>
132 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
133 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
134 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
135 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
136 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
137 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
138 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
139 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
140 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
141 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
142 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
143 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
144 </ol>
145 <li><a href="#otherops">Other Operations</a>
146 <ol>
147 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
148 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
Nate Begeman646fa482008-05-12 19:01:56 +0000149 <li><a href="#i_vicmp">'<tt>vicmp</tt>' Instruction</a></li>
150 <li><a href="#i_vfcmp">'<tt>vfcmp</tt>' Instruction</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000151 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
152 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
153 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
154 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
Devang Patela3cc5372008-03-10 20:49:15 +0000155 <li><a href="#i_getresult">'<tt>getresult</tt>' Instruction</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000156 </ol>
157 </li>
158 </ol>
159 </li>
160 <li><a href="#intrinsics">Intrinsic Functions</a>
161 <ol>
162 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
163 <ol>
164 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
165 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
166 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
167 </ol>
168 </li>
169 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
170 <ol>
171 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
172 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
173 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
174 </ol>
175 </li>
176 <li><a href="#int_codegen">Code Generator Intrinsics</a>
177 <ol>
178 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
179 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
180 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
181 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
182 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
183 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
184 <li><a href="#int_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
185 </ol>
186 </li>
187 <li><a href="#int_libc">Standard C Library Intrinsics</a>
188 <ol>
189 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
190 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
191 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
192 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
193 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
Dan Gohman361079c2007-10-15 20:30:11 +0000194 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
195 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
196 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000197 </ol>
198 </li>
199 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
200 <ol>
201 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
202 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
203 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
204 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
205 <li><a href="#int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic </a></li>
206 <li><a href="#int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic </a></li>
207 </ol>
208 </li>
209 <li><a href="#int_debugger">Debugger intrinsics</a></li>
210 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
Duncan Sands7407a9f2007-09-11 14:10:23 +0000211 <li><a href="#int_trampoline">Trampoline Intrinsic</a>
Duncan Sands38947cd2007-07-27 12:58:54 +0000212 <ol>
213 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
Duncan Sands38947cd2007-07-27 12:58:54 +0000214 </ol>
215 </li>
Andrew Lenharth785610d2008-02-16 01:24:58 +0000216 <li><a href="#int_atomics">Atomic intrinsics</a>
217 <ol>
Andrew Lenharthe44f3902008-02-21 06:45:13 +0000218 <li><a href="#int_memory_barrier"><tt>llvm.memory_barrier</tt></a></li>
219 <li><a href="#int_atomic_lcs"><tt>llvm.atomic.lcs</tt></a></li>
220 <li><a href="#int_atomic_las"><tt>llvm.atomic.las</tt></a></li>
221 <li><a href="#int_atomic_swap"><tt>llvm.atomic.swap</tt></a></li>
Andrew Lenharth785610d2008-02-16 01:24:58 +0000222 </ol>
223 </li>
Reid Spencerb043f672007-07-20 19:59:11 +0000224 <li><a href="#int_general">General intrinsics</a>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000225 <ol>
Reid Spencerb043f672007-07-20 19:59:11 +0000226 <li><a href="#int_var_annotation">
Tanya Lattner51369f32007-09-22 00:01:26 +0000227 <tt>llvm.var.annotation</tt>' Intrinsic</a></li>
Tanya Lattnerb306a9e2007-09-21 22:59:12 +0000228 <li><a href="#int_annotation">
Tanya Lattner51369f32007-09-22 00:01:26 +0000229 <tt>llvm.annotation.*</tt>' Intrinsic</a></li>
Anton Korobeynikove6e764f2008-01-15 22:31:34 +0000230 <li><a href="#int_trap">
231 <tt>llvm.trap</tt>' Intrinsic</a></li>
Tanya Lattnerb306a9e2007-09-21 22:59:12 +0000232 </ol>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000233 </li>
234 </ol>
235 </li>
236</ol>
237
238<div class="doc_author">
239 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
240 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
241</div>
242
243<!-- *********************************************************************** -->
244<div class="doc_section"> <a name="abstract">Abstract </a></div>
245<!-- *********************************************************************** -->
246
247<div class="doc_text">
248<p>This document is a reference manual for the LLVM assembly language.
249LLVM is an SSA based representation that provides type safety,
250low-level operations, flexibility, and the capability of representing
251'all' high-level languages cleanly. It is the common code
252representation used throughout all phases of the LLVM compilation
253strategy.</p>
254</div>
255
256<!-- *********************************************************************** -->
257<div class="doc_section"> <a name="introduction">Introduction</a> </div>
258<!-- *********************************************************************** -->
259
260<div class="doc_text">
261
262<p>The LLVM code representation is designed to be used in three
263different forms: as an in-memory compiler IR, as an on-disk bitcode
264representation (suitable for fast loading by a Just-In-Time compiler),
265and as a human readable assembly language representation. This allows
266LLVM to provide a powerful intermediate representation for efficient
267compiler transformations and analysis, while providing a natural means
268to debug and visualize the transformations. The three different forms
269of LLVM are all equivalent. This document describes the human readable
270representation and notation.</p>
271
272<p>The LLVM representation aims to be light-weight and low-level
273while being expressive, typed, and extensible at the same time. It
274aims to be a "universal IR" of sorts, by being at a low enough level
275that high-level ideas may be cleanly mapped to it (similar to how
276microprocessors are "universal IR's", allowing many source languages to
277be mapped to them). By providing type information, LLVM can be used as
278the target of optimizations: for example, through pointer analysis, it
279can be proven that a C automatic variable is never accessed outside of
280the current function... allowing it to be promoted to a simple SSA
281value instead of a memory location.</p>
282
283</div>
284
285<!-- _______________________________________________________________________ -->
286<div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
287
288<div class="doc_text">
289
290<p>It is important to note that this document describes 'well formed'
291LLVM assembly language. There is a difference between what the parser
292accepts and what is considered 'well formed'. For example, the
293following instruction is syntactically okay, but not well formed:</p>
294
295<div class="doc_code">
296<pre>
297%x = <a href="#i_add">add</a> i32 1, %x
298</pre>
299</div>
300
301<p>...because the definition of <tt>%x</tt> does not dominate all of
302its uses. The LLVM infrastructure provides a verification pass that may
303be used to verify that an LLVM module is well formed. This pass is
304automatically run by the parser after parsing input assembly and by
305the optimizer before it outputs bitcode. The violations pointed out
306by the verifier pass indicate bugs in transformation passes or input to
307the parser.</p>
308</div>
309
Chris Lattnera83fdc02007-10-03 17:34:29 +0000310<!-- Describe the typesetting conventions here. -->
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000311
312<!-- *********************************************************************** -->
313<div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
314<!-- *********************************************************************** -->
315
316<div class="doc_text">
317
Reid Spencerc8245b02007-08-07 14:34:28 +0000318 <p>LLVM identifiers come in two basic types: global and local. Global
319 identifiers (functions, global variables) begin with the @ character. Local
320 identifiers (register names, types) begin with the % character. Additionally,
321 there are three different formats for identifiers, for different purposes:
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000322
323<ol>
Reid Spencerc8245b02007-08-07 14:34:28 +0000324 <li>Named values are represented as a string of characters with their prefix.
325 For example, %foo, @DivisionByZero, %a.really.long.identifier. The actual
326 regular expression used is '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000327 Identifiers which require other characters in their names can be surrounded
Reid Spencerc8245b02007-08-07 14:34:28 +0000328 with quotes. In this way, anything except a <tt>&quot;</tt> character can
329 be used in a named value.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000330
Reid Spencerc8245b02007-08-07 14:34:28 +0000331 <li>Unnamed values are represented as an unsigned numeric value with their
332 prefix. For example, %12, @2, %44.</li>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000333
334 <li>Constants, which are described in a <a href="#constants">section about
335 constants</a>, below.</li>
336</ol>
337
Reid Spencerc8245b02007-08-07 14:34:28 +0000338<p>LLVM requires that values start with a prefix for two reasons: Compilers
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000339don't need to worry about name clashes with reserved words, and the set of
340reserved words may be expanded in the future without penalty. Additionally,
341unnamed identifiers allow a compiler to quickly come up with a temporary
342variable without having to avoid symbol table conflicts.</p>
343
344<p>Reserved words in LLVM are very similar to reserved words in other
345languages. There are keywords for different opcodes
346('<tt><a href="#i_add">add</a></tt>',
347 '<tt><a href="#i_bitcast">bitcast</a></tt>',
348 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
349href="#t_void">void</a></tt>', '<tt><a href="#t_primitive">i32</a></tt>', etc...),
350and others. These reserved words cannot conflict with variable names, because
Reid Spencerc8245b02007-08-07 14:34:28 +0000351none of them start with a prefix character ('%' or '@').</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000352
353<p>Here is an example of LLVM code to multiply the integer variable
354'<tt>%X</tt>' by 8:</p>
355
356<p>The easy way:</p>
357
358<div class="doc_code">
359<pre>
360%result = <a href="#i_mul">mul</a> i32 %X, 8
361</pre>
362</div>
363
364<p>After strength reduction:</p>
365
366<div class="doc_code">
367<pre>
368%result = <a href="#i_shl">shl</a> i32 %X, i8 3
369</pre>
370</div>
371
372<p>And the hard way:</p>
373
374<div class="doc_code">
375<pre>
376<a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
377<a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
378%result = <a href="#i_add">add</a> i32 %1, %1
379</pre>
380</div>
381
382<p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
383important lexical features of LLVM:</p>
384
385<ol>
386
387 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
388 line.</li>
389
390 <li>Unnamed temporaries are created when the result of a computation is not
391 assigned to a named value.</li>
392
393 <li>Unnamed temporaries are numbered sequentially</li>
394
395</ol>
396
397<p>...and it also shows a convention that we follow in this document. When
398demonstrating instructions, we will follow an instruction with a comment that
399defines the type and name of value produced. Comments are shown in italic
400text.</p>
401
402</div>
403
404<!-- *********************************************************************** -->
405<div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
406<!-- *********************************************************************** -->
407
408<!-- ======================================================================= -->
409<div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
410</div>
411
412<div class="doc_text">
413
414<p>LLVM programs are composed of "Module"s, each of which is a
415translation unit of the input programs. Each module consists of
416functions, global variables, and symbol table entries. Modules may be
417combined together with the LLVM linker, which merges function (and
418global variable) definitions, resolves forward declarations, and merges
419symbol table entries. Here is an example of the "hello world" module:</p>
420
421<div class="doc_code">
422<pre><i>; Declare the string constant as a global constant...</i>
423<a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a
424 href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
425
426<i>; External declaration of the puts function</i>
427<a href="#functionstructure">declare</a> i32 @puts(i8 *) <i>; i32(i8 *)* </i>
428
429<i>; Definition of main function</i>
430define i32 @main() { <i>; i32()* </i>
431 <i>; Convert [13x i8 ]* to i8 *...</i>
432 %cast210 = <a
433 href="#i_getelementptr">getelementptr</a> [13 x i8 ]* @.LC0, i64 0, i64 0 <i>; i8 *</i>
434
435 <i>; Call puts function to write out the string to stdout...</i>
436 <a
437 href="#i_call">call</a> i32 @puts(i8 * %cast210) <i>; i32</i>
438 <a
439 href="#i_ret">ret</a> i32 0<br>}<br>
440</pre>
441</div>
442
443<p>This example is made up of a <a href="#globalvars">global variable</a>
444named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
445function, and a <a href="#functionstructure">function definition</a>
446for "<tt>main</tt>".</p>
447
448<p>In general, a module is made up of a list of global values,
449where both functions and global variables are global values. Global values are
450represented by a pointer to a memory location (in this case, a pointer to an
451array of char, and a pointer to a function), and have one of the following <a
452href="#linkage">linkage types</a>.</p>
453
454</div>
455
456<!-- ======================================================================= -->
457<div class="doc_subsection">
458 <a name="linkage">Linkage Types</a>
459</div>
460
461<div class="doc_text">
462
463<p>
464All Global Variables and Functions have one of the following types of linkage:
465</p>
466
467<dl>
468
469 <dt><tt><b><a name="linkage_internal">internal</a></b></tt> </dt>
470
471 <dd>Global values with internal linkage are only directly accessible by
472 objects in the current module. In particular, linking code into a module with
473 an internal global value may cause the internal to be renamed as necessary to
474 avoid collisions. Because the symbol is internal to the module, all
475 references can be updated. This corresponds to the notion of the
476 '<tt>static</tt>' keyword in C.
477 </dd>
478
479 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
480
481 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
482 the same name when linkage occurs. This is typically used to implement
483 inline functions, templates, or other code which must be generated in each
484 translation unit that uses it. Unreferenced <tt>linkonce</tt> globals are
485 allowed to be discarded.
486 </dd>
487
488 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
489
490 <dd>"<tt>weak</tt>" linkage is exactly the same as <tt>linkonce</tt> linkage,
491 except that unreferenced <tt>weak</tt> globals may not be discarded. This is
492 used for globals that may be emitted in multiple translation units, but that
493 are not guaranteed to be emitted into every translation unit that uses them.
494 One example of this are common globals in C, such as "<tt>int X;</tt>" at
495 global scope.
496 </dd>
497
498 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
499
500 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
501 pointer to array type. When two global variables with appending linkage are
502 linked together, the two global arrays are appended together. This is the
503 LLVM, typesafe, equivalent of having the system linker append together
504 "sections" with identical names when .o files are linked.
505 </dd>
506
507 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt>: </dt>
508 <dd>The semantics of this linkage follow the ELF model: the symbol is weak
509 until linked, if not linked, the symbol becomes null instead of being an
510 undefined reference.
511 </dd>
512
513 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
514
515 <dd>If none of the above identifiers are used, the global is externally
516 visible, meaning that it participates in linkage and can be used to resolve
517 external symbol references.
518 </dd>
519</dl>
520
521 <p>
522 The next two types of linkage are targeted for Microsoft Windows platform
523 only. They are designed to support importing (exporting) symbols from (to)
524 DLLs.
525 </p>
526
527 <dl>
528 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt>: </dt>
529
530 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
531 or variable via a global pointer to a pointer that is set up by the DLL
532 exporting the symbol. On Microsoft Windows targets, the pointer name is
533 formed by combining <code>_imp__</code> and the function or variable name.
534 </dd>
535
536 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt>: </dt>
537
538 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
539 pointer to a pointer in a DLL, so that it can be referenced with the
540 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
541 name is formed by combining <code>_imp__</code> and the function or variable
542 name.
543 </dd>
544
545</dl>
546
547<p><a name="linkage_external"></a>For example, since the "<tt>.LC0</tt>"
548variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
549variable and was linked with this one, one of the two would be renamed,
550preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
551external (i.e., lacking any linkage declarations), they are accessible
552outside of the current module.</p>
553<p>It is illegal for a function <i>declaration</i>
554to have any linkage type other than "externally visible", <tt>dllimport</tt>,
555or <tt>extern_weak</tt>.</p>
556<p>Aliases can have only <tt>external</tt>, <tt>internal</tt> and <tt>weak</tt>
557linkages.
558</div>
559
560<!-- ======================================================================= -->
561<div class="doc_subsection">
562 <a name="callingconv">Calling Conventions</a>
563</div>
564
565<div class="doc_text">
566
567<p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
568and <a href="#i_invoke">invokes</a> can all have an optional calling convention
569specified for the call. The calling convention of any pair of dynamic
570caller/callee must match, or the behavior of the program is undefined. The
571following calling conventions are supported by LLVM, and more may be added in
572the future:</p>
573
574<dl>
575 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
576
577 <dd>This calling convention (the default if no other calling convention is
578 specified) matches the target C calling conventions. This calling convention
579 supports varargs function calls and tolerates some mismatch in the declared
580 prototype and implemented declaration of the function (as does normal C).
581 </dd>
582
583 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
584
585 <dd>This calling convention attempts to make calls as fast as possible
586 (e.g. by passing things in registers). This calling convention allows the
587 target to use whatever tricks it wants to produce fast code for the target,
588 without having to conform to an externally specified ABI. Implementations of
589 this convention should allow arbitrary tail call optimization to be supported.
590 This calling convention does not support varargs and requires the prototype of
591 all callees to exactly match the prototype of the function definition.
592 </dd>
593
594 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
595
596 <dd>This calling convention attempts to make code in the caller as efficient
597 as possible under the assumption that the call is not commonly executed. As
598 such, these calls often preserve all registers so that the call does not break
599 any live ranges in the caller side. This calling convention does not support
600 varargs and requires the prototype of all callees to exactly match the
601 prototype of the function definition.
602 </dd>
603
604 <dt><b>"<tt>cc &lt;<em>n</em>&gt;</tt>" - Numbered convention</b>:</dt>
605
606 <dd>Any calling convention may be specified by number, allowing
607 target-specific calling conventions to be used. Target specific calling
608 conventions start at 64.
609 </dd>
610</dl>
611
612<p>More calling conventions can be added/defined on an as-needed basis, to
613support pascal conventions or any other well-known target-independent
614convention.</p>
615
616</div>
617
618<!-- ======================================================================= -->
619<div class="doc_subsection">
620 <a name="visibility">Visibility Styles</a>
621</div>
622
623<div class="doc_text">
624
625<p>
626All Global Variables and Functions have one of the following visibility styles:
627</p>
628
629<dl>
630 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
631
632 <dd>On ELF, default visibility means that the declaration is visible to other
633 modules and, in shared libraries, means that the declared entity may be
634 overridden. On Darwin, default visibility means that the declaration is
635 visible to other modules. Default visibility corresponds to "external
636 linkage" in the language.
637 </dd>
638
639 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
640
641 <dd>Two declarations of an object with hidden visibility refer to the same
642 object if they are in the same shared object. Usually, hidden visibility
643 indicates that the symbol will not be placed into the dynamic symbol table,
644 so no other module (executable or shared library) can reference it
645 directly.
646 </dd>
647
648 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
649
650 <dd>On ELF, protected visibility indicates that the symbol will be placed in
651 the dynamic symbol table, but that references within the defining module will
652 bind to the local symbol. That is, the symbol cannot be overridden by another
653 module.
654 </dd>
655</dl>
656
657</div>
658
659<!-- ======================================================================= -->
660<div class="doc_subsection">
661 <a name="globalvars">Global Variables</a>
662</div>
663
664<div class="doc_text">
665
666<p>Global variables define regions of memory allocated at compilation time
667instead of run-time. Global variables may optionally be initialized, may have
668an explicit section to be placed in, and may have an optional explicit alignment
669specified. A variable may be defined as "thread_local", which means that it
670will not be shared by threads (each thread will have a separated copy of the
671variable). A variable may be defined as a global "constant," which indicates
672that the contents of the variable will <b>never</b> be modified (enabling better
673optimization, allowing the global data to be placed in the read-only section of
674an executable, etc). Note that variables that need runtime initialization
675cannot be marked "constant" as there is a store to the variable.</p>
676
677<p>
678LLVM explicitly allows <em>declarations</em> of global variables to be marked
679constant, even if the final definition of the global is not. This capability
680can be used to enable slightly better optimization of the program, but requires
681the language definition to guarantee that optimizations based on the
682'constantness' are valid for the translation units that do not include the
683definition.
684</p>
685
686<p>As SSA values, global variables define pointer values that are in
687scope (i.e. they dominate) all basic blocks in the program. Global
688variables always define a pointer to their "content" type because they
689describe a region of memory, and all memory objects in LLVM are
690accessed through pointers.</p>
691
Christopher Lambdd0049d2007-12-11 09:31:00 +0000692<p>A global variable may be declared to reside in a target-specifc numbered
693address space. For targets that support them, address spaces may affect how
694optimizations are performed and/or what target instructions are used to access
Christopher Lamb20a39e92007-12-12 08:44:39 +0000695the variable. The default address space is zero. The address space qualifier
696must precede any other attributes.</p>
Christopher Lambdd0049d2007-12-11 09:31:00 +0000697
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000698<p>LLVM allows an explicit section to be specified for globals. If the target
699supports it, it will emit globals to the section specified.</p>
700
701<p>An explicit alignment may be specified for a global. If not present, or if
702the alignment is set to zero, the alignment of the global is set by the target
703to whatever it feels convenient. If an explicit alignment is specified, the
704global is forced to have at least that much alignment. All alignments must be
705a power of 2.</p>
706
Christopher Lambdd0049d2007-12-11 09:31:00 +0000707<p>For example, the following defines a global in a numbered address space with
708an initializer, section, and alignment:</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000709
710<div class="doc_code">
711<pre>
Christopher Lambdd0049d2007-12-11 09:31:00 +0000712@G = constant float 1.0 addrspace(5), section "foo", align 4
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000713</pre>
714</div>
715
716</div>
717
718
719<!-- ======================================================================= -->
720<div class="doc_subsection">
721 <a name="functionstructure">Functions</a>
722</div>
723
724<div class="doc_text">
725
726<p>LLVM function definitions consist of the "<tt>define</tt>" keyord,
727an optional <a href="#linkage">linkage type</a>, an optional
728<a href="#visibility">visibility style</a>, an optional
729<a href="#callingconv">calling convention</a>, a return type, an optional
730<a href="#paramattrs">parameter attribute</a> for the return type, a function
731name, a (possibly empty) argument list (each with optional
732<a href="#paramattrs">parameter attributes</a>), an optional section, an
Gordon Henriksend606f5b2007-12-10 03:30:21 +0000733optional alignment, an optional <a href="#gc">garbage collector name</a>, an
Gordon Henriksen13fe5e32007-12-10 03:18:06 +0000734opening curly brace, a list of basic blocks, and a closing curly brace.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000735
736LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
737optional <a href="#linkage">linkage type</a>, an optional
738<a href="#visibility">visibility style</a>, an optional
739<a href="#callingconv">calling convention</a>, a return type, an optional
740<a href="#paramattrs">parameter attribute</a> for the return type, a function
Gordon Henriksen13fe5e32007-12-10 03:18:06 +0000741name, a possibly empty list of arguments, an optional alignment, and an optional
Gordon Henriksend606f5b2007-12-10 03:30:21 +0000742<a href="#gc">garbage collector name</a>.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000743
744<p>A function definition contains a list of basic blocks, forming the CFG for
745the function. Each basic block may optionally start with a label (giving the
746basic block a symbol table entry), contains a list of instructions, and ends
747with a <a href="#terminators">terminator</a> instruction (such as a branch or
748function return).</p>
749
750<p>The first basic block in a function is special in two ways: it is immediately
751executed on entrance to the function, and it is not allowed to have predecessor
752basic blocks (i.e. there can not be any branches to the entry block of a
753function). Because the block can have no predecessors, it also cannot have any
754<a href="#i_phi">PHI nodes</a>.</p>
755
756<p>LLVM allows an explicit section to be specified for functions. If the target
757supports it, it will emit functions to the section specified.</p>
758
759<p>An explicit alignment may be specified for a function. If not present, or if
760the alignment is set to zero, the alignment of the function is set by the target
761to whatever it feels convenient. If an explicit alignment is specified, the
762function is forced to have at least that much alignment. All alignments must be
763a power of 2.</p>
764
765</div>
766
767
768<!-- ======================================================================= -->
769<div class="doc_subsection">
770 <a name="aliasstructure">Aliases</a>
771</div>
772<div class="doc_text">
773 <p>Aliases act as "second name" for the aliasee value (which can be either
Anton Korobeynikov96822812008-03-22 08:36:14 +0000774 function, global variable, another alias or bitcast of global value). Aliases
775 may have an optional <a href="#linkage">linkage type</a>, and an
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000776 optional <a href="#visibility">visibility style</a>.</p>
777
778 <h5>Syntax:</h5>
779
780<div class="doc_code">
781<pre>
782@&lt;Name&gt; = [Linkage] [Visibility] alias &lt;AliaseeTy&gt; @&lt;Aliasee&gt;
783</pre>
784</div>
785
786</div>
787
788
789
790<!-- ======================================================================= -->
791<div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
792<div class="doc_text">
793 <p>The return type and each parameter of a function type may have a set of
794 <i>parameter attributes</i> associated with them. Parameter attributes are
795 used to communicate additional information about the result or parameters of
Duncan Sandsf5588dc2007-11-27 13:23:08 +0000796 a function. Parameter attributes are considered to be part of the function,
797 not of the function type, so functions with different parameter attributes
798 can have the same function type.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000799
800 <p>Parameter attributes are simple keywords that follow the type specified. If
801 multiple parameter attributes are needed, they are space separated. For
802 example:</p>
803
804<div class="doc_code">
805<pre>
Duncan Sandsf5588dc2007-11-27 13:23:08 +0000806declare i32 @printf(i8* noalias , ...) nounwind
807declare i32 @atoi(i8*) nounwind readonly
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000808</pre>
809</div>
810
Duncan Sandsf5588dc2007-11-27 13:23:08 +0000811 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
812 <tt>readonly</tt>) come immediately after the argument list.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000813
814 <p>Currently, only the following parameter attributes are defined:</p>
815 <dl>
Reid Spencerf234bed2007-07-19 23:13:04 +0000816 <dt><tt>zeroext</tt></dt>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000817 <dd>This indicates that the parameter should be zero extended just before
818 a call to this function.</dd>
Chris Lattner275e6be2008-01-11 06:20:47 +0000819
Reid Spencerf234bed2007-07-19 23:13:04 +0000820 <dt><tt>signext</tt></dt>
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000821 <dd>This indicates that the parameter should be sign extended just before
822 a call to this function.</dd>
Chris Lattner275e6be2008-01-11 06:20:47 +0000823
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000824 <dt><tt>inreg</tt></dt>
825 <dd>This indicates that the parameter should be placed in register (if
826 possible) during assembling function call. Support for this attribute is
827 target-specific</dd>
Chris Lattner275e6be2008-01-11 06:20:47 +0000828
829 <dt><tt>byval</tt></dt>
Chris Lattner04c86182008-01-15 04:34:22 +0000830 <dd>This indicates that the pointer parameter should really be passed by
831 value to the function. The attribute implies that a hidden copy of the
832 pointee is made between the caller and the callee, so the callee is unable
833 to modify the value in the callee. This attribute is only valid on llvm
834 pointer arguments. It is generally used to pass structs and arrays by
835 value, but is also valid on scalars (even though this is silly).</dd>
Chris Lattner275e6be2008-01-11 06:20:47 +0000836
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000837 <dt><tt>sret</tt></dt>
Duncan Sands616cc032008-02-18 04:19:38 +0000838 <dd>This indicates that the pointer parameter specifies the address of a
839 structure that is the return value of the function in the source program.
Duncan Sandsef0e9e42008-03-17 12:17:41 +0000840 Loads and stores to the structure are assumed not to trap.
Duncan Sands616cc032008-02-18 04:19:38 +0000841 May only be applied to the first parameter.</dd>
Chris Lattner275e6be2008-01-11 06:20:47 +0000842
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000843 <dt><tt>noalias</tt></dt>
Owen Andersonc4fc4cd2008-02-18 04:09:01 +0000844 <dd>This indicates that the parameter does not alias any global or any other
845 parameter. The caller is responsible for ensuring that this is the case,
846 usually by placing the value in a stack allocation.</dd>
Chris Lattner275e6be2008-01-11 06:20:47 +0000847
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000848 <dt><tt>noreturn</tt></dt>
849 <dd>This function attribute indicates that the function never returns. This
850 indicates to LLVM that every call to this function should be treated as if
851 an <tt>unreachable</tt> instruction immediately followed the call.</dd>
Chris Lattner275e6be2008-01-11 06:20:47 +0000852
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000853 <dt><tt>nounwind</tt></dt>
Duncan Sandsef0e9e42008-03-17 12:17:41 +0000854 <dd>This function attribute indicates that no exceptions unwind out of the
855 function. Usually this is because the function makes no use of exceptions,
856 but it may also be that the function catches any exceptions thrown when
857 executing it.</dd>
858
Duncan Sands4ee46812007-07-27 19:57:41 +0000859 <dt><tt>nest</tt></dt>
860 <dd>This indicates that the parameter can be excised using the
861 <a href="#int_trampoline">trampoline intrinsics</a>.</dd>
Duncan Sands13e13f82007-11-22 20:23:04 +0000862 <dt><tt>readonly</tt></dt>
Duncan Sandsd69c0e62007-11-14 21:14:02 +0000863 <dd>This function attribute indicates that the function has no side-effects
Duncan Sands13e13f82007-11-22 20:23:04 +0000864 except for producing a return value or throwing an exception. The value
865 returned must only depend on the function arguments and/or global variables.
866 It may use values obtained by dereferencing pointers.</dd>
867 <dt><tt>readnone</tt></dt>
868 <dd>A <tt>readnone</tt> function has the same restrictions as a <tt>readonly</tt>
Duncan Sandsd69c0e62007-11-14 21:14:02 +0000869 function, but in addition it is not allowed to dereference any pointer arguments
870 or global variables.
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000871 </dl>
872
873</div>
874
875<!-- ======================================================================= -->
876<div class="doc_subsection">
Gordon Henriksen13fe5e32007-12-10 03:18:06 +0000877 <a name="gc">Garbage Collector Names</a>
878</div>
879
880<div class="doc_text">
881<p>Each function may specify a garbage collector name, which is simply a
882string.</p>
883
884<div class="doc_code"><pre
885>define void @f() gc "name" { ...</pre></div>
886
887<p>The compiler declares the supported values of <i>name</i>. Specifying a
888collector which will cause the compiler to alter its output in order to support
889the named garbage collection algorithm.</p>
890</div>
891
892<!-- ======================================================================= -->
893<div class="doc_subsection">
Dan Gohmanf17a25c2007-07-18 16:29:46 +0000894 <a name="moduleasm">Module-Level Inline Assembly</a>
895</div>
896
897<div class="doc_text">
898<p>
899Modules may contain "module-level inline asm" blocks, which corresponds to the
900GCC "file scope inline asm" blocks. These blocks are internally concatenated by
901LLVM and treated as a single unit, but may be separated in the .ll file if
902desired. The syntax is very simple:
903</p>
904
905<div class="doc_code">
906<pre>
907module asm "inline asm code goes here"
908module asm "more can go here"
909</pre>
910</div>
911
912<p>The strings can contain any character by escaping non-printable characters.
913 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
914 for the number.
915</p>
916
917<p>
918 The inline asm code is simply printed to the machine code .s file when
919 assembly code is generated.
920</p>
921</div>
922
923<!-- ======================================================================= -->
924<div class="doc_subsection">
925 <a name="datalayout">Data Layout</a>
926</div>
927
928<div class="doc_text">
929<p>A module may specify a target specific data layout string that specifies how
930data is to be laid out in memory. The syntax for the data layout is simply:</p>
931<pre> target datalayout = "<i>layout specification</i>"</pre>
932<p>The <i>layout specification</i> consists of a list of specifications
933separated by the minus sign character ('-'). Each specification starts with a
934letter and may include other information after the letter to define some
935aspect of the data layout. The specifications accepted are as follows: </p>
936<dl>
937 <dt><tt>E</tt></dt>
938 <dd>Specifies that the target lays out data in big-endian form. That is, the
939 bits with the most significance have the lowest address location.</dd>
940 <dt><tt>e</tt></dt>
941 <dd>Specifies that hte target lays out data in little-endian form. That is,
942 the bits with the least significance have the lowest address location.</dd>
943 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
944 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
945 <i>preferred</i> alignments. All sizes are in bits. Specifying the <i>pref</i>
946 alignment is optional. If omitted, the preceding <tt>:</tt> should be omitted
947 too.</dd>
948 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
949 <dd>This specifies the alignment for an integer type of a given bit
950 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
951 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
952 <dd>This specifies the alignment for a vector type of a given bit
953 <i>size</i>.</dd>
954 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
955 <dd>This specifies the alignment for a floating point type of a given bit
956 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
957 (double).</dd>
958 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
959 <dd>This specifies the alignment for an aggregate type of a given bit
960 <i>size</i>.</dd>
961</dl>
962<p>When constructing the data layout for a given target, LLVM starts with a
963default set of specifications which are then (possibly) overriden by the
964specifications in the <tt>datalayout</tt> keyword. The default specifications
965are given in this list:</p>
966<ul>
967 <li><tt>E</tt> - big endian</li>
968 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
969 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
970 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
971 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
972 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
973 <li><tt>i64:32:64</tt> - i64 has abi alignment of 32-bits but preferred
974 alignment of 64-bits</li>
975 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
976 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
977 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
978 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
979 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
980</ul>
981<p>When llvm is determining the alignment for a given type, it uses the
982following rules:
983<ol>
984 <li>If the type sought is an exact match for one of the specifications, that
985 specification is used.</li>
986 <li>If no match is found, and the type sought is an integer type, then the
987 smallest integer type that is larger than the bitwidth of the sought type is
988 used. If none of the specifications are larger than the bitwidth then the the
989 largest integer type is used. For example, given the default specifications
990 above, the i7 type will use the alignment of i8 (next largest) while both
991 i65 and i256 will use the alignment of i64 (largest specified).</li>
992 <li>If no match is found, and the type sought is a vector type, then the
993 largest vector type that is smaller than the sought vector type will be used
994 as a fall back. This happens because <128 x double> can be implemented in
995 terms of 64 <2 x double>, for example.</li>
996</ol>
997</div>
998
999<!-- *********************************************************************** -->
1000<div class="doc_section"> <a name="typesystem">Type System</a> </div>
1001<!-- *********************************************************************** -->
1002
1003<div class="doc_text">
1004
1005<p>The LLVM type system is one of the most important features of the
1006intermediate representation. Being typed enables a number of
1007optimizations to be performed on the IR directly, without having to do
1008extra analyses on the side before the transformation. A strong type
1009system makes it easier to read the generated code and enables novel
1010analyses and transformations that are not feasible to perform on normal
1011three address code representations.</p>
1012
1013</div>
1014
1015<!-- ======================================================================= -->
Chris Lattner488772f2008-01-04 04:32:38 +00001016<div class="doc_subsection"> <a name="t_classifications">Type
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001017Classifications</a> </div>
1018<div class="doc_text">
Chris Lattner488772f2008-01-04 04:32:38 +00001019<p>The types fall into a few useful
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001020classifications:</p>
1021
1022<table border="1" cellspacing="0" cellpadding="4">
1023 <tbody>
1024 <tr><th>Classification</th><th>Types</th></tr>
1025 <tr>
Chris Lattner488772f2008-01-04 04:32:38 +00001026 <td><a href="#t_integer">integer</a></td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001027 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1028 </tr>
1029 <tr>
Chris Lattner488772f2008-01-04 04:32:38 +00001030 <td><a href="#t_floating">floating point</a></td>
1031 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001032 </tr>
1033 <tr>
1034 <td><a name="t_firstclass">first class</a></td>
Chris Lattner488772f2008-01-04 04:32:38 +00001035 <td><a href="#t_integer">integer</a>,
1036 <a href="#t_floating">floating point</a>,
1037 <a href="#t_pointer">pointer</a>,
1038 <a href="#t_vector">vector</a>
Dan Gohman74d6faf2008-05-12 23:51:09 +00001039 <a href="#t_struct">structure</a>,
1040 <a href="#t_array">array</a>,
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001041 </td>
1042 </tr>
Chris Lattner488772f2008-01-04 04:32:38 +00001043 <tr>
1044 <td><a href="#t_primitive">primitive</a></td>
1045 <td><a href="#t_label">label</a>,
1046 <a href="#t_void">void</a>,
1047 <a href="#t_integer">integer</a>,
1048 <a href="#t_floating">floating point</a>.</td>
1049 </tr>
1050 <tr>
1051 <td><a href="#t_derived">derived</a></td>
1052 <td><a href="#t_integer">integer</a>,
1053 <a href="#t_array">array</a>,
1054 <a href="#t_function">function</a>,
1055 <a href="#t_pointer">pointer</a>,
1056 <a href="#t_struct">structure</a>,
1057 <a href="#t_pstruct">packed structure</a>,
1058 <a href="#t_vector">vector</a>,
1059 <a href="#t_opaque">opaque</a>.
1060 </tr>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001061 </tbody>
1062</table>
1063
1064<p>The <a href="#t_firstclass">first class</a> types are perhaps the
1065most important. Values of these types are the only ones which can be
1066produced by instructions, passed as arguments, or used as operands to
1067instructions. This means that all structures and arrays must be
1068manipulated either by pointer or by component.</p>
1069</div>
1070
1071<!-- ======================================================================= -->
Chris Lattner488772f2008-01-04 04:32:38 +00001072<div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
Chris Lattner86437612008-01-04 04:34:14 +00001073
Chris Lattner488772f2008-01-04 04:32:38 +00001074<div class="doc_text">
1075<p>The primitive types are the fundamental building blocks of the LLVM
1076system.</p>
1077
Chris Lattner86437612008-01-04 04:34:14 +00001078</div>
1079
Chris Lattner488772f2008-01-04 04:32:38 +00001080<!-- _______________________________________________________________________ -->
1081<div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1082
1083<div class="doc_text">
1084 <table>
1085 <tbody>
1086 <tr><th>Type</th><th>Description</th></tr>
1087 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1088 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1089 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1090 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1091 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1092 </tbody>
1093 </table>
1094</div>
1095
1096<!-- _______________________________________________________________________ -->
1097<div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1098
1099<div class="doc_text">
1100<h5>Overview:</h5>
1101<p>The void type does not represent any value and has no size.</p>
1102
1103<h5>Syntax:</h5>
1104
1105<pre>
1106 void
1107</pre>
1108</div>
1109
1110<!-- _______________________________________________________________________ -->
1111<div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1112
1113<div class="doc_text">
1114<h5>Overview:</h5>
1115<p>The label type represents code labels.</p>
1116
1117<h5>Syntax:</h5>
1118
1119<pre>
1120 label
1121</pre>
1122</div>
1123
1124
1125<!-- ======================================================================= -->
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001126<div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1127
1128<div class="doc_text">
1129
1130<p>The real power in LLVM comes from the derived types in the system.
1131This is what allows a programmer to represent arrays, functions,
1132pointers, and other useful types. Note that these derived types may be
1133recursive: For example, it is possible to have a two dimensional array.</p>
1134
1135</div>
1136
1137<!-- _______________________________________________________________________ -->
1138<div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1139
1140<div class="doc_text">
1141
1142<h5>Overview:</h5>
1143<p>The integer type is a very simple derived type that simply specifies an
1144arbitrary bit width for the integer type desired. Any bit width from 1 bit to
11452^23-1 (about 8 million) can be specified.</p>
1146
1147<h5>Syntax:</h5>
1148
1149<pre>
1150 iN
1151</pre>
1152
1153<p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1154value.</p>
1155
1156<h5>Examples:</h5>
1157<table class="layout">
Chris Lattner251ab812007-12-18 06:18:21 +00001158 <tbody>
1159 <tr>
1160 <td><tt>i1</tt></td>
1161 <td>a single-bit integer.</td>
1162 </tr><tr>
1163 <td><tt>i32</tt></td>
1164 <td>a 32-bit integer.</td>
1165 </tr><tr>
1166 <td><tt>i1942652</tt></td>
1167 <td>a really big integer of over 1 million bits.</td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001168 </tr>
Chris Lattner251ab812007-12-18 06:18:21 +00001169 </tbody>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001170</table>
1171</div>
1172
1173<!-- _______________________________________________________________________ -->
1174<div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1175
1176<div class="doc_text">
1177
1178<h5>Overview:</h5>
1179
1180<p>The array type is a very simple derived type that arranges elements
1181sequentially in memory. The array type requires a size (number of
1182elements) and an underlying data type.</p>
1183
1184<h5>Syntax:</h5>
1185
1186<pre>
1187 [&lt;# elements&gt; x &lt;elementtype&gt;]
1188</pre>
1189
1190<p>The number of elements is a constant integer value; elementtype may
1191be any type with a size.</p>
1192
1193<h5>Examples:</h5>
1194<table class="layout">
1195 <tr class="layout">
Chris Lattner7311d222007-12-19 05:04:11 +00001196 <td class="left"><tt>[40 x i32]</tt></td>
1197 <td class="left">Array of 40 32-bit integer values.</td>
1198 </tr>
1199 <tr class="layout">
1200 <td class="left"><tt>[41 x i32]</tt></td>
1201 <td class="left">Array of 41 32-bit integer values.</td>
1202 </tr>
1203 <tr class="layout">
1204 <td class="left"><tt>[4 x i8]</tt></td>
1205 <td class="left">Array of 4 8-bit integer values.</td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001206 </tr>
1207</table>
1208<p>Here are some examples of multidimensional arrays:</p>
1209<table class="layout">
1210 <tr class="layout">
Chris Lattner7311d222007-12-19 05:04:11 +00001211 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1212 <td class="left">3x4 array of 32-bit integer values.</td>
1213 </tr>
1214 <tr class="layout">
1215 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1216 <td class="left">12x10 array of single precision floating point values.</td>
1217 </tr>
1218 <tr class="layout">
1219 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1220 <td class="left">2x3x4 array of 16-bit integer values.</td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001221 </tr>
1222</table>
1223
1224<p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1225length array. Normally, accesses past the end of an array are undefined in
1226LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
1227As a special case, however, zero length arrays are recognized to be variable
1228length. This allows implementation of 'pascal style arrays' with the LLVM
1229type "{ i32, [0 x float]}", for example.</p>
1230
1231</div>
1232
1233<!-- _______________________________________________________________________ -->
1234<div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1235<div class="doc_text">
Chris Lattner43030e72008-04-23 04:59:35 +00001236
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001237<h5>Overview:</h5>
Chris Lattner43030e72008-04-23 04:59:35 +00001238
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001239<p>The function type can be thought of as a function signature. It
Devang Pateld4ba41d2008-03-24 05:35:41 +00001240consists of a return type and a list of formal parameter types. The
Chris Lattner43030e72008-04-23 04:59:35 +00001241return type of a function type is a scalar type, a void type, or a struct type.
Devang Pateld5404c02008-03-24 20:52:42 +00001242If the return type is a struct type then all struct elements must be of first
Chris Lattner43030e72008-04-23 04:59:35 +00001243class types, and the struct must have at least one element.</p>
Devang Patela3cc5372008-03-10 20:49:15 +00001244
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001245<h5>Syntax:</h5>
Chris Lattner43030e72008-04-23 04:59:35 +00001246
1247<pre>
1248 &lt;returntype list&gt; (&lt;parameter list&gt;)
1249</pre>
1250
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001251<p>...where '<tt>&lt;parameter list&gt;</tt>' is a comma-separated list of type
1252specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1253which indicates that the function takes a variable number of arguments.
1254Variable argument functions can access their arguments with the <a
Devang Patela3cc5372008-03-10 20:49:15 +00001255 href="#int_varargs">variable argument handling intrinsic</a> functions.
1256'<tt>&lt;returntype list&gt;</tt>' is a comma-separated list of
1257<a href="#t_firstclass">first class</a> type specifiers.</p>
Chris Lattner43030e72008-04-23 04:59:35 +00001258
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001259<h5>Examples:</h5>
1260<table class="layout">
1261 <tr class="layout">
1262 <td class="left"><tt>i32 (i32)</tt></td>
1263 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1264 </td>
1265 </tr><tr class="layout">
Reid Spencerf234bed2007-07-19 23:13:04 +00001266 <td class="left"><tt>float&nbsp;(i16&nbsp;signext,&nbsp;i32&nbsp;*)&nbsp;*
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001267 </tt></td>
1268 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1269 an <tt>i16</tt> that should be sign extended and a
1270 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1271 <tt>float</tt>.
1272 </td>
1273 </tr><tr class="layout">
1274 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1275 <td class="left">A vararg function that takes at least one
1276 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1277 which returns an integer. This is the signature for <tt>printf</tt> in
1278 LLVM.
1279 </td>
Devang Pateld4ba41d2008-03-24 05:35:41 +00001280 </tr><tr class="layout">
1281 <td class="left"><tt>{i32, i32} (i32)</tt></td>
Devang Patel6dddba22008-03-24 18:10:52 +00001282 <td class="left">A function taking an <tt>i32></tt>, returning two
1283 <tt> i32 </tt> values as an aggregate of type <tt>{ i32, i32 }</tt>
Devang Pateld4ba41d2008-03-24 05:35:41 +00001284 </td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001285 </tr>
1286</table>
1287
1288</div>
1289<!-- _______________________________________________________________________ -->
1290<div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1291<div class="doc_text">
1292<h5>Overview:</h5>
1293<p>The structure type is used to represent a collection of data members
1294together in memory. The packing of the field types is defined to match
1295the ABI of the underlying processor. The elements of a structure may
1296be any type that has a size.</p>
1297<p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1298and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1299field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1300instruction.</p>
1301<h5>Syntax:</h5>
1302<pre> { &lt;type list&gt; }<br></pre>
1303<h5>Examples:</h5>
1304<table class="layout">
1305 <tr class="layout">
1306 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1307 <td class="left">A triple of three <tt>i32</tt> values</td>
1308 </tr><tr class="layout">
1309 <td class="left"><tt>{&nbsp;float,&nbsp;i32&nbsp;(i32)&nbsp;*&nbsp;}</tt></td>
1310 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1311 second element is a <a href="#t_pointer">pointer</a> to a
1312 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1313 an <tt>i32</tt>.</td>
1314 </tr>
1315</table>
1316</div>
1317
1318<!-- _______________________________________________________________________ -->
1319<div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1320</div>
1321<div class="doc_text">
1322<h5>Overview:</h5>
1323<p>The packed structure type is used to represent a collection of data members
1324together in memory. There is no padding between fields. Further, the alignment
1325of a packed structure is 1 byte. The elements of a packed structure may
1326be any type that has a size.</p>
1327<p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1328and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1329field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1330instruction.</p>
1331<h5>Syntax:</h5>
1332<pre> &lt; { &lt;type list&gt; } &gt; <br></pre>
1333<h5>Examples:</h5>
1334<table class="layout">
1335 <tr class="layout">
1336 <td class="left"><tt>&lt; { i32, i32, i32 } &gt;</tt></td>
1337 <td class="left">A triple of three <tt>i32</tt> values</td>
1338 </tr><tr class="layout">
Chris Lattner7311d222007-12-19 05:04:11 +00001339 <td class="left"><tt>&lt; { float, i32 (i32)* } &gt;</tt></td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001340 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1341 second element is a <a href="#t_pointer">pointer</a> to a
1342 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1343 an <tt>i32</tt>.</td>
1344 </tr>
1345</table>
1346</div>
1347
1348<!-- _______________________________________________________________________ -->
1349<div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1350<div class="doc_text">
1351<h5>Overview:</h5>
1352<p>As in many languages, the pointer type represents a pointer or
Christopher Lambdd0049d2007-12-11 09:31:00 +00001353reference to another object, which must live in memory. Pointer types may have
1354an optional address space attribute defining the target-specific numbered
1355address space where the pointed-to object resides. The default address space is
1356zero.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001357<h5>Syntax:</h5>
1358<pre> &lt;type&gt; *<br></pre>
1359<h5>Examples:</h5>
1360<table class="layout">
1361 <tr class="layout">
Chris Lattner7311d222007-12-19 05:04:11 +00001362 <td class="left"><tt>[4x i32]*</tt></td>
1363 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1364 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1365 </tr>
1366 <tr class="layout">
1367 <td class="left"><tt>i32 (i32 *) *</tt></td>
1368 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001369 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
Chris Lattner7311d222007-12-19 05:04:11 +00001370 <tt>i32</tt>.</td>
1371 </tr>
1372 <tr class="layout">
1373 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1374 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1375 that resides in address space #5.</td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001376 </tr>
1377</table>
1378</div>
1379
1380<!-- _______________________________________________________________________ -->
1381<div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1382<div class="doc_text">
1383
1384<h5>Overview:</h5>
1385
1386<p>A vector type is a simple derived type that represents a vector
1387of elements. Vector types are used when multiple primitive data
1388are operated in parallel using a single instruction (SIMD).
1389A vector type requires a size (number of
1390elements) and an underlying primitive data type. Vectors must have a power
1391of two length (1, 2, 4, 8, 16 ...). Vector types are
1392considered <a href="#t_firstclass">first class</a>.</p>
1393
1394<h5>Syntax:</h5>
1395
1396<pre>
1397 &lt; &lt;# elements&gt; x &lt;elementtype&gt; &gt;
1398</pre>
1399
1400<p>The number of elements is a constant integer value; elementtype may
1401be any integer or floating point type.</p>
1402
1403<h5>Examples:</h5>
1404
1405<table class="layout">
1406 <tr class="layout">
Chris Lattner7311d222007-12-19 05:04:11 +00001407 <td class="left"><tt>&lt;4 x i32&gt;</tt></td>
1408 <td class="left">Vector of 4 32-bit integer values.</td>
1409 </tr>
1410 <tr class="layout">
1411 <td class="left"><tt>&lt;8 x float&gt;</tt></td>
1412 <td class="left">Vector of 8 32-bit floating-point values.</td>
1413 </tr>
1414 <tr class="layout">
1415 <td class="left"><tt>&lt;2 x i64&gt;</tt></td>
1416 <td class="left">Vector of 2 64-bit integer values.</td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001417 </tr>
1418</table>
1419</div>
1420
1421<!-- _______________________________________________________________________ -->
1422<div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1423<div class="doc_text">
1424
1425<h5>Overview:</h5>
1426
1427<p>Opaque types are used to represent unknown types in the system. This
Gordon Henriksenda0706e2007-10-14 00:34:53 +00001428corresponds (for example) to the C notion of a forward declared structure type.
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001429In LLVM, opaque types can eventually be resolved to any type (not just a
1430structure type).</p>
1431
1432<h5>Syntax:</h5>
1433
1434<pre>
1435 opaque
1436</pre>
1437
1438<h5>Examples:</h5>
1439
1440<table class="layout">
1441 <tr class="layout">
Chris Lattner7311d222007-12-19 05:04:11 +00001442 <td class="left"><tt>opaque</tt></td>
1443 <td class="left">An opaque type.</td>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001444 </tr>
1445</table>
1446</div>
1447
1448
1449<!-- *********************************************************************** -->
1450<div class="doc_section"> <a name="constants">Constants</a> </div>
1451<!-- *********************************************************************** -->
1452
1453<div class="doc_text">
1454
1455<p>LLVM has several different basic types of constants. This section describes
1456them all and their syntax.</p>
1457
1458</div>
1459
1460<!-- ======================================================================= -->
1461<div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1462
1463<div class="doc_text">
1464
1465<dl>
1466 <dt><b>Boolean constants</b></dt>
1467
1468 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1469 constants of the <tt><a href="#t_primitive">i1</a></tt> type.
1470 </dd>
1471
1472 <dt><b>Integer constants</b></dt>
1473
1474 <dd>Standard integers (such as '4') are constants of the <a
1475 href="#t_integer">integer</a> type. Negative numbers may be used with
1476 integer types.
1477 </dd>
1478
1479 <dt><b>Floating point constants</b></dt>
1480
1481 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1482 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
Chris Lattnerdc15b1d2008-04-01 18:45:27 +00001483 notation (see below). The assembler requires the exact decimal value of
1484 a floating-point constant. For example, the assembler accepts 1.25 but
1485 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
1486 constants must have a <a href="#t_floating">floating point</a> type. </dd>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001487
1488 <dt><b>Null pointer constants</b></dt>
1489
1490 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1491 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1492
1493</dl>
1494
1495<p>The one non-intuitive notation for constants is the optional hexadecimal form
1496of floating point constants. For example, the form '<tt>double
14970x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
14984.5e+15</tt>'. The only time hexadecimal floating point constants are required
1499(and the only time that they are generated by the disassembler) is when a
1500floating point constant must be emitted but it cannot be represented as a
1501decimal floating point number. For example, NaN's, infinities, and other
1502special values are represented in their IEEE hexadecimal format so that
1503assembly and disassembly do not cause any bits to change in the constants.</p>
1504
1505</div>
1506
1507<!-- ======================================================================= -->
1508<div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1509</div>
1510
1511<div class="doc_text">
1512<p>Aggregate constants arise from aggregation of simple constants
1513and smaller aggregate constants.</p>
1514
1515<dl>
1516 <dt><b>Structure constants</b></dt>
1517
1518 <dd>Structure constants are represented with notation similar to structure
1519 type definitions (a comma separated list of elements, surrounded by braces
Chris Lattner6c8de962007-12-25 20:34:52 +00001520 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
1521 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>". Structure constants
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001522 must have <a href="#t_struct">structure type</a>, and the number and
1523 types of elements must match those specified by the type.
1524 </dd>
1525
1526 <dt><b>Array constants</b></dt>
1527
1528 <dd>Array constants are represented with notation similar to array type
1529 definitions (a comma separated list of elements, surrounded by square brackets
1530 (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74 ]</tt>". Array
1531 constants must have <a href="#t_array">array type</a>, and the number and
1532 types of elements must match those specified by the type.
1533 </dd>
1534
1535 <dt><b>Vector constants</b></dt>
1536
1537 <dd>Vector constants are represented with notation similar to vector type
1538 definitions (a comma separated list of elements, surrounded by
1539 less-than/greater-than's (<tt>&lt;&gt;</tt>)). For example: "<tt>&lt; i32 42,
1540 i32 11, i32 74, i32 100 &gt;</tt>". Vector constants must have <a
1541 href="#t_vector">vector type</a>, and the number and types of elements must
1542 match those specified by the type.
1543 </dd>
1544
1545 <dt><b>Zero initialization</b></dt>
1546
1547 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1548 value to zero of <em>any</em> type, including scalar and aggregate types.
1549 This is often used to avoid having to print large zero initializers (e.g. for
1550 large arrays) and is always exactly equivalent to using explicit zero
1551 initializers.
1552 </dd>
1553</dl>
1554
1555</div>
1556
1557<!-- ======================================================================= -->
1558<div class="doc_subsection">
1559 <a name="globalconstants">Global Variable and Function Addresses</a>
1560</div>
1561
1562<div class="doc_text">
1563
1564<p>The addresses of <a href="#globalvars">global variables</a> and <a
1565href="#functionstructure">functions</a> are always implicitly valid (link-time)
1566constants. These constants are explicitly referenced when the <a
1567href="#identifiers">identifier for the global</a> is used and always have <a
1568href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1569file:</p>
1570
1571<div class="doc_code">
1572<pre>
1573@X = global i32 17
1574@Y = global i32 42
1575@Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1576</pre>
1577</div>
1578
1579</div>
1580
1581<!-- ======================================================================= -->
1582<div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1583<div class="doc_text">
1584 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1585 no specific value. Undefined values may be of any type and be used anywhere
1586 a constant is permitted.</p>
1587
1588 <p>Undefined values indicate to the compiler that the program is well defined
1589 no matter what value is used, giving the compiler more freedom to optimize.
1590 </p>
1591</div>
1592
1593<!-- ======================================================================= -->
1594<div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1595</div>
1596
1597<div class="doc_text">
1598
1599<p>Constant expressions are used to allow expressions involving other constants
1600to be used as constants. Constant expressions may be of any <a
1601href="#t_firstclass">first class</a> type and may involve any LLVM operation
1602that does not have side effects (e.g. load and call are not supported). The
1603following is the syntax for constant expressions:</p>
1604
1605<dl>
1606 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1607 <dd>Truncate a constant to another type. The bit size of CST must be larger
1608 than the bit size of TYPE. Both types must be integers.</dd>
1609
1610 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1611 <dd>Zero extend a constant to another type. The bit size of CST must be
1612 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1613
1614 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1615 <dd>Sign extend a constant to another type. The bit size of CST must be
1616 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1617
1618 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1619 <dd>Truncate a floating point constant to another floating point type. The
1620 size of CST must be larger than the size of TYPE. Both types must be
1621 floating point.</dd>
1622
1623 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1624 <dd>Floating point extend a constant to another type. The size of CST must be
1625 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1626
Reid Spencere6adee82007-07-31 14:40:14 +00001627 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001628 <dd>Convert a floating point constant to the corresponding unsigned integer
Nate Begeman78246ca2007-11-17 03:58:34 +00001629 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1630 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1631 of the same number of elements. If the value won't fit in the integer type,
1632 the results are undefined.</dd>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001633
1634 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
1635 <dd>Convert a floating point constant to the corresponding signed integer
Nate Begeman78246ca2007-11-17 03:58:34 +00001636 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1637 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1638 of the same number of elements. If the value won't fit in the integer type,
1639 the results are undefined.</dd>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001640
1641 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
1642 <dd>Convert an unsigned integer constant to the corresponding floating point
Nate Begeman78246ca2007-11-17 03:58:34 +00001643 constant. TYPE must be a scalar or vector floating point type. CST must be of
1644 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1645 of the same number of elements. If the value won't fit in the floating point
1646 type, the results are undefined.</dd>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001647
1648 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
1649 <dd>Convert a signed integer constant to the corresponding floating point
Nate Begeman78246ca2007-11-17 03:58:34 +00001650 constant. TYPE must be a scalar or vector floating point type. CST must be of
1651 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1652 of the same number of elements. If the value won't fit in the floating point
1653 type, the results are undefined.</dd>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001654
1655 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
1656 <dd>Convert a pointer typed constant to the corresponding integer constant
1657 TYPE must be an integer type. CST must be of pointer type. The CST value is
1658 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1659
1660 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
1661 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1662 pointer type. CST must be of integer type. The CST value is zero extended,
1663 truncated, or unchanged to make it fit in a pointer size. This one is
1664 <i>really</i> dangerous!</dd>
1665
1666 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
1667 <dd>Convert a constant, CST, to another TYPE. The size of CST and TYPE must be
1668 identical (same number of bits). The conversion is done as if the CST value
1669 was stored to memory and read back as TYPE. In other words, no bits change
1670 with this operator, just the type. This can be used for conversion of
1671 vector types to any other type, as long as they have the same bit width. For
1672 pointers it is only valid to cast to another pointer type.
1673 </dd>
1674
1675 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1676
1677 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1678 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1679 instruction, the index list may have zero or more indexes, which are required
1680 to make sense for the type of "CSTPTR".</dd>
1681
1682 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1683
1684 <dd>Perform the <a href="#i_select">select operation</a> on
1685 constants.</dd>
1686
1687 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
1688 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
1689
1690 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
1691 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
1692
Nate Begeman646fa482008-05-12 19:01:56 +00001693 <dt><b><tt>vicmp COND ( VAL1, VAL2 )</tt></b></dt>
1694 <dd>Performs the <a href="#i_vicmp">vicmp operation</a> on constants.</dd>
1695
1696 <dt><b><tt>vfcmp COND ( VAL1, VAL2 )</tt></b></dt>
1697 <dd>Performs the <a href="#i_vfcmp">vfcmp operation</a> on constants.</dd>
1698
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001699 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1700
1701 <dd>Perform the <a href="#i_extractelement">extractelement
1702 operation</a> on constants.
1703
1704 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1705
1706 <dd>Perform the <a href="#i_insertelement">insertelement
1707 operation</a> on constants.</dd>
1708
1709
1710 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
1711
1712 <dd>Perform the <a href="#i_shufflevector">shufflevector
1713 operation</a> on constants.</dd>
1714
1715 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1716
1717 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1718 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1719 binary</a> operations. The constraints on operands are the same as those for
1720 the corresponding instruction (e.g. no bitwise operations on floating point
1721 values are allowed).</dd>
1722</dl>
1723</div>
1724
1725<!-- *********************************************************************** -->
1726<div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1727<!-- *********************************************************************** -->
1728
1729<!-- ======================================================================= -->
1730<div class="doc_subsection">
1731<a name="inlineasm">Inline Assembler Expressions</a>
1732</div>
1733
1734<div class="doc_text">
1735
1736<p>
1737LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
1738Module-Level Inline Assembly</a>) through the use of a special value. This
1739value represents the inline assembler as a string (containing the instructions
1740to emit), a list of operand constraints (stored as a string), and a flag that
1741indicates whether or not the inline asm expression has side effects. An example
1742inline assembler expression is:
1743</p>
1744
1745<div class="doc_code">
1746<pre>
1747i32 (i32) asm "bswap $0", "=r,r"
1748</pre>
1749</div>
1750
1751<p>
1752Inline assembler expressions may <b>only</b> be used as the callee operand of
1753a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
1754</p>
1755
1756<div class="doc_code">
1757<pre>
1758%X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
1759</pre>
1760</div>
1761
1762<p>
1763Inline asms with side effects not visible in the constraint list must be marked
1764as having side effects. This is done through the use of the
1765'<tt>sideeffect</tt>' keyword, like so:
1766</p>
1767
1768<div class="doc_code">
1769<pre>
1770call void asm sideeffect "eieio", ""()
1771</pre>
1772</div>
1773
1774<p>TODO: The format of the asm and constraints string still need to be
1775documented here. Constraints on what can be done (e.g. duplication, moving, etc
1776need to be documented).
1777</p>
1778
1779</div>
1780
1781<!-- *********************************************************************** -->
1782<div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1783<!-- *********************************************************************** -->
1784
1785<div class="doc_text">
1786
1787<p>The LLVM instruction set consists of several different
1788classifications of instructions: <a href="#terminators">terminator
1789instructions</a>, <a href="#binaryops">binary instructions</a>,
1790<a href="#bitwiseops">bitwise binary instructions</a>, <a
1791 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1792instructions</a>.</p>
1793
1794</div>
1795
1796<!-- ======================================================================= -->
1797<div class="doc_subsection"> <a name="terminators">Terminator
1798Instructions</a> </div>
1799
1800<div class="doc_text">
1801
1802<p>As mentioned <a href="#functionstructure">previously</a>, every
1803basic block in a program ends with a "Terminator" instruction, which
1804indicates which block should be executed after the current block is
1805finished. These terminator instructions typically yield a '<tt>void</tt>'
1806value: they produce control flow, not values (the one exception being
1807the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1808<p>There are six different terminator instructions: the '<a
1809 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1810instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1811the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1812 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1813 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1814
1815</div>
1816
1817<!-- _______________________________________________________________________ -->
1818<div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1819Instruction</a> </div>
1820<div class="doc_text">
1821<h5>Syntax:</h5>
1822<pre> ret &lt;type&gt; &lt;value&gt; <i>; Return a value from a non-void function</i>
1823 ret void <i>; Return from void function</i>
Devang Patela3cc5372008-03-10 20:49:15 +00001824 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 +00001825</pre>
Chris Lattner43030e72008-04-23 04:59:35 +00001826
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001827<h5>Overview:</h5>
Chris Lattner43030e72008-04-23 04:59:35 +00001828
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001829<p>The '<tt>ret</tt>' instruction is used to return control flow (and a
1830value) from a function back to the caller.</p>
1831<p>There are two forms of the '<tt>ret</tt>' instruction: one that
Chris Lattner43030e72008-04-23 04:59:35 +00001832returns value(s) and then causes control flow, and one that just causes
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001833control flow to occur.</p>
Chris Lattner43030e72008-04-23 04:59:35 +00001834
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001835<h5>Arguments:</h5>
Chris Lattner43030e72008-04-23 04:59:35 +00001836
1837<p>The '<tt>ret</tt>' instruction may return zero, one or multiple values.
1838The type of each return value must be a '<a href="#t_firstclass">first
1839class</a>' type. Note that a function is not <a href="#wellformed">well
1840formed</a> if there exists a '<tt>ret</tt>' instruction inside of the
1841function that returns values that do not match the return type of the
1842function.</p>
1843
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001844<h5>Semantics:</h5>
Chris Lattner43030e72008-04-23 04:59:35 +00001845
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001846<p>When the '<tt>ret</tt>' instruction is executed, control flow
1847returns back to the calling function's context. If the caller is a "<a
1848 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1849the instruction after the call. If the caller was an "<a
1850 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1851at the beginning of the "normal" destination block. If the instruction
1852returns a value, that value shall set the call or invoke instruction's
Devang Patela3cc5372008-03-10 20:49:15 +00001853return value. If the instruction returns multiple values then these
Devang Patelec8a5b02008-03-11 05:51:59 +00001854values can only be accessed through a '<a href="#i_getresult"><tt>getresult</tt>
1855</a>' instruction.</p>
Chris Lattner43030e72008-04-23 04:59:35 +00001856
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001857<h5>Example:</h5>
Chris Lattner43030e72008-04-23 04:59:35 +00001858
1859<pre>
1860 ret i32 5 <i>; Return an integer value of 5</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001861 ret void <i>; Return from a void function</i>
Devang Patela3cc5372008-03-10 20:49:15 +00001862 ret i32 4, i8 2 <i>; Return two values 4 and 2 </i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001863</pre>
1864</div>
1865<!-- _______________________________________________________________________ -->
1866<div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1867<div class="doc_text">
1868<h5>Syntax:</h5>
1869<pre> br i1 &lt;cond&gt;, label &lt;iftrue&gt;, label &lt;iffalse&gt;<br> br label &lt;dest&gt; <i>; Unconditional branch</i>
1870</pre>
1871<h5>Overview:</h5>
1872<p>The '<tt>br</tt>' instruction is used to cause control flow to
1873transfer to a different basic block in the current function. There are
1874two forms of this instruction, corresponding to a conditional branch
1875and an unconditional branch.</p>
1876<h5>Arguments:</h5>
1877<p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1878single '<tt>i1</tt>' value and two '<tt>label</tt>' values. The
1879unconditional form of the '<tt>br</tt>' instruction takes a single
1880'<tt>label</tt>' value as a target.</p>
1881<h5>Semantics:</h5>
1882<p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
1883argument is evaluated. If the value is <tt>true</tt>, control flows
1884to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1885control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1886<h5>Example:</h5>
1887<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
1888 href="#i_ret">ret</a> i32 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> i32 0<br></pre>
1889</div>
1890<!-- _______________________________________________________________________ -->
1891<div class="doc_subsubsection">
1892 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
1893</div>
1894
1895<div class="doc_text">
1896<h5>Syntax:</h5>
1897
1898<pre>
1899 switch &lt;intty&gt; &lt;value&gt;, label &lt;defaultdest&gt; [ &lt;intty&gt; &lt;val&gt;, label &lt;dest&gt; ... ]
1900</pre>
1901
1902<h5>Overview:</h5>
1903
1904<p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
1905several different places. It is a generalization of the '<tt>br</tt>'
1906instruction, allowing a branch to occur to one of many possible
1907destinations.</p>
1908
1909
1910<h5>Arguments:</h5>
1911
1912<p>The '<tt>switch</tt>' instruction uses three parameters: an integer
1913comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
1914an array of pairs of comparison value constants and '<tt>label</tt>'s. The
1915table is not allowed to contain duplicate constant entries.</p>
1916
1917<h5>Semantics:</h5>
1918
1919<p>The <tt>switch</tt> instruction specifies a table of values and
1920destinations. When the '<tt>switch</tt>' instruction is executed, this
1921table is searched for the given value. If the value is found, control flow is
1922transfered to the corresponding destination; otherwise, control flow is
1923transfered to the default destination.</p>
1924
1925<h5>Implementation:</h5>
1926
1927<p>Depending on properties of the target machine and the particular
1928<tt>switch</tt> instruction, this instruction may be code generated in different
1929ways. For example, it could be generated as a series of chained conditional
1930branches or with a lookup table.</p>
1931
1932<h5>Example:</h5>
1933
1934<pre>
1935 <i>; Emulate a conditional br instruction</i>
1936 %Val = <a href="#i_zext">zext</a> i1 %value to i32
1937 switch i32 %Val, label %truedest [i32 0, label %falsedest ]
1938
1939 <i>; Emulate an unconditional br instruction</i>
1940 switch i32 0, label %dest [ ]
1941
1942 <i>; Implement a jump table:</i>
1943 switch i32 %val, label %otherwise [ i32 0, label %onzero
1944 i32 1, label %onone
1945 i32 2, label %ontwo ]
1946</pre>
1947</div>
1948
1949<!-- _______________________________________________________________________ -->
1950<div class="doc_subsubsection">
1951 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
1952</div>
1953
1954<div class="doc_text">
1955
1956<h5>Syntax:</h5>
1957
1958<pre>
Nick Lewyckya1c11a12008-03-16 07:18:12 +00001959 &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 +00001960 to label &lt;normal label&gt; unwind label &lt;exception label&gt;
1961</pre>
1962
1963<h5>Overview:</h5>
1964
1965<p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
1966function, with the possibility of control flow transfer to either the
1967'<tt>normal</tt>' label or the
1968'<tt>exception</tt>' label. If the callee function returns with the
1969"<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
1970"normal" label. If the callee (or any indirect callees) returns with the "<a
1971href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
Devang Patela3cc5372008-03-10 20:49:15 +00001972continued at the dynamically nearest "exception" label. If the callee function
Devang Patelec8a5b02008-03-11 05:51:59 +00001973returns multiple values then individual return values are only accessible through
1974a '<tt><a href="#i_getresult">getresult</a></tt>' instruction.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00001975
1976<h5>Arguments:</h5>
1977
1978<p>This instruction requires several arguments:</p>
1979
1980<ol>
1981 <li>
1982 The optional "cconv" marker indicates which <a href="#callingconv">calling
1983 convention</a> the call should use. If none is specified, the call defaults
1984 to using C calling conventions.
1985 </li>
1986 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
1987 function value being invoked. In most cases, this is a direct function
1988 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
1989 an arbitrary pointer to function value.
1990 </li>
1991
1992 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
1993 function to be invoked. </li>
1994
1995 <li>'<tt>function args</tt>': argument list whose types match the function
1996 signature argument types. If the function signature indicates the function
1997 accepts a variable number of arguments, the extra arguments can be
1998 specified. </li>
1999
2000 <li>'<tt>normal label</tt>': the label reached when the called function
2001 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
2002
2003 <li>'<tt>exception label</tt>': the label reached when a callee returns with
2004 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
2005
2006</ol>
2007
2008<h5>Semantics:</h5>
2009
2010<p>This instruction is designed to operate as a standard '<tt><a
2011href="#i_call">call</a></tt>' instruction in most regards. The primary
2012difference is that it establishes an association with a label, which is used by
2013the runtime library to unwind the stack.</p>
2014
2015<p>This instruction is used in languages with destructors to ensure that proper
2016cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
2017exception. Additionally, this is important for implementation of
2018'<tt>catch</tt>' clauses in high-level languages that support them.</p>
2019
2020<h5>Example:</h5>
2021<pre>
Nick Lewyckya1c11a12008-03-16 07:18:12 +00002022 %retval = invoke i32 @Test(i32 15) to label %Continue
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002023 unwind label %TestCleanup <i>; {i32}:retval set</i>
Nick Lewyckya1c11a12008-03-16 07:18:12 +00002024 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002025 unwind label %TestCleanup <i>; {i32}:retval set</i>
2026</pre>
2027</div>
2028
2029
2030<!-- _______________________________________________________________________ -->
2031
2032<div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
2033Instruction</a> </div>
2034
2035<div class="doc_text">
2036
2037<h5>Syntax:</h5>
2038<pre>
2039 unwind
2040</pre>
2041
2042<h5>Overview:</h5>
2043
2044<p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
2045at the first callee in the dynamic call stack which used an <a
2046href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
2047primarily used to implement exception handling.</p>
2048
2049<h5>Semantics:</h5>
2050
Chris Lattner8b094fc2008-04-19 21:01:16 +00002051<p>The '<tt>unwind</tt>' instruction causes execution of the current function to
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002052immediately halt. The dynamic call stack is then searched for the first <a
2053href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
2054execution continues at the "exceptional" destination block specified by the
2055<tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
2056dynamic call chain, undefined behavior results.</p>
2057</div>
2058
2059<!-- _______________________________________________________________________ -->
2060
2061<div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
2062Instruction</a> </div>
2063
2064<div class="doc_text">
2065
2066<h5>Syntax:</h5>
2067<pre>
2068 unreachable
2069</pre>
2070
2071<h5>Overview:</h5>
2072
2073<p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
2074instruction is used to inform the optimizer that a particular portion of the
2075code is not reachable. This can be used to indicate that the code after a
2076no-return function cannot be reached, and other facts.</p>
2077
2078<h5>Semantics:</h5>
2079
2080<p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
2081</div>
2082
2083
2084
2085<!-- ======================================================================= -->
2086<div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
2087<div class="doc_text">
2088<p>Binary operators are used to do most of the computation in a
Chris Lattnerab596d92008-04-01 18:47:32 +00002089program. They require two operands of the same type, execute an operation on them, and
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002090produce a single value. The operands might represent
2091multiple data, as is the case with the <a href="#t_vector">vector</a> data type.
Chris Lattnerab596d92008-04-01 18:47:32 +00002092The result value has the same type as its operands.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002093<p>There are several different binary operators:</p>
2094</div>
2095<!-- _______________________________________________________________________ -->
2096<div class="doc_subsubsection"> <a name="i_add">'<tt>add</tt>'
2097Instruction</a> </div>
2098<div class="doc_text">
2099<h5>Syntax:</h5>
2100<pre> &lt;result&gt; = add &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2101</pre>
2102<h5>Overview:</h5>
2103<p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
2104<h5>Arguments:</h5>
2105<p>The two arguments to the '<tt>add</tt>' instruction must be either <a
2106 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a> values.
2107 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2108Both arguments must have identical types.</p>
2109<h5>Semantics:</h5>
2110<p>The value produced is the integer or floating point sum of the two
2111operands.</p>
Chris Lattner9aba1e22008-01-28 00:36:27 +00002112<p>If an integer sum has unsigned overflow, the result returned is the
2113mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2114the result.</p>
2115<p>Because LLVM integers use a two's complement representation, this
2116instruction is appropriate for both signed and unsigned integers.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002117<h5>Example:</h5>
2118<pre> &lt;result&gt; = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
2119</pre>
2120</div>
2121<!-- _______________________________________________________________________ -->
2122<div class="doc_subsubsection"> <a name="i_sub">'<tt>sub</tt>'
2123Instruction</a> </div>
2124<div class="doc_text">
2125<h5>Syntax:</h5>
2126<pre> &lt;result&gt; = sub &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2127</pre>
2128<h5>Overview:</h5>
2129<p>The '<tt>sub</tt>' instruction returns the difference of its two
2130operands.</p>
2131<p>Note that the '<tt>sub</tt>' instruction is used to represent the '<tt>neg</tt>'
2132instruction present in most other intermediate representations.</p>
2133<h5>Arguments:</h5>
2134<p>The two arguments to the '<tt>sub</tt>' instruction must be either <a
2135 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
2136values.
2137This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2138Both arguments must have identical types.</p>
2139<h5>Semantics:</h5>
2140<p>The value produced is the integer or floating point difference of
2141the two operands.</p>
Chris Lattner9aba1e22008-01-28 00:36:27 +00002142<p>If an integer difference has unsigned overflow, the result returned is the
2143mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2144the result.</p>
2145<p>Because LLVM integers use a two's complement representation, this
2146instruction is appropriate for both signed and unsigned integers.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002147<h5>Example:</h5>
2148<pre>
2149 &lt;result&gt; = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
2150 &lt;result&gt; = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
2151</pre>
2152</div>
2153<!-- _______________________________________________________________________ -->
2154<div class="doc_subsubsection"> <a name="i_mul">'<tt>mul</tt>'
2155Instruction</a> </div>
2156<div class="doc_text">
2157<h5>Syntax:</h5>
2158<pre> &lt;result&gt; = mul &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2159</pre>
2160<h5>Overview:</h5>
2161<p>The '<tt>mul</tt>' instruction returns the product of its two
2162operands.</p>
2163<h5>Arguments:</h5>
2164<p>The two arguments to the '<tt>mul</tt>' instruction must be either <a
2165 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
2166values.
2167This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2168Both arguments must have identical types.</p>
2169<h5>Semantics:</h5>
2170<p>The value produced is the integer or floating point product of the
2171two operands.</p>
Chris Lattner9aba1e22008-01-28 00:36:27 +00002172<p>If the result of an integer multiplication has unsigned overflow,
2173the result returned is the mathematical result modulo
21742<sup>n</sup>, where n is the bit width of the result.</p>
2175<p>Because LLVM integers use a two's complement representation, and the
2176result is the same width as the operands, this instruction returns the
2177correct result for both signed and unsigned integers. If a full product
2178(e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands
2179should be sign-extended or zero-extended as appropriate to the
2180width of the full product.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002181<h5>Example:</h5>
2182<pre> &lt;result&gt; = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
2183</pre>
2184</div>
2185<!-- _______________________________________________________________________ -->
2186<div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
2187</a></div>
2188<div class="doc_text">
2189<h5>Syntax:</h5>
2190<pre> &lt;result&gt; = udiv &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2191</pre>
2192<h5>Overview:</h5>
2193<p>The '<tt>udiv</tt>' instruction returns the quotient of its two
2194operands.</p>
2195<h5>Arguments:</h5>
2196<p>The two arguments to the '<tt>udiv</tt>' instruction must be
2197<a href="#t_integer">integer</a> values. Both arguments must have identical
2198types. This instruction can also take <a href="#t_vector">vector</a> versions
2199of the values in which case the elements must be integers.</p>
2200<h5>Semantics:</h5>
Chris Lattner9aba1e22008-01-28 00:36:27 +00002201<p>The value produced is the unsigned integer quotient of the two operands.</p>
2202<p>Note that unsigned integer division and signed integer division are distinct
2203operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
2204<p>Division by zero leads to undefined behavior.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002205<h5>Example:</h5>
2206<pre> &lt;result&gt; = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2207</pre>
2208</div>
2209<!-- _______________________________________________________________________ -->
2210<div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
2211</a> </div>
2212<div class="doc_text">
2213<h5>Syntax:</h5>
2214<pre> &lt;result&gt; = sdiv &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2215</pre>
2216<h5>Overview:</h5>
2217<p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
2218operands.</p>
2219<h5>Arguments:</h5>
2220<p>The two arguments to the '<tt>sdiv</tt>' instruction must be
2221<a href="#t_integer">integer</a> values. Both arguments must have identical
2222types. This instruction can also take <a href="#t_vector">vector</a> versions
2223of the values in which case the elements must be integers.</p>
2224<h5>Semantics:</h5>
Chris Lattnerdc15b1d2008-04-01 18:45:27 +00002225<p>The value produced is the signed integer quotient of the two operands rounded towards zero.</p>
Chris Lattner9aba1e22008-01-28 00:36:27 +00002226<p>Note that signed integer division and unsigned integer division are distinct
2227operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
2228<p>Division by zero leads to undefined behavior. Overflow also leads to
2229undefined behavior; this is a rare case, but can occur, for example,
2230by doing a 32-bit division of -2147483648 by -1.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002231<h5>Example:</h5>
2232<pre> &lt;result&gt; = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2233</pre>
2234</div>
2235<!-- _______________________________________________________________________ -->
2236<div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
2237Instruction</a> </div>
2238<div class="doc_text">
2239<h5>Syntax:</h5>
2240<pre> &lt;result&gt; = fdiv &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2241</pre>
2242<h5>Overview:</h5>
2243<p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
2244operands.</p>
2245<h5>Arguments:</h5>
2246<p>The two arguments to the '<tt>fdiv</tt>' instruction must be
2247<a href="#t_floating">floating point</a> values. Both arguments must have
2248identical types. This instruction can also take <a href="#t_vector">vector</a>
2249versions of floating point values.</p>
2250<h5>Semantics:</h5>
2251<p>The value produced is the floating point quotient of the two operands.</p>
2252<h5>Example:</h5>
2253<pre> &lt;result&gt; = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
2254</pre>
2255</div>
2256<!-- _______________________________________________________________________ -->
2257<div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
2258</div>
2259<div class="doc_text">
2260<h5>Syntax:</h5>
2261<pre> &lt;result&gt; = urem &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2262</pre>
2263<h5>Overview:</h5>
2264<p>The '<tt>urem</tt>' instruction returns the remainder from the
2265unsigned division of its two arguments.</p>
2266<h5>Arguments:</h5>
2267<p>The two arguments to the '<tt>urem</tt>' instruction must be
2268<a href="#t_integer">integer</a> values. Both arguments must have identical
Dan Gohman3e3fd8c2007-11-05 23:35:22 +00002269types. This instruction can also take <a href="#t_vector">vector</a> versions
2270of the values in which case the elements must be integers.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002271<h5>Semantics:</h5>
2272<p>This instruction returns the unsigned integer <i>remainder</i> of a division.
Chris Lattnerdc15b1d2008-04-01 18:45:27 +00002273This instruction always performs an unsigned division to get the remainder.</p>
Chris Lattner9aba1e22008-01-28 00:36:27 +00002274<p>Note that unsigned integer remainder and signed integer remainder are
2275distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
2276<p>Taking the remainder of a division by zero leads to undefined behavior.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002277<h5>Example:</h5>
2278<pre> &lt;result&gt; = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2279</pre>
2280
2281</div>
2282<!-- _______________________________________________________________________ -->
2283<div class="doc_subsubsection"> <a name="i_srem">'<tt>srem</tt>'
2284Instruction</a> </div>
2285<div class="doc_text">
2286<h5>Syntax:</h5>
2287<pre> &lt;result&gt; = srem &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2288</pre>
2289<h5>Overview:</h5>
2290<p>The '<tt>srem</tt>' instruction returns the remainder from the
Dan Gohman3e3fd8c2007-11-05 23:35:22 +00002291signed division of its two operands. This instruction can also take
2292<a href="#t_vector">vector</a> versions of the values in which case
2293the elements must be integers.</p>
Chris Lattner08497ce2008-01-04 04:33:49 +00002294
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002295<h5>Arguments:</h5>
2296<p>The two arguments to the '<tt>srem</tt>' instruction must be
2297<a href="#t_integer">integer</a> values. Both arguments must have identical
2298types.</p>
2299<h5>Semantics:</h5>
2300<p>This instruction returns the <i>remainder</i> of a division (where the result
2301has the same sign as the dividend, <tt>var1</tt>), not the <i>modulo</i>
2302operator (where the result has the same sign as the divisor, <tt>var2</tt>) of
2303a value. For more information about the difference, see <a
2304 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
2305Math Forum</a>. For a table of how this is implemented in various languages,
2306please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
2307Wikipedia: modulo operation</a>.</p>
Chris Lattner9aba1e22008-01-28 00:36:27 +00002308<p>Note that signed integer remainder and unsigned integer remainder are
2309distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
2310<p>Taking the remainder of a division by zero leads to undefined behavior.
2311Overflow also leads to undefined behavior; this is a rare case, but can occur,
2312for example, by taking the remainder of a 32-bit division of -2147483648 by -1.
2313(The remainder doesn't actually overflow, but this rule lets srem be
2314implemented using instructions that return both the result of the division
2315and the remainder.)</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002316<h5>Example:</h5>
2317<pre> &lt;result&gt; = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2318</pre>
2319
2320</div>
2321<!-- _______________________________________________________________________ -->
2322<div class="doc_subsubsection"> <a name="i_frem">'<tt>frem</tt>'
2323Instruction</a> </div>
2324<div class="doc_text">
2325<h5>Syntax:</h5>
2326<pre> &lt;result&gt; = frem &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2327</pre>
2328<h5>Overview:</h5>
2329<p>The '<tt>frem</tt>' instruction returns the remainder from the
2330division of its two operands.</p>
2331<h5>Arguments:</h5>
2332<p>The two arguments to the '<tt>frem</tt>' instruction must be
2333<a href="#t_floating">floating point</a> values. Both arguments must have
Dan Gohman3e3fd8c2007-11-05 23:35:22 +00002334identical types. This instruction can also take <a href="#t_vector">vector</a>
2335versions of floating point values.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002336<h5>Semantics:</h5>
Chris Lattnerdc15b1d2008-04-01 18:45:27 +00002337<p>This instruction returns the <i>remainder</i> of a division.
2338The remainder has the same sign as the dividend.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002339<h5>Example:</h5>
2340<pre> &lt;result&gt; = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
2341</pre>
2342</div>
2343
2344<!-- ======================================================================= -->
2345<div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
2346Operations</a> </div>
2347<div class="doc_text">
2348<p>Bitwise binary operators are used to do various forms of
2349bit-twiddling in a program. They are generally very efficient
2350instructions and can commonly be strength reduced from other
Chris Lattnerdc15b1d2008-04-01 18:45:27 +00002351instructions. They require two operands of the same type, execute an operation on them,
2352and produce a single value. The resulting value is the same type as its operands.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002353</div>
2354
2355<!-- _______________________________________________________________________ -->
2356<div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
2357Instruction</a> </div>
2358<div class="doc_text">
2359<h5>Syntax:</h5>
2360<pre> &lt;result&gt; = shl &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2361</pre>
Chris Lattnerd939d9f2007-10-03 21:01:14 +00002362
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002363<h5>Overview:</h5>
Chris Lattnerd939d9f2007-10-03 21:01:14 +00002364
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002365<p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2366the left a specified number of bits.</p>
Chris Lattnerd939d9f2007-10-03 21:01:14 +00002367
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002368<h5>Arguments:</h5>
Chris Lattnerd939d9f2007-10-03 21:01:14 +00002369
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002370<p>Both arguments to the '<tt>shl</tt>' instruction must be the same <a
Chris Lattner8b094fc2008-04-19 21:01:16 +00002371 href="#t_integer">integer</a> type. '<tt>var2</tt>' is treated as an
2372unsigned value.</p>
Chris Lattnerd939d9f2007-10-03 21:01:14 +00002373
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002374<h5>Semantics:</h5>
Chris Lattnerd939d9f2007-10-03 21:01:14 +00002375
Chris Lattnerdc15b1d2008-04-01 18:45:27 +00002376<p>The value produced is <tt>var1</tt> * 2<sup><tt>var2</tt></sup> mod 2<sup>n</sup>,
2377where n is the width of the result. If <tt>var2</tt> is (statically or dynamically) negative or
2378equal to or larger than the number of bits in <tt>var1</tt>, the result is undefined.</p>
Chris Lattnerd939d9f2007-10-03 21:01:14 +00002379
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002380<h5>Example:</h5><pre>
2381 &lt;result&gt; = shl i32 4, %var <i>; yields {i32}: 4 &lt;&lt; %var</i>
2382 &lt;result&gt; = shl i32 4, 2 <i>; yields {i32}: 16</i>
2383 &lt;result&gt; = shl i32 1, 10 <i>; yields {i32}: 1024</i>
Chris Lattnerd939d9f2007-10-03 21:01:14 +00002384 &lt;result&gt; = shl i32 1, 32 <i>; undefined</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002385</pre>
2386</div>
2387<!-- _______________________________________________________________________ -->
2388<div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
2389Instruction</a> </div>
2390<div class="doc_text">
2391<h5>Syntax:</h5>
2392<pre> &lt;result&gt; = lshr &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2393</pre>
2394
2395<h5>Overview:</h5>
2396<p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
2397operand shifted to the right a specified number of bits with zero fill.</p>
2398
2399<h5>Arguments:</h5>
2400<p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
Chris Lattner8b094fc2008-04-19 21:01:16 +00002401<a href="#t_integer">integer</a> type. '<tt>var2</tt>' is treated as an
2402unsigned value.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002403
2404<h5>Semantics:</h5>
Chris Lattnerd939d9f2007-10-03 21:01:14 +00002405
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002406<p>This instruction always performs a logical shift right operation. The most
2407significant bits of the result will be filled with zero bits after the
Chris Lattnerd939d9f2007-10-03 21:01:14 +00002408shift. If <tt>var2</tt> is (statically or dynamically) equal to or larger than
2409the number of bits in <tt>var1</tt>, the result is undefined.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002410
2411<h5>Example:</h5>
2412<pre>
2413 &lt;result&gt; = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
2414 &lt;result&gt; = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
2415 &lt;result&gt; = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
2416 &lt;result&gt; = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
Chris Lattnerd939d9f2007-10-03 21:01:14 +00002417 &lt;result&gt; = lshr i32 1, 32 <i>; undefined</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002418</pre>
2419</div>
2420
2421<!-- _______________________________________________________________________ -->
2422<div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
2423Instruction</a> </div>
2424<div class="doc_text">
2425
2426<h5>Syntax:</h5>
2427<pre> &lt;result&gt; = ashr &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2428</pre>
2429
2430<h5>Overview:</h5>
2431<p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
2432operand shifted to the right a specified number of bits with sign extension.</p>
2433
2434<h5>Arguments:</h5>
2435<p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
Chris Lattner8b094fc2008-04-19 21:01:16 +00002436<a href="#t_integer">integer</a> type. '<tt>var2</tt>' is treated as an
2437unsigned value.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002438
2439<h5>Semantics:</h5>
2440<p>This instruction always performs an arithmetic shift right operation,
2441The most significant bits of the result will be filled with the sign bit
Chris Lattnerd939d9f2007-10-03 21:01:14 +00002442of <tt>var1</tt>. If <tt>var2</tt> is (statically or dynamically) equal to or
2443larger than the number of bits in <tt>var1</tt>, the result is undefined.
2444</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002445
2446<h5>Example:</h5>
2447<pre>
2448 &lt;result&gt; = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
2449 &lt;result&gt; = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
2450 &lt;result&gt; = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
2451 &lt;result&gt; = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
Chris Lattnerd939d9f2007-10-03 21:01:14 +00002452 &lt;result&gt; = ashr i32 1, 32 <i>; undefined</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002453</pre>
2454</div>
2455
2456<!-- _______________________________________________________________________ -->
2457<div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
2458Instruction</a> </div>
2459<div class="doc_text">
2460<h5>Syntax:</h5>
2461<pre> &lt;result&gt; = and &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2462</pre>
2463<h5>Overview:</h5>
2464<p>The '<tt>and</tt>' instruction returns the bitwise logical and of
2465its two operands.</p>
2466<h5>Arguments:</h5>
2467<p>The two arguments to the '<tt>and</tt>' instruction must be <a
2468 href="#t_integer">integer</a> values. Both arguments must have
2469identical types.</p>
2470<h5>Semantics:</h5>
2471<p>The truth table used for the '<tt>and</tt>' instruction is:</p>
2472<p> </p>
2473<div style="align: center">
2474<table border="1" cellspacing="0" cellpadding="4">
2475 <tbody>
2476 <tr>
2477 <td>In0</td>
2478 <td>In1</td>
2479 <td>Out</td>
2480 </tr>
2481 <tr>
2482 <td>0</td>
2483 <td>0</td>
2484 <td>0</td>
2485 </tr>
2486 <tr>
2487 <td>0</td>
2488 <td>1</td>
2489 <td>0</td>
2490 </tr>
2491 <tr>
2492 <td>1</td>
2493 <td>0</td>
2494 <td>0</td>
2495 </tr>
2496 <tr>
2497 <td>1</td>
2498 <td>1</td>
2499 <td>1</td>
2500 </tr>
2501 </tbody>
2502</table>
2503</div>
2504<h5>Example:</h5>
2505<pre> &lt;result&gt; = and i32 4, %var <i>; yields {i32}:result = 4 &amp; %var</i>
2506 &lt;result&gt; = and i32 15, 40 <i>; yields {i32}:result = 8</i>
2507 &lt;result&gt; = and i32 4, 8 <i>; yields {i32}:result = 0</i>
2508</pre>
2509</div>
2510<!-- _______________________________________________________________________ -->
2511<div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
2512<div class="doc_text">
2513<h5>Syntax:</h5>
2514<pre> &lt;result&gt; = or &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2515</pre>
2516<h5>Overview:</h5>
2517<p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
2518or of its two operands.</p>
2519<h5>Arguments:</h5>
2520<p>The two arguments to the '<tt>or</tt>' instruction must be <a
2521 href="#t_integer">integer</a> values. Both arguments must have
2522identical types.</p>
2523<h5>Semantics:</h5>
2524<p>The truth table used for the '<tt>or</tt>' instruction is:</p>
2525<p> </p>
2526<div style="align: center">
2527<table border="1" cellspacing="0" cellpadding="4">
2528 <tbody>
2529 <tr>
2530 <td>In0</td>
2531 <td>In1</td>
2532 <td>Out</td>
2533 </tr>
2534 <tr>
2535 <td>0</td>
2536 <td>0</td>
2537 <td>0</td>
2538 </tr>
2539 <tr>
2540 <td>0</td>
2541 <td>1</td>
2542 <td>1</td>
2543 </tr>
2544 <tr>
2545 <td>1</td>
2546 <td>0</td>
2547 <td>1</td>
2548 </tr>
2549 <tr>
2550 <td>1</td>
2551 <td>1</td>
2552 <td>1</td>
2553 </tr>
2554 </tbody>
2555</table>
2556</div>
2557<h5>Example:</h5>
2558<pre> &lt;result&gt; = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
2559 &lt;result&gt; = or i32 15, 40 <i>; yields {i32}:result = 47</i>
2560 &lt;result&gt; = or i32 4, 8 <i>; yields {i32}:result = 12</i>
2561</pre>
2562</div>
2563<!-- _______________________________________________________________________ -->
2564<div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
2565Instruction</a> </div>
2566<div class="doc_text">
2567<h5>Syntax:</h5>
2568<pre> &lt;result&gt; = xor &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
2569</pre>
2570<h5>Overview:</h5>
2571<p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
2572or of its two operands. The <tt>xor</tt> is used to implement the
2573"one's complement" operation, which is the "~" operator in C.</p>
2574<h5>Arguments:</h5>
2575<p>The two arguments to the '<tt>xor</tt>' instruction must be <a
2576 href="#t_integer">integer</a> values. Both arguments must have
2577identical types.</p>
2578<h5>Semantics:</h5>
2579<p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
2580<p> </p>
2581<div style="align: center">
2582<table border="1" cellspacing="0" cellpadding="4">
2583 <tbody>
2584 <tr>
2585 <td>In0</td>
2586 <td>In1</td>
2587 <td>Out</td>
2588 </tr>
2589 <tr>
2590 <td>0</td>
2591 <td>0</td>
2592 <td>0</td>
2593 </tr>
2594 <tr>
2595 <td>0</td>
2596 <td>1</td>
2597 <td>1</td>
2598 </tr>
2599 <tr>
2600 <td>1</td>
2601 <td>0</td>
2602 <td>1</td>
2603 </tr>
2604 <tr>
2605 <td>1</td>
2606 <td>1</td>
2607 <td>0</td>
2608 </tr>
2609 </tbody>
2610</table>
2611</div>
2612<p> </p>
2613<h5>Example:</h5>
2614<pre> &lt;result&gt; = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
2615 &lt;result&gt; = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
2616 &lt;result&gt; = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
2617 &lt;result&gt; = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
2618</pre>
2619</div>
2620
2621<!-- ======================================================================= -->
2622<div class="doc_subsection">
2623 <a name="vectorops">Vector Operations</a>
2624</div>
2625
2626<div class="doc_text">
2627
2628<p>LLVM supports several instructions to represent vector operations in a
2629target-independent manner. These instructions cover the element-access and
2630vector-specific operations needed to process vectors effectively. While LLVM
2631does directly support these vector operations, many sophisticated algorithms
2632will want to use target-specific intrinsics to take full advantage of a specific
2633target.</p>
2634
2635</div>
2636
2637<!-- _______________________________________________________________________ -->
2638<div class="doc_subsubsection">
2639 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2640</div>
2641
2642<div class="doc_text">
2643
2644<h5>Syntax:</h5>
2645
2646<pre>
2647 &lt;result&gt; = extractelement &lt;n x &lt;ty&gt;&gt; &lt;val&gt;, i32 &lt;idx&gt; <i>; yields &lt;ty&gt;</i>
2648</pre>
2649
2650<h5>Overview:</h5>
2651
2652<p>
2653The '<tt>extractelement</tt>' instruction extracts a single scalar
2654element from a vector at a specified index.
2655</p>
2656
2657
2658<h5>Arguments:</h5>
2659
2660<p>
2661The first operand of an '<tt>extractelement</tt>' instruction is a
2662value of <a href="#t_vector">vector</a> type. The second operand is
2663an index indicating the position from which to extract the element.
2664The index may be a variable.</p>
2665
2666<h5>Semantics:</h5>
2667
2668<p>
2669The result is a scalar of the same type as the element type of
2670<tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2671<tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2672results are undefined.
2673</p>
2674
2675<h5>Example:</h5>
2676
2677<pre>
2678 %result = extractelement &lt;4 x i32&gt; %vec, i32 0 <i>; yields i32</i>
2679</pre>
2680</div>
2681
2682
2683<!-- _______________________________________________________________________ -->
2684<div class="doc_subsubsection">
2685 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
2686</div>
2687
2688<div class="doc_text">
2689
2690<h5>Syntax:</h5>
2691
2692<pre>
Dan Gohmanbcc3c502008-05-12 23:38:42 +00002693 &lt;result&gt; = insertelement &lt;n x &lt;ty&gt;&gt; &lt;val&gt;, &lt;ty&gt; &lt;elt&gt;, i32 &lt;idx&gt; <i>; yields &lt;n x &lt;ty&gt;&gt;</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002694</pre>
2695
2696<h5>Overview:</h5>
2697
2698<p>
2699The '<tt>insertelement</tt>' instruction inserts a scalar
2700element into a vector at a specified index.
2701</p>
2702
2703
2704<h5>Arguments:</h5>
2705
2706<p>
2707The first operand of an '<tt>insertelement</tt>' instruction is a
2708value of <a href="#t_vector">vector</a> type. The second operand is a
2709scalar value whose type must equal the element type of the first
2710operand. The third operand is an index indicating the position at
2711which to insert the value. The index may be a variable.</p>
2712
2713<h5>Semantics:</h5>
2714
2715<p>
2716The result is a vector of the same type as <tt>val</tt>. Its
2717element values are those of <tt>val</tt> except at position
2718<tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
2719exceeds the length of <tt>val</tt>, the results are undefined.
2720</p>
2721
2722<h5>Example:</h5>
2723
2724<pre>
2725 %result = insertelement &lt;4 x i32&gt; %vec, i32 1, i32 0 <i>; yields &lt;4 x i32&gt;</i>
2726</pre>
2727</div>
2728
2729<!-- _______________________________________________________________________ -->
2730<div class="doc_subsubsection">
2731 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
2732</div>
2733
2734<div class="doc_text">
2735
2736<h5>Syntax:</h5>
2737
2738<pre>
2739 &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>
2740</pre>
2741
2742<h5>Overview:</h5>
2743
2744<p>
2745The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
2746from two input vectors, returning a vector of the same type.
2747</p>
2748
2749<h5>Arguments:</h5>
2750
2751<p>
2752The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
2753with types that match each other and types that match the result of the
2754instruction. The third argument is a shuffle mask, which has the same number
2755of elements as the other vector type, but whose element type is always 'i32'.
2756</p>
2757
2758<p>
2759The shuffle mask operand is required to be a constant vector with either
2760constant integer or undef values.
2761</p>
2762
2763<h5>Semantics:</h5>
2764
2765<p>
2766The elements of the two input vectors are numbered from left to right across
2767both of the vectors. The shuffle mask operand specifies, for each element of
2768the result vector, which element of the two input registers the result element
2769gets. The element selector may be undef (meaning "don't care") and the second
2770operand may be undef if performing a shuffle from only one vector.
2771</p>
2772
2773<h5>Example:</h5>
2774
2775<pre>
2776 %result = shufflevector &lt;4 x i32&gt; %v1, &lt;4 x i32&gt; %v2,
2777 &lt;4 x i32&gt; &lt;i32 0, i32 4, i32 1, i32 5&gt; <i>; yields &lt;4 x i32&gt;</i>
2778 %result = shufflevector &lt;4 x i32&gt; %v1, &lt;4 x i32&gt; undef,
2779 &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.
2780</pre>
2781</div>
2782
2783
2784<!-- ======================================================================= -->
2785<div class="doc_subsection">
Dan Gohman74d6faf2008-05-12 23:51:09 +00002786 <a name="aggregateops">Aggregate Operations</a>
2787</div>
2788
2789<div class="doc_text">
2790
2791<p>LLVM supports several instructions for working with aggregate values.
2792</p>
2793
2794</div>
2795
2796<!-- _______________________________________________________________________ -->
2797<div class="doc_subsubsection">
2798 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
2799</div>
2800
2801<div class="doc_text">
2802
2803<h5>Syntax:</h5>
2804
2805<pre>
2806 &lt;result&gt; = extractvalue &lt;aggregate type&gt; &lt;val&gt;, &lt;idx&gt;{, &lt;idx&gt;}*
2807</pre>
2808
2809<h5>Overview:</h5>
2810
2811<p>
2812The '<tt>extractvalue</tt>' instruction extracts a value
2813from an aggregate value.
2814</p>
2815
2816
2817<h5>Arguments:</h5>
2818
2819<p>
2820The first operand of an '<tt>extractvalue</tt>' instruction is a
2821value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a>
2822type. The operands are constant indicies to specify which value to extract
2823in the same manner as indicies in a
2824'<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
2825</p>
2826
2827<h5>Semantics:</h5>
2828
2829<p>
2830The result is the value at the position in the aggregate specified by
2831the index operands.
2832</p>
2833
2834<h5>Example:</h5>
2835
2836<pre>
2837 %result = extractvalue {i32, float} %agg, i32 0 <i>; yields i32</i>
2838</pre>
2839</div>
2840
2841
2842<!-- _______________________________________________________________________ -->
2843<div class="doc_subsubsection">
2844 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
2845</div>
2846
2847<div class="doc_text">
2848
2849<h5>Syntax:</h5>
2850
2851<pre>
2852 &lt;result&gt; = insertvalue &lt;aggregate type&gt; &lt;val&gt;, &lt;ty&gt; &lt;val&gt;, i32 &lt;idx&gt; <i>; yields &lt;n x &lt;ty&gt;&gt;</i>
2853</pre>
2854
2855<h5>Overview:</h5>
2856
2857<p>
2858The '<tt>insertvalue</tt>' instruction inserts a value
2859into a aggregate.
2860</p>
2861
2862
2863<h5>Arguments:</h5>
2864
2865<p>
2866The first operand of an '<tt>insertvalue</tt>' instruction is a
2867value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a> type.
2868The second operand is a first-class value to insert.
2869type of the first operand. The following operands are constant indicies
2870indicating the position at which to insert the value in the same manner as
2871indicies in a
2872'<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
2873The value to insert must have the same type as the value identified
2874by the indicies.
2875
2876<h5>Semantics:</h5>
2877
2878<p>
2879The result is an aggregate of the same type as <tt>val</tt>. Its
2880value is that of <tt>val</tt> except that the value at the position
2881specified by the indicies is that of <tt>elt</tt>.
2882</p>
2883
2884<h5>Example:</h5>
2885
2886<pre>
2887 %result = insertvalue {i32, float} %agg, i32 1, i32 0 <i>; yields {i32, float}</i>
2888</pre>
2889</div>
2890
2891
2892<!-- ======================================================================= -->
2893<div class="doc_subsection">
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002894 <a name="memoryops">Memory Access and Addressing Operations</a>
2895</div>
2896
2897<div class="doc_text">
2898
2899<p>A key design point of an SSA-based representation is how it
2900represents memory. In LLVM, no memory locations are in SSA form, which
2901makes things very simple. This section describes how to read, write,
2902allocate, and free memory in LLVM.</p>
2903
2904</div>
2905
2906<!-- _______________________________________________________________________ -->
2907<div class="doc_subsubsection">
2908 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
2909</div>
2910
2911<div class="doc_text">
2912
2913<h5>Syntax:</h5>
2914
2915<pre>
2916 &lt;result&gt; = malloc &lt;type&gt;[, i32 &lt;NumElements&gt;][, align &lt;alignment&gt;] <i>; yields {type*}:result</i>
2917</pre>
2918
2919<h5>Overview:</h5>
2920
2921<p>The '<tt>malloc</tt>' instruction allocates memory from the system
Christopher Lambcfe00962007-12-17 01:00:21 +00002922heap and returns a pointer to it. The object is always allocated in the generic
2923address space (address space zero).</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002924
2925<h5>Arguments:</h5>
2926
2927<p>The '<tt>malloc</tt>' instruction allocates
2928<tt>sizeof(&lt;type&gt;)*NumElements</tt>
2929bytes of memory from the operating system and returns a pointer of the
2930appropriate type to the program. If "NumElements" is specified, it is the
Gabor Greif5082cf42008-02-09 22:24:34 +00002931number of elements allocated, otherwise "NumElements" is defaulted to be one.
Chris Lattner10368b62008-04-02 00:38:26 +00002932If a constant alignment is specified, the value result of the allocation is guaranteed to
Gabor Greif5082cf42008-02-09 22:24:34 +00002933be aligned to at least that boundary. If not specified, or if zero, the target can
2934choose to align the allocation on any convenient boundary.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002935
2936<p>'<tt>type</tt>' must be a sized type.</p>
2937
2938<h5>Semantics:</h5>
2939
2940<p>Memory is allocated using the system "<tt>malloc</tt>" function, and
Chris Lattner8b094fc2008-04-19 21:01:16 +00002941a pointer is returned. The result of a zero byte allocattion is undefined. The
2942result is null if there is insufficient memory available.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002943
2944<h5>Example:</h5>
2945
2946<pre>
2947 %array = malloc [4 x i8 ] <i>; yields {[%4 x i8]*}:array</i>
2948
2949 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
2950 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
2951 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
2952 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
2953 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
2954</pre>
2955</div>
2956
2957<!-- _______________________________________________________________________ -->
2958<div class="doc_subsubsection">
2959 <a name="i_free">'<tt>free</tt>' Instruction</a>
2960</div>
2961
2962<div class="doc_text">
2963
2964<h5>Syntax:</h5>
2965
2966<pre>
2967 free &lt;type&gt; &lt;value&gt; <i>; yields {void}</i>
2968</pre>
2969
2970<h5>Overview:</h5>
2971
2972<p>The '<tt>free</tt>' instruction returns memory back to the unused
2973memory heap to be reallocated in the future.</p>
2974
2975<h5>Arguments:</h5>
2976
2977<p>'<tt>value</tt>' shall be a pointer value that points to a value
2978that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
2979instruction.</p>
2980
2981<h5>Semantics:</h5>
2982
2983<p>Access to the memory pointed to by the pointer is no longer defined
Chris Lattner329d6532008-04-19 22:41:32 +00002984after this instruction executes. If the pointer is null, the operation
2985is a noop.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00002986
2987<h5>Example:</h5>
2988
2989<pre>
2990 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
2991 free [4 x i8]* %array
2992</pre>
2993</div>
2994
2995<!-- _______________________________________________________________________ -->
2996<div class="doc_subsubsection">
2997 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
2998</div>
2999
3000<div class="doc_text">
3001
3002<h5>Syntax:</h5>
3003
3004<pre>
3005 &lt;result&gt; = alloca &lt;type&gt;[, i32 &lt;NumElements&gt;][, align &lt;alignment&gt;] <i>; yields {type*}:result</i>
3006</pre>
3007
3008<h5>Overview:</h5>
3009
3010<p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
3011currently executing function, to be automatically released when this function
Christopher Lambcfe00962007-12-17 01:00:21 +00003012returns to its caller. The object is always allocated in the generic address
3013space (address space zero).</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003014
3015<h5>Arguments:</h5>
3016
3017<p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(&lt;type&gt;)*NumElements</tt>
3018bytes of memory on the runtime stack, returning a pointer of the
Gabor Greif5082cf42008-02-09 22:24:34 +00003019appropriate type to the program. If "NumElements" is specified, it is the
3020number of elements allocated, otherwise "NumElements" is defaulted to be one.
Chris Lattner10368b62008-04-02 00:38:26 +00003021If a constant alignment is specified, the value result of the allocation is guaranteed
Gabor Greif5082cf42008-02-09 22:24:34 +00003022to be aligned to at least that boundary. If not specified, or if zero, the target
3023can choose to align the allocation on any convenient boundary.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003024
3025<p>'<tt>type</tt>' may be any sized type.</p>
3026
3027<h5>Semantics:</h5>
3028
Chris Lattner8b094fc2008-04-19 21:01:16 +00003029<p>Memory is allocated; a pointer is returned. The operation is undefiend if
3030there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003031memory is automatically released when the function returns. The '<tt>alloca</tt>'
3032instruction is commonly used to represent automatic variables that must
3033have an address available. When the function returns (either with the <tt><a
3034 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
Chris Lattner10368b62008-04-02 00:38:26 +00003035instructions), the memory is reclaimed. Allocating zero bytes
3036is legal, but the result is undefined.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003037
3038<h5>Example:</h5>
3039
3040<pre>
3041 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
3042 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
3043 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
3044 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
3045</pre>
3046</div>
3047
3048<!-- _______________________________________________________________________ -->
3049<div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
3050Instruction</a> </div>
3051<div class="doc_text">
3052<h5>Syntax:</h5>
3053<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>
3054<h5>Overview:</h5>
3055<p>The '<tt>load</tt>' instruction is used to read from memory.</p>
3056<h5>Arguments:</h5>
3057<p>The argument to the '<tt>load</tt>' instruction specifies the memory
3058address from which to load. The pointer must point to a <a
3059 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
3060marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
3061the number or order of execution of this <tt>load</tt> with other
3062volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
3063instructions. </p>
Chris Lattner21003c82008-01-06 21:04:43 +00003064<p>
Chris Lattner10368b62008-04-02 00:38:26 +00003065The optional constant "align" argument specifies the alignment of the operation
Chris Lattner21003c82008-01-06 21:04:43 +00003066(that is, the alignment of the memory address). A value of 0 or an
3067omitted "align" argument means that the operation has the preferential
3068alignment for the target. It is the responsibility of the code emitter
3069to ensure that the alignment information is correct. Overestimating
3070the alignment results in an undefined behavior. Underestimating the
3071alignment may produce less efficient code. An alignment of 1 is always
3072safe.
3073</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003074<h5>Semantics:</h5>
3075<p>The location of memory pointed to is loaded.</p>
3076<h5>Examples:</h5>
3077<pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3078 <a
3079 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
3080 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
3081</pre>
3082</div>
3083<!-- _______________________________________________________________________ -->
3084<div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
3085Instruction</a> </div>
3086<div class="doc_text">
3087<h5>Syntax:</h5>
3088<pre> store &lt;ty&gt; &lt;value&gt;, &lt;ty&gt;* &lt;pointer&gt;[, align &lt;alignment&gt;] <i>; yields {void}</i>
3089 volatile store &lt;ty&gt; &lt;value&gt;, &lt;ty&gt;* &lt;pointer&gt;[, align &lt;alignment&gt;] <i>; yields {void}</i>
3090</pre>
3091<h5>Overview:</h5>
3092<p>The '<tt>store</tt>' instruction is used to write to memory.</p>
3093<h5>Arguments:</h5>
3094<p>There are two arguments to the '<tt>store</tt>' instruction: a value
3095to 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 +00003096operand must be a pointer to the <a href="#t_firstclass">first class</a> type
3097of the '<tt>&lt;value&gt;</tt>'
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003098operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
3099optimizer is not allowed to modify the number or order of execution of
3100this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
3101 href="#i_store">store</a></tt> instructions.</p>
Chris Lattner21003c82008-01-06 21:04:43 +00003102<p>
Chris Lattner10368b62008-04-02 00:38:26 +00003103The optional constant "align" argument specifies the alignment of the operation
Chris Lattner21003c82008-01-06 21:04:43 +00003104(that is, the alignment of the memory address). A value of 0 or an
3105omitted "align" argument means that the operation has the preferential
3106alignment for the target. It is the responsibility of the code emitter
3107to ensure that the alignment information is correct. Overestimating
3108the alignment results in an undefined behavior. Underestimating the
3109alignment may produce less efficient code. An alignment of 1 is always
3110safe.
3111</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003112<h5>Semantics:</h5>
3113<p>The contents of memory are updated to contain '<tt>&lt;value&gt;</tt>'
3114at the location specified by the '<tt>&lt;pointer&gt;</tt>' operand.</p>
3115<h5>Example:</h5>
3116<pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
Bill Wendling63ffa142007-10-22 05:10:05 +00003117 store i32 3, i32* %ptr <i>; yields {void}</i>
3118 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003119</pre>
3120</div>
3121
3122<!-- _______________________________________________________________________ -->
3123<div class="doc_subsubsection">
3124 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
3125</div>
3126
3127<div class="doc_text">
3128<h5>Syntax:</h5>
3129<pre>
3130 &lt;result&gt; = getelementptr &lt;ty&gt;* &lt;ptrval&gt;{, &lt;ty&gt; &lt;idx&gt;}*
3131</pre>
3132
3133<h5>Overview:</h5>
3134
3135<p>
3136The '<tt>getelementptr</tt>' instruction is used to get the address of a
3137subelement of an aggregate data structure.</p>
3138
3139<h5>Arguments:</h5>
3140
3141<p>This instruction takes a list of integer operands that indicate what
3142elements of the aggregate object to index to. The actual types of the arguments
3143provided depend on the type of the first pointer argument. The
3144'<tt>getelementptr</tt>' instruction is used to index down through the type
3145levels of a structure or to a specific index in an array. When indexing into a
3146structure, only <tt>i32</tt> integer constants are allowed. When indexing
Chris Lattnera7d94ba2008-04-24 05:59:56 +00003147into an array or pointer, only integers of 32 or 64 bits are allowed; 32-bit
3148values will be sign extended to 64-bits if required.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003149
3150<p>For example, let's consider a C code fragment and how it gets
3151compiled to LLVM:</p>
3152
3153<div class="doc_code">
3154<pre>
3155struct RT {
3156 char A;
3157 int B[10][20];
3158 char C;
3159};
3160struct ST {
3161 int X;
3162 double Y;
3163 struct RT Z;
3164};
3165
3166int *foo(struct ST *s) {
3167 return &amp;s[1].Z.B[5][13];
3168}
3169</pre>
3170</div>
3171
3172<p>The LLVM code generated by the GCC frontend is:</p>
3173
3174<div class="doc_code">
3175<pre>
3176%RT = type { i8 , [10 x [20 x i32]], i8 }
3177%ST = type { i32, double, %RT }
3178
3179define i32* %foo(%ST* %s) {
3180entry:
3181 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
3182 ret i32* %reg
3183}
3184</pre>
3185</div>
3186
3187<h5>Semantics:</h5>
3188
3189<p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
3190on the pointer type that is being indexed into. <a href="#t_pointer">Pointer</a>
3191and <a href="#t_array">array</a> types can use a 32-bit or 64-bit
3192<a href="#t_integer">integer</a> type but the value will always be sign extended
Chris Lattner10368b62008-04-02 00:38:26 +00003193to 64-bits. <a href="#t_struct">Structure</a> and <a href="#t_pstruct">packed
3194structure</a> types require <tt>i32</tt> <b>constants</b>.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003195
3196<p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
3197type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
3198}</tt>' type, a structure. The second index indexes into the third element of
3199the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
3200i8 }</tt>' type, another structure. The third index indexes into the second
3201element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
3202array. The two dimensions of the array are subscripted into, yielding an
3203'<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
3204to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
3205
3206<p>Note that it is perfectly legal to index partially through a
3207structure, returning a pointer to an inner element. Because of this,
3208the LLVM code for the given testcase is equivalent to:</p>
3209
3210<pre>
3211 define i32* %foo(%ST* %s) {
3212 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
3213 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
3214 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
3215 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
3216 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
3217 ret i32* %t5
3218 }
3219</pre>
3220
3221<p>Note that it is undefined to access an array out of bounds: array and
3222pointer indexes must always be within the defined bounds of the array type.
Chris Lattnera7d94ba2008-04-24 05:59:56 +00003223The one exception for this rule is zero length arrays. These arrays are
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003224defined to be accessible as variable length arrays, which requires access
3225beyond the zero'th element.</p>
3226
3227<p>The getelementptr instruction is often confusing. For some more insight
3228into how it works, see <a href="GetElementPtr.html">the getelementptr
3229FAQ</a>.</p>
3230
3231<h5>Example:</h5>
3232
3233<pre>
3234 <i>; yields [12 x i8]*:aptr</i>
3235 %aptr = getelementptr {i32, [12 x i8]}* %sptr, i64 0, i32 1
3236</pre>
3237</div>
3238
3239<!-- ======================================================================= -->
3240<div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
3241</div>
3242<div class="doc_text">
3243<p>The instructions in this category are the conversion instructions (casting)
3244which all take a single operand and a type. They perform various bit conversions
3245on the operand.</p>
3246</div>
3247
3248<!-- _______________________________________________________________________ -->
3249<div class="doc_subsubsection">
3250 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
3251</div>
3252<div class="doc_text">
3253
3254<h5>Syntax:</h5>
3255<pre>
3256 &lt;result&gt; = trunc &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3257</pre>
3258
3259<h5>Overview:</h5>
3260<p>
3261The '<tt>trunc</tt>' instruction truncates its operand to the type <tt>ty2</tt>.
3262</p>
3263
3264<h5>Arguments:</h5>
3265<p>
3266The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
3267be an <a href="#t_integer">integer</a> type, and a type that specifies the size
3268and type of the result, which must be an <a href="#t_integer">integer</a>
3269type. The bit size of <tt>value</tt> must be larger than the bit size of
3270<tt>ty2</tt>. Equal sized types are not allowed.</p>
3271
3272<h5>Semantics:</h5>
3273<p>
3274The '<tt>trunc</tt>' instruction truncates the high order bits in <tt>value</tt>
3275and converts the remaining bits to <tt>ty2</tt>. Since the source size must be
3276larger than the destination size, <tt>trunc</tt> cannot be a <i>no-op cast</i>.
3277It will always truncate bits.</p>
3278
3279<h5>Example:</h5>
3280<pre>
3281 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
3282 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
3283 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
3284</pre>
3285</div>
3286
3287<!-- _______________________________________________________________________ -->
3288<div class="doc_subsubsection">
3289 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
3290</div>
3291<div class="doc_text">
3292
3293<h5>Syntax:</h5>
3294<pre>
3295 &lt;result&gt; = zext &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3296</pre>
3297
3298<h5>Overview:</h5>
3299<p>The '<tt>zext</tt>' instruction zero extends its operand to type
3300<tt>ty2</tt>.</p>
3301
3302
3303<h5>Arguments:</h5>
3304<p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
3305<a href="#t_integer">integer</a> type, and a type to cast it to, which must
3306also be of <a href="#t_integer">integer</a> type. The bit size of the
3307<tt>value</tt> must be smaller than the bit size of the destination type,
3308<tt>ty2</tt>.</p>
3309
3310<h5>Semantics:</h5>
3311<p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
3312bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
3313
3314<p>When zero extending from i1, the result will always be either 0 or 1.</p>
3315
3316<h5>Example:</h5>
3317<pre>
3318 %X = zext i32 257 to i64 <i>; yields i64:257</i>
3319 %Y = zext i1 true to i32 <i>; yields i32:1</i>
3320</pre>
3321</div>
3322
3323<!-- _______________________________________________________________________ -->
3324<div class="doc_subsubsection">
3325 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
3326</div>
3327<div class="doc_text">
3328
3329<h5>Syntax:</h5>
3330<pre>
3331 &lt;result&gt; = sext &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3332</pre>
3333
3334<h5>Overview:</h5>
3335<p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
3336
3337<h5>Arguments:</h5>
3338<p>
3339The '<tt>sext</tt>' instruction takes a value to cast, which must be of
3340<a href="#t_integer">integer</a> type, and a type to cast it to, which must
3341also be of <a href="#t_integer">integer</a> type. The bit size of the
3342<tt>value</tt> must be smaller than the bit size of the destination type,
3343<tt>ty2</tt>.</p>
3344
3345<h5>Semantics:</h5>
3346<p>
3347The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
3348bit (highest order bit) of the <tt>value</tt> until it reaches the bit size of
3349the type <tt>ty2</tt>.</p>
3350
3351<p>When sign extending from i1, the extension always results in -1 or 0.</p>
3352
3353<h5>Example:</h5>
3354<pre>
3355 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
3356 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
3357</pre>
3358</div>
3359
3360<!-- _______________________________________________________________________ -->
3361<div class="doc_subsubsection">
3362 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
3363</div>
3364
3365<div class="doc_text">
3366
3367<h5>Syntax:</h5>
3368
3369<pre>
3370 &lt;result&gt; = fptrunc &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3371</pre>
3372
3373<h5>Overview:</h5>
3374<p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
3375<tt>ty2</tt>.</p>
3376
3377
3378<h5>Arguments:</h5>
3379<p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
3380 point</a> value to cast and a <a href="#t_floating">floating point</a> type to
3381cast it to. The size of <tt>value</tt> must be larger than the size of
3382<tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
3383<i>no-op cast</i>.</p>
3384
3385<h5>Semantics:</h5>
3386<p> The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
3387<a href="#t_floating">floating point</a> type to a smaller
3388<a href="#t_floating">floating point</a> type. If the value cannot fit within
3389the destination type, <tt>ty2</tt>, then the results are undefined.</p>
3390
3391<h5>Example:</h5>
3392<pre>
3393 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
3394 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
3395</pre>
3396</div>
3397
3398<!-- _______________________________________________________________________ -->
3399<div class="doc_subsubsection">
3400 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
3401</div>
3402<div class="doc_text">
3403
3404<h5>Syntax:</h5>
3405<pre>
3406 &lt;result&gt; = fpext &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>fpext</tt>' extends a floating point <tt>value</tt> to a larger
3411floating point value.</p>
3412
3413<h5>Arguments:</h5>
3414<p>The '<tt>fpext</tt>' instruction takes a
3415<a href="#t_floating">floating point</a> <tt>value</tt> to cast,
3416and a <a href="#t_floating">floating point</a> type to cast it to. The source
3417type must be smaller than the destination type.</p>
3418
3419<h5>Semantics:</h5>
3420<p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
3421<a href="#t_floating">floating point</a> type to a larger
3422<a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
3423used to make a <i>no-op cast</i> because it always changes bits. Use
3424<tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
3425
3426<h5>Example:</h5>
3427<pre>
3428 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
3429 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
3430</pre>
3431</div>
3432
3433<!-- _______________________________________________________________________ -->
3434<div class="doc_subsubsection">
3435 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
3436</div>
3437<div class="doc_text">
3438
3439<h5>Syntax:</h5>
3440<pre>
Reid Spencere6adee82007-07-31 14:40:14 +00003441 &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 +00003442</pre>
3443
3444<h5>Overview:</h5>
Reid Spencere6adee82007-07-31 14:40:14 +00003445<p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003446unsigned integer equivalent of type <tt>ty2</tt>.
3447</p>
3448
3449<h5>Arguments:</h5>
Reid Spencere6adee82007-07-31 14:40:14 +00003450<p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
Nate Begeman78246ca2007-11-17 03:58:34 +00003451scalar or vector <a href="#t_floating">floating point</a> value, and a type
3452to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3453type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3454vector integer type with the same number of elements as <tt>ty</tt></p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003455
3456<h5>Semantics:</h5>
Reid Spencere6adee82007-07-31 14:40:14 +00003457<p> The '<tt>fptoui</tt>' instruction converts its
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003458<a href="#t_floating">floating point</a> operand into the nearest (rounding
3459towards zero) unsigned integer value. If the value cannot fit in <tt>ty2</tt>,
3460the results are undefined.</p>
3461
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003462<h5>Example:</h5>
3463<pre>
Reid Spencere6adee82007-07-31 14:40:14 +00003464 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
Chris Lattner681f1e82007-09-22 03:17:52 +00003465 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
Reid Spencere6adee82007-07-31 14:40:14 +00003466 %X = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003467</pre>
3468</div>
3469
3470<!-- _______________________________________________________________________ -->
3471<div class="doc_subsubsection">
3472 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
3473</div>
3474<div class="doc_text">
3475
3476<h5>Syntax:</h5>
3477<pre>
3478 &lt;result&gt; = fptosi &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3479</pre>
3480
3481<h5>Overview:</h5>
3482<p>The '<tt>fptosi</tt>' instruction converts
3483<a href="#t_floating">floating point</a> <tt>value</tt> to type <tt>ty2</tt>.
3484</p>
3485
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003486<h5>Arguments:</h5>
3487<p> The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
Nate Begeman78246ca2007-11-17 03:58:34 +00003488scalar or vector <a href="#t_floating">floating point</a> value, and a type
3489to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3490type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3491vector integer type with the same number of elements as <tt>ty</tt></p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003492
3493<h5>Semantics:</h5>
3494<p>The '<tt>fptosi</tt>' instruction converts its
3495<a href="#t_floating">floating point</a> operand into the nearest (rounding
3496towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
3497the results are undefined.</p>
3498
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003499<h5>Example:</h5>
3500<pre>
3501 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
Chris Lattner681f1e82007-09-22 03:17:52 +00003502 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003503 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
3504</pre>
3505</div>
3506
3507<!-- _______________________________________________________________________ -->
3508<div class="doc_subsubsection">
3509 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
3510</div>
3511<div class="doc_text">
3512
3513<h5>Syntax:</h5>
3514<pre>
3515 &lt;result&gt; = uitofp &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3516</pre>
3517
3518<h5>Overview:</h5>
3519<p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
3520integer and converts that value to the <tt>ty2</tt> type.</p>
3521
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003522<h5>Arguments:</h5>
Nate Begeman78246ca2007-11-17 03:58:34 +00003523<p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
3524scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3525to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3526type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3527floating point type with the same number of elements as <tt>ty</tt></p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003528
3529<h5>Semantics:</h5>
3530<p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
3531integer quantity and converts it to the corresponding floating point value. If
3532the value cannot fit in the floating point value, the results are undefined.</p>
3533
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003534<h5>Example:</h5>
3535<pre>
3536 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
3537 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
3538</pre>
3539</div>
3540
3541<!-- _______________________________________________________________________ -->
3542<div class="doc_subsubsection">
3543 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
3544</div>
3545<div class="doc_text">
3546
3547<h5>Syntax:</h5>
3548<pre>
3549 &lt;result&gt; = sitofp &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3550</pre>
3551
3552<h5>Overview:</h5>
3553<p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed
3554integer and converts that value to the <tt>ty2</tt> type.</p>
3555
3556<h5>Arguments:</h5>
Nate Begeman78246ca2007-11-17 03:58:34 +00003557<p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
3558scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3559to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3560type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3561floating point type with the same number of elements as <tt>ty</tt></p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003562
3563<h5>Semantics:</h5>
3564<p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed
3565integer quantity and converts it to the corresponding floating point value. If
3566the value cannot fit in the floating point value, the results are undefined.</p>
3567
3568<h5>Example:</h5>
3569<pre>
3570 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
3571 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
3572</pre>
3573</div>
3574
3575<!-- _______________________________________________________________________ -->
3576<div class="doc_subsubsection">
3577 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
3578</div>
3579<div class="doc_text">
3580
3581<h5>Syntax:</h5>
3582<pre>
3583 &lt;result&gt; = ptrtoint &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3584</pre>
3585
3586<h5>Overview:</h5>
3587<p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
3588the integer type <tt>ty2</tt>.</p>
3589
3590<h5>Arguments:</h5>
3591<p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
3592must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
3593<tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.
3594
3595<h5>Semantics:</h5>
3596<p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
3597<tt>ty2</tt> by interpreting the pointer value as an integer and either
3598truncating or zero extending that value to the size of the integer type. If
3599<tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
3600<tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
3601are the same size, then nothing is done (<i>no-op cast</i>) other than a type
3602change.</p>
3603
3604<h5>Example:</h5>
3605<pre>
3606 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
3607 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
3608</pre>
3609</div>
3610
3611<!-- _______________________________________________________________________ -->
3612<div class="doc_subsubsection">
3613 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
3614</div>
3615<div class="doc_text">
3616
3617<h5>Syntax:</h5>
3618<pre>
3619 &lt;result&gt; = inttoptr &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3620</pre>
3621
3622<h5>Overview:</h5>
3623<p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to
3624a pointer type, <tt>ty2</tt>.</p>
3625
3626<h5>Arguments:</h5>
3627<p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
3628value to cast, and a type to cast it to, which must be a
3629<a href="#t_pointer">pointer</a> type.
3630
3631<h5>Semantics:</h5>
3632<p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
3633<tt>ty2</tt> by applying either a zero extension or a truncation depending on
3634the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
3635size of a pointer then a truncation is done. If <tt>value</tt> is smaller than
3636the size of a pointer then a zero extension is done. If they are the same size,
3637nothing is done (<i>no-op cast</i>).</p>
3638
3639<h5>Example:</h5>
3640<pre>
3641 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
3642 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
3643 %Y = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
3644</pre>
3645</div>
3646
3647<!-- _______________________________________________________________________ -->
3648<div class="doc_subsubsection">
3649 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
3650</div>
3651<div class="doc_text">
3652
3653<h5>Syntax:</h5>
3654<pre>
3655 &lt;result&gt; = bitcast &lt;ty&gt; &lt;value&gt; to &lt;ty2&gt; <i>; yields ty2</i>
3656</pre>
3657
3658<h5>Overview:</h5>
3659<p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3660<tt>ty2</tt> without changing any bits.</p>
3661
3662<h5>Arguments:</h5>
3663<p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be
3664a first class value, and a type to cast it to, which must also be a <a
3665 href="#t_firstclass">first class</a> type. The bit sizes of <tt>value</tt>
3666and the destination type, <tt>ty2</tt>, must be identical. If the source
3667type is a pointer, the destination type must also be a pointer.</p>
3668
3669<h5>Semantics:</h5>
3670<p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3671<tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
3672this conversion. The conversion is done as if the <tt>value</tt> had been
3673stored to memory and read back as type <tt>ty2</tt>. Pointer types may only be
3674converted to other pointer types with this instruction. To convert pointers to
3675other types, use the <a href="#i_inttoptr">inttoptr</a> or
3676<a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
3677
3678<h5>Example:</h5>
3679<pre>
3680 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
3681 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
3682 %Z = bitcast <2xint> %V to i64; <i>; yields i64: %V</i>
3683</pre>
3684</div>
3685
3686<!-- ======================================================================= -->
3687<div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
3688<div class="doc_text">
3689<p>The instructions in this category are the "miscellaneous"
3690instructions, which defy better classification.</p>
3691</div>
3692
3693<!-- _______________________________________________________________________ -->
3694<div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
3695</div>
3696<div class="doc_text">
3697<h5>Syntax:</h5>
3698<pre> &lt;result&gt; = icmp &lt;cond&gt; &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {i1}:result</i>
3699</pre>
3700<h5>Overview:</h5>
3701<p>The '<tt>icmp</tt>' instruction returns a boolean value based on comparison
Chris Lattner10368b62008-04-02 00:38:26 +00003702of its two integer or pointer operands.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003703<h5>Arguments:</h5>
3704<p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
3705the condition code indicating the kind of comparison to perform. It is not
3706a value, just a keyword. The possible condition code are:
3707<ol>
3708 <li><tt>eq</tt>: equal</li>
3709 <li><tt>ne</tt>: not equal </li>
3710 <li><tt>ugt</tt>: unsigned greater than</li>
3711 <li><tt>uge</tt>: unsigned greater or equal</li>
3712 <li><tt>ult</tt>: unsigned less than</li>
3713 <li><tt>ule</tt>: unsigned less or equal</li>
3714 <li><tt>sgt</tt>: signed greater than</li>
3715 <li><tt>sge</tt>: signed greater or equal</li>
3716 <li><tt>slt</tt>: signed less than</li>
3717 <li><tt>sle</tt>: signed less or equal</li>
3718</ol>
3719<p>The remaining two arguments must be <a href="#t_integer">integer</a> or
3720<a href="#t_pointer">pointer</a> typed. They must also be identical types.</p>
3721<h5>Semantics:</h5>
3722<p>The '<tt>icmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3723the condition code given as <tt>cond</tt>. The comparison performed always
3724yields a <a href="#t_primitive">i1</a> result, as follows:
3725<ol>
3726 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
3727 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3728 </li>
3729 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
3730 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3731 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
3732 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3733 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
3734 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3735 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
3736 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3737 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
3738 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3739 <li><tt>sgt</tt>: interprets the operands as signed values and yields
3740 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3741 <li><tt>sge</tt>: interprets the operands as signed values and yields
3742 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3743 <li><tt>slt</tt>: interprets the operands as signed values and yields
3744 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3745 <li><tt>sle</tt>: interprets the operands as signed values and yields
3746 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3747</ol>
3748<p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
3749values are compared as if they were integers.</p>
3750
3751<h5>Example:</h5>
3752<pre> &lt;result&gt; = icmp eq i32 4, 5 <i>; yields: result=false</i>
3753 &lt;result&gt; = icmp ne float* %X, %X <i>; yields: result=false</i>
3754 &lt;result&gt; = icmp ult i16 4, 5 <i>; yields: result=true</i>
3755 &lt;result&gt; = icmp sgt i16 4, 5 <i>; yields: result=false</i>
3756 &lt;result&gt; = icmp ule i16 -4, 5 <i>; yields: result=false</i>
3757 &lt;result&gt; = icmp sge i16 4, 5 <i>; yields: result=false</i>
3758</pre>
3759</div>
3760
3761<!-- _______________________________________________________________________ -->
3762<div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
3763</div>
3764<div class="doc_text">
3765<h5>Syntax:</h5>
3766<pre> &lt;result&gt; = fcmp &lt;cond&gt; &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {i1}:result</i>
3767</pre>
3768<h5>Overview:</h5>
3769<p>The '<tt>fcmp</tt>' instruction returns a boolean value based on comparison
3770of its floating point operands.</p>
3771<h5>Arguments:</h5>
3772<p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
3773the condition code indicating the kind of comparison to perform. It is not
3774a value, just a keyword. The possible condition code are:
3775<ol>
3776 <li><tt>false</tt>: no comparison, always returns false</li>
3777 <li><tt>oeq</tt>: ordered and equal</li>
3778 <li><tt>ogt</tt>: ordered and greater than </li>
3779 <li><tt>oge</tt>: ordered and greater than or equal</li>
3780 <li><tt>olt</tt>: ordered and less than </li>
3781 <li><tt>ole</tt>: ordered and less than or equal</li>
3782 <li><tt>one</tt>: ordered and not equal</li>
3783 <li><tt>ord</tt>: ordered (no nans)</li>
3784 <li><tt>ueq</tt>: unordered or equal</li>
3785 <li><tt>ugt</tt>: unordered or greater than </li>
3786 <li><tt>uge</tt>: unordered or greater than or equal</li>
3787 <li><tt>ult</tt>: unordered or less than </li>
3788 <li><tt>ule</tt>: unordered or less than or equal</li>
3789 <li><tt>une</tt>: unordered or not equal</li>
3790 <li><tt>uno</tt>: unordered (either nans)</li>
3791 <li><tt>true</tt>: no comparison, always returns true</li>
3792</ol>
3793<p><i>Ordered</i> means that neither operand is a QNAN while
3794<i>unordered</i> means that either operand may be a QNAN.</p>
3795<p>The <tt>val1</tt> and <tt>val2</tt> arguments must be
3796<a href="#t_floating">floating point</a> typed. They must have identical
3797types.</p>
3798<h5>Semantics:</h5>
Nate Begeman646fa482008-05-12 19:01:56 +00003799<p>The '<tt>fcmp</tt>' instruction compares <tt>var1</tt> and <tt>var2</tt>
3800according to the condition code given as <tt>cond</tt>. The comparison performed
3801always yields a <a href="#t_primitive">i1</a> result, as follows:
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003802<ol>
3803 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
3804 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3805 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3806 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3807 <tt>var1</tt> is greather than <tt>var2</tt>.</li>
3808 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3809 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3810 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3811 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3812 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3813 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3814 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3815 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3816 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
3817 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
3818 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3819 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
3820 <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3821 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
3822 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3823 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
3824 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3825 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
3826 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3827 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
3828 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3829 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
3830 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
3831</ol>
3832
3833<h5>Example:</h5>
3834<pre> &lt;result&gt; = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
3835 &lt;result&gt; = icmp one float 4.0, 5.0 <i>; yields: result=true</i>
3836 &lt;result&gt; = icmp olt float 4.0, 5.0 <i>; yields: result=true</i>
3837 &lt;result&gt; = icmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
3838</pre>
3839</div>
3840
3841<!-- _______________________________________________________________________ -->
Nate Begeman646fa482008-05-12 19:01:56 +00003842<div class="doc_subsubsection">
3843 <a name="i_vicmp">'<tt>vicmp</tt>' Instruction</a>
3844</div>
3845<div class="doc_text">
3846<h5>Syntax:</h5>
3847<pre> &lt;result&gt; = vicmp &lt;cond&gt; &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt; <i>; yields {ty}:result</i>
3848</pre>
3849<h5>Overview:</h5>
3850<p>The '<tt>vicmp</tt>' instruction returns an integer vector value based on
3851element-wise comparison of its two integer vector operands.</p>
3852<h5>Arguments:</h5>
3853<p>The '<tt>vicmp</tt>' instruction takes three operands. The first operand is
3854the condition code indicating the kind of comparison to perform. It is not
3855a value, just a keyword. The possible condition code are:
3856<ol>
3857 <li><tt>eq</tt>: equal</li>
3858 <li><tt>ne</tt>: not equal </li>
3859 <li><tt>ugt</tt>: unsigned greater than</li>
3860 <li><tt>uge</tt>: unsigned greater or equal</li>
3861 <li><tt>ult</tt>: unsigned less than</li>
3862 <li><tt>ule</tt>: unsigned less or equal</li>
3863 <li><tt>sgt</tt>: signed greater than</li>
3864 <li><tt>sge</tt>: signed greater or equal</li>
3865 <li><tt>slt</tt>: signed less than</li>
3866 <li><tt>sle</tt>: signed less or equal</li>
3867</ol>
3868<p>The remaining two arguments must be <a href="#t_vector">vector</a> of
3869<a href="#t_integer">integer</a> typed. They must also be identical types.</p>
3870<h5>Semantics:</h5>
3871<p>The '<tt>vicmp</tt>' instruction compares <tt>var1</tt> and <tt>var2</tt>
3872according to the condition code given as <tt>cond</tt>. The comparison yields a
3873<a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, of
3874identical type as the values being compared. The most significant bit in each
3875element is 1 if the element-wise comparison evaluates to true, and is 0
3876otherwise. All other bits of the result are undefined. The condition codes
3877are evaluated identically to the <a href="#i_icmp">'<tt>icmp</tt>'
3878instruction</a>.
3879
3880<h5>Example:</h5>
3881<pre>
3882 &lt;result&gt; = vicmp eq <2 x i32> < i32 4, i32 0 >, < i32 5, i32 0 > <i>; yields: result=<2 x i32> < i32 0, i32 -1 ></i>
3883 &lt;result&gt; = vicmp ult <2 x i8> < i8 1, i8 2 >, < i8 2, i8 2> <i>; yields: result=<2 x i8> < i8 -1, i8 0 ></i>
3884</pre>
3885</div>
3886
3887<!-- _______________________________________________________________________ -->
3888<div class="doc_subsubsection">
3889 <a name="i_vfcmp">'<tt>vfcmp</tt>' Instruction</a>
3890</div>
3891<div class="doc_text">
3892<h5>Syntax:</h5>
3893<pre> &lt;result&gt; = vfcmp &lt;cond&gt; &lt;ty&gt; &lt;var1&gt;, &lt;var2&gt;</pre>
3894<h5>Overview:</h5>
3895<p>The '<tt>vfcmp</tt>' instruction returns an integer vector value based on
3896element-wise comparison of its two floating point vector operands. The output
3897elements have the same width as the input elements.</p>
3898<h5>Arguments:</h5>
3899<p>The '<tt>vfcmp</tt>' instruction takes three operands. The first operand is
3900the condition code indicating the kind of comparison to perform. It is not
3901a value, just a keyword. The possible condition code are:
3902<ol>
3903 <li><tt>false</tt>: no comparison, always returns false</li>
3904 <li><tt>oeq</tt>: ordered and equal</li>
3905 <li><tt>ogt</tt>: ordered and greater than </li>
3906 <li><tt>oge</tt>: ordered and greater than or equal</li>
3907 <li><tt>olt</tt>: ordered and less than </li>
3908 <li><tt>ole</tt>: ordered and less than or equal</li>
3909 <li><tt>one</tt>: ordered and not equal</li>
3910 <li><tt>ord</tt>: ordered (no nans)</li>
3911 <li><tt>ueq</tt>: unordered or equal</li>
3912 <li><tt>ugt</tt>: unordered or greater than </li>
3913 <li><tt>uge</tt>: unordered or greater than or equal</li>
3914 <li><tt>ult</tt>: unordered or less than </li>
3915 <li><tt>ule</tt>: unordered or less than or equal</li>
3916 <li><tt>une</tt>: unordered or not equal</li>
3917 <li><tt>uno</tt>: unordered (either nans)</li>
3918 <li><tt>true</tt>: no comparison, always returns true</li>
3919</ol>
3920<p>The remaining two arguments must be <a href="#t_vector">vector</a> of
3921<a href="#t_floating">floating point</a> typed. They must also be identical
3922types.</p>
3923<h5>Semantics:</h5>
3924<p>The '<tt>vfcmp</tt>' instruction compares <tt>var1</tt> and <tt>var2</tt>
3925according to the condition code given as <tt>cond</tt>. The comparison yields a
3926<a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, with
3927an identical number of elements as the values being compared, and each element
3928having identical with to the width of the floating point elements. The most
3929significant bit in each element is 1 if the element-wise comparison evaluates to
3930true, and is 0 otherwise. All other bits of the result are undefined. The
3931condition codes are evaluated identically to the
3932<a href="#i_fcmp">'<tt>fcmp</tt>' instruction</a>.
3933
3934<h5>Example:</h5>
3935<pre>
3936 &lt;result&gt; = vfcmp oeq <2 x float> < float 4, float 0 >, < float 5, float 0 > <i>; yields: result=<2 x i32> < i32 0, i32 -1 ></i>
3937 &lt;result&gt; = vfcmp ult <2 x double> < double 1, double 2 >, < double 2, double 2> <i>; yields: result=<2 x i64> < i64 -1, i64 0 ></i>
3938</pre>
3939</div>
3940
3941<!-- _______________________________________________________________________ -->
Dan Gohmanf17a25c2007-07-18 16:29:46 +00003942<div class="doc_subsubsection"> <a name="i_phi">'<tt>phi</tt>'
3943Instruction</a> </div>
3944<div class="doc_text">
3945<h5>Syntax:</h5>
3946<pre> &lt;result&gt; = phi &lt;ty&gt; [ &lt;val0&gt;, &lt;label0&gt;], ...<br></pre>
3947<h5>Overview:</h5>
3948<p>The '<tt>phi</tt>' instruction is used to implement the &#966; node in
3949the SSA graph representing the function.</p>
3950<h5>Arguments:</h5>
3951<p>The type of the incoming values is specified with the first type
3952field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
3953as arguments, with one pair for each predecessor basic block of the
3954current block. Only values of <a href="#t_firstclass">first class</a>
3955type may be used as the value arguments to the PHI node. Only labels
3956may be used as the label arguments.</p>
3957<p>There must be no non-phi instructions between the start of a basic
3958block and the PHI instructions: i.e. PHI instructions must be first in
3959a basic block.</p>
3960<h5>Semantics:</h5>
3961<p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
3962specified by the pair corresponding to the predecessor basic block that executed
3963just prior to the current block.</p>
3964<h5>Example:</h5>
3965<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>
3966</div>
3967
3968<!-- _______________________________________________________________________ -->
3969<div class="doc_subsubsection">
3970 <a name="i_select">'<tt>select</tt>' Instruction</a>
3971</div>
3972
3973<div class="doc_text">
3974
3975<h5>Syntax:</h5>
3976
3977<pre>
3978 &lt;result&gt; = select i1 &lt;cond&gt;, &lt;ty&gt; &lt;val1&gt;, &lt;ty&gt; &lt;val2&gt; <i>; yields ty</i>
3979</pre>
3980
3981<h5>Overview:</h5>
3982
3983<p>
3984The '<tt>select</tt>' instruction is used to choose one value based on a
3985condition, without branching.
3986</p>
3987
3988
3989<h5>Arguments:</h5>
3990
3991<p>
3992The '<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.
3993</p>
3994
3995<h5>Semantics:</h5>
3996
3997<p>
3998If the boolean condition evaluates to true, the instruction returns the first
3999value argument; otherwise, it returns the second value argument.
4000</p>
4001
4002<h5>Example:</h5>
4003
4004<pre>
4005 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
4006</pre>
4007</div>
4008
4009
4010<!-- _______________________________________________________________________ -->
4011<div class="doc_subsubsection">
4012 <a name="i_call">'<tt>call</tt>' Instruction</a>
4013</div>
4014
4015<div class="doc_text">
4016
4017<h5>Syntax:</h5>
4018<pre>
Nick Lewycky93082fc2007-09-08 13:57:50 +00004019 &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 +00004020</pre>
4021
4022<h5>Overview:</h5>
4023
4024<p>The '<tt>call</tt>' instruction represents a simple function call.</p>
4025
4026<h5>Arguments:</h5>
4027
4028<p>This instruction requires several arguments:</p>
4029
4030<ol>
4031 <li>
4032 <p>The optional "tail" marker indicates whether the callee function accesses
4033 any allocas or varargs in the caller. If the "tail" marker is present, the
4034 function call is eligible for tail call optimization. Note that calls may
4035 be marked "tail" even if they do not occur before a <a
4036 href="#i_ret"><tt>ret</tt></a> instruction.
4037 </li>
4038 <li>
4039 <p>The optional "cconv" marker indicates which <a href="#callingconv">calling
4040 convention</a> the call should use. If none is specified, the call defaults
4041 to using C calling conventions.
4042 </li>
4043 <li>
Nick Lewycky93082fc2007-09-08 13:57:50 +00004044 <p>'<tt>ty</tt>': the type of the call instruction itself which is also
4045 the type of the return value. Functions that return no value are marked
4046 <tt><a href="#t_void">void</a></tt>.</p>
4047 </li>
4048 <li>
4049 <p>'<tt>fnty</tt>': shall be the signature of the pointer to function
4050 value being invoked. The argument types must match the types implied by
4051 this signature. This type can be omitted if the function is not varargs
4052 and if the function type does not return a pointer to a function.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004053 </li>
4054 <li>
4055 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
4056 be invoked. In most cases, this is a direct function invocation, but
4057 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
4058 to function value.</p>
4059 </li>
4060 <li>
4061 <p>'<tt>function args</tt>': argument list whose types match the
4062 function signature argument types. All arguments must be of
4063 <a href="#t_firstclass">first class</a> type. If the function signature
4064 indicates the function accepts a variable number of arguments, the extra
4065 arguments can be specified.</p>
4066 </li>
4067</ol>
4068
4069<h5>Semantics:</h5>
4070
4071<p>The '<tt>call</tt>' instruction is used to cause control flow to
4072transfer to a specified function, with its incoming arguments bound to
4073the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
4074instruction in the called function, control flow continues with the
4075instruction after the function call, and the return value of the
Chris Lattner5e893ef2008-03-21 17:24:17 +00004076function is bound to the result argument. If the callee returns multiple
4077values then the return values of the function are only accessible through
4078the '<tt><a href="#i_getresult">getresult</a></tt>' instruction.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004079
4080<h5>Example:</h5>
4081
4082<pre>
Nick Lewycky93082fc2007-09-08 13:57:50 +00004083 %retval = call i32 @test(i32 %argc)
Chris Lattner5e893ef2008-03-21 17:24:17 +00004084 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42) <i>; yields i32</i>
4085 %X = tail call i32 @foo() <i>; yields i32</i>
4086 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
4087 call void %foo(i8 97 signext)
Devang Patela3cc5372008-03-10 20:49:15 +00004088
4089 %struct.A = type { i32, i8 }
Chris Lattner5e893ef2008-03-21 17:24:17 +00004090 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
4091 %gr = getresult %struct.A %r, 0 <i>; yields i32</i>
4092 %gr1 = getresult %struct.A %r, 1 <i>; yields i8</i>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004093</pre>
4094
4095</div>
4096
4097<!-- _______________________________________________________________________ -->
4098<div class="doc_subsubsection">
4099 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
4100</div>
4101
4102<div class="doc_text">
4103
4104<h5>Syntax:</h5>
4105
4106<pre>
4107 &lt;resultval&gt; = va_arg &lt;va_list*&gt; &lt;arglist&gt;, &lt;argty&gt;
4108</pre>
4109
4110<h5>Overview:</h5>
4111
4112<p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
4113the "variable argument" area of a function call. It is used to implement the
4114<tt>va_arg</tt> macro in C.</p>
4115
4116<h5>Arguments:</h5>
4117
4118<p>This instruction takes a <tt>va_list*</tt> value and the type of
4119the argument. It returns a value of the specified argument type and
4120increments the <tt>va_list</tt> to point to the next argument. The
4121actual type of <tt>va_list</tt> is target specific.</p>
4122
4123<h5>Semantics:</h5>
4124
4125<p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
4126type from the specified <tt>va_list</tt> and causes the
4127<tt>va_list</tt> to point to the next argument. For more information,
4128see the variable argument handling <a href="#int_varargs">Intrinsic
4129Functions</a>.</p>
4130
4131<p>It is legal for this instruction to be called in a function which does not
4132take a variable number of arguments, for example, the <tt>vfprintf</tt>
4133function.</p>
4134
4135<p><tt>va_arg</tt> is an LLVM instruction instead of an <a
4136href="#intrinsics">intrinsic function</a> because it takes a type as an
4137argument.</p>
4138
4139<h5>Example:</h5>
4140
4141<p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
4142
4143</div>
4144
Devang Patela3cc5372008-03-10 20:49:15 +00004145<!-- _______________________________________________________________________ -->
4146<div class="doc_subsubsection">
4147 <a name="i_getresult">'<tt>getresult</tt>' Instruction</a>
4148</div>
4149
4150<div class="doc_text">
4151
4152<h5>Syntax:</h5>
4153<pre>
Chris Lattneree9da3f2008-03-21 17:20:51 +00004154 &lt;resultval&gt; = getresult &lt;type&gt; &lt;retval&gt;, &lt;index&gt;
Devang Patela3cc5372008-03-10 20:49:15 +00004155</pre>
Chris Lattneree9da3f2008-03-21 17:20:51 +00004156
Devang Patela3cc5372008-03-10 20:49:15 +00004157<h5>Overview:</h5>
4158
4159<p> The '<tt>getresult</tt>' instruction is used to extract individual values
Chris Lattneree9da3f2008-03-21 17:20:51 +00004160from a '<tt><a href="#i_call">call</a></tt>'
4161or '<tt><a href="#i_invoke">invoke</a></tt>' instruction that returns multiple
4162results.</p>
Devang Patela3cc5372008-03-10 20:49:15 +00004163
4164<h5>Arguments:</h5>
4165
Chris Lattneree9da3f2008-03-21 17:20:51 +00004166<p>The '<tt>getresult</tt>' instruction takes a call or invoke value as its
Chris Lattnerd8dd3522008-04-23 04:06:52 +00004167first argument, or an undef value. The value must have <a
4168href="#t_struct">structure type</a>. The second argument is a constant
4169unsigned index value which must be in range for the number of values returned
4170by the call.</p>
Devang Patela3cc5372008-03-10 20:49:15 +00004171
4172<h5>Semantics:</h5>
4173
Chris Lattneree9da3f2008-03-21 17:20:51 +00004174<p>The '<tt>getresult</tt>' instruction extracts the element identified by
4175'<tt>index</tt>' from the aggregate value.</p>
Devang Patela3cc5372008-03-10 20:49:15 +00004176
4177<h5>Example:</h5>
4178
4179<pre>
4180 %struct.A = type { i32, i8 }
4181
4182 %r = call %struct.A @foo()
Chris Lattneree9da3f2008-03-21 17:20:51 +00004183 %gr = getresult %struct.A %r, 0 <i>; yields i32:%gr</i>
4184 %gr1 = getresult %struct.A %r, 1 <i>; yields i8:%gr1</i>
Devang Patela3cc5372008-03-10 20:49:15 +00004185 add i32 %gr, 42
4186 add i8 %gr1, 41
4187</pre>
4188
4189</div>
4190
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004191<!-- *********************************************************************** -->
4192<div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
4193<!-- *********************************************************************** -->
4194
4195<div class="doc_text">
4196
4197<p>LLVM supports the notion of an "intrinsic function". These functions have
4198well known names and semantics and are required to follow certain restrictions.
4199Overall, these intrinsics represent an extension mechanism for the LLVM
4200language that does not require changing all of the transformations in LLVM when
4201adding to the language (or the bitcode reader/writer, the parser, etc...).</p>
4202
4203<p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
4204prefix is reserved in LLVM for intrinsic names; thus, function names may not
4205begin with this prefix. Intrinsic functions must always be external functions:
4206you cannot define the body of intrinsic functions. Intrinsic functions may
4207only be used in call or invoke instructions: it is illegal to take the address
4208of an intrinsic function. Additionally, because intrinsic functions are part
4209of the LLVM language, it is required if any are added that they be documented
4210here.</p>
4211
Chandler Carrutha228e392007-08-04 01:51:18 +00004212<p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
4213a family of functions that perform the same operation but on different data
4214types. Because LLVM can represent over 8 million different integer types,
4215overloading is used commonly to allow an intrinsic function to operate on any
4216integer type. One or more of the argument types or the result type can be
4217overloaded to accept any integer type. Argument types may also be defined as
4218exactly matching a previous argument's type or the result type. This allows an
4219intrinsic function which accepts multiple arguments, but needs all of them to
4220be of the same type, to only be overloaded with respect to a single argument or
4221the result.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004222
Chandler Carrutha228e392007-08-04 01:51:18 +00004223<p>Overloaded intrinsics will have the names of its overloaded argument types
4224encoded into its function name, each preceded by a period. Only those types
4225which are overloaded result in a name suffix. Arguments whose type is matched
4226against another type do not. For example, the <tt>llvm.ctpop</tt> function can
4227take an integer of any width and returns an integer of exactly the same integer
4228width. This leads to a family of functions such as
4229<tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29 %val)</tt>.
4230Only one type, the return type, is overloaded, and only one type suffix is
4231required. Because the argument's type is matched against the return type, it
4232does not require its own name suffix.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004233
4234<p>To learn how to add an intrinsic function, please see the
4235<a href="ExtendingLLVM.html">Extending LLVM Guide</a>.
4236</p>
4237
4238</div>
4239
4240<!-- ======================================================================= -->
4241<div class="doc_subsection">
4242 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
4243</div>
4244
4245<div class="doc_text">
4246
4247<p>Variable argument support is defined in LLVM with the <a
4248 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
4249intrinsic functions. These functions are related to the similarly
4250named macros defined in the <tt>&lt;stdarg.h&gt;</tt> header file.</p>
4251
4252<p>All of these functions operate on arguments that use a
4253target-specific value type "<tt>va_list</tt>". The LLVM assembly
4254language reference manual does not define what this type is, so all
4255transformations should be prepared to handle these functions regardless of
4256the type used.</p>
4257
4258<p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
4259instruction and the variable argument handling intrinsic functions are
4260used.</p>
4261
4262<div class="doc_code">
4263<pre>
4264define i32 @test(i32 %X, ...) {
4265 ; Initialize variable argument processing
4266 %ap = alloca i8*
4267 %ap2 = bitcast i8** %ap to i8*
4268 call void @llvm.va_start(i8* %ap2)
4269
4270 ; Read a single integer argument
4271 %tmp = va_arg i8** %ap, i32
4272
4273 ; Demonstrate usage of llvm.va_copy and llvm.va_end
4274 %aq = alloca i8*
4275 %aq2 = bitcast i8** %aq to i8*
4276 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
4277 call void @llvm.va_end(i8* %aq2)
4278
4279 ; Stop processing of arguments.
4280 call void @llvm.va_end(i8* %ap2)
4281 ret i32 %tmp
4282}
4283
4284declare void @llvm.va_start(i8*)
4285declare void @llvm.va_copy(i8*, i8*)
4286declare void @llvm.va_end(i8*)
4287</pre>
4288</div>
4289
4290</div>
4291
4292<!-- _______________________________________________________________________ -->
4293<div class="doc_subsubsection">
4294 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
4295</div>
4296
4297
4298<div class="doc_text">
4299<h5>Syntax:</h5>
4300<pre> declare void %llvm.va_start(i8* &lt;arglist&gt;)<br></pre>
4301<h5>Overview:</h5>
4302<P>The '<tt>llvm.va_start</tt>' intrinsic initializes
4303<tt>*&lt;arglist&gt;</tt> for subsequent use by <tt><a
4304href="#i_va_arg">va_arg</a></tt>.</p>
4305
4306<h5>Arguments:</h5>
4307
4308<P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
4309
4310<h5>Semantics:</h5>
4311
4312<P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
4313macro available in C. In a target-dependent way, it initializes the
4314<tt>va_list</tt> element to which the argument points, so that the next call to
4315<tt>va_arg</tt> will produce the first variable argument passed to the function.
4316Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
4317last argument of the function as the compiler can figure that out.</p>
4318
4319</div>
4320
4321<!-- _______________________________________________________________________ -->
4322<div class="doc_subsubsection">
4323 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
4324</div>
4325
4326<div class="doc_text">
4327<h5>Syntax:</h5>
4328<pre> declare void @llvm.va_end(i8* &lt;arglist&gt;)<br></pre>
4329<h5>Overview:</h5>
4330
4331<p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*&lt;arglist&gt;</tt>,
4332which has been initialized previously with <tt><a href="#int_va_start">llvm.va_start</a></tt>
4333or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
4334
4335<h5>Arguments:</h5>
4336
4337<p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
4338
4339<h5>Semantics:</h5>
4340
4341<p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
4342macro available in C. In a target-dependent way, it destroys the
4343<tt>va_list</tt> element to which the argument points. Calls to <a
4344href="#int_va_start"><tt>llvm.va_start</tt></a> and <a href="#int_va_copy">
4345<tt>llvm.va_copy</tt></a> must be matched exactly with calls to
4346<tt>llvm.va_end</tt>.</p>
4347
4348</div>
4349
4350<!-- _______________________________________________________________________ -->
4351<div class="doc_subsubsection">
4352 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
4353</div>
4354
4355<div class="doc_text">
4356
4357<h5>Syntax:</h5>
4358
4359<pre>
4360 declare void @llvm.va_copy(i8* &lt;destarglist&gt;, i8* &lt;srcarglist&gt;)
4361</pre>
4362
4363<h5>Overview:</h5>
4364
4365<p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
4366from the source argument list to the destination argument list.</p>
4367
4368<h5>Arguments:</h5>
4369
4370<p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
4371The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
4372
4373
4374<h5>Semantics:</h5>
4375
4376<p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
4377macro available in C. In a target-dependent way, it copies the source
4378<tt>va_list</tt> element into the destination <tt>va_list</tt> element. This
4379intrinsic is necessary because the <tt><a href="#int_va_start">
4380llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
4381example, memory allocation.</p>
4382
4383</div>
4384
4385<!-- ======================================================================= -->
4386<div class="doc_subsection">
4387 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
4388</div>
4389
4390<div class="doc_text">
4391
4392<p>
4393LLVM support for <a href="GarbageCollection.html">Accurate Garbage
4394Collection</a> requires the implementation and generation of these intrinsics.
4395These intrinsics allow identification of <a href="#int_gcroot">GC roots on the
4396stack</a>, as well as garbage collector implementations that require <a
4397href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a> barriers.
4398Front-ends for type-safe garbage collected languages should generate these
4399intrinsics to make use of the LLVM garbage collectors. For more details, see <a
4400href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
4401</p>
Christopher Lambcfe00962007-12-17 01:00:21 +00004402
4403<p>The garbage collection intrinsics only operate on objects in the generic
4404 address space (address space zero).</p>
4405
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004406</div>
4407
4408<!-- _______________________________________________________________________ -->
4409<div class="doc_subsubsection">
4410 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
4411</div>
4412
4413<div class="doc_text">
4414
4415<h5>Syntax:</h5>
4416
4417<pre>
Chris Lattner38bd5dd2007-09-21 17:30:40 +00004418 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004419</pre>
4420
4421<h5>Overview:</h5>
4422
4423<p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
4424the code generator, and allows some metadata to be associated with it.</p>
4425
4426<h5>Arguments:</h5>
4427
4428<p>The first argument specifies the address of a stack object that contains the
4429root pointer. The second pointer (which must be either a constant or a global
4430value address) contains the meta-data to be associated with the root.</p>
4431
4432<h5>Semantics:</h5>
4433
Chris Lattnera7d94ba2008-04-24 05:59:56 +00004434<p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004435location. At compile-time, the code generator generates information to allow
Gordon Henriksen40393542007-12-25 02:31:26 +00004436the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
4437intrinsic may only be used in a function which <a href="#gc">specifies a GC
4438algorithm</a>.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004439
4440</div>
4441
4442
4443<!-- _______________________________________________________________________ -->
4444<div class="doc_subsubsection">
4445 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
4446</div>
4447
4448<div class="doc_text">
4449
4450<h5>Syntax:</h5>
4451
4452<pre>
Chris Lattner38bd5dd2007-09-21 17:30:40 +00004453 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004454</pre>
4455
4456<h5>Overview:</h5>
4457
4458<p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
4459locations, allowing garbage collector implementations that require read
4460barriers.</p>
4461
4462<h5>Arguments:</h5>
4463
4464<p>The second argument is the address to read from, which should be an address
4465allocated from the garbage collector. The first object is a pointer to the
4466start of the referenced object, if needed by the language runtime (otherwise
4467null).</p>
4468
4469<h5>Semantics:</h5>
4470
4471<p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
4472instruction, but may be replaced with substantially more complex code by the
Gordon Henriksen40393542007-12-25 02:31:26 +00004473garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
4474may only be used in a function which <a href="#gc">specifies a GC
4475algorithm</a>.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004476
4477</div>
4478
4479
4480<!-- _______________________________________________________________________ -->
4481<div class="doc_subsubsection">
4482 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
4483</div>
4484
4485<div class="doc_text">
4486
4487<h5>Syntax:</h5>
4488
4489<pre>
Chris Lattner38bd5dd2007-09-21 17:30:40 +00004490 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004491</pre>
4492
4493<h5>Overview:</h5>
4494
4495<p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
4496locations, allowing garbage collector implementations that require write
4497barriers (such as generational or reference counting collectors).</p>
4498
4499<h5>Arguments:</h5>
4500
4501<p>The first argument is the reference to store, the second is the start of the
4502object to store it to, and the third is the address of the field of Obj to
4503store to. If the runtime does not require a pointer to the object, Obj may be
4504null.</p>
4505
4506<h5>Semantics:</h5>
4507
4508<p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
4509instruction, but may be replaced with substantially more complex code by the
Gordon Henriksen40393542007-12-25 02:31:26 +00004510garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
4511may only be used in a function which <a href="#gc">specifies a GC
4512algorithm</a>.</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004513
4514</div>
4515
4516
4517
4518<!-- ======================================================================= -->
4519<div class="doc_subsection">
4520 <a name="int_codegen">Code Generator Intrinsics</a>
4521</div>
4522
4523<div class="doc_text">
4524<p>
4525These intrinsics are provided by LLVM to expose special features that may only
4526be implemented with code generator support.
4527</p>
4528
4529</div>
4530
4531<!-- _______________________________________________________________________ -->
4532<div class="doc_subsubsection">
4533 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
4534</div>
4535
4536<div class="doc_text">
4537
4538<h5>Syntax:</h5>
4539<pre>
4540 declare i8 *@llvm.returnaddress(i32 &lt;level&gt;)
4541</pre>
4542
4543<h5>Overview:</h5>
4544
4545<p>
4546The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
4547target-specific value indicating the return address of the current function
4548or one of its callers.
4549</p>
4550
4551<h5>Arguments:</h5>
4552
4553<p>
4554The argument to this intrinsic indicates which function to return the address
4555for. Zero indicates the calling function, one indicates its caller, etc. The
4556argument is <b>required</b> to be a constant integer value.
4557</p>
4558
4559<h5>Semantics:</h5>
4560
4561<p>
4562The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
4563the return address of the specified call frame, or zero if it cannot be
4564identified. The value returned by this intrinsic is likely to be incorrect or 0
4565for arguments other than zero, so it should only be used for debugging purposes.
4566</p>
4567
4568<p>
4569Note that calling this intrinsic does not prevent function inlining or other
4570aggressive transformations, so the value returned may not be that of the obvious
4571source-language caller.
4572</p>
4573</div>
4574
4575
4576<!-- _______________________________________________________________________ -->
4577<div class="doc_subsubsection">
4578 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
4579</div>
4580
4581<div class="doc_text">
4582
4583<h5>Syntax:</h5>
4584<pre>
Chris Lattner38bd5dd2007-09-21 17:30:40 +00004585 declare i8 *@llvm.frameaddress(i32 &lt;level&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004586</pre>
4587
4588<h5>Overview:</h5>
4589
4590<p>
4591The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
4592target-specific frame pointer value for the specified stack frame.
4593</p>
4594
4595<h5>Arguments:</h5>
4596
4597<p>
4598The argument to this intrinsic indicates which function to return the frame
4599pointer for. Zero indicates the calling function, one indicates its caller,
4600etc. The argument is <b>required</b> to be a constant integer value.
4601</p>
4602
4603<h5>Semantics:</h5>
4604
4605<p>
4606The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
4607the frame address of the specified call frame, or zero if it cannot be
4608identified. The value returned by this intrinsic is likely to be incorrect or 0
4609for arguments other than zero, so it should only be used for debugging purposes.
4610</p>
4611
4612<p>
4613Note that calling this intrinsic does not prevent function inlining or other
4614aggressive transformations, so the value returned may not be that of the obvious
4615source-language caller.
4616</p>
4617</div>
4618
4619<!-- _______________________________________________________________________ -->
4620<div class="doc_subsubsection">
4621 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
4622</div>
4623
4624<div class="doc_text">
4625
4626<h5>Syntax:</h5>
4627<pre>
Chris Lattner38bd5dd2007-09-21 17:30:40 +00004628 declare i8 *@llvm.stacksave()
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004629</pre>
4630
4631<h5>Overview:</h5>
4632
4633<p>
4634The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
4635the function stack, for use with <a href="#int_stackrestore">
4636<tt>llvm.stackrestore</tt></a>. This is useful for implementing language
4637features like scoped automatic variable sized arrays in C99.
4638</p>
4639
4640<h5>Semantics:</h5>
4641
4642<p>
4643This intrinsic returns a opaque pointer value that can be passed to <a
4644href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
4645<tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
4646<tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
4647state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
4648practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
4649that were allocated after the <tt>llvm.stacksave</tt> was executed.
4650</p>
4651
4652</div>
4653
4654<!-- _______________________________________________________________________ -->
4655<div class="doc_subsubsection">
4656 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
4657</div>
4658
4659<div class="doc_text">
4660
4661<h5>Syntax:</h5>
4662<pre>
4663 declare void @llvm.stackrestore(i8 * %ptr)
4664</pre>
4665
4666<h5>Overview:</h5>
4667
4668<p>
4669The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
4670the function stack to the state it was in when the corresponding <a
4671href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
4672useful for implementing language features like scoped automatic variable sized
4673arrays in C99.
4674</p>
4675
4676<h5>Semantics:</h5>
4677
4678<p>
4679See the description for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.
4680</p>
4681
4682</div>
4683
4684
4685<!-- _______________________________________________________________________ -->
4686<div class="doc_subsubsection">
4687 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
4688</div>
4689
4690<div class="doc_text">
4691
4692<h5>Syntax:</h5>
4693<pre>
Chris Lattner38bd5dd2007-09-21 17:30:40 +00004694 declare void @llvm.prefetch(i8* &lt;address&gt;, i32 &lt;rw&gt;, i32 &lt;locality&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004695</pre>
4696
4697<h5>Overview:</h5>
4698
4699
4700<p>
4701The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
4702a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
4703no
4704effect on the behavior of the program but can change its performance
4705characteristics.
4706</p>
4707
4708<h5>Arguments:</h5>
4709
4710<p>
4711<tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
4712determining if the fetch should be for a read (0) or write (1), and
4713<tt>locality</tt> is a temporal locality specifier ranging from (0) - no
4714locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
4715<tt>locality</tt> arguments must be constant integers.
4716</p>
4717
4718<h5>Semantics:</h5>
4719
4720<p>
4721This intrinsic does not modify the behavior of the program. In particular,
4722prefetches cannot trap and do not produce a value. On targets that support this
4723intrinsic, the prefetch can provide hints to the processor cache for better
4724performance.
4725</p>
4726
4727</div>
4728
4729<!-- _______________________________________________________________________ -->
4730<div class="doc_subsubsection">
4731 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
4732</div>
4733
4734<div class="doc_text">
4735
4736<h5>Syntax:</h5>
4737<pre>
Chris Lattner38bd5dd2007-09-21 17:30:40 +00004738 declare void @llvm.pcmarker(i32 &lt;id&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004739</pre>
4740
4741<h5>Overview:</h5>
4742
4743
4744<p>
4745The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
4746(PC) in a region of
4747code to simulators and other tools. The method is target specific, but it is
4748expected that the marker will use exported symbols to transmit the PC of the marker.
4749The marker makes no guarantees that it will remain with any specific instruction
4750after optimizations. It is possible that the presence of a marker will inhibit
4751optimizations. The intended use is to be inserted after optimizations to allow
4752correlations of simulation runs.
4753</p>
4754
4755<h5>Arguments:</h5>
4756
4757<p>
4758<tt>id</tt> is a numerical id identifying the marker.
4759</p>
4760
4761<h5>Semantics:</h5>
4762
4763<p>
4764This intrinsic does not modify the behavior of the program. Backends that do not
4765support this intrinisic may ignore it.
4766</p>
4767
4768</div>
4769
4770<!-- _______________________________________________________________________ -->
4771<div class="doc_subsubsection">
4772 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
4773</div>
4774
4775<div class="doc_text">
4776
4777<h5>Syntax:</h5>
4778<pre>
4779 declare i64 @llvm.readcyclecounter( )
4780</pre>
4781
4782<h5>Overview:</h5>
4783
4784
4785<p>
4786The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
4787counter register (or similar low latency, high accuracy clocks) on those targets
4788that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
4789As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
4790should only be used for small timings.
4791</p>
4792
4793<h5>Semantics:</h5>
4794
4795<p>
4796When directly supported, reading the cycle counter should not modify any memory.
4797Implementations are allowed to either return a application specific value or a
4798system wide value. On backends without support, this is lowered to a constant 0.
4799</p>
4800
4801</div>
4802
4803<!-- ======================================================================= -->
4804<div class="doc_subsection">
4805 <a name="int_libc">Standard C Library Intrinsics</a>
4806</div>
4807
4808<div class="doc_text">
4809<p>
4810LLVM provides intrinsics for a few important standard C library functions.
4811These intrinsics allow source-language front-ends to pass information about the
4812alignment of the pointer arguments to the code generator, providing opportunity
4813for more efficient code generation.
4814</p>
4815
4816</div>
4817
4818<!-- _______________________________________________________________________ -->
4819<div class="doc_subsubsection">
4820 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
4821</div>
4822
4823<div class="doc_text">
4824
4825<h5>Syntax:</h5>
4826<pre>
4827 declare void @llvm.memcpy.i32(i8 * &lt;dest&gt;, i8 * &lt;src&gt;,
4828 i32 &lt;len&gt;, i32 &lt;align&gt;)
4829 declare void @llvm.memcpy.i64(i8 * &lt;dest&gt;, i8 * &lt;src&gt;,
4830 i64 &lt;len&gt;, i32 &lt;align&gt;)
4831</pre>
4832
4833<h5>Overview:</h5>
4834
4835<p>
4836The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4837location to the destination location.
4838</p>
4839
4840<p>
4841Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
4842intrinsics do not return a value, and takes an extra alignment argument.
4843</p>
4844
4845<h5>Arguments:</h5>
4846
4847<p>
4848The first argument is a pointer to the destination, the second is a pointer to
4849the source. The third argument is an integer argument
4850specifying the number of bytes to copy, and the fourth argument is the alignment
4851of the source and destination locations.
4852</p>
4853
4854<p>
4855If the call to this intrinisic has an alignment value that is not 0 or 1, then
4856the caller guarantees that both the source and destination pointers are aligned
4857to that boundary.
4858</p>
4859
4860<h5>Semantics:</h5>
4861
4862<p>
4863The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4864location to the destination location, which are not allowed to overlap. It
4865copies "len" bytes of memory over. If the argument is known to be aligned to
4866some boundary, this can be specified as the fourth argument, otherwise it should
4867be set to 0 or 1.
4868</p>
4869</div>
4870
4871
4872<!-- _______________________________________________________________________ -->
4873<div class="doc_subsubsection">
4874 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
4875</div>
4876
4877<div class="doc_text">
4878
4879<h5>Syntax:</h5>
4880<pre>
4881 declare void @llvm.memmove.i32(i8 * &lt;dest&gt;, i8 * &lt;src&gt;,
4882 i32 &lt;len&gt;, i32 &lt;align&gt;)
4883 declare void @llvm.memmove.i64(i8 * &lt;dest&gt;, i8 * &lt;src&gt;,
4884 i64 &lt;len&gt;, i32 &lt;align&gt;)
4885</pre>
4886
4887<h5>Overview:</h5>
4888
4889<p>
4890The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
4891location to the destination location. It is similar to the
Chris Lattnerdba16ea2008-01-06 19:51:52 +00004892'<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to overlap.
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004893</p>
4894
4895<p>
4896Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
4897intrinsics do not return a value, and takes an extra alignment argument.
4898</p>
4899
4900<h5>Arguments:</h5>
4901
4902<p>
4903The first argument is a pointer to the destination, the second is a pointer to
4904the source. The third argument is an integer argument
4905specifying the number of bytes to copy, and the fourth argument is the alignment
4906of the source and destination locations.
4907</p>
4908
4909<p>
4910If the call to this intrinisic has an alignment value that is not 0 or 1, then
4911the caller guarantees that the source and destination pointers are aligned to
4912that boundary.
4913</p>
4914
4915<h5>Semantics:</h5>
4916
4917<p>
4918The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
4919location to the destination location, which may overlap. It
4920copies "len" bytes of memory over. If the argument is known to be aligned to
4921some boundary, this can be specified as the fourth argument, otherwise it should
4922be set to 0 or 1.
4923</p>
4924</div>
4925
4926
4927<!-- _______________________________________________________________________ -->
4928<div class="doc_subsubsection">
4929 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
4930</div>
4931
4932<div class="doc_text">
4933
4934<h5>Syntax:</h5>
4935<pre>
4936 declare void @llvm.memset.i32(i8 * &lt;dest&gt;, i8 &lt;val&gt;,
4937 i32 &lt;len&gt;, i32 &lt;align&gt;)
4938 declare void @llvm.memset.i64(i8 * &lt;dest&gt;, i8 &lt;val&gt;,
4939 i64 &lt;len&gt;, i32 &lt;align&gt;)
4940</pre>
4941
4942<h5>Overview:</h5>
4943
4944<p>
4945The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
4946byte value.
4947</p>
4948
4949<p>
4950Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
4951does not return a value, and takes an extra alignment argument.
4952</p>
4953
4954<h5>Arguments:</h5>
4955
4956<p>
4957The first argument is a pointer to the destination to fill, the second is the
4958byte value to fill it with, the third argument is an integer
4959argument specifying the number of bytes to fill, and the fourth argument is the
4960known alignment of destination location.
4961</p>
4962
4963<p>
4964If the call to this intrinisic has an alignment value that is not 0 or 1, then
4965the caller guarantees that the destination pointer is aligned to that boundary.
4966</p>
4967
4968<h5>Semantics:</h5>
4969
4970<p>
4971The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
4972the
4973destination location. If the argument is known to be aligned to some boundary,
4974this can be specified as the fourth argument, otherwise it should be set to 0 or
49751.
4976</p>
4977</div>
4978
4979
4980<!-- _______________________________________________________________________ -->
4981<div class="doc_subsubsection">
4982 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
4983</div>
4984
4985<div class="doc_text">
4986
4987<h5>Syntax:</h5>
Dale Johannesenf9adbb62007-10-02 17:47:38 +00004988<p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
Dan Gohman361079c2007-10-15 20:30:11 +00004989floating point or vector of floating point type. Not all targets support all
4990types however.
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004991<pre>
Dale Johannesenf9adbb62007-10-02 17:47:38 +00004992 declare float @llvm.sqrt.f32(float %Val)
4993 declare double @llvm.sqrt.f64(double %Val)
4994 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
4995 declare fp128 @llvm.sqrt.f128(fp128 %Val)
4996 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
Dan Gohmanf17a25c2007-07-18 16:29:46 +00004997</pre>
4998
4999<h5>Overview:</h5>
5000
5001<p>
5002The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
Dan Gohman361079c2007-10-15 20:30:11 +00005003returning the same value as the libm '<tt>sqrt</tt>' functions would. Unlike
Dan Gohmanf17a25c2007-07-18 16:29:46 +00005004<tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
Chris Lattnerf914a002008-01-29 07:00:44 +00005005negative numbers other than -0.0 (which allows for better optimization, because
5006there is no need to worry about errno being set). <tt>llvm.sqrt(-0.0)</tt> is
5007defined to return -0.0 like IEEE sqrt.
Dan Gohmanf17a25c2007-07-18 16:29:46 +00005008</p>
5009
5010<h5>Arguments:</h5>
5011
5012<p>
5013The argument and return value are floating point numbers of the same type.
5014</p>
5015
5016<h5>Semantics:</h5>
5017
5018<p>
5019This function returns the sqrt of the specified operand if it is a nonnegative
5020floating point number.
5021</p>
5022</div>
5023
5024<!-- _______________________________________________________________________ -->
5025<div class="doc_subsubsection">
5026 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
5027</div>
5028
5029<div class="doc_text">
5030
5031<h5>Syntax:</h5>
Dale Johannesenf9adbb62007-10-02 17:47:38 +00005032<p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
Dan Gohman361079c2007-10-15 20:30:11 +00005033floating point or vector of floating point type. Not all targets support all
5034types however.
Dan Gohmanf17a25c2007-07-18 16:29:46 +00005035<pre>
Dale Johannesenf9adbb62007-10-02 17:47:38 +00005036 declare float @llvm.powi.f32(float %Val, i32 %power)
5037 declare double @llvm.powi.f64(double %Val, i32 %power)
5038 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
5039 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
5040 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
Dan Gohmanf17a25c2007-07-18 16:29:46 +00005041</pre>
5042
5043<h5>Overview:</h5>
5044
5045<p>
5046The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
5047specified (positive or negative) power. The order of evaluation of
Dan Gohman361079c2007-10-15 20:30:11 +00005048multiplications is not defined. When a vector of floating point type is
5049used, the second argument remains a scalar integer value.
Dan Gohmanf17a25c2007-07-18 16:29:46 +00005050</p>
5051
5052<h5>Arguments:</h5>
5053
5054<p>
5055The second argument is an integer power, and the first is a value to raise to
5056that power.
5057</p>
5058
5059<h5>Semantics:</h5>
5060
5061<p>
5062This function returns the first value raised to the second power with an
5063unspecified sequence of rounding operations.</p>
5064</div>
5065
Dan Gohman361079c2007-10-15 20:30:11 +00005066<!-- _______________________________________________________________________ -->
5067<div class="doc_subsubsection">
5068 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
5069</div>
5070
5071<div class="doc_text">
5072
5073<h5>Syntax:</h5>
5074<p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
5075floating point or vector of floating point type. Not all targets support all
5076types however.
5077<pre>
5078 declare float @llvm.sin.f32(float %Val)
5079 declare double @llvm.sin.f64(double %Val)
5080 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
5081 declare fp128 @llvm.sin.f128(fp128 %Val)
5082 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
5083</pre>
5084
5085<h5>Overview:</h5>
5086
5087<p>
5088The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.
5089</p>
5090
5091<h5>Arguments:</h5>
5092
5093<p>
5094The argument and return value are floating point numbers of the same type.
5095</p>
5096
5097<h5>Semantics:</h5>
5098
5099<p>
5100This function returns the sine of the specified operand, returning the
5101same values as the libm <tt>sin</tt> functions would, and handles error
Dan Gohmaneaba92e2007-10-17 18:05:13 +00005102conditions in the same way.</p>
Dan Gohman361079c2007-10-15 20:30:11 +00005103</div>
5104
5105<!-- _______________________________________________________________________ -->
5106<div class="doc_subsubsection">
5107 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
5108</div>
5109
5110<div class="doc_text">
5111
5112<h5>Syntax:</h5>
5113<p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
5114floating point or vector of floating point type. Not all targets support all
5115types however.
5116<pre>
5117 declare float @llvm.cos.f32(float %Val)
5118 declare double @llvm.cos.f64(double %Val)
5119 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
5120 declare fp128 @llvm.cos.f128(fp128 %Val)
5121 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
5122</pre>
5123
5124<h5>Overview:</h5>
5125
5126<p>
5127The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.
5128</p>
5129
5130<h5>Arguments:</h5>
5131
5132<p>
5133The argument and return value are floating point numbers of the same type.
5134</p>
5135
5136<h5>Semantics:</h5>
5137
5138<p>
5139This function returns the cosine of the specified operand, returning the
5140same values as the libm <tt>cos</tt> functions would, and handles error
Dan Gohmaneaba92e2007-10-17 18:05:13 +00005141conditions in the same way.</p>
Dan Gohman361079c2007-10-15 20:30:11 +00005142</div>
5143
5144<!-- _______________________________________________________________________ -->
5145<div class="doc_subsubsection">
5146 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
5147</div>
5148
5149<div class="doc_text">
5150
5151<h5>Syntax:</h5>
5152<p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
5153floating point or vector of floating point type. Not all targets support all
5154types however.
5155<pre>
5156 declare float @llvm.pow.f32(float %Val, float %Power)
5157 declare double @llvm.pow.f64(double %Val, double %Power)
5158 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
5159 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
5160 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
5161</pre>
5162
5163<h5>Overview:</h5>
5164
5165<p>
5166The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
5167specified (positive or negative) power.
5168</p>
5169
5170<h5>Arguments:</h5>
5171
5172<p>
5173The second argument is a floating point power, and the first is a value to
5174raise to that power.
5175</p>
5176
5177<h5>Semantics:</h5>
5178
5179<p>
5180This function returns the first value raised to the second power,
5181returning the
5182same values as the libm <tt>pow</tt> functions would, and handles error
Dan Gohmaneaba92e2007-10-17 18:05:13 +00005183conditions in the same way.</p>
Dan Gohman361079c2007-10-15 20:30:11 +00005184</div>
5185
Dan Gohmanf17a25c2007-07-18 16:29:46 +00005186
5187<!-- ======================================================================= -->
5188<div class="doc_subsection">
5189 <a name="int_manip">Bit Manipulation Intrinsics</a>
5190</div>
5191
5192<div class="doc_text">
5193<p>
5194LLVM provides intrinsics for a few important bit manipulation operations.
5195These allow efficient code generation for some algorithms.
5196</p>
5197
5198</div>
5199
5200<!-- _______________________________________________________________________ -->
5201<div class="doc_subsubsection">
5202 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
5203</div>
5204
5205<div class="doc_text">
5206
5207<h5>Syntax:</h5>
5208<p>This is an overloaded intrinsic function. You can use bswap on any integer
Chandler Carrutha228e392007-08-04 01:51:18 +00005209type that is an even number of bytes (i.e. BitWidth % 16 == 0).
Dan Gohmanf17a25c2007-07-18 16:29:46 +00005210<pre>
Chandler Carrutha228e392007-08-04 01:51:18 +00005211 declare i16 @llvm.bswap.i16(i16 &lt;id&gt;)
5212 declare i32 @llvm.bswap.i32(i32 &lt;id&gt;)
5213 declare i64 @llvm.bswap.i64(i64 &lt;id&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +00005214</pre>
5215
5216<h5>Overview:</h5>
5217
5218<p>
5219The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
5220values with an even number of bytes (positive multiple of 16 bits). These are
5221useful for performing operations on data that is not in the target's native
5222byte order.
5223</p>
5224
5225<h5>Semantics:</h5>
5226
5227<p>
Chandler Carrutha228e392007-08-04 01:51:18 +00005228The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
Dan Gohmanf17a25c2007-07-18 16:29:46 +00005229and low byte of the input i16 swapped. Similarly, the <tt>llvm.bswap.i32</tt>
5230intrinsic returns an i32 value that has the four bytes of the input i32
5231swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the returned
Chandler Carrutha228e392007-08-04 01:51:18 +00005232i32 will have its bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i48</tt>,
5233<tt>llvm.bswap.i64</tt> and other intrinsics extend this concept to
Dan Gohmanf17a25c2007-07-18 16:29:46 +00005234additional even-byte lengths (6 bytes, 8 bytes and more, respectively).
5235</p>
5236
5237</div>
5238
5239<!-- _______________________________________________________________________ -->
5240<div class="doc_subsubsection">
5241 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
5242</div>
5243
5244<div class="doc_text">
5245
5246<h5>Syntax:</h5>
5247<p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
5248width. Not all targets support all bit widths however.
5249<pre>
Chandler Carrutha228e392007-08-04 01:51:18 +00005250 declare i8 @llvm.ctpop.i8 (i8 &lt;src&gt;)
5251 declare i16 @llvm.ctpop.i16(i16 &lt;src&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +00005252 declare i32 @llvm.ctpop.i32(i32 &lt;src&gt;)
Chandler Carrutha228e392007-08-04 01:51:18 +00005253 declare i64 @llvm.ctpop.i64(i64 &lt;src&gt;)
5254 declare i256 @llvm.ctpop.i256(i256 &lt;src&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +00005255</pre>
5256
5257<h5>Overview:</h5>
5258
5259<p>
5260The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
5261value.
5262</p>
5263
5264<h5>Arguments:</h5>
5265
5266<p>
5267The only argument is the value to be counted. The argument may be of any
5268integer type. The return type must match the argument type.
5269</p>
5270
5271<h5>Semantics:</h5>
5272
5273<p>
5274The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
5275</p>
5276</div>
5277
5278<!-- _______________________________________________________________________ -->
5279<div class="doc_subsubsection">
5280 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
5281</div>
5282
5283<div class="doc_text">
5284
5285<h5>Syntax:</h5>
5286<p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
5287integer bit width. Not all targets support all bit widths however.
5288<pre>
Chandler Carrutha228e392007-08-04 01:51:18 +00005289 declare i8 @llvm.ctlz.i8 (i8 &lt;src&gt;)
5290 declare i16 @llvm.ctlz.i16(i16 &lt;src&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +00005291 declare i32 @llvm.ctlz.i32(i32 &lt;src&gt;)
Chandler Carrutha228e392007-08-04 01:51:18 +00005292 declare i64 @llvm.ctlz.i64(i64 &lt;src&gt;)
5293 declare i256 @llvm.ctlz.i256(i256 &lt;src&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +00005294</pre>
5295
5296<h5>Overview:</h5>
5297
5298<p>
5299The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
5300leading zeros in a variable.
5301</p>
5302
5303<h5>Arguments:</h5>
5304
5305<p>
5306The only argument is the value to be counted. The argument may be of any
5307integer type. The return type must match the argument type.
5308</p>
5309
5310<h5>Semantics:</h5>
5311
5312<p>
5313The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
5314in a variable. If the src == 0 then the result is the size in bits of the type
5315of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.
5316</p>
5317</div>
5318
5319
5320
5321<!-- _______________________________________________________________________ -->
5322<div class="doc_subsubsection">
5323 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
5324</div>
5325
5326<div class="doc_text">
5327
5328<h5>Syntax:</h5>
5329<p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
5330integer bit width. Not all targets support all bit widths however.
5331<pre>
Chandler Carrutha228e392007-08-04 01:51:18 +00005332 declare i8 @llvm.cttz.i8 (i8 &lt;src&gt;)
5333 declare i16 @llvm.cttz.i16(i16 &lt;src&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +00005334 declare i32 @llvm.cttz.i32(i32 &lt;src&gt;)
Chandler Carrutha228e392007-08-04 01:51:18 +00005335 declare i64 @llvm.cttz.i64(i64 &lt;src&gt;)
5336 declare i256 @llvm.cttz.i256(i256 &lt;src&gt;)
Dan Gohmanf17a25c2007-07-18 16:29:46 +00005337</pre>
5338
5339<h5>Overview:</h5>
5340
5341<p>
5342The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
5343trailing zeros.
5344</p>
5345
5346<h5>Arguments:</h5>
5347
5348<p>
5349The only argument is the value to be counted. The argument may be of any
5350integer type. The return type must match the argument type.
5351</p>
5352
5353<h5>Semantics:</h5>
5354
5355<p>
5356The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
5357in a variable. If the src == 0 then the result is the size in bits of the type
5358of src. For example, <tt>llvm.cttz(2) = 1</tt>.
5359</p>
5360</div>
5361
5362<!-- _______________________________________________________________________ -->
5363<div class="doc_subsubsection">
5364 <a name="int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic</a>
5365</div>
5366
5367<div class="doc_text">
5368
5369<h5>Syntax:</h5>
5370<p>This is an overloaded intrinsic. You can use <tt>llvm.part.select</tt>
5371on any integer bit width.
5372<pre>
Chandler Carrutha228e392007-08-04 01:51:18 +00005373 declare i17 @llvm.part.select.i17 (i17 %val, i32 %loBit, i32 %hiBit)
5374 declare i29 @llvm.part.select.i29 (i29 %val, i32 %loBit, i32 %hiBit)
Dan Gohmanf17a25c2007-07-18 16:29:46 +00005375</pre>
5376
5377<h5>Overview:</h5>
5378<p>The '<tt>llvm.part.select</tt>' family of intrinsic functions selects a
5379range of bits from an integer value and returns them in the same bit width as
5380the original value.</p>
5381
5382<h5>Arguments:</h5>
5383<p>The first argument, <tt>%val</tt> and the result may be integer types of
5384any bit width but they must have the same bit width. The second and third
5385arguments must be <tt>i32</tt> type since they specify only a bit index.</p>
5386
5387<h5>Semantics:</h5>
5388<p>The operation of the '<tt>llvm.part.select</tt>' intrinsic has two modes
5389of operation: forwards and reverse. If <tt>%loBit</tt> is greater than
5390<tt>%hiBits</tt> then the intrinsic operates in reverse mode. Otherwise it
5391operates in forward mode.</p>
5392<p>In forward mode, this intrinsic is the equivalent of shifting <tt>%val</tt>
5393right by <tt>%loBit</tt> bits and then ANDing it with a mask with
5394only the <tt>%hiBit - %loBit</tt> bits set, as follows:</p>
5395<ol>
5396 <li>The <tt>%val</tt> is shifted right (LSHR) by the number of bits specified
5397 by <tt>%loBits</tt>. This normalizes the value to the low order bits.</li>
5398 <li>The <tt>%loBits</tt> value is subtracted from the <tt>%hiBits</tt> value
5399 to determine the number of bits to retain.</li>
5400 <li>A mask of the retained bits is created by shifting a -1 value.</li>
5401 <li>The mask is ANDed with <tt>%val</tt> to produce the result.
5402</ol>
5403<p>In reverse mode, a similar computation is made except that the bits are
5404returned in the reverse order. So, for example, if <tt>X</tt> has the value
5405<tt>i16 0x0ACF (101011001111)</tt> and we apply
5406<tt>part.select(i16 X, 8, 3)</tt> to it, we get back the value
5407<tt>i16 0x0026 (000000100110)</tt>.</p>
5408</div>
5409
5410<div class="doc_subsubsection">
5411 <a name="int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic</a>
5412</div>
5413
5414<div class="doc_text">
5415
5416<h5>Syntax:</h5>
5417<p>This is an overloaded intrinsic. You can use <tt>llvm.part.set</tt>
5418on any integer bit width.
5419<pre>
Chandler Carrutha228e392007-08-04 01:51:18 +00005420 declare i17 @llvm.part.set.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
5421 declare i29 @llvm.part.set.i29.i9 (i29 %val, i9 %repl, i32 %lo, i32 %hi)
Dan Gohmanf17a25c2007-07-18 16:29:46 +00005422</pre>
5423
5424<h5>Overview:</h5>
5425<p>The '<tt>llvm.part.set</tt>' family of intrinsic functions replaces a range
5426of bits in an integer value with another integer value. It returns the integer
5427with the replaced bits.</p>
5428
5429<h5>Arguments:</h5>
5430<p>The first argument, <tt>%val</tt> and the result may be integer types of
5431any bit width but they must have the same bit width. <tt>%val</tt> is the value
5432whose bits will be replaced. The second argument, <tt>%repl</tt> may be an
5433integer of any bit width. The third and fourth arguments must be <tt>i32</tt>
5434type since they specify only a bit index.</p>
5435
5436<h5>Semantics:</h5>
5437<p>The operation of the '<tt>llvm.part.set</tt>' intrinsic has two modes
5438of operation: forwards and reverse. If <tt>%lo</tt> is greater than
5439<tt>%hi</tt> then the intrinsic operates in reverse mode. Otherwise it
5440operates in forward mode.</p>
5441<p>For both modes, the <tt>%repl</tt> value is prepared for use by either
5442truncating it down to the size of the replacement area or zero extending it
5443up to that size.</p>
5444<p>In forward mode, the bits between <tt>%lo</tt> and <tt>%hi</tt> (inclusive)
5445are replaced with corresponding bits from <tt>%repl</tt>. That is the 0th bit
5446in <tt>%repl</tt> replaces the <tt>%lo</tt>th bit in <tt>%val</tt> and etc. up
5447to the <tt>%hi</tt>th bit.
5448<p>In reverse mode, a similar computation is made except that the bits are
5449reversed. That is, the <tt>0</tt>th bit in <tt>%repl</tt> replaces the
5450<tt>%hi</tt> bit in <tt>%val</tt> and etc. down to the <tt>%lo</tt>th bit.
5451<h5>Examples:</h5>
5452<pre>
5453 llvm.part.set(0xFFFF, 0, 4, 7) -&gt; 0xFF0F
5454 llvm.part.set(0xFFFF, 0, 7, 4) -&gt; 0xFF0F
5455 llvm.part.set(0xFFFF, 1, 7, 4) -&gt; 0xFF8F
5456 llvm.part.set(0xFFFF, F, 8, 3) -&gt; 0xFFE7
5457 llvm.part.set(0xFFFF, 0, 3, 8) -&gt; 0xFE07
5458</pre>
5459</div>
5460
5461<!-- ======================================================================= -->
5462<div class="doc_subsection">
5463 <a name="int_debugger">Debugger Intrinsics</a>
5464</div>
5465
5466<div class="doc_text">
5467<p>
5468The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
5469are described in the <a
5470href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
5471Debugging</a> document.
5472</p>
5473</div>
5474
5475
5476<!-- ======================================================================= -->
5477<div class="doc_subsection">
5478 <a name="int_eh">Exception Handling Intrinsics</a>
5479</div>
5480
5481<div class="doc_text">
5482<p> The LLVM exception handling intrinsics (which all start with
5483<tt>llvm.eh.</tt> prefix), are described in the <a
5484href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
5485Handling</a> document. </p>
5486</div>
5487
5488<!-- ======================================================================= -->
5489<div class="doc_subsection">
Duncan Sands7407a9f2007-09-11 14:10:23 +00005490 <a name="int_trampoline">Trampoline Intrinsic</a>
Duncan Sands38947cd2007-07-27 12:58:54 +00005491</div>
5492
5493<div class="doc_text">
5494<p>
Duncan Sands7407a9f2007-09-11 14:10:23 +00005495 This intrinsic makes it possible to excise one parameter, marked with
Duncan Sands38947cd2007-07-27 12:58:54 +00005496 the <tt>nest</tt> attribute, from a function. The result is a callable
5497 function pointer lacking the nest parameter - the caller does not need
5498 to provide a value for it. Instead, the value to use is stored in
5499 advance in a "trampoline", a block of memory usually allocated
5500 on the stack, which also contains code to splice the nest value into the
5501 argument list. This is used to implement the GCC nested function address
5502 extension.
5503</p>
5504<p>
5505 For example, if the function is
5506 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
Bill Wendlinge262b4c2007-09-22 09:23:55 +00005507 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as follows:</p>
Duncan Sands38947cd2007-07-27 12:58:54 +00005508<pre>
Duncan Sands7407a9f2007-09-11 14:10:23 +00005509 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
5510 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
5511 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
5512 %fp = bitcast i8* %p to i32 (i32, i32)*
Duncan Sands38947cd2007-07-27 12:58:54 +00005513</pre>
Bill Wendlinge262b4c2007-09-22 09:23:55 +00005514 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
5515 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
Duncan Sands38947cd2007-07-27 12:58:54 +00005516</div>
5517
5518<!-- _______________________________________________________________________ -->
5519<div class="doc_subsubsection">
5520 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
5521</div>
5522<div class="doc_text">
5523<h5>Syntax:</h5>
5524<pre>
Duncan Sands7407a9f2007-09-11 14:10:23 +00005525declare i8* @llvm.init.trampoline(i8* &lt;tramp&gt;, i8* &lt;func&gt;, i8* &lt;nval&gt;)
Duncan Sands38947cd2007-07-27 12:58:54 +00005526</pre>
5527<h5>Overview:</h5>
5528<p>
Duncan Sands7407a9f2007-09-11 14:10:23 +00005529 This fills the memory pointed to by <tt>tramp</tt> with code
5530 and returns a function pointer suitable for executing it.
Duncan Sands38947cd2007-07-27 12:58:54 +00005531</p>
5532<h5>Arguments:</h5>
5533<p>
5534 The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
5535 pointers. The <tt>tramp</tt> argument must point to a sufficiently large
5536 and sufficiently aligned block of memory; this memory is written to by the
Duncan Sands35012212007-08-22 23:39:54 +00005537 intrinsic. Note that the size and the alignment are target-specific - LLVM
5538 currently provides no portable way of determining them, so a front-end that
5539 generates this intrinsic needs to have some target-specific knowledge.
5540 The <tt>func</tt> argument must hold a function bitcast to an <tt>i8*</tt>.
Duncan Sands38947cd2007-07-27 12:58:54 +00005541</p>
5542<h5>Semantics:</h5>
5543<p>
5544 The block of memory pointed to by <tt>tramp</tt> is filled with target
Duncan Sands7407a9f2007-09-11 14:10:23 +00005545 dependent code, turning it into a function. A pointer to this function is
5546 returned, but needs to be bitcast to an
Duncan Sands38947cd2007-07-27 12:58:54 +00005547 <a href="#int_trampoline">appropriate function pointer type</a>
Duncan Sands7407a9f2007-09-11 14:10:23 +00005548 before being called. The new function's signature is the same as that of
5549 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
5550 removed. At most one such <tt>nest</tt> argument is allowed, and it must be
5551 of pointer type. Calling the new function is equivalent to calling
5552 <tt>func</tt> with the same argument list, but with <tt>nval</tt> used for the
5553 missing <tt>nest</tt> argument. If, after calling
5554 <tt>llvm.init.trampoline</tt>, the memory pointed to by <tt>tramp</tt> is
5555 modified, then the effect of any later call to the returned function pointer is
5556 undefined.
Duncan Sands38947cd2007-07-27 12:58:54 +00005557</p>
5558</div>
5559
5560<!-- ======================================================================= -->
5561<div class="doc_subsection">
Andrew Lenharth785610d2008-02-16 01:24:58 +00005562 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
5563</div>
5564
5565<div class="doc_text">
5566<p>
5567 These intrinsic functions expand the "universal IR" of LLVM to represent
5568 hardware constructs for atomic operations and memory synchronization. This
5569 provides an interface to the hardware, not an interface to the programmer. It
5570 is aimed at a low enough level to allow any programming models or APIs which
5571 need atomic behaviors to map cleanly onto it. It is also modeled primarily on
5572 hardware behavior. Just as hardware provides a "universal IR" for source
5573 languages, it also provides a starting point for developing a "universal"
5574 atomic operation and synchronization IR.
5575</p>
5576<p>
5577 These do <em>not</em> form an API such as high-level threading libraries,
5578 software transaction memory systems, atomic primitives, and intrinsic
5579 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
5580 application libraries. The hardware interface provided by LLVM should allow
5581 a clean implementation of all of these APIs and parallel programming models.
5582 No one model or paradigm should be selected above others unless the hardware
5583 itself ubiquitously does so.
5584
5585</p>
5586</div>
5587
5588<!-- _______________________________________________________________________ -->
5589<div class="doc_subsubsection">
5590 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
5591</div>
5592<div class="doc_text">
5593<h5>Syntax:</h5>
5594<pre>
5595declare void @llvm.memory.barrier( i1 &lt;ll&gt;, i1 &lt;ls&gt;, i1 &lt;sl&gt;, i1 &lt;ss&gt;,
5596i1 &lt;device&gt; )
5597
5598</pre>
5599<h5>Overview:</h5>
5600<p>
5601 The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
5602 specific pairs of memory access types.
5603</p>
5604<h5>Arguments:</h5>
5605<p>
5606 The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
5607 The first four arguments enables a specific barrier as listed below. The fith
5608 argument specifies that the barrier applies to io or device or uncached memory.
5609
5610</p>
5611 <ul>
5612 <li><tt>ll</tt>: load-load barrier</li>
5613 <li><tt>ls</tt>: load-store barrier</li>
5614 <li><tt>sl</tt>: store-load barrier</li>
5615 <li><tt>ss</tt>: store-store barrier</li>
5616 <li><tt>device</tt>: barrier applies to device and uncached memory also.
5617 </ul>
5618<h5>Semantics:</h5>
5619<p>
5620 This intrinsic causes the system to enforce some ordering constraints upon
5621 the loads and stores of the program. This barrier does not indicate
5622 <em>when</em> any events will occur, it only enforces an <em>order</em> in
5623 which they occur. For any of the specified pairs of load and store operations
5624 (f.ex. load-load, or store-load), all of the first operations preceding the
5625 barrier will complete before any of the second operations succeeding the
5626 barrier begin. Specifically the semantics for each pairing is as follows:
5627</p>
5628 <ul>
5629 <li><tt>ll</tt>: All loads before the barrier must complete before any load
5630 after the barrier begins.</li>
5631
5632 <li><tt>ls</tt>: All loads before the barrier must complete before any
5633 store after the barrier begins.</li>
5634 <li><tt>ss</tt>: All stores before the barrier must complete before any
5635 store after the barrier begins.</li>
5636 <li><tt>sl</tt>: All stores before the barrier must complete before any
5637 load after the barrier begins.</li>
5638 </ul>
5639<p>
5640 These semantics are applied with a logical "and" behavior when more than one
5641 is enabled in a single memory barrier intrinsic.
5642</p>
5643<p>
5644 Backends may implement stronger barriers than those requested when they do not
5645 support as fine grained a barrier as requested. Some architectures do not
5646 need all types of barriers and on such architectures, these become noops.
5647</p>
5648<h5>Example:</h5>
5649<pre>
5650%ptr = malloc i32
5651 store i32 4, %ptr
5652
5653%result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
5654 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
5655 <i>; guarantee the above finishes</i>
5656 store i32 8, %ptr <i>; before this begins</i>
5657</pre>
5658</div>
5659
Andrew Lenharthe44f3902008-02-21 06:45:13 +00005660<!-- _______________________________________________________________________ -->
5661<div class="doc_subsubsection">
5662 <a name="int_atomic_lcs">'<tt>llvm.atomic.lcs.*</tt>' Intrinsic</a>
5663</div>
5664<div class="doc_text">
5665<h5>Syntax:</h5>
5666<p>
5667 This is an overloaded intrinsic. You can use <tt>llvm.atomic.lcs</tt> on any
5668 integer bit width. Not all targets support all bit widths however.</p>
5669
5670<pre>
5671declare i8 @llvm.atomic.lcs.i8( i8* &lt;ptr&gt;, i8 &lt;cmp&gt;, i8 &lt;val&gt; )
5672declare i16 @llvm.atomic.lcs.i16( i16* &lt;ptr&gt;, i16 &lt;cmp&gt;, i16 &lt;val&gt; )
5673declare i32 @llvm.atomic.lcs.i32( i32* &lt;ptr&gt;, i32 &lt;cmp&gt;, i32 &lt;val&gt; )
5674declare i64 @llvm.atomic.lcs.i64( i64* &lt;ptr&gt;, i64 &lt;cmp&gt;, i64 &lt;val&gt; )
5675
5676</pre>
5677<h5>Overview:</h5>
5678<p>
5679 This loads a value in memory and compares it to a given value. If they are
5680 equal, it stores a new value into the memory.
5681</p>
5682<h5>Arguments:</h5>
5683<p>
5684 The <tt>llvm.atomic.lcs</tt> intrinsic takes three arguments. The result as
5685 well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
5686 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
5687 this integer type. While any bit width integer may be used, targets may only
5688 lower representations they support in hardware.
5689
5690</p>
5691<h5>Semantics:</h5>
5692<p>
5693 This entire intrinsic must be executed atomically. It first loads the value
5694 in memory pointed to by <tt>ptr</tt> and compares it with the value
5695 <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the memory. The
5696 loaded value is yielded in all cases. This provides the equivalent of an
5697 atomic compare-and-swap operation within the SSA framework.
5698</p>
5699<h5>Examples:</h5>
5700
5701<pre>
5702%ptr = malloc i32
5703 store i32 4, %ptr
5704
5705%val1 = add i32 4, 4
5706%result1 = call i32 @llvm.atomic.lcs.i32( i32* %ptr, i32 4, %val1 )
5707 <i>; yields {i32}:result1 = 4</i>
5708%stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
5709%memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
5710
5711%val2 = add i32 1, 1
5712%result2 = call i32 @llvm.atomic.lcs.i32( i32* %ptr, i32 5, %val2 )
5713 <i>; yields {i32}:result2 = 8</i>
5714%stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
5715
5716%memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
5717</pre>
5718</div>
5719
5720<!-- _______________________________________________________________________ -->
5721<div class="doc_subsubsection">
5722 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
5723</div>
5724<div class="doc_text">
5725<h5>Syntax:</h5>
5726
5727<p>
5728 This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
5729 integer bit width. Not all targets support all bit widths however.</p>
5730<pre>
5731declare i8 @llvm.atomic.swap.i8( i8* &lt;ptr&gt;, i8 &lt;val&gt; )
5732declare i16 @llvm.atomic.swap.i16( i16* &lt;ptr&gt;, i16 &lt;val&gt; )
5733declare i32 @llvm.atomic.swap.i32( i32* &lt;ptr&gt;, i32 &lt;val&gt; )
5734declare i64 @llvm.atomic.swap.i64( i64* &lt;ptr&gt;, i64 &lt;val&gt; )
5735
5736</pre>
5737<h5>Overview:</h5>
5738<p>
5739 This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
5740 the value from memory. It then stores the value in <tt>val</tt> in the memory
5741 at <tt>ptr</tt>.
5742</p>
5743<h5>Arguments:</h5>
5744
5745<p>
5746 The <tt>llvm.atomic.ls</tt> intrinsic takes two arguments. Both the
5747 <tt>val</tt> argument and the result must be integers of the same bit width.
5748 The first argument, <tt>ptr</tt>, must be a pointer to a value of this
5749 integer type. The targets may only lower integer representations they
5750 support.
5751</p>
5752<h5>Semantics:</h5>
5753<p>
5754 This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
5755 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
5756 equivalent of an atomic swap operation within the SSA framework.
5757
5758</p>
5759<h5>Examples:</h5>
5760<pre>
5761%ptr = malloc i32
5762 store i32 4, %ptr
5763
5764%val1 = add i32 4, 4
5765%result1 = call i32 @llvm.atomic.swap.i32( i32* %ptr, i32 %val1 )
5766 <i>; yields {i32}:result1 = 4</i>
5767%stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
5768%memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
5769
5770%val2 = add i32 1, 1
5771%result2 = call i32 @llvm.atomic.swap.i32( i32* %ptr, i32 %val2 )
5772 <i>; yields {i32}:result2 = 8</i>
5773
5774%stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
5775%memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
5776</pre>
5777</div>
5778
5779<!-- _______________________________________________________________________ -->
5780<div class="doc_subsubsection">
5781 <a name="int_atomic_las">'<tt>llvm.atomic.las.*</tt>' Intrinsic</a>
5782
5783</div>
5784<div class="doc_text">
5785<h5>Syntax:</h5>
5786<p>
5787 This is an overloaded intrinsic. You can use <tt>llvm.atomic.las</tt> on any
5788 integer bit width. Not all targets support all bit widths however.</p>
5789<pre>
5790declare i8 @llvm.atomic.las.i8.( i8* &lt;ptr&gt;, i8 &lt;delta&gt; )
5791declare i16 @llvm.atomic.las.i16.( i16* &lt;ptr&gt;, i16 &lt;delta&gt; )
5792declare i32 @llvm.atomic.las.i32.( i32* &lt;ptr&gt;, i32 &lt;delta&gt; )
5793declare i64 @llvm.atomic.las.i64.( i64* &lt;ptr&gt;, i64 &lt;delta&gt; )
5794
5795</pre>
5796<h5>Overview:</h5>
5797<p>
5798 This intrinsic adds <tt>delta</tt> to the value stored in memory at
5799 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
5800</p>
5801<h5>Arguments:</h5>
5802<p>
5803
5804 The intrinsic takes two arguments, the first a pointer to an integer value
5805 and the second an integer value. The result is also an integer value. These
5806 integer types can have any bit width, but they must all have the same bit
5807 width. The targets may only lower integer representations they support.
5808</p>
5809<h5>Semantics:</h5>
5810<p>
5811 This intrinsic does a series of operations atomically. It first loads the
5812 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
5813 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
5814</p>
5815
5816<h5>Examples:</h5>
5817<pre>
5818%ptr = malloc i32
5819 store i32 4, %ptr
5820%result1 = call i32 @llvm.atomic.las.i32( i32* %ptr, i32 4 )
5821 <i>; yields {i32}:result1 = 4</i>
5822%result2 = call i32 @llvm.atomic.las.i32( i32* %ptr, i32 2 )
5823 <i>; yields {i32}:result2 = 8</i>
5824%result3 = call i32 @llvm.atomic.las.i32( i32* %ptr, i32 5 )
5825 <i>; yields {i32}:result3 = 10</i>
5826%memval = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
5827</pre>
5828</div>
5829
Andrew Lenharth785610d2008-02-16 01:24:58 +00005830
5831<!-- ======================================================================= -->
5832<div class="doc_subsection">
Dan Gohmanf17a25c2007-07-18 16:29:46 +00005833 <a name="int_general">General Intrinsics</a>
5834</div>
5835
5836<div class="doc_text">
5837<p> This class of intrinsics is designed to be generic and has
5838no specific purpose. </p>
5839</div>
5840
5841<!-- _______________________________________________________________________ -->
5842<div class="doc_subsubsection">
5843 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
5844</div>
5845
5846<div class="doc_text">
5847
5848<h5>Syntax:</h5>
5849<pre>
5850 declare void @llvm.var.annotation(i8* &lt;val&gt;, i8* &lt;str&gt;, i8* &lt;str&gt;, i32 &lt;int&gt; )
5851</pre>
5852
5853<h5>Overview:</h5>
5854
5855<p>
5856The '<tt>llvm.var.annotation</tt>' intrinsic
5857</p>
5858
5859<h5>Arguments:</h5>
5860
5861<p>
5862The first argument is a pointer to a value, the second is a pointer to a
5863global string, the third is a pointer to a global string which is the source
5864file name, and the last argument is the line number.
5865</p>
5866
5867<h5>Semantics:</h5>
5868
5869<p>
Anton Korobeynikove6e764f2008-01-15 22:31:34 +00005870This intrinsic allows annotation of local variables with arbitrary strings.
Dan Gohmanf17a25c2007-07-18 16:29:46 +00005871This can be useful for special purpose optimizations that want to look for these
Anton Korobeynikove6e764f2008-01-15 22:31:34 +00005872annotations. These have no other defined use, they are ignored by code
5873generation and optimization.
5874</p>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00005875</div>
5876
Tanya Lattnerb306a9e2007-09-21 22:59:12 +00005877<!-- _______________________________________________________________________ -->
5878<div class="doc_subsubsection">
Tanya Lattnerc9869b12007-09-21 23:57:59 +00005879 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
Tanya Lattnerb306a9e2007-09-21 22:59:12 +00005880</div>
5881
5882<div class="doc_text">
5883
5884<h5>Syntax:</h5>
Tanya Lattnere545be72007-09-21 23:56:27 +00005885<p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
5886any integer bit width.
5887</p>
Tanya Lattnerb306a9e2007-09-21 22:59:12 +00005888<pre>
Tanya Lattner09161fe2007-09-22 00:03:01 +00005889 declare i8 @llvm.annotation.i8(i8 &lt;val&gt;, i8* &lt;str&gt;, i8* &lt;str&gt;, i32 &lt;int&gt; )
5890 declare i16 @llvm.annotation.i16(i16 &lt;val&gt;, i8* &lt;str&gt;, i8* &lt;str&gt;, i32 &lt;int&gt; )
5891 declare i32 @llvm.annotation.i32(i32 &lt;val&gt;, i8* &lt;str&gt;, i8* &lt;str&gt;, i32 &lt;int&gt; )
5892 declare i64 @llvm.annotation.i64(i64 &lt;val&gt;, i8* &lt;str&gt;, i8* &lt;str&gt;, i32 &lt;int&gt; )
5893 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 +00005894</pre>
5895
5896<h5>Overview:</h5>
Tanya Lattnere545be72007-09-21 23:56:27 +00005897
5898<p>
5899The '<tt>llvm.annotation</tt>' intrinsic.
Tanya Lattnerb306a9e2007-09-21 22:59:12 +00005900</p>
5901
5902<h5>Arguments:</h5>
5903
5904<p>
5905The first argument is an integer value (result of some expression),
5906the second is a pointer to a global string, the third is a pointer to a global
5907string which is the source file name, and the last argument is the line number.
Tanya Lattnere545be72007-09-21 23:56:27 +00005908It returns the value of the first argument.
Tanya Lattnerb306a9e2007-09-21 22:59:12 +00005909</p>
5910
5911<h5>Semantics:</h5>
5912
5913<p>
5914This intrinsic allows annotations to be put on arbitrary expressions
5915with arbitrary strings. This can be useful for special purpose optimizations
5916that want to look for these annotations. These have no other defined use, they
5917are ignored by code generation and optimization.
5918</div>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00005919
Anton Korobeynikove6e764f2008-01-15 22:31:34 +00005920<!-- _______________________________________________________________________ -->
5921<div class="doc_subsubsection">
5922 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
5923</div>
5924
5925<div class="doc_text">
5926
5927<h5>Syntax:</h5>
5928<pre>
5929 declare void @llvm.trap()
5930</pre>
5931
5932<h5>Overview:</h5>
5933
5934<p>
5935The '<tt>llvm.trap</tt>' intrinsic
5936</p>
5937
5938<h5>Arguments:</h5>
5939
5940<p>
5941None
5942</p>
5943
5944<h5>Semantics:</h5>
5945
5946<p>
5947This intrinsics is lowered to the target dependent trap instruction. If the
5948target does not have a trap instruction, this intrinsic will be lowered to the
5949call of the abort() function.
5950</p>
5951</div>
5952
Dan Gohmanf17a25c2007-07-18 16:29:46 +00005953<!-- *********************************************************************** -->
5954<hr>
5955<address>
5956 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
5957 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
5958 <a href="http://validator.w3.org/check/referer"><img
Chris Lattner08497ce2008-01-04 04:33:49 +00005959 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
Dan Gohmanf17a25c2007-07-18 16:29:46 +00005960
5961 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
5962 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
5963 Last modified: $Date$
5964</address>
Chris Lattner08497ce2008-01-04 04:33:49 +00005965
Dan Gohmanf17a25c2007-07-18 16:29:46 +00005966</body>
5967</html>