1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
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">
15 <div class="doc_title"> LLVM Language Reference Manual </div>
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>
22 <li><a href="#modulestructure">Module Structure</a></li>
23 <li><a href="#linkage">Linkage Types</a>
25 <li><a href="#linkage_private">'<tt>private</tt>' Linkage</a></li>
26 <li><a href="#linkage_linker_private">'<tt>linker_private</tt>' Linkage</a></li>
27 <li><a href="#linkage_internal">'<tt>internal</tt>' Linkage</a></li>
28 <li><a href="#linkage_available_externally">'<tt>available_externally</tt>' Linkage</a></li>
29 <li><a href="#linkage_linkonce">'<tt>linkonce</tt>' Linkage</a></li>
30 <li><a href="#linkage_common">'<tt>common</tt>' Linkage</a></li>
31 <li><a href="#linkage_weak">'<tt>weak</tt>' Linkage</a></li>
32 <li><a href="#linkage_appending">'<tt>appending</tt>' Linkage</a></li>
33 <li><a href="#linkage_externweak">'<tt>extern_weak</tt>' Linkage</a></li>
34 <li><a href="#linkage_linkonce">'<tt>linkonce_odr</tt>' Linkage</a></li>
35 <li><a href="#linkage_weak">'<tt>weak_odr</tt>' Linkage</a></li>
36 <li><a href="#linkage_external">'<tt>externally visible</tt>' Linkage</a></li>
37 <li><a href="#linkage_dllimport">'<tt>dllimport</tt>' Linkage</a></li>
38 <li><a href="#linkage_dllexport">'<tt>dllexport</tt>' Linkage</a></li>
41 <li><a href="#callingconv">Calling Conventions</a></li>
42 <li><a href="#namedtypes">Named Types</a></li>
43 <li><a href="#globalvars">Global Variables</a></li>
44 <li><a href="#functionstructure">Functions</a></li>
45 <li><a href="#aliasstructure">Aliases</a></li>
46 <li><a href="#paramattrs">Parameter Attributes</a></li>
47 <li><a href="#fnattrs">Function Attributes</a></li>
48 <li><a href="#gc">Garbage Collector Names</a></li>
49 <li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
50 <li><a href="#datalayout">Data Layout</a></li>
51 <li><a href="#pointeraliasing">Pointer Aliasing Rules</a></li>
54 <li><a href="#typesystem">Type System</a>
56 <li><a href="#t_classifications">Type Classifications</a></li>
57 <li><a href="#t_primitive">Primitive Types</a>
59 <li><a href="#t_floating">Floating Point Types</a></li>
60 <li><a href="#t_void">Void Type</a></li>
61 <li><a href="#t_label">Label Type</a></li>
62 <li><a href="#t_metadata">Metadata Type</a></li>
65 <li><a href="#t_derived">Derived Types</a>
67 <li><a href="#t_integer">Integer Type</a></li>
68 <li><a href="#t_array">Array Type</a></li>
69 <li><a href="#t_function">Function Type</a></li>
70 <li><a href="#t_pointer">Pointer Type</a></li>
71 <li><a href="#t_struct">Structure Type</a></li>
72 <li><a href="#t_pstruct">Packed Structure Type</a></li>
73 <li><a href="#t_vector">Vector Type</a></li>
74 <li><a href="#t_opaque">Opaque Type</a></li>
77 <li><a href="#t_uprefs">Type Up-references</a></li>
80 <li><a href="#constants">Constants</a>
82 <li><a href="#simpleconstants">Simple Constants</a></li>
83 <li><a href="#complexconstants">Complex Constants</a></li>
84 <li><a href="#globalconstants">Global Variable and Function Addresses</a></li>
85 <li><a href="#undefvalues">Undefined Values</a></li>
86 <li><a href="#constantexprs">Constant Expressions</a></li>
87 <li><a href="#metadata">Embedded Metadata</a></li>
90 <li><a href="#othervalues">Other Values</a>
92 <li><a href="#inlineasm">Inline Assembler Expressions</a></li>
95 <li><a href="#intrinsic_globals">Intrinsic Global Variables</a>
97 <li><a href="#intg_used">The '<tt>llvm.used</tt>' Global Variable</a></li>
98 <li><a href="#intg_compiler_used">The '<tt>llvm.compiler.used</tt>'
99 Global Variable</a></li>
100 <li><a href="#intg_global_ctors">The '<tt>llvm.global_ctors</tt>'
101 Global Variable</a></li>
102 <li><a href="#intg_global_dtors">The '<tt>llvm.global_dtors</tt>'
103 Global Variable</a></li>
106 <li><a href="#instref">Instruction Reference</a>
108 <li><a href="#terminators">Terminator Instructions</a>
110 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
111 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
112 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
113 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
114 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
115 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
118 <li><a href="#binaryops">Binary Operations</a>
120 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
121 <li><a href="#i_fadd">'<tt>fadd</tt>' Instruction</a></li>
122 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
123 <li><a href="#i_fsub">'<tt>fsub</tt>' Instruction</a></li>
124 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
125 <li><a href="#i_fmul">'<tt>fmul</tt>' Instruction</a></li>
126 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
127 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
128 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
129 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
130 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
131 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
134 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
136 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
137 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
138 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
139 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
140 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
141 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
144 <li><a href="#vectorops">Vector Operations</a>
146 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
147 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
148 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
151 <li><a href="#aggregateops">Aggregate Operations</a>
153 <li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li>
154 <li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li>
157 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
159 <li><a href="#i_malloc">'<tt>malloc</tt>' Instruction</a></li>
160 <li><a href="#i_free">'<tt>free</tt>' Instruction</a></li>
161 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
162 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
163 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
164 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
167 <li><a href="#convertops">Conversion Operations</a>
169 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
170 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
171 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
172 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
173 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
174 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
175 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
176 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
177 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
178 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
179 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
180 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
183 <li><a href="#otherops">Other Operations</a>
185 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
186 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
187 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
188 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
189 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
190 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
195 <li><a href="#intrinsics">Intrinsic Functions</a>
197 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
199 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
200 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
201 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
204 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
206 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
207 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
208 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
211 <li><a href="#int_codegen">Code Generator Intrinsics</a>
213 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
214 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
215 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
216 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
217 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
218 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
219 <li><a href="#int_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
222 <li><a href="#int_libc">Standard C Library Intrinsics</a>
224 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
225 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
226 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
227 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
228 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
229 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
230 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
231 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
234 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
236 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
237 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
238 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
239 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
242 <li><a href="#int_overflow">Arithmetic with Overflow Intrinsics</a>
244 <li><a href="#int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt> Intrinsics</a></li>
245 <li><a href="#int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt> Intrinsics</a></li>
246 <li><a href="#int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt> Intrinsics</a></li>
247 <li><a href="#int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt> Intrinsics</a></li>
248 <li><a href="#int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt> Intrinsics</a></li>
249 <li><a href="#int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt> Intrinsics</a></li>
252 <li><a href="#int_debugger">Debugger intrinsics</a></li>
253 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
254 <li><a href="#int_trampoline">Trampoline Intrinsic</a>
256 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
259 <li><a href="#int_atomics">Atomic intrinsics</a>
261 <li><a href="#int_memory_barrier"><tt>llvm.memory_barrier</tt></a></li>
262 <li><a href="#int_atomic_cmp_swap"><tt>llvm.atomic.cmp.swap</tt></a></li>
263 <li><a href="#int_atomic_swap"><tt>llvm.atomic.swap</tt></a></li>
264 <li><a href="#int_atomic_load_add"><tt>llvm.atomic.load.add</tt></a></li>
265 <li><a href="#int_atomic_load_sub"><tt>llvm.atomic.load.sub</tt></a></li>
266 <li><a href="#int_atomic_load_and"><tt>llvm.atomic.load.and</tt></a></li>
267 <li><a href="#int_atomic_load_nand"><tt>llvm.atomic.load.nand</tt></a></li>
268 <li><a href="#int_atomic_load_or"><tt>llvm.atomic.load.or</tt></a></li>
269 <li><a href="#int_atomic_load_xor"><tt>llvm.atomic.load.xor</tt></a></li>
270 <li><a href="#int_atomic_load_max"><tt>llvm.atomic.load.max</tt></a></li>
271 <li><a href="#int_atomic_load_min"><tt>llvm.atomic.load.min</tt></a></li>
272 <li><a href="#int_atomic_load_umax"><tt>llvm.atomic.load.umax</tt></a></li>
273 <li><a href="#int_atomic_load_umin"><tt>llvm.atomic.load.umin</tt></a></li>
276 <li><a href="#int_general">General intrinsics</a>
278 <li><a href="#int_var_annotation">
279 '<tt>llvm.var.annotation</tt>' Intrinsic</a></li>
280 <li><a href="#int_annotation">
281 '<tt>llvm.annotation.*</tt>' Intrinsic</a></li>
282 <li><a href="#int_trap">
283 '<tt>llvm.trap</tt>' Intrinsic</a></li>
284 <li><a href="#int_stackprotector">
285 '<tt>llvm.stackprotector</tt>' Intrinsic</a></li>
292 <div class="doc_author">
293 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
294 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
297 <!-- *********************************************************************** -->
298 <div class="doc_section"> <a name="abstract">Abstract </a></div>
299 <!-- *********************************************************************** -->
301 <div class="doc_text">
303 <p>This document is a reference manual for the LLVM assembly language. LLVM is
304 a Static Single Assignment (SSA) based representation that provides type
305 safety, low-level operations, flexibility, and the capability of representing
306 'all' high-level languages cleanly. It is the common code representation
307 used throughout all phases of the LLVM compilation strategy.</p>
311 <!-- *********************************************************************** -->
312 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
313 <!-- *********************************************************************** -->
315 <div class="doc_text">
317 <p>The LLVM code representation is designed to be used in three different forms:
318 as an in-memory compiler IR, as an on-disk bitcode representation (suitable
319 for fast loading by a Just-In-Time compiler), and as a human readable
320 assembly language representation. This allows LLVM to provide a powerful
321 intermediate representation for efficient compiler transformations and
322 analysis, while providing a natural means to debug and visualize the
323 transformations. The three different forms of LLVM are all equivalent. This
324 document describes the human readable representation and notation.</p>
326 <p>The LLVM representation aims to be light-weight and low-level while being
327 expressive, typed, and extensible at the same time. It aims to be a
328 "universal IR" of sorts, by being at a low enough level that high-level ideas
329 may be cleanly mapped to it (similar to how microprocessors are "universal
330 IR's", allowing many source languages to be mapped to them). By providing
331 type information, LLVM can be used as the target of optimizations: for
332 example, through pointer analysis, it can be proven that a C automatic
333 variable is never accessed outside of the current function... allowing it to
334 be promoted to a simple SSA value instead of a memory location.</p>
338 <!-- _______________________________________________________________________ -->
339 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
341 <div class="doc_text">
343 <p>It is important to note that this document describes 'well formed' LLVM
344 assembly language. There is a difference between what the parser accepts and
345 what is considered 'well formed'. For example, the following instruction is
346 syntactically okay, but not well formed:</p>
348 <div class="doc_code">
350 %x = <a href="#i_add">add</a> i32 1, %x
354 <p>...because the definition of <tt>%x</tt> does not dominate all of its
355 uses. The LLVM infrastructure provides a verification pass that may be used
356 to verify that an LLVM module is well formed. This pass is automatically run
357 by the parser after parsing input assembly and by the optimizer before it
358 outputs bitcode. The violations pointed out by the verifier pass indicate
359 bugs in transformation passes or input to the parser.</p>
363 <!-- Describe the typesetting conventions here. -->
365 <!-- *********************************************************************** -->
366 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
367 <!-- *********************************************************************** -->
369 <div class="doc_text">
371 <p>LLVM identifiers come in two basic types: global and local. Global
372 identifiers (functions, global variables) begin with the <tt>'@'</tt>
373 character. Local identifiers (register names, types) begin with
374 the <tt>'%'</tt> character. Additionally, there are three different formats
375 for identifiers, for different purposes:</p>
378 <li>Named values are represented as a string of characters with their prefix.
379 For example, <tt>%foo</tt>, <tt>@DivisionByZero</tt>,
380 <tt>%a.really.long.identifier</tt>. The actual regular expression used is
381 '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'. Identifiers which require
382 other characters in their names can be surrounded with quotes. Special
383 characters may be escaped using <tt>"\xx"</tt> where <tt>xx</tt> is the
384 ASCII code for the character in hexadecimal. In this way, any character
385 can be used in a name value, even quotes themselves.</li>
387 <li>Unnamed values are represented as an unsigned numeric value with their
388 prefix. For example, <tt>%12</tt>, <tt>@2</tt>, <tt>%44</tt>.</li>
390 <li>Constants, which are described in a <a href="#constants">section about
391 constants</a>, below.</li>
394 <p>LLVM requires that values start with a prefix for two reasons: Compilers
395 don't need to worry about name clashes with reserved words, and the set of
396 reserved words may be expanded in the future without penalty. Additionally,
397 unnamed identifiers allow a compiler to quickly come up with a temporary
398 variable without having to avoid symbol table conflicts.</p>
400 <p>Reserved words in LLVM are very similar to reserved words in other
401 languages. There are keywords for different opcodes
402 ('<tt><a href="#i_add">add</a></tt>',
403 '<tt><a href="#i_bitcast">bitcast</a></tt>',
404 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names
405 ('<tt><a href="#t_void">void</a></tt>',
406 '<tt><a href="#t_primitive">i32</a></tt>', etc...), and others. These
407 reserved words cannot conflict with variable names, because none of them
408 start with a prefix character (<tt>'%'</tt> or <tt>'@'</tt>).</p>
410 <p>Here is an example of LLVM code to multiply the integer variable
411 '<tt>%X</tt>' by 8:</p>
415 <div class="doc_code">
417 %result = <a href="#i_mul">mul</a> i32 %X, 8
421 <p>After strength reduction:</p>
423 <div class="doc_code">
425 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
429 <p>And the hard way:</p>
431 <div class="doc_code">
433 <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
434 <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
435 %result = <a href="#i_add">add</a> i32 %1, %1
439 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several important
440 lexical features of LLVM:</p>
443 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
446 <li>Unnamed temporaries are created when the result of a computation is not
447 assigned to a named value.</li>
449 <li>Unnamed temporaries are numbered sequentially</li>
452 <p>...and it also shows a convention that we follow in this document. When
453 demonstrating instructions, we will follow an instruction with a comment that
454 defines the type and name of value produced. Comments are shown in italic
459 <!-- *********************************************************************** -->
460 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
461 <!-- *********************************************************************** -->
463 <!-- ======================================================================= -->
464 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
467 <div class="doc_text">
469 <p>LLVM programs are composed of "Module"s, each of which is a translation unit
470 of the input programs. Each module consists of functions, global variables,
471 and symbol table entries. Modules may be combined together with the LLVM
472 linker, which merges function (and global variable) definitions, resolves
473 forward declarations, and merges symbol table entries. Here is an example of
474 the "hello world" module:</p>
476 <div class="doc_code">
477 <pre><i>; Declare the string constant as a global constant...</i>
478 <a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a
479 href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
481 <i>; External declaration of the puts function</i>
482 <a href="#functionstructure">declare</a> i32 @puts(i8 *) <i>; i32(i8 *)* </i>
484 <i>; Definition of main function</i>
485 define i32 @main() { <i>; i32()* </i>
486 <i>; Convert [13 x i8]* to i8 *...</i>
488 href="#i_getelementptr">getelementptr</a> [13 x i8]* @.LC0, i64 0, i64 0 <i>; i8 *</i>
490 <i>; Call puts function to write out the string to stdout...</i>
492 href="#i_call">call</a> i32 @puts(i8 * %cast210) <i>; i32</i>
494 href="#i_ret">ret</a> i32 0<br>}<br>
498 <p>This example is made up of a <a href="#globalvars">global variable</a> named
499 "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>" function, and
500 a <a href="#functionstructure">function definition</a> for
503 <p>In general, a module is made up of a list of global values, where both
504 functions and global variables are global values. Global values are
505 represented by a pointer to a memory location (in this case, a pointer to an
506 array of char, and a pointer to a function), and have one of the
507 following <a href="#linkage">linkage types</a>.</p>
511 <!-- ======================================================================= -->
512 <div class="doc_subsection">
513 <a name="linkage">Linkage Types</a>
516 <div class="doc_text">
518 <p>All Global Variables and Functions have one of the following types of
522 <dt><tt><b><a name="linkage_private">private</a></b></tt>: </dt>
523 <dd>Global values with private linkage are only directly accessible by objects
524 in the current module. In particular, linking code into a module with an
525 private global value may cause the private to be renamed as necessary to
526 avoid collisions. Because the symbol is private to the module, all
527 references can be updated. This doesn't show up in any symbol table in the
530 <dt><tt><b><a name="linkage_linker_private">linker_private</a></b></tt>: </dt>
531 <dd>Similar to private, but the symbol is passed through the assembler and
532 removed by the linker after evaluation. Note that (unlike private
533 symbols) linker_private symbols are subject to coalescing by the linker:
534 weak symbols get merged and redefinitions are rejected. However, unlike
535 normal strong symbols, they are removed by the linker from the final
536 linked image (executable or dynamic library).</dd>
538 <dt><tt><b><a name="linkage_internal">internal</a></b></tt>: </dt>
539 <dd>Similar to private, but the value shows as a local symbol
540 (<tt>STB_LOCAL</tt> in the case of ELF) in the object file. This
541 corresponds to the notion of the '<tt>static</tt>' keyword in C.</dd>
543 <dt><tt><b><a name="linkage_available_externally">available_externally</a></b></tt>: </dt>
544 <dd>Globals with "<tt>available_externally</tt>" linkage are never emitted
545 into the object file corresponding to the LLVM module. They exist to
546 allow inlining and other optimizations to take place given knowledge of
547 the definition of the global, which is known to be somewhere outside the
548 module. Globals with <tt>available_externally</tt> linkage are allowed to
549 be discarded at will, and are otherwise the same as <tt>linkonce_odr</tt>.
550 This linkage type is only allowed on definitions, not declarations.</dd>
552 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
553 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
554 the same name when linkage occurs. This is typically used to implement
555 inline functions, templates, or other code which must be generated in each
556 translation unit that uses it. Unreferenced <tt>linkonce</tt> globals are
557 allowed to be discarded.</dd>
559 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
560 <dd>"<tt>weak</tt>" linkage has the same merging semantics as
561 <tt>linkonce</tt> linkage, except that unreferenced globals with
562 <tt>weak</tt> linkage may not be discarded. This is used for globals that
563 are declared "weak" in C source code.</dd>
565 <dt><tt><b><a name="linkage_common">common</a></b></tt>: </dt>
566 <dd>"<tt>common</tt>" linkage is most similar to "<tt>weak</tt>" linkage, but
567 they are used for tentative definitions in C, such as "<tt>int X;</tt>" at
569 Symbols with "<tt>common</tt>" linkage are merged in the same way as
570 <tt>weak symbols</tt>, and they may not be deleted if unreferenced.
571 <tt>common</tt> symbols may not have an explicit section,
572 must have a zero initializer, and may not be marked '<a
573 href="#globalvars"><tt>constant</tt></a>'. Functions and aliases may not
574 have common linkage.</dd>
577 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
578 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
579 pointer to array type. When two global variables with appending linkage
580 are linked together, the two global arrays are appended together. This is
581 the LLVM, typesafe, equivalent of having the system linker append together
582 "sections" with identical names when .o files are linked.</dd>
584 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt>: </dt>
585 <dd>The semantics of this linkage follow the ELF object file model: the symbol
586 is weak until linked, if not linked, the symbol becomes null instead of
587 being an undefined reference.</dd>
589 <dt><tt><b><a name="linkage_linkonce">linkonce_odr</a></b></tt>: </dt>
590 <dt><tt><b><a name="linkage_weak">weak_odr</a></b></tt>: </dt>
591 <dd>Some languages allow differing globals to be merged, such as two functions
592 with different semantics. Other languages, such as <tt>C++</tt>, ensure
593 that only equivalent globals are ever merged (the "one definition rule" -
594 "ODR"). Such languages can use the <tt>linkonce_odr</tt>
595 and <tt>weak_odr</tt> linkage types to indicate that the global will only
596 be merged with equivalent globals. These linkage types are otherwise the
597 same as their non-<tt>odr</tt> versions.</dd>
599 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
600 <dd>If none of the above identifiers are used, the global is externally
601 visible, meaning that it participates in linkage and can be used to
602 resolve external symbol references.</dd>
605 <p>The next two types of linkage are targeted for Microsoft Windows platform
606 only. They are designed to support importing (exporting) symbols from (to)
607 DLLs (Dynamic Link Libraries).</p>
610 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt>: </dt>
611 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
612 or variable via a global pointer to a pointer that is set up by the DLL
613 exporting the symbol. On Microsoft Windows targets, the pointer name is
614 formed by combining <code>__imp_</code> and the function or variable
617 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt>: </dt>
618 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
619 pointer to a pointer in a DLL, so that it can be referenced with the
620 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
621 name is formed by combining <code>__imp_</code> and the function or
625 <p>For example, since the "<tt>.LC0</tt>" variable is defined to be internal, if
626 another module defined a "<tt>.LC0</tt>" variable and was linked with this
627 one, one of the two would be renamed, preventing a collision. Since
628 "<tt>main</tt>" and "<tt>puts</tt>" are external (i.e., lacking any linkage
629 declarations), they are accessible outside of the current module.</p>
631 <p>It is illegal for a function <i>declaration</i> to have any linkage type
632 other than "externally visible", <tt>dllimport</tt>
633 or <tt>extern_weak</tt>.</p>
635 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt>, <tt>weak</tt>
636 or <tt>weak_odr</tt> linkages.</p>
640 <!-- ======================================================================= -->
641 <div class="doc_subsection">
642 <a name="callingconv">Calling Conventions</a>
645 <div class="doc_text">
647 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
648 and <a href="#i_invoke">invokes</a> can all have an optional calling
649 convention specified for the call. The calling convention of any pair of
650 dynamic caller/callee must match, or the behavior of the program is
651 undefined. The following calling conventions are supported by LLVM, and more
652 may be added in the future:</p>
655 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
656 <dd>This calling convention (the default if no other calling convention is
657 specified) matches the target C calling conventions. This calling
658 convention supports varargs function calls and tolerates some mismatch in
659 the declared prototype and implemented declaration of the function (as
662 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
663 <dd>This calling convention attempts to make calls as fast as possible
664 (e.g. by passing things in registers). This calling convention allows the
665 target to use whatever tricks it wants to produce fast code for the
666 target, without having to conform to an externally specified ABI
667 (Application Binary Interface). Implementations of this convention should
668 allow arbitrary <a href="CodeGenerator.html#tailcallopt">tail call
669 optimization</a> to be supported. This calling convention does not
670 support varargs and requires the prototype of all callees to exactly match
671 the prototype of the function definition.</dd>
673 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
674 <dd>This calling convention attempts to make code in the caller as efficient
675 as possible under the assumption that the call is not commonly executed.
676 As such, these calls often preserve all registers so that the call does
677 not break any live ranges in the caller side. This calling convention
678 does not support varargs and requires the prototype of all callees to
679 exactly match the prototype of the function definition.</dd>
681 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
682 <dd>Any calling convention may be specified by number, allowing
683 target-specific calling conventions to be used. Target specific calling
684 conventions start at 64.</dd>
687 <p>More calling conventions can be added/defined on an as-needed basis, to
688 support Pascal conventions or any other well-known target-independent
693 <!-- ======================================================================= -->
694 <div class="doc_subsection">
695 <a name="visibility">Visibility Styles</a>
698 <div class="doc_text">
700 <p>All Global Variables and Functions have one of the following visibility
704 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
705 <dd>On targets that use the ELF object file format, default visibility means
706 that the declaration is visible to other modules and, in shared libraries,
707 means that the declared entity may be overridden. On Darwin, default
708 visibility means that the declaration is visible to other modules. Default
709 visibility corresponds to "external linkage" in the language.</dd>
711 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
712 <dd>Two declarations of an object with hidden visibility refer to the same
713 object if they are in the same shared object. Usually, hidden visibility
714 indicates that the symbol will not be placed into the dynamic symbol
715 table, so no other module (executable or shared library) can reference it
718 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
719 <dd>On ELF, protected visibility indicates that the symbol will be placed in
720 the dynamic symbol table, but that references within the defining module
721 will bind to the local symbol. That is, the symbol cannot be overridden by
727 <!-- ======================================================================= -->
728 <div class="doc_subsection">
729 <a name="namedtypes">Named Types</a>
732 <div class="doc_text">
734 <p>LLVM IR allows you to specify name aliases for certain types. This can make
735 it easier to read the IR and make the IR more condensed (particularly when
736 recursive types are involved). An example of a name specification is:</p>
738 <div class="doc_code">
740 %mytype = type { %mytype*, i32 }
744 <p>You may give a name to any <a href="#typesystem">type</a> except
745 "<a href="t_void">void</a>". Type name aliases may be used anywhere a type
746 is expected with the syntax "%mytype".</p>
748 <p>Note that type names are aliases for the structural type that they indicate,
749 and that you can therefore specify multiple names for the same type. This
750 often leads to confusing behavior when dumping out a .ll file. Since LLVM IR
751 uses structural typing, the name is not part of the type. When printing out
752 LLVM IR, the printer will pick <em>one name</em> to render all types of a
753 particular shape. This means that if you have code where two different
754 source types end up having the same LLVM type, that the dumper will sometimes
755 print the "wrong" or unexpected type. This is an important design point and
756 isn't going to change.</p>
760 <!-- ======================================================================= -->
761 <div class="doc_subsection">
762 <a name="globalvars">Global Variables</a>
765 <div class="doc_text">
767 <p>Global variables define regions of memory allocated at compilation time
768 instead of run-time. Global variables may optionally be initialized, may
769 have an explicit section to be placed in, and may have an optional explicit
770 alignment specified. A variable may be defined as "thread_local", which
771 means that it will not be shared by threads (each thread will have a
772 separated copy of the variable). A variable may be defined as a global
773 "constant," which indicates that the contents of the variable
774 will <b>never</b> be modified (enabling better optimization, allowing the
775 global data to be placed in the read-only section of an executable, etc).
776 Note that variables that need runtime initialization cannot be marked
777 "constant" as there is a store to the variable.</p>
779 <p>LLVM explicitly allows <em>declarations</em> of global variables to be marked
780 constant, even if the final definition of the global is not. This capability
781 can be used to enable slightly better optimization of the program, but
782 requires the language definition to guarantee that optimizations based on the
783 'constantness' are valid for the translation units that do not include the
786 <p>As SSA values, global variables define pointer values that are in scope
787 (i.e. they dominate) all basic blocks in the program. Global variables
788 always define a pointer to their "content" type because they describe a
789 region of memory, and all memory objects in LLVM are accessed through
792 <p>A global variable may be declared to reside in a target-specific numbered
793 address space. For targets that support them, address spaces may affect how
794 optimizations are performed and/or what target instructions are used to
795 access the variable. The default address space is zero. The address space
796 qualifier must precede any other attributes.</p>
798 <p>LLVM allows an explicit section to be specified for globals. If the target
799 supports it, it will emit globals to the section specified.</p>
801 <p>An explicit alignment may be specified for a global. If not present, or if
802 the alignment is set to zero, the alignment of the global is set by the
803 target to whatever it feels convenient. If an explicit alignment is
804 specified, the global is forced to have at least that much alignment. All
805 alignments must be a power of 2.</p>
807 <p>For example, the following defines a global in a numbered address space with
808 an initializer, section, and alignment:</p>
810 <div class="doc_code">
812 @G = addrspace(5) constant float 1.0, section "foo", align 4
819 <!-- ======================================================================= -->
820 <div class="doc_subsection">
821 <a name="functionstructure">Functions</a>
824 <div class="doc_text">
826 <p>LLVM function definitions consist of the "<tt>define</tt>" keyord, an
827 optional <a href="#linkage">linkage type</a>, an optional
828 <a href="#visibility">visibility style</a>, an optional
829 <a href="#callingconv">calling convention</a>, a return type, an optional
830 <a href="#paramattrs">parameter attribute</a> for the return type, a function
831 name, a (possibly empty) argument list (each with optional
832 <a href="#paramattrs">parameter attributes</a>), optional
833 <a href="#fnattrs">function attributes</a>, an optional section, an optional
834 alignment, an optional <a href="#gc">garbage collector name</a>, an opening
835 curly brace, a list of basic blocks, and a closing curly brace.</p>
837 <p>LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
838 optional <a href="#linkage">linkage type</a>, an optional
839 <a href="#visibility">visibility style</a>, an optional
840 <a href="#callingconv">calling convention</a>, a return type, an optional
841 <a href="#paramattrs">parameter attribute</a> for the return type, a function
842 name, a possibly empty list of arguments, an optional alignment, and an
843 optional <a href="#gc">garbage collector name</a>.</p>
845 <p>A function definition contains a list of basic blocks, forming the CFG
846 (Control Flow Graph) for the function. Each basic block may optionally start
847 with a label (giving the basic block a symbol table entry), contains a list
848 of instructions, and ends with a <a href="#terminators">terminator</a>
849 instruction (such as a branch or function return).</p>
851 <p>The first basic block in a function is special in two ways: it is immediately
852 executed on entrance to the function, and it is not allowed to have
853 predecessor basic blocks (i.e. there can not be any branches to the entry
854 block of a function). Because the block can have no predecessors, it also
855 cannot have any <a href="#i_phi">PHI nodes</a>.</p>
857 <p>LLVM allows an explicit section to be specified for functions. If the target
858 supports it, it will emit functions to the section specified.</p>
860 <p>An explicit alignment may be specified for a function. If not present, or if
861 the alignment is set to zero, the alignment of the function is set by the
862 target to whatever it feels convenient. If an explicit alignment is
863 specified, the function is forced to have at least that much alignment. All
864 alignments must be a power of 2.</p>
867 <div class="doc_code">
869 define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
870 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
871 <ResultType> @<FunctionName> ([argument list])
872 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
873 [<a href="#gc">gc</a>] { ... }
879 <!-- ======================================================================= -->
880 <div class="doc_subsection">
881 <a name="aliasstructure">Aliases</a>
884 <div class="doc_text">
886 <p>Aliases act as "second name" for the aliasee value (which can be either
887 function, global variable, another alias or bitcast of global value). Aliases
888 may have an optional <a href="#linkage">linkage type</a>, and an
889 optional <a href="#visibility">visibility style</a>.</p>
892 <div class="doc_code">
894 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
900 <!-- ======================================================================= -->
901 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
903 <div class="doc_text">
905 <p>The return type and each parameter of a function type may have a set of
906 <i>parameter attributes</i> associated with them. Parameter attributes are
907 used to communicate additional information about the result or parameters of
908 a function. Parameter attributes are considered to be part of the function,
909 not of the function type, so functions with different parameter attributes
910 can have the same function type.</p>
912 <p>Parameter attributes are simple keywords that follow the type specified. If
913 multiple parameter attributes are needed, they are space separated. For
916 <div class="doc_code">
918 declare i32 @printf(i8* noalias nocapture, ...)
919 declare i32 @atoi(i8 zeroext)
920 declare signext i8 @returns_signed_char()
924 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
925 <tt>readonly</tt>) come immediately after the argument list.</p>
927 <p>Currently, only the following parameter attributes are defined:</p>
930 <dt><tt>zeroext</tt></dt>
931 <dd>This indicates to the code generator that the parameter or return value
932 should be zero-extended to a 32-bit value by the caller (for a parameter)
933 or the callee (for a return value).</dd>
935 <dt><tt>signext</tt></dt>
936 <dd>This indicates to the code generator that the parameter or return value
937 should be sign-extended to a 32-bit value by the caller (for a parameter)
938 or the callee (for a return value).</dd>
940 <dt><tt>inreg</tt></dt>
941 <dd>This indicates that this parameter or return value should be treated in a
942 special target-dependent fashion during while emitting code for a function
943 call or return (usually, by putting it in a register as opposed to memory,
944 though some targets use it to distinguish between two different kinds of
945 registers). Use of this attribute is target-specific.</dd>
947 <dt><tt><a name="byval">byval</a></tt></dt>
948 <dd>This indicates that the pointer parameter should really be passed by value
949 to the function. The attribute implies that a hidden copy of the pointee
950 is made between the caller and the callee, so the callee is unable to
951 modify the value in the callee. This attribute is only valid on LLVM
952 pointer arguments. It is generally used to pass structs and arrays by
953 value, but is also valid on pointers to scalars. The copy is considered
954 to belong to the caller not the callee (for example,
955 <tt><a href="#readonly">readonly</a></tt> functions should not write to
956 <tt>byval</tt> parameters). This is not a valid attribute for return
957 values. The byval attribute also supports specifying an alignment with
958 the align attribute. This has a target-specific effect on the code
959 generator that usually indicates a desired alignment for the synthesized
962 <dt><tt>sret</tt></dt>
963 <dd>This indicates that the pointer parameter specifies the address of a
964 structure that is the return value of the function in the source program.
965 This pointer must be guaranteed by the caller to be valid: loads and
966 stores to the structure may be assumed by the callee to not to trap. This
967 may only be applied to the first parameter. This is not a valid attribute
968 for return values. </dd>
970 <dt><tt>noalias</tt></dt>
971 <dd>This indicates that the pointer does not alias any global or any other
972 parameter. The caller is responsible for ensuring that this is the
973 case. On a function return value, <tt>noalias</tt> additionally indicates
974 that the pointer does not alias any other pointers visible to the
975 caller. For further details, please see the discussion of the NoAlias
977 <a href="http://llvm.org/docs/AliasAnalysis.html#MustMayNo">alias
980 <dt><tt>nocapture</tt></dt>
981 <dd>This indicates that the callee does not make any copies of the pointer
982 that outlive the callee itself. This is not a valid attribute for return
985 <dt><tt>nest</tt></dt>
986 <dd>This indicates that the pointer parameter can be excised using the
987 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
988 attribute for return values.</dd>
993 <!-- ======================================================================= -->
994 <div class="doc_subsection">
995 <a name="gc">Garbage Collector Names</a>
998 <div class="doc_text">
1000 <p>Each function may specify a garbage collector name, which is simply a
1003 <div class="doc_code">
1005 define void @f() gc "name" { ...
1009 <p>The compiler declares the supported values of <i>name</i>. Specifying a
1010 collector which will cause the compiler to alter its output in order to
1011 support the named garbage collection algorithm.</p>
1015 <!-- ======================================================================= -->
1016 <div class="doc_subsection">
1017 <a name="fnattrs">Function Attributes</a>
1020 <div class="doc_text">
1022 <p>Function attributes are set to communicate additional information about a
1023 function. Function attributes are considered to be part of the function, not
1024 of the function type, so functions with different parameter attributes can
1025 have the same function type.</p>
1027 <p>Function attributes are simple keywords that follow the type specified. If
1028 multiple attributes are needed, they are space separated. For example:</p>
1030 <div class="doc_code">
1032 define void @f() noinline { ... }
1033 define void @f() alwaysinline { ... }
1034 define void @f() alwaysinline optsize { ... }
1035 define void @f() optsize
1040 <dt><tt>alwaysinline</tt></dt>
1041 <dd>This attribute indicates that the inliner should attempt to inline this
1042 function into callers whenever possible, ignoring any active inlining size
1043 threshold for this caller.</dd>
1045 <dt><tt>inlinehint</tt></dt>
1046 <dd>This attribute indicates that the source code contained a hint that inlining
1047 this function is desirable (such as the "inline" keyword in C/C++). It
1048 is just a hint; it imposes no requirements on the inliner.</dd>
1050 <dt><tt>noinline</tt></dt>
1051 <dd>This attribute indicates that the inliner should never inline this
1052 function in any situation. This attribute may not be used together with
1053 the <tt>alwaysinline</tt> attribute.</dd>
1055 <dt><tt>optsize</tt></dt>
1056 <dd>This attribute suggests that optimization passes and code generator passes
1057 make choices that keep the code size of this function low, and otherwise
1058 do optimizations specifically to reduce code size.</dd>
1060 <dt><tt>noreturn</tt></dt>
1061 <dd>This function attribute indicates that the function never returns
1062 normally. This produces undefined behavior at runtime if the function
1063 ever does dynamically return.</dd>
1065 <dt><tt>nounwind</tt></dt>
1066 <dd>This function attribute indicates that the function never returns with an
1067 unwind or exceptional control flow. If the function does unwind, its
1068 runtime behavior is undefined.</dd>
1070 <dt><tt>readnone</tt></dt>
1071 <dd>This attribute indicates that the function computes its result (or decides
1072 to unwind an exception) based strictly on its arguments, without
1073 dereferencing any pointer arguments or otherwise accessing any mutable
1074 state (e.g. memory, control registers, etc) visible to caller functions.
1075 It does not write through any pointer arguments
1076 (including <tt><a href="#byval">byval</a></tt> arguments) and never
1077 changes any state visible to callers. This means that it cannot unwind
1078 exceptions by calling the <tt>C++</tt> exception throwing methods, but
1079 could use the <tt>unwind</tt> instruction.</dd>
1081 <dt><tt><a name="readonly">readonly</a></tt></dt>
1082 <dd>This attribute indicates that the function does not write through any
1083 pointer arguments (including <tt><a href="#byval">byval</a></tt>
1084 arguments) or otherwise modify any state (e.g. memory, control registers,
1085 etc) visible to caller functions. It may dereference pointer arguments
1086 and read state that may be set in the caller. A readonly function always
1087 returns the same value (or unwinds an exception identically) when called
1088 with the same set of arguments and global state. It cannot unwind an
1089 exception by calling the <tt>C++</tt> exception throwing methods, but may
1090 use the <tt>unwind</tt> instruction.</dd>
1092 <dt><tt><a name="ssp">ssp</a></tt></dt>
1093 <dd>This attribute indicates that the function should emit a stack smashing
1094 protector. It is in the form of a "canary"—a random value placed on
1095 the stack before the local variables that's checked upon return from the
1096 function to see if it has been overwritten. A heuristic is used to
1097 determine if a function needs stack protectors or not.<br>
1099 If a function that has an <tt>ssp</tt> attribute is inlined into a
1100 function that doesn't have an <tt>ssp</tt> attribute, then the resulting
1101 function will have an <tt>ssp</tt> attribute.</dd>
1103 <dt><tt>sspreq</tt></dt>
1104 <dd>This attribute indicates that the function should <em>always</em> emit a
1105 stack smashing protector. This overrides
1106 the <tt><a href="#ssp">ssp</a></tt> function attribute.<br>
1108 If a function that has an <tt>sspreq</tt> attribute is inlined into a
1109 function that doesn't have an <tt>sspreq</tt> attribute or which has
1110 an <tt>ssp</tt> attribute, then the resulting function will have
1111 an <tt>sspreq</tt> attribute.</dd>
1113 <dt><tt>noredzone</tt></dt>
1114 <dd>This attribute indicates that the code generator should not use a red
1115 zone, even if the target-specific ABI normally permits it.</dd>
1117 <dt><tt>noimplicitfloat</tt></dt>
1118 <dd>This attributes disables implicit floating point instructions.</dd>
1120 <dt><tt>naked</tt></dt>
1121 <dd>This attribute disables prologue / epilogue emission for the function.
1122 This can have very system-specific consequences.</dd>
1127 <!-- ======================================================================= -->
1128 <div class="doc_subsection">
1129 <a name="moduleasm">Module-Level Inline Assembly</a>
1132 <div class="doc_text">
1134 <p>Modules may contain "module-level inline asm" blocks, which corresponds to
1135 the GCC "file scope inline asm" blocks. These blocks are internally
1136 concatenated by LLVM and treated as a single unit, but may be separated in
1137 the <tt>.ll</tt> file if desired. The syntax is very simple:</p>
1139 <div class="doc_code">
1141 module asm "inline asm code goes here"
1142 module asm "more can go here"
1146 <p>The strings can contain any character by escaping non-printable characters.
1147 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1150 <p>The inline asm code is simply printed to the machine code .s file when
1151 assembly code is generated.</p>
1155 <!-- ======================================================================= -->
1156 <div class="doc_subsection">
1157 <a name="datalayout">Data Layout</a>
1160 <div class="doc_text">
1162 <p>A module may specify a target specific data layout string that specifies how
1163 data is to be laid out in memory. The syntax for the data layout is
1166 <div class="doc_code">
1168 target datalayout = "<i>layout specification</i>"
1172 <p>The <i>layout specification</i> consists of a list of specifications
1173 separated by the minus sign character ('-'). Each specification starts with
1174 a letter and may include other information after the letter to define some
1175 aspect of the data layout. The specifications accepted are as follows:</p>
1179 <dd>Specifies that the target lays out data in big-endian form. That is, the
1180 bits with the most significance have the lowest address location.</dd>
1183 <dd>Specifies that the target lays out data in little-endian form. That is,
1184 the bits with the least significance have the lowest address
1187 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1188 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1189 <i>preferred</i> alignments. All sizes are in bits. Specifying
1190 the <i>pref</i> alignment is optional. If omitted, the
1191 preceding <tt>:</tt> should be omitted too.</dd>
1193 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1194 <dd>This specifies the alignment for an integer type of a given bit
1195 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1197 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1198 <dd>This specifies the alignment for a vector type of a given bit
1201 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1202 <dd>This specifies the alignment for a floating point type of a given bit
1203 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
1206 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1207 <dd>This specifies the alignment for an aggregate type of a given bit
1210 <dt><tt>s<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1211 <dd>This specifies the alignment for a stack object of a given bit
1215 <p>When constructing the data layout for a given target, LLVM starts with a
1216 default set of specifications which are then (possibly) overriden by the
1217 specifications in the <tt>datalayout</tt> keyword. The default specifications
1218 are given in this list:</p>
1221 <li><tt>E</tt> - big endian</li>
1222 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
1223 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1224 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1225 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1226 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1227 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1228 alignment of 64-bits</li>
1229 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1230 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1231 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1232 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1233 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1234 <li><tt>s0:64:64</tt> - stack objects are 64-bit aligned</li>
1237 <p>When LLVM is determining the alignment for a given type, it uses the
1238 following rules:</p>
1241 <li>If the type sought is an exact match for one of the specifications, that
1242 specification is used.</li>
1244 <li>If no match is found, and the type sought is an integer type, then the
1245 smallest integer type that is larger than the bitwidth of the sought type
1246 is used. If none of the specifications are larger than the bitwidth then
1247 the the largest integer type is used. For example, given the default
1248 specifications above, the i7 type will use the alignment of i8 (next
1249 largest) while both i65 and i256 will use the alignment of i64 (largest
1252 <li>If no match is found, and the type sought is a vector type, then the
1253 largest vector type that is smaller than the sought vector type will be
1254 used as a fall back. This happens because <128 x double> can be
1255 implemented in terms of 64 <2 x double>, for example.</li>
1260 <!-- ======================================================================= -->
1261 <div class="doc_subsection">
1262 <a name="pointeraliasing">Pointer Aliasing Rules</a>
1265 <div class="doc_text">
1267 <p>Any memory access must be done through a pointer value associated
1268 with an address range of the memory access, otherwise the behavior
1269 is undefined. Pointer values are associated with address ranges
1270 according to the following rules:</p>
1273 <li>A pointer value formed from a
1274 <tt><a href="#i_getelementptr">getelementptr</a></tt> instruction
1275 is associated with the addresses associated with the first operand
1276 of the <tt>getelementptr</tt>.</li>
1277 <li>An address of a global variable is associated with the address
1278 range of the variable's storage.</li>
1279 <li>The result value of an allocation instruction is associated with
1280 the address range of the allocated storage.</li>
1281 <li>A null pointer in the default address-space is associated with
1283 <li>A pointer value formed by an
1284 <tt><a href="#i_inttoptr">inttoptr</a></tt> is associated with all
1285 address ranges of all pointer values that contribute (directly or
1286 indirectly) to the computation of the pointer's value.</li>
1287 <li>The result value of a
1288 <tt><a href="#i_bitcast">bitcast</a></tt> is associated with all
1289 addresses associated with the operand of the <tt>bitcast</tt>.</li>
1290 <li>An integer constant other than zero or a pointer value returned
1291 from a function not defined within LLVM may be associated with address
1292 ranges allocated through mechanisms other than those provided by
1293 LLVM. Such ranges shall not overlap with any ranges of addresses
1294 allocated by mechanisms provided by LLVM.</li>
1297 <p>LLVM IR does not associate types with memory. The result type of a
1298 <tt><a href="#i_load">load</a></tt> merely indicates the size and
1299 alignment of the memory from which to load, as well as the
1300 interpretation of the value. The first operand of a
1301 <tt><a href="#i_store">store</a></tt> similarly only indicates the size
1302 and alignment of the store.</p>
1304 <p>Consequently, type-based alias analysis, aka TBAA, aka
1305 <tt>-fstrict-aliasing</tt>, is not applicable to general unadorned
1306 LLVM IR. <a href="#metadata">Metadata</a> may be used to encode
1307 additional information which specialized optimization passes may use
1308 to implement type-based alias analysis.</p>
1312 <!-- *********************************************************************** -->
1313 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
1314 <!-- *********************************************************************** -->
1316 <div class="doc_text">
1318 <p>The LLVM type system is one of the most important features of the
1319 intermediate representation. Being typed enables a number of optimizations
1320 to be performed on the intermediate representation directly, without having
1321 to do extra analyses on the side before the transformation. A strong type
1322 system makes it easier to read the generated code and enables novel analyses
1323 and transformations that are not feasible to perform on normal three address
1324 code representations.</p>
1328 <!-- ======================================================================= -->
1329 <div class="doc_subsection"> <a name="t_classifications">Type
1330 Classifications</a> </div>
1332 <div class="doc_text">
1334 <p>The types fall into a few useful classifications:</p>
1336 <table border="1" cellspacing="0" cellpadding="4">
1338 <tr><th>Classification</th><th>Types</th></tr>
1340 <td><a href="#t_integer">integer</a></td>
1341 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1344 <td><a href="#t_floating">floating point</a></td>
1345 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1348 <td><a name="t_firstclass">first class</a></td>
1349 <td><a href="#t_integer">integer</a>,
1350 <a href="#t_floating">floating point</a>,
1351 <a href="#t_pointer">pointer</a>,
1352 <a href="#t_vector">vector</a>,
1353 <a href="#t_struct">structure</a>,
1354 <a href="#t_array">array</a>,
1355 <a href="#t_label">label</a>,
1356 <a href="#t_metadata">metadata</a>.
1360 <td><a href="#t_primitive">primitive</a></td>
1361 <td><a href="#t_label">label</a>,
1362 <a href="#t_void">void</a>,
1363 <a href="#t_floating">floating point</a>,
1364 <a href="#t_metadata">metadata</a>.</td>
1367 <td><a href="#t_derived">derived</a></td>
1368 <td><a href="#t_integer">integer</a>,
1369 <a href="#t_array">array</a>,
1370 <a href="#t_function">function</a>,
1371 <a href="#t_pointer">pointer</a>,
1372 <a href="#t_struct">structure</a>,
1373 <a href="#t_pstruct">packed structure</a>,
1374 <a href="#t_vector">vector</a>,
1375 <a href="#t_opaque">opaque</a>.
1381 <p>The <a href="#t_firstclass">first class</a> types are perhaps the most
1382 important. Values of these types are the only ones which can be produced by
1383 instructions, passed as arguments, or used as operands to instructions.</p>
1387 <!-- ======================================================================= -->
1388 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
1390 <div class="doc_text">
1392 <p>The primitive types are the fundamental building blocks of the LLVM
1397 <!-- _______________________________________________________________________ -->
1398 <div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1400 <div class="doc_text">
1404 <tr><th>Type</th><th>Description</th></tr>
1405 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1406 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1407 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1408 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1409 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1415 <!-- _______________________________________________________________________ -->
1416 <div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1418 <div class="doc_text">
1421 <p>The void type does not represent any value and has no size.</p>
1430 <!-- _______________________________________________________________________ -->
1431 <div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1433 <div class="doc_text">
1436 <p>The label type represents code labels.</p>
1445 <!-- _______________________________________________________________________ -->
1446 <div class="doc_subsubsection"> <a name="t_metadata">Metadata Type</a> </div>
1448 <div class="doc_text">
1451 <p>The metadata type represents embedded metadata. The only derived type that
1452 may contain metadata is <tt>metadata*</tt> or a function type that returns or
1453 takes metadata typed parameters, but not pointer to metadata types.</p>
1463 <!-- ======================================================================= -->
1464 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1466 <div class="doc_text">
1468 <p>The real power in LLVM comes from the derived types in the system. This is
1469 what allows a programmer to represent arrays, functions, pointers, and other
1470 useful types. Note that these derived types may be recursive: For example,
1471 it is possible to have a two dimensional array.</p>
1475 <!-- _______________________________________________________________________ -->
1476 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1478 <div class="doc_text">
1481 <p>The integer type is a very simple derived type that simply specifies an
1482 arbitrary bit width for the integer type desired. Any bit width from 1 bit to
1483 2^23-1 (about 8 million) can be specified.</p>
1490 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1494 <table class="layout">
1496 <td class="left"><tt>i1</tt></td>
1497 <td class="left">a single-bit integer.</td>
1500 <td class="left"><tt>i32</tt></td>
1501 <td class="left">a 32-bit integer.</td>
1504 <td class="left"><tt>i1942652</tt></td>
1505 <td class="left">a really big integer of over 1 million bits.</td>
1509 <p>Note that the code generator does not yet support large integer types to be
1510 used as function return types. The specific limit on how large a return type
1511 the code generator can currently handle is target-dependent; currently it's
1512 often 64 bits for 32-bit targets and 128 bits for 64-bit targets.</p>
1516 <!-- _______________________________________________________________________ -->
1517 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1519 <div class="doc_text">
1522 <p>The array type is a very simple derived type that arranges elements
1523 sequentially in memory. The array type requires a size (number of elements)
1524 and an underlying data type.</p>
1528 [<# elements> x <elementtype>]
1531 <p>The number of elements is a constant integer value; <tt>elementtype</tt> may
1532 be any type with a size.</p>
1535 <table class="layout">
1537 <td class="left"><tt>[40 x i32]</tt></td>
1538 <td class="left">Array of 40 32-bit integer values.</td>
1541 <td class="left"><tt>[41 x i32]</tt></td>
1542 <td class="left">Array of 41 32-bit integer values.</td>
1545 <td class="left"><tt>[4 x i8]</tt></td>
1546 <td class="left">Array of 4 8-bit integer values.</td>
1549 <p>Here are some examples of multidimensional arrays:</p>
1550 <table class="layout">
1552 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1553 <td class="left">3x4 array of 32-bit integer values.</td>
1556 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1557 <td class="left">12x10 array of single precision floating point values.</td>
1560 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1561 <td class="left">2x3x4 array of 16-bit integer values.</td>
1565 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1566 length array. Normally, accesses past the end of an array are undefined in
1567 LLVM (e.g. it is illegal to access the 5th element of a 3 element array). As
1568 a special case, however, zero length arrays are recognized to be variable
1569 length. This allows implementation of 'pascal style arrays' with the LLVM
1570 type "<tt>{ i32, [0 x float]}</tt>", for example.</p>
1572 <p>Note that the code generator does not yet support large aggregate types to be
1573 used as function return types. The specific limit on how large an aggregate
1574 return type the code generator can currently handle is target-dependent, and
1575 also dependent on the aggregate element types.</p>
1579 <!-- _______________________________________________________________________ -->
1580 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1582 <div class="doc_text">
1585 <p>The function type can be thought of as a function signature. It consists of
1586 a return type and a list of formal parameter types. The return type of a
1587 function type is a scalar type, a void type, or a struct type. If the return
1588 type is a struct type then all struct elements must be of first class types,
1589 and the struct must have at least one element.</p>
1593 <returntype list> (<parameter list>)
1596 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1597 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1598 which indicates that the function takes a variable number of arguments.
1599 Variable argument functions can access their arguments with
1600 the <a href="#int_varargs">variable argument handling intrinsic</a>
1601 functions. '<tt><returntype list></tt>' is a comma-separated list of
1602 <a href="#t_firstclass">first class</a> type specifiers.</p>
1605 <table class="layout">
1607 <td class="left"><tt>i32 (i32)</tt></td>
1608 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1610 </tr><tr class="layout">
1611 <td class="left"><tt>float (i16 signext, i32 *) *
1613 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1614 an <tt>i16</tt> that should be sign extended and a
1615 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1618 </tr><tr class="layout">
1619 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1620 <td class="left">A vararg function that takes at least one
1621 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1622 which returns an integer. This is the signature for <tt>printf</tt> in
1625 </tr><tr class="layout">
1626 <td class="left"><tt>{i32, i32} (i32)</tt></td>
1627 <td class="left">A function taking an <tt>i32</tt>, returning two
1628 <tt>i32</tt> values as an aggregate of type <tt>{ i32, i32 }</tt>
1635 <!-- _______________________________________________________________________ -->
1636 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1638 <div class="doc_text">
1641 <p>The structure type is used to represent a collection of data members together
1642 in memory. The packing of the field types is defined to match the ABI of the
1643 underlying processor. The elements of a structure may be any type that has a
1646 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt> and
1647 '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field with
1648 the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
1652 { <type list> }
1656 <table class="layout">
1658 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1659 <td class="left">A triple of three <tt>i32</tt> values</td>
1660 </tr><tr class="layout">
1661 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1662 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1663 second element is a <a href="#t_pointer">pointer</a> to a
1664 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1665 an <tt>i32</tt>.</td>
1669 <p>Note that the code generator does not yet support large aggregate types to be
1670 used as function return types. The specific limit on how large an aggregate
1671 return type the code generator can currently handle is target-dependent, and
1672 also dependent on the aggregate element types.</p>
1676 <!-- _______________________________________________________________________ -->
1677 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1680 <div class="doc_text">
1683 <p>The packed structure type is used to represent a collection of data members
1684 together in memory. There is no padding between fields. Further, the
1685 alignment of a packed structure is 1 byte. The elements of a packed
1686 structure may be any type that has a size.</p>
1688 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt> and
1689 '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field with
1690 the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
1694 < { <type list> } >
1698 <table class="layout">
1700 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1701 <td class="left">A triple of three <tt>i32</tt> values</td>
1702 </tr><tr class="layout">
1704 <tt>< { float, i32 (i32)* } ></tt></td>
1705 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1706 second element is a <a href="#t_pointer">pointer</a> to a
1707 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1708 an <tt>i32</tt>.</td>
1714 <!-- _______________________________________________________________________ -->
1715 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1717 <div class="doc_text">
1720 <p>As in many languages, the pointer type represents a pointer or reference to
1721 another object, which must live in memory. Pointer types may have an optional
1722 address space attribute defining the target-specific numbered address space
1723 where the pointed-to object resides. The default address space is zero.</p>
1725 <p>Note that LLVM does not permit pointers to void (<tt>void*</tt>) nor does it
1726 permit pointers to labels (<tt>label*</tt>). Use <tt>i8*</tt> instead.</p>
1734 <table class="layout">
1736 <td class="left"><tt>[4 x i32]*</tt></td>
1737 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1738 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1741 <td class="left"><tt>i32 (i32 *) *</tt></td>
1742 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
1743 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1747 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1748 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1749 that resides in address space #5.</td>
1755 <!-- _______________________________________________________________________ -->
1756 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1758 <div class="doc_text">
1761 <p>A vector type is a simple derived type that represents a vector of elements.
1762 Vector types are used when multiple primitive data are operated in parallel
1763 using a single instruction (SIMD). A vector type requires a size (number of
1764 elements) and an underlying primitive data type. Vectors must have a power
1765 of two length (1, 2, 4, 8, 16 ...). Vector types are considered
1766 <a href="#t_firstclass">first class</a>.</p>
1770 < <# elements> x <elementtype> >
1773 <p>The number of elements is a constant integer value; elementtype may be any
1774 integer or floating point type.</p>
1777 <table class="layout">
1779 <td class="left"><tt><4 x i32></tt></td>
1780 <td class="left">Vector of 4 32-bit integer values.</td>
1783 <td class="left"><tt><8 x float></tt></td>
1784 <td class="left">Vector of 8 32-bit floating-point values.</td>
1787 <td class="left"><tt><2 x i64></tt></td>
1788 <td class="left">Vector of 2 64-bit integer values.</td>
1792 <p>Note that the code generator does not yet support large vector types to be
1793 used as function return types. The specific limit on how large a vector
1794 return type codegen can currently handle is target-dependent; currently it's
1795 often a few times longer than a hardware vector register.</p>
1799 <!-- _______________________________________________________________________ -->
1800 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1801 <div class="doc_text">
1804 <p>Opaque types are used to represent unknown types in the system. This
1805 corresponds (for example) to the C notion of a forward declared structure
1806 type. In LLVM, opaque types can eventually be resolved to any type (not just
1807 a structure type).</p>
1815 <table class="layout">
1817 <td class="left"><tt>opaque</tt></td>
1818 <td class="left">An opaque type.</td>
1824 <!-- ======================================================================= -->
1825 <div class="doc_subsection">
1826 <a name="t_uprefs">Type Up-references</a>
1829 <div class="doc_text">
1832 <p>An "up reference" allows you to refer to a lexically enclosing type without
1833 requiring it to have a name. For instance, a structure declaration may
1834 contain a pointer to any of the types it is lexically a member of. Example
1835 of up references (with their equivalent as named type declarations)
1839 { \2 * } %x = type { %x* }
1840 { \2 }* %y = type { %y }*
1844 <p>An up reference is needed by the asmprinter for printing out cyclic types
1845 when there is no declared name for a type in the cycle. Because the
1846 asmprinter does not want to print out an infinite type string, it needs a
1847 syntax to handle recursive types that have no names (all names are optional
1855 <p>The level is the count of the lexical type that is being referred to.</p>
1858 <table class="layout">
1860 <td class="left"><tt>\1*</tt></td>
1861 <td class="left">Self-referential pointer.</td>
1864 <td class="left"><tt>{ { \3*, i8 }, i32 }</tt></td>
1865 <td class="left">Recursive structure where the upref refers to the out-most
1872 <!-- *********************************************************************** -->
1873 <div class="doc_section"> <a name="constants">Constants</a> </div>
1874 <!-- *********************************************************************** -->
1876 <div class="doc_text">
1878 <p>LLVM has several different basic types of constants. This section describes
1879 them all and their syntax.</p>
1883 <!-- ======================================================================= -->
1884 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1886 <div class="doc_text">
1889 <dt><b>Boolean constants</b></dt>
1890 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1891 constants of the <tt><a href="#t_primitive">i1</a></tt> type.</dd>
1893 <dt><b>Integer constants</b></dt>
1894 <dd>Standard integers (such as '4') are constants of
1895 the <a href="#t_integer">integer</a> type. Negative numbers may be used
1896 with integer types.</dd>
1898 <dt><b>Floating point constants</b></dt>
1899 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1900 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1901 notation (see below). The assembler requires the exact decimal value of a
1902 floating-point constant. For example, the assembler accepts 1.25 but
1903 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
1904 constants must have a <a href="#t_floating">floating point</a> type. </dd>
1906 <dt><b>Null pointer constants</b></dt>
1907 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1908 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1911 <p>The one non-intuitive notation for constants is the hexadecimal form of
1912 floating point constants. For example, the form '<tt>double
1913 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than)
1914 '<tt>double 4.5e+15</tt>'. The only time hexadecimal floating point
1915 constants are required (and the only time that they are generated by the
1916 disassembler) is when a floating point constant must be emitted but it cannot
1917 be represented as a decimal floating point number in a reasonable number of
1918 digits. For example, NaN's, infinities, and other special values are
1919 represented in their IEEE hexadecimal format so that assembly and disassembly
1920 do not cause any bits to change in the constants.</p>
1922 <p>When using the hexadecimal form, constants of types float and double are
1923 represented using the 16-digit form shown above (which matches the IEEE754
1924 representation for double); float values must, however, be exactly
1925 representable as IEE754 single precision. Hexadecimal format is always used
1926 for long double, and there are three forms of long double. The 80-bit format
1927 used by x86 is represented as <tt>0xK</tt> followed by 20 hexadecimal digits.
1928 The 128-bit format used by PowerPC (two adjacent doubles) is represented
1929 by <tt>0xM</tt> followed by 32 hexadecimal digits. The IEEE 128-bit format
1930 is represented by <tt>0xL</tt> followed by 32 hexadecimal digits; no
1931 currently supported target uses this format. Long doubles will only work if
1932 they match the long double format on your target. All hexadecimal formats
1933 are big-endian (sign bit at the left).</p>
1937 <!-- ======================================================================= -->
1938 <div class="doc_subsection">
1939 <a name="aggregateconstants"></a> <!-- old anchor -->
1940 <a name="complexconstants">Complex Constants</a>
1943 <div class="doc_text">
1945 <p>Complex constants are a (potentially recursive) combination of simple
1946 constants and smaller complex constants.</p>
1949 <dt><b>Structure constants</b></dt>
1950 <dd>Structure constants are represented with notation similar to structure
1951 type definitions (a comma separated list of elements, surrounded by braces
1952 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
1953 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>".
1954 Structure constants must have <a href="#t_struct">structure type</a>, and
1955 the number and types of elements must match those specified by the
1958 <dt><b>Array constants</b></dt>
1959 <dd>Array constants are represented with notation similar to array type
1960 definitions (a comma separated list of elements, surrounded by square
1961 brackets (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74
1962 ]</tt>". Array constants must have <a href="#t_array">array type</a>, and
1963 the number and types of elements must match those specified by the
1966 <dt><b>Vector constants</b></dt>
1967 <dd>Vector constants are represented with notation similar to vector type
1968 definitions (a comma separated list of elements, surrounded by
1969 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32
1970 42, i32 11, i32 74, i32 100 ></tt>". Vector constants must
1971 have <a href="#t_vector">vector type</a>, and the number and types of
1972 elements must match those specified by the type.</dd>
1974 <dt><b>Zero initialization</b></dt>
1975 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1976 value to zero of <em>any</em> type, including scalar and aggregate types.
1977 This is often used to avoid having to print large zero initializers
1978 (e.g. for large arrays) and is always exactly equivalent to using explicit
1979 zero initializers.</dd>
1981 <dt><b>Metadata node</b></dt>
1982 <dd>A metadata node is a structure-like constant with
1983 <a href="#t_metadata">metadata type</a>. For example: "<tt>metadata !{
1984 i32 0, metadata !"test" }</tt>". Unlike other constants that are meant to
1985 be interpreted as part of the instruction stream, metadata is a place to
1986 attach additional information such as debug info.</dd>
1991 <!-- ======================================================================= -->
1992 <div class="doc_subsection">
1993 <a name="globalconstants">Global Variable and Function Addresses</a>
1996 <div class="doc_text">
1998 <p>The addresses of <a href="#globalvars">global variables</a>
1999 and <a href="#functionstructure">functions</a> are always implicitly valid
2000 (link-time) constants. These constants are explicitly referenced when
2001 the <a href="#identifiers">identifier for the global</a> is used and always
2002 have <a href="#t_pointer">pointer</a> type. For example, the following is a
2003 legal LLVM file:</p>
2005 <div class="doc_code">
2009 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2015 <!-- ======================================================================= -->
2016 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
2017 <div class="doc_text">
2019 <p>The string '<tt>undef</tt>' can be used anywhere a constant is expected, and
2020 indicates that the user of the value may recieve an unspecified bit-pattern.
2021 Undefined values may be of any type (other than label or void) and be used
2022 anywhere a constant is permitted.</p>
2024 <p>Undefined values are useful, because it indicates to the compiler that the
2025 program is well defined no matter what value is used. This gives the
2026 compiler more freedom to optimize. Here are some examples of (potentially
2027 surprising) transformations that are valid (in pseudo IR):</p>
2030 <div class="doc_code">
2042 <p>This is safe because all of the output bits are affected by the undef bits.
2043 Any output bit can have a zero or one depending on the input bits.</p>
2045 <div class="doc_code">
2058 <p>These logical operations have bits that are not always affected by the input.
2059 For example, if "%X" has a zero bit, then the output of the 'and' operation will
2060 always be a zero, no matter what the corresponding bit from the undef is. As
2061 such, it is unsafe to optimizer or assume that the result of the and is undef.
2062 However, it is safe to assume that all bits of the undef are 0, and optimize the
2063 and to 0. Likewise, it is safe to assume that all the bits of the undef operand
2064 to the or could be set, allowing the or to be folded to -1.</p>
2066 <div class="doc_code">
2068 %A = select undef, %X, %Y
2069 %B = select undef, 42, %Y
2070 %C = select %X, %Y, undef
2082 <p>This set of examples show that undefined select (and conditional branch)
2083 conditions can go "either way" but they have to come from one of the two
2084 operands. In the %A example, if %X and %Y were both known to have a clear low
2085 bit, then %A would have to have a cleared low bit. However, in the %C example,
2086 the optimizer is allowed to assume that the undef operand could be the same as
2087 %Y, allowing the whole select to be eliminated.</p>
2090 <div class="doc_code">
2092 %A = xor undef, undef
2111 <p>This example points out that two undef operands are not necessarily the same.
2112 This can be surprising to people (and also matches C semantics) where they
2113 assume that "X^X" is always zero, even if X is undef. This isn't true for a
2114 number of reasons, but the short answer is that an undef "variable" can
2115 arbitrarily change its value over its "live range". This is true because the
2116 "variable" doesn't actually <em>have a live range</em>. Instead, the value is
2117 logically read from arbitrary registers that happen to be around when needed,
2118 so the value is not neccesarily consistent over time. In fact, %A and %C need
2119 to have the same semantics or the core LLVM "replace all uses with" concept
2122 <div class="doc_code">
2132 <p>These examples show the crucial difference between an <em>undefined
2133 value</em> and <em>undefined behavior</em>. An undefined value (like undef) is
2134 allowed to have an arbitrary bit-pattern. This means that the %A operation
2135 can be constant folded to undef because the undef could be an SNaN, and fdiv is
2136 not (currently) defined on SNaN's. However, in the second example, we can make
2137 a more aggressive assumption: because the undef is allowed to be an arbitrary
2138 value, we are allowed to assume that it could be zero. Since a divide by zero
2139 has <em>undefined behavior</em>, we are allowed to assume that the operation
2140 does not execute at all. This allows us to delete the divide and all code after
2141 it: since the undefined operation "can't happen", the optimizer can assume that
2142 it occurs in dead code.
2145 <div class="doc_code">
2147 a: store undef -> %X
2148 b: store %X -> undef
2155 <p>These examples reiterate the fdiv example: a store "of" an undefined value
2156 can be assumed to not have any effect: we can assume that the value is
2157 overwritten with bits that happen to match what was already there. However, a
2158 store "to" an undefined location could clobber arbitrary memory, therefore, it
2159 has undefined behavior.</p>
2163 <!-- ======================================================================= -->
2164 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
2167 <div class="doc_text">
2169 <p>Constant expressions are used to allow expressions involving other constants
2170 to be used as constants. Constant expressions may be of
2171 any <a href="#t_firstclass">first class</a> type and may involve any LLVM
2172 operation that does not have side effects (e.g. load and call are not
2173 supported). The following is the syntax for constant expressions:</p>
2176 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
2177 <dd>Truncate a constant to another type. The bit size of CST must be larger
2178 than the bit size of TYPE. Both types must be integers.</dd>
2180 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
2181 <dd>Zero extend a constant to another type. The bit size of CST must be
2182 smaller or equal to the bit size of TYPE. Both types must be
2185 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
2186 <dd>Sign extend a constant to another type. The bit size of CST must be
2187 smaller or equal to the bit size of TYPE. Both types must be
2190 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
2191 <dd>Truncate a floating point constant to another floating point type. The
2192 size of CST must be larger than the size of TYPE. Both types must be
2193 floating point.</dd>
2195 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
2196 <dd>Floating point extend a constant to another type. The size of CST must be
2197 smaller or equal to the size of TYPE. Both types must be floating
2200 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
2201 <dd>Convert a floating point constant to the corresponding unsigned integer
2202 constant. TYPE must be a scalar or vector integer type. CST must be of
2203 scalar or vector floating point type. Both CST and TYPE must be scalars,
2204 or vectors of the same number of elements. If the value won't fit in the
2205 integer type, the results are undefined.</dd>
2207 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
2208 <dd>Convert a floating point constant to the corresponding signed integer
2209 constant. TYPE must be a scalar or vector integer type. CST must be of
2210 scalar or vector floating point type. Both CST and TYPE must be scalars,
2211 or vectors of the same number of elements. If the value won't fit in the
2212 integer type, the results are undefined.</dd>
2214 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
2215 <dd>Convert an unsigned integer constant to the corresponding floating point
2216 constant. TYPE must be a scalar or vector floating point type. CST must be
2217 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2218 vectors of the same number of elements. If the value won't fit in the
2219 floating point type, the results are undefined.</dd>
2221 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
2222 <dd>Convert a signed integer constant to the corresponding floating point
2223 constant. TYPE must be a scalar or vector floating point type. CST must be
2224 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2225 vectors of the same number of elements. If the value won't fit in the
2226 floating point type, the results are undefined.</dd>
2228 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
2229 <dd>Convert a pointer typed constant to the corresponding integer constant
2230 <tt>TYPE</tt> must be an integer type. <tt>CST</tt> must be of pointer
2231 type. The <tt>CST</tt> value is zero extended, truncated, or unchanged to
2232 make it fit in <tt>TYPE</tt>.</dd>
2234 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
2235 <dd>Convert a integer constant to a pointer constant. TYPE must be a pointer
2236 type. CST must be of integer type. The CST value is zero extended,
2237 truncated, or unchanged to make it fit in a pointer size. This one is
2238 <i>really</i> dangerous!</dd>
2240 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
2241 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
2242 are the same as those for the <a href="#i_bitcast">bitcast
2243 instruction</a>.</dd>
2245 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
2246 <dt><b><tt>getelementptr inbounds ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
2247 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
2248 constants. As with the <a href="#i_getelementptr">getelementptr</a>
2249 instruction, the index list may have zero or more indexes, which are
2250 required to make sense for the type of "CSTPTR".</dd>
2252 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
2253 <dd>Perform the <a href="#i_select">select operation</a> on constants.</dd>
2255 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
2256 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
2258 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
2259 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
2261 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
2262 <dd>Perform the <a href="#i_extractelement">extractelement operation</a> on
2265 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
2266 <dd>Perform the <a href="#i_insertelement">insertelement operation</a> on
2269 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
2270 <dd>Perform the <a href="#i_shufflevector">shufflevector operation</a> on
2273 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
2274 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2275 be any of the <a href="#binaryops">binary</a>
2276 or <a href="#bitwiseops">bitwise binary</a> operations. The constraints
2277 on operands are the same as those for the corresponding instruction
2278 (e.g. no bitwise operations on floating point values are allowed).</dd>
2283 <!-- ======================================================================= -->
2284 <div class="doc_subsection"><a name="metadata">Embedded Metadata</a>
2287 <div class="doc_text">
2289 <p>Embedded metadata provides a way to attach arbitrary data to the instruction
2290 stream without affecting the behaviour of the program. There are two
2291 metadata primitives, strings and nodes. All metadata has the
2292 <tt>metadata</tt> type and is identified in syntax by a preceding exclamation
2293 point ('<tt>!</tt>').</p>
2295 <p>A metadata string is a string surrounded by double quotes. It can contain
2296 any character by escaping non-printable characters with "\xx" where "xx" is
2297 the two digit hex code. For example: "<tt>!"test\00"</tt>".</p>
2299 <p>Metadata nodes are represented with notation similar to structure constants
2300 (a comma separated list of elements, surrounded by braces and preceeded by an
2301 exclamation point). For example: "<tt>!{ metadata !"test\00", i32
2304 <p>A metadata node will attempt to track changes to the values it holds. In the
2305 event that a value is deleted, it will be replaced with a typeless
2306 "<tt>null</tt>", such as "<tt>metadata !{null, i32 10}</tt>".</p>
2308 <p>Optimizations may rely on metadata to provide additional information about
2309 the program that isn't available in the instructions, or that isn't easily
2310 computable. Similarly, the code generator may expect a certain metadata
2311 format to be used to express debugging information.</p>
2315 <!-- *********************************************************************** -->
2316 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
2317 <!-- *********************************************************************** -->
2319 <!-- ======================================================================= -->
2320 <div class="doc_subsection">
2321 <a name="inlineasm">Inline Assembler Expressions</a>
2324 <div class="doc_text">
2326 <p>LLVM supports inline assembler expressions (as opposed
2327 to <a href="#moduleasm"> Module-Level Inline Assembly</a>) through the use of
2328 a special value. This value represents the inline assembler as a string
2329 (containing the instructions to emit), a list of operand constraints (stored
2330 as a string), and a flag that indicates whether or not the inline asm
2331 expression has side effects. An example inline assembler expression is:</p>
2333 <div class="doc_code">
2335 i32 (i32) asm "bswap $0", "=r,r"
2339 <p>Inline assembler expressions may <b>only</b> be used as the callee operand of
2340 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we
2343 <div class="doc_code">
2345 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
2349 <p>Inline asms with side effects not visible in the constraint list must be
2350 marked as having side effects. This is done through the use of the
2351 '<tt>sideeffect</tt>' keyword, like so:</p>
2353 <div class="doc_code">
2355 call void asm sideeffect "eieio", ""()
2359 <p>TODO: The format of the asm and constraints string still need to be
2360 documented here. Constraints on what can be done (e.g. duplication, moving,
2361 etc need to be documented). This is probably best done by reference to
2362 another document that covers inline asm from a holistic perspective.</p>
2367 <!-- *********************************************************************** -->
2368 <div class="doc_section">
2369 <a name="intrinsic_globals">Intrinsic Global Variables</a>
2371 <!-- *********************************************************************** -->
2373 <p>LLVM has a number of "magic" global variables that contain data that affect
2374 code generation or other IR semantics. These are documented here. All globals
2375 of this sort should have a section specified as "<tt>llvm.metadata</tt>". This
2376 section and all globals that start with "<tt>llvm.</tt>" are reserved for use
2379 <!-- ======================================================================= -->
2380 <div class="doc_subsection">
2381 <a name="intg_used">The '<tt>llvm.used</tt>' Global Variable</a>
2384 <div class="doc_text">
2386 <p>The <tt>@llvm.used</tt> global is an array with i8* element type which has <a
2387 href="#linkage_appending">appending linkage</a>. This array contains a list of
2388 pointers to global variables and functions which may optionally have a pointer
2389 cast formed of bitcast or getelementptr. For example, a legal use of it is:</p>
2395 @llvm.used = appending global [2 x i8*] [
2397 i8* bitcast (i32* @Y to i8*)
2398 ], section "llvm.metadata"
2401 <p>If a global variable appears in the <tt>@llvm.used</tt> list, then the
2402 compiler, assembler, and linker are required to treat the symbol as if there is
2403 a reference to the global that it cannot see. For example, if a variable has
2404 internal linkage and no references other than that from the <tt>@llvm.used</tt>
2405 list, it cannot be deleted. This is commonly used to represent references from
2406 inline asms and other things the compiler cannot "see", and corresponds to
2407 "attribute((used))" in GNU C.</p>
2409 <p>On some targets, the code generator must emit a directive to the assembler or
2410 object file to prevent the assembler and linker from molesting the symbol.</p>
2414 <!-- ======================================================================= -->
2415 <div class="doc_subsection">
2416 <a name="intg_compiler_used">The '<tt>llvm.compiler.used</tt>' Global Variable</a>
2419 <div class="doc_text">
2421 <p>The <tt>@llvm.compiler.used</tt> directive is the same as the
2422 <tt>@llvm.used</tt> directive, except that it only prevents the compiler from
2423 touching the symbol. On targets that support it, this allows an intelligent
2424 linker to optimize references to the symbol without being impeded as it would be
2425 by <tt>@llvm.used</tt>.</p>
2427 <p>This is a rare construct that should only be used in rare circumstances, and
2428 should not be exposed to source languages.</p>
2432 <!-- ======================================================================= -->
2433 <div class="doc_subsection">
2434 <a name="intg_global_ctors">The '<tt>llvm.global_ctors</tt>' Global Variable</a>
2437 <div class="doc_text">
2439 <p>TODO: Describe this.</p>
2443 <!-- ======================================================================= -->
2444 <div class="doc_subsection">
2445 <a name="intg_global_dtors">The '<tt>llvm.global_dtors</tt>' Global Variable</a>
2448 <div class="doc_text">
2450 <p>TODO: Describe this.</p>
2455 <!-- *********************************************************************** -->
2456 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
2457 <!-- *********************************************************************** -->
2459 <div class="doc_text">
2461 <p>The LLVM instruction set consists of several different classifications of
2462 instructions: <a href="#terminators">terminator
2463 instructions</a>, <a href="#binaryops">binary instructions</a>,
2464 <a href="#bitwiseops">bitwise binary instructions</a>,
2465 <a href="#memoryops">memory instructions</a>, and
2466 <a href="#otherops">other instructions</a>.</p>
2470 <!-- ======================================================================= -->
2471 <div class="doc_subsection"> <a name="terminators">Terminator
2472 Instructions</a> </div>
2474 <div class="doc_text">
2476 <p>As mentioned <a href="#functionstructure">previously</a>, every basic block
2477 in a program ends with a "Terminator" instruction, which indicates which
2478 block should be executed after the current block is finished. These
2479 terminator instructions typically yield a '<tt>void</tt>' value: they produce
2480 control flow, not values (the one exception being the
2481 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
2483 <p>There are six different terminator instructions: the
2484 '<a href="#i_ret"><tt>ret</tt></a>' instruction, the
2485 '<a href="#i_br"><tt>br</tt></a>' instruction, the
2486 '<a href="#i_switch"><tt>switch</tt></a>' instruction, the
2487 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the
2488 '<a href="#i_unwind"><tt>unwind</tt></a>' instruction, and the
2489 '<a href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
2493 <!-- _______________________________________________________________________ -->
2494 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
2495 Instruction</a> </div>
2497 <div class="doc_text">
2501 ret <type> <value> <i>; Return a value from a non-void function</i>
2502 ret void <i>; Return from void function</i>
2506 <p>The '<tt>ret</tt>' instruction is used to return control flow (and optionally
2507 a value) from a function back to the caller.</p>
2509 <p>There are two forms of the '<tt>ret</tt>' instruction: one that returns a
2510 value and then causes control flow, and one that just causes control flow to
2514 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument, the
2515 return value. The type of the return value must be a
2516 '<a href="#t_firstclass">first class</a>' type.</p>
2518 <p>A function is not <a href="#wellformed">well formed</a> if it it has a
2519 non-void return type and contains a '<tt>ret</tt>' instruction with no return
2520 value or a return value with a type that does not match its type, or if it
2521 has a void return type and contains a '<tt>ret</tt>' instruction with a
2525 <p>When the '<tt>ret</tt>' instruction is executed, control flow returns back to
2526 the calling function's context. If the caller is a
2527 "<a href="#i_call"><tt>call</tt></a>" instruction, execution continues at the
2528 instruction after the call. If the caller was an
2529 "<a href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues at
2530 the beginning of the "normal" destination block. If the instruction returns
2531 a value, that value shall set the call or invoke instruction's return
2536 ret i32 5 <i>; Return an integer value of 5</i>
2537 ret void <i>; Return from a void function</i>
2538 ret { i32, i8 } { i32 4, i8 2 } <i>; Return a struct of values 4 and 2</i>
2541 <p>Note that the code generator does not yet fully support large
2542 return values. The specific sizes that are currently supported are
2543 dependent on the target. For integers, on 32-bit targets the limit
2544 is often 64 bits, and on 64-bit targets the limit is often 128 bits.
2545 For aggregate types, the current limits are dependent on the element
2546 types; for example targets are often limited to 2 total integer
2547 elements and 2 total floating-point elements.</p>
2550 <!-- _______________________________________________________________________ -->
2551 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
2553 <div class="doc_text">
2557 br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
2561 <p>The '<tt>br</tt>' instruction is used to cause control flow to transfer to a
2562 different basic block in the current function. There are two forms of this
2563 instruction, corresponding to a conditional branch and an unconditional
2567 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a single
2568 '<tt>i1</tt>' value and two '<tt>label</tt>' values. The unconditional form
2569 of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>' value as a
2573 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
2574 argument is evaluated. If the value is <tt>true</tt>, control flows to the
2575 '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
2576 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
2581 %cond = <a href="#i_icmp">icmp</a> eq i32 %a, %b
2582 br i1 %cond, label %IfEqual, label %IfUnequal
2584 <a href="#i_ret">ret</a> i32 1
2586 <a href="#i_ret">ret</a> i32 0
2591 <!-- _______________________________________________________________________ -->
2592 <div class="doc_subsubsection">
2593 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
2596 <div class="doc_text">
2600 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
2604 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
2605 several different places. It is a generalization of the '<tt>br</tt>'
2606 instruction, allowing a branch to occur to one of many possible
2610 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
2611 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination,
2612 and an array of pairs of comparison value constants and '<tt>label</tt>'s.
2613 The table is not allowed to contain duplicate constant entries.</p>
2616 <p>The <tt>switch</tt> instruction specifies a table of values and
2617 destinations. When the '<tt>switch</tt>' instruction is executed, this table
2618 is searched for the given value. If the value is found, control flow is
2619 transfered to the corresponding destination; otherwise, control flow is
2620 transfered to the default destination.</p>
2622 <h5>Implementation:</h5>
2623 <p>Depending on properties of the target machine and the particular
2624 <tt>switch</tt> instruction, this instruction may be code generated in
2625 different ways. For example, it could be generated as a series of chained
2626 conditional branches or with a lookup table.</p>
2630 <i>; Emulate a conditional br instruction</i>
2631 %Val = <a href="#i_zext">zext</a> i1 %value to i32
2632 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
2634 <i>; Emulate an unconditional br instruction</i>
2635 switch i32 0, label %dest [ ]
2637 <i>; Implement a jump table:</i>
2638 switch i32 %val, label %otherwise [ i32 0, label %onzero
2640 i32 2, label %ontwo ]
2645 <!-- _______________________________________________________________________ -->
2646 <div class="doc_subsubsection">
2647 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
2650 <div class="doc_text">
2654 <result> = invoke [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>] <ptr to function ty> <function ptr val>(<function args>) [<a href="#fnattrs">fn attrs</a>]
2655 to label <normal label> unwind label <exception label>
2659 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
2660 function, with the possibility of control flow transfer to either the
2661 '<tt>normal</tt>' label or the '<tt>exception</tt>' label. If the callee
2662 function returns with the "<tt><a href="#i_ret">ret</a></tt>" instruction,
2663 control flow will return to the "normal" label. If the callee (or any
2664 indirect callees) returns with the "<a href="#i_unwind"><tt>unwind</tt></a>"
2665 instruction, control is interrupted and continued at the dynamically nearest
2666 "exception" label.</p>
2669 <p>This instruction requires several arguments:</p>
2672 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
2673 convention</a> the call should use. If none is specified, the call
2674 defaults to using C calling conventions.</li>
2676 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
2677 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
2678 '<tt>inreg</tt>' attributes are valid here.</li>
2680 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
2681 function value being invoked. In most cases, this is a direct function
2682 invocation, but indirect <tt>invoke</tt>s are just as possible, branching
2683 off an arbitrary pointer to function value.</li>
2685 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
2686 function to be invoked. </li>
2688 <li>'<tt>function args</tt>': argument list whose types match the function
2689 signature argument types. If the function signature indicates the
2690 function accepts a variable number of arguments, the extra arguments can
2693 <li>'<tt>normal label</tt>': the label reached when the called function
2694 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
2696 <li>'<tt>exception label</tt>': the label reached when a callee returns with
2697 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
2699 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
2700 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
2701 '<tt>readnone</tt>' attributes are valid here.</li>
2705 <p>This instruction is designed to operate as a standard
2706 '<tt><a href="#i_call">call</a></tt>' instruction in most regards. The
2707 primary difference is that it establishes an association with a label, which
2708 is used by the runtime library to unwind the stack.</p>
2710 <p>This instruction is used in languages with destructors to ensure that proper
2711 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
2712 exception. Additionally, this is important for implementation of
2713 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
2715 <p>For the purposes of the SSA form, the definition of the value returned by the
2716 '<tt>invoke</tt>' instruction is deemed to occur on the edge from the current
2717 block to the "normal" label. If the callee unwinds then no return value is
2722 %retval = invoke i32 @Test(i32 15) to label %Continue
2723 unwind label %TestCleanup <i>; {i32}:retval set</i>
2724 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
2725 unwind label %TestCleanup <i>; {i32}:retval set</i>
2730 <!-- _______________________________________________________________________ -->
2732 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
2733 Instruction</a> </div>
2735 <div class="doc_text">
2743 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
2744 at the first callee in the dynamic call stack which used
2745 an <a href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call.
2746 This is primarily used to implement exception handling.</p>
2749 <p>The '<tt>unwind</tt>' instruction causes execution of the current function to
2750 immediately halt. The dynamic call stack is then searched for the
2751 first <a href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack.
2752 Once found, execution continues at the "exceptional" destination block
2753 specified by the <tt>invoke</tt> instruction. If there is no <tt>invoke</tt>
2754 instruction in the dynamic call chain, undefined behavior results.</p>
2758 <!-- _______________________________________________________________________ -->
2760 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
2761 Instruction</a> </div>
2763 <div class="doc_text">
2771 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
2772 instruction is used to inform the optimizer that a particular portion of the
2773 code is not reachable. This can be used to indicate that the code after a
2774 no-return function cannot be reached, and other facts.</p>
2777 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
2781 <!-- ======================================================================= -->
2782 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
2784 <div class="doc_text">
2786 <p>Binary operators are used to do most of the computation in a program. They
2787 require two operands of the same type, execute an operation on them, and
2788 produce a single value. The operands might represent multiple data, as is
2789 the case with the <a href="#t_vector">vector</a> data type. The result value
2790 has the same type as its operands.</p>
2792 <p>There are several different binary operators:</p>
2796 <!-- _______________________________________________________________________ -->
2797 <div class="doc_subsubsection">
2798 <a name="i_add">'<tt>add</tt>' Instruction</a>
2801 <div class="doc_text">
2805 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2806 <result> = add nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2807 <result> = add nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2808 <result> = add nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2812 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
2815 <p>The two arguments to the '<tt>add</tt>' instruction must
2816 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
2817 integer values. Both arguments must have identical types.</p>
2820 <p>The value produced is the integer sum of the two operands.</p>
2822 <p>If the sum has unsigned overflow, the result returned is the mathematical
2823 result modulo 2<sup>n</sup>, where n is the bit width of the result.</p>
2825 <p>Because LLVM integers use a two's complement representation, this instruction
2826 is appropriate for both signed and unsigned integers.</p>
2828 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
2829 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
2830 <tt>nsw</tt> keywords are present, the result value of the <tt>add</tt>
2831 is undefined if unsigned and/or signed overflow, respectively, occurs.</p>
2835 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
2840 <!-- _______________________________________________________________________ -->
2841 <div class="doc_subsubsection">
2842 <a name="i_fadd">'<tt>fadd</tt>' Instruction</a>
2845 <div class="doc_text">
2849 <result> = fadd <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2853 <p>The '<tt>fadd</tt>' instruction returns the sum of its two operands.</p>
2856 <p>The two arguments to the '<tt>fadd</tt>' instruction must be
2857 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
2858 floating point values. Both arguments must have identical types.</p>
2861 <p>The value produced is the floating point sum of the two operands.</p>
2865 <result> = fadd float 4.0, %var <i>; yields {float}:result = 4.0 + %var</i>
2870 <!-- _______________________________________________________________________ -->
2871 <div class="doc_subsubsection">
2872 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
2875 <div class="doc_text">
2879 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2880 <result> = sub nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2881 <result> = sub nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2882 <result> = sub nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2886 <p>The '<tt>sub</tt>' instruction returns the difference of its two
2889 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
2890 '<tt>neg</tt>' instruction present in most other intermediate
2891 representations.</p>
2894 <p>The two arguments to the '<tt>sub</tt>' instruction must
2895 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
2896 integer values. Both arguments must have identical types.</p>
2899 <p>The value produced is the integer difference of the two operands.</p>
2901 <p>If the difference has unsigned overflow, the result returned is the
2902 mathematical result modulo 2<sup>n</sup>, where n is the bit width of the
2905 <p>Because LLVM integers use a two's complement representation, this instruction
2906 is appropriate for both signed and unsigned integers.</p>
2908 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
2909 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
2910 <tt>nsw</tt> keywords are present, the result value of the <tt>sub</tt>
2911 is undefined if unsigned and/or signed overflow, respectively, occurs.</p>
2915 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
2916 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
2921 <!-- _______________________________________________________________________ -->
2922 <div class="doc_subsubsection">
2923 <a name="i_fsub">'<tt>fsub</tt>' Instruction</a>
2926 <div class="doc_text">
2930 <result> = fsub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2934 <p>The '<tt>fsub</tt>' instruction returns the difference of its two
2937 <p>Note that the '<tt>fsub</tt>' instruction is used to represent the
2938 '<tt>fneg</tt>' instruction present in most other intermediate
2939 representations.</p>
2942 <p>The two arguments to the '<tt>fsub</tt>' instruction must be
2943 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
2944 floating point values. Both arguments must have identical types.</p>
2947 <p>The value produced is the floating point difference of the two operands.</p>
2951 <result> = fsub float 4.0, %var <i>; yields {float}:result = 4.0 - %var</i>
2952 <result> = fsub float -0.0, %val <i>; yields {float}:result = -%var</i>
2957 <!-- _______________________________________________________________________ -->
2958 <div class="doc_subsubsection">
2959 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
2962 <div class="doc_text">
2966 <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2967 <result> = mul nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2968 <result> = mul nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2969 <result> = mul nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2973 <p>The '<tt>mul</tt>' instruction returns the product of its two operands.</p>
2976 <p>The two arguments to the '<tt>mul</tt>' instruction must
2977 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
2978 integer values. Both arguments must have identical types.</p>
2981 <p>The value produced is the integer product of the two operands.</p>
2983 <p>If the result of the multiplication has unsigned overflow, the result
2984 returned is the mathematical result modulo 2<sup>n</sup>, where n is the bit
2985 width of the result.</p>
2987 <p>Because LLVM integers use a two's complement representation, and the result
2988 is the same width as the operands, this instruction returns the correct
2989 result for both signed and unsigned integers. If a full product
2990 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands should
2991 be sign-extended or zero-extended as appropriate to the width of the full
2994 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
2995 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
2996 <tt>nsw</tt> keywords are present, the result value of the <tt>mul</tt>
2997 is undefined if unsigned and/or signed overflow, respectively, occurs.</p>
3001 <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
3006 <!-- _______________________________________________________________________ -->
3007 <div class="doc_subsubsection">
3008 <a name="i_fmul">'<tt>fmul</tt>' Instruction</a>
3011 <div class="doc_text">
3015 <result> = fmul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3019 <p>The '<tt>fmul</tt>' instruction returns the product of its two operands.</p>
3022 <p>The two arguments to the '<tt>fmul</tt>' instruction must be
3023 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3024 floating point values. Both arguments must have identical types.</p>
3027 <p>The value produced is the floating point product of the two operands.</p>
3031 <result> = fmul float 4.0, %var <i>; yields {float}:result = 4.0 * %var</i>
3036 <!-- _______________________________________________________________________ -->
3037 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
3040 <div class="doc_text">
3044 <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3048 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two operands.</p>
3051 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
3052 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3053 values. Both arguments must have identical types.</p>
3056 <p>The value produced is the unsigned integer quotient of the two operands.</p>
3058 <p>Note that unsigned integer division and signed integer division are distinct
3059 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
3061 <p>Division by zero leads to undefined behavior.</p>
3065 <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
3070 <!-- _______________________________________________________________________ -->
3071 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
3074 <div class="doc_text">
3078 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3079 <result> = sdiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3083 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two operands.</p>
3086 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
3087 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3088 values. Both arguments must have identical types.</p>
3091 <p>The value produced is the signed integer quotient of the two operands rounded
3094 <p>Note that signed integer division and unsigned integer division are distinct
3095 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
3097 <p>Division by zero leads to undefined behavior. Overflow also leads to
3098 undefined behavior; this is a rare case, but can occur, for example, by doing
3099 a 32-bit division of -2147483648 by -1.</p>
3101 <p>If the <tt>exact</tt> keyword is present, the result value of the
3102 <tt>sdiv</tt> is undefined if the result would be rounded or if overflow
3107 <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
3112 <!-- _______________________________________________________________________ -->
3113 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
3114 Instruction</a> </div>
3116 <div class="doc_text">
3120 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3124 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two operands.</p>
3127 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
3128 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3129 floating point values. Both arguments must have identical types.</p>
3132 <p>The value produced is the floating point quotient of the two operands.</p>
3136 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
3141 <!-- _______________________________________________________________________ -->
3142 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
3145 <div class="doc_text">
3149 <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3153 <p>The '<tt>urem</tt>' instruction returns the remainder from the unsigned
3154 division of its two arguments.</p>
3157 <p>The two arguments to the '<tt>urem</tt>' instruction must be
3158 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3159 values. Both arguments must have identical types.</p>
3162 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
3163 This instruction always performs an unsigned division to get the
3166 <p>Note that unsigned integer remainder and signed integer remainder are
3167 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
3169 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
3173 <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
3178 <!-- _______________________________________________________________________ -->
3179 <div class="doc_subsubsection">
3180 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
3183 <div class="doc_text">
3187 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3191 <p>The '<tt>srem</tt>' instruction returns the remainder from the signed
3192 division of its two operands. This instruction can also take
3193 <a href="#t_vector">vector</a> versions of the values in which case the
3194 elements must be integers.</p>
3197 <p>The two arguments to the '<tt>srem</tt>' instruction must be
3198 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3199 values. Both arguments must have identical types.</p>
3202 <p>This instruction returns the <i>remainder</i> of a division (where the result
3203 has the same sign as the dividend, <tt>op1</tt>), not the <i>modulo</i>
3204 operator (where the result has the same sign as the divisor, <tt>op2</tt>) of
3205 a value. For more information about the difference,
3206 see <a href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
3207 Math Forum</a>. For a table of how this is implemented in various languages,
3208 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
3209 Wikipedia: modulo operation</a>.</p>
3211 <p>Note that signed integer remainder and unsigned integer remainder are
3212 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
3214 <p>Taking the remainder of a division by zero leads to undefined behavior.
3215 Overflow also leads to undefined behavior; this is a rare case, but can
3216 occur, for example, by taking the remainder of a 32-bit division of
3217 -2147483648 by -1. (The remainder doesn't actually overflow, but this rule
3218 lets srem be implemented using instructions that return both the result of
3219 the division and the remainder.)</p>
3223 <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
3228 <!-- _______________________________________________________________________ -->
3229 <div class="doc_subsubsection">
3230 <a name="i_frem">'<tt>frem</tt>' Instruction</a> </div>
3232 <div class="doc_text">
3236 <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3240 <p>The '<tt>frem</tt>' instruction returns the remainder from the division of
3241 its two operands.</p>
3244 <p>The two arguments to the '<tt>frem</tt>' instruction must be
3245 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3246 floating point values. Both arguments must have identical types.</p>
3249 <p>This instruction returns the <i>remainder</i> of a division. The remainder
3250 has the same sign as the dividend.</p>
3254 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
3259 <!-- ======================================================================= -->
3260 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
3261 Operations</a> </div>
3263 <div class="doc_text">
3265 <p>Bitwise binary operators are used to do various forms of bit-twiddling in a
3266 program. They are generally very efficient instructions and can commonly be
3267 strength reduced from other instructions. They require two operands of the
3268 same type, execute an operation on them, and produce a single value. The
3269 resulting value is the same type as its operands.</p>
3273 <!-- _______________________________________________________________________ -->
3274 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
3275 Instruction</a> </div>
3277 <div class="doc_text">
3281 <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3285 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to the left
3286 a specified number of bits.</p>
3289 <p>Both arguments to the '<tt>shl</tt>' instruction must be the
3290 same <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3291 integer type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3294 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod
3295 2<sup>n</sup>, where <tt>n</tt> is the width of the result. If <tt>op2</tt>
3296 is (statically or dynamically) negative or equal to or larger than the number
3297 of bits in <tt>op1</tt>, the result is undefined. If the arguments are
3298 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
3299 shift amount in <tt>op2</tt>.</p>
3303 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
3304 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
3305 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
3306 <result> = shl i32 1, 32 <i>; undefined</i>
3307 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
3312 <!-- _______________________________________________________________________ -->
3313 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
3314 Instruction</a> </div>
3316 <div class="doc_text">
3320 <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3324 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
3325 operand shifted to the right a specified number of bits with zero fill.</p>
3328 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
3329 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3330 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3333 <p>This instruction always performs a logical shift right operation. The most
3334 significant bits of the result will be filled with zero bits after the shift.
3335 If <tt>op2</tt> is (statically or dynamically) equal to or larger than the
3336 number of bits in <tt>op1</tt>, the result is undefined. If the arguments are
3337 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
3338 shift amount in <tt>op2</tt>.</p>
3342 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
3343 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
3344 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
3345 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
3346 <result> = lshr i32 1, 32 <i>; undefined</i>
3347 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
3352 <!-- _______________________________________________________________________ -->
3353 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
3354 Instruction</a> </div>
3355 <div class="doc_text">
3359 <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3363 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
3364 operand shifted to the right a specified number of bits with sign
3368 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
3369 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3370 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3373 <p>This instruction always performs an arithmetic shift right operation, The
3374 most significant bits of the result will be filled with the sign bit
3375 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
3376 larger than the number of bits in <tt>op1</tt>, the result is undefined. If
3377 the arguments are vectors, each vector element of <tt>op1</tt> is shifted by
3378 the corresponding shift amount in <tt>op2</tt>.</p>
3382 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
3383 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
3384 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
3385 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
3386 <result> = ashr i32 1, 32 <i>; undefined</i>
3387 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
3392 <!-- _______________________________________________________________________ -->
3393 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
3394 Instruction</a> </div>
3396 <div class="doc_text">
3400 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3404 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of its two
3408 <p>The two arguments to the '<tt>and</tt>' instruction must be
3409 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3410 values. Both arguments must have identical types.</p>
3413 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
3415 <table border="1" cellspacing="0" cellpadding="4">
3447 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
3448 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
3449 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
3452 <!-- _______________________________________________________________________ -->
3453 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
3455 <div class="doc_text">
3459 <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3463 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive or of its
3467 <p>The two arguments to the '<tt>or</tt>' instruction must be
3468 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3469 values. Both arguments must have identical types.</p>
3472 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
3474 <table border="1" cellspacing="0" cellpadding="4">
3506 <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
3507 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
3508 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
3513 <!-- _______________________________________________________________________ -->
3514 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
3515 Instruction</a> </div>
3517 <div class="doc_text">
3521 <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3525 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive or of
3526 its two operands. The <tt>xor</tt> is used to implement the "one's
3527 complement" operation, which is the "~" operator in C.</p>
3530 <p>The two arguments to the '<tt>xor</tt>' instruction must be
3531 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3532 values. Both arguments must have identical types.</p>
3535 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
3537 <table border="1" cellspacing="0" cellpadding="4">
3569 <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
3570 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
3571 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
3572 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
3577 <!-- ======================================================================= -->
3578 <div class="doc_subsection">
3579 <a name="vectorops">Vector Operations</a>
3582 <div class="doc_text">
3584 <p>LLVM supports several instructions to represent vector operations in a
3585 target-independent manner. These instructions cover the element-access and
3586 vector-specific operations needed to process vectors effectively. While LLVM
3587 does directly support these vector operations, many sophisticated algorithms
3588 will want to use target-specific intrinsics to take full advantage of a
3589 specific target.</p>
3593 <!-- _______________________________________________________________________ -->
3594 <div class="doc_subsubsection">
3595 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
3598 <div class="doc_text">
3602 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
3606 <p>The '<tt>extractelement</tt>' instruction extracts a single scalar element
3607 from a vector at a specified index.</p>
3611 <p>The first operand of an '<tt>extractelement</tt>' instruction is a value
3612 of <a href="#t_vector">vector</a> type. The second operand is an index
3613 indicating the position from which to extract the element. The index may be
3617 <p>The result is a scalar of the same type as the element type of
3618 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
3619 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
3620 results are undefined.</p>
3624 %result = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
3629 <!-- _______________________________________________________________________ -->
3630 <div class="doc_subsubsection">
3631 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
3634 <div class="doc_text">
3638 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
3642 <p>The '<tt>insertelement</tt>' instruction inserts a scalar element into a
3643 vector at a specified index.</p>
3646 <p>The first operand of an '<tt>insertelement</tt>' instruction is a value
3647 of <a href="#t_vector">vector</a> type. The second operand is a scalar value
3648 whose type must equal the element type of the first operand. The third
3649 operand is an index indicating the position at which to insert the value.
3650 The index may be a variable.</p>
3653 <p>The result is a vector of the same type as <tt>val</tt>. Its element values
3654 are those of <tt>val</tt> except at position <tt>idx</tt>, where it gets the
3655 value <tt>elt</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
3656 results are undefined.</p>
3660 %result = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
3665 <!-- _______________________________________________________________________ -->
3666 <div class="doc_subsubsection">
3667 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
3670 <div class="doc_text">
3674 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
3678 <p>The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
3679 from two input vectors, returning a vector with the same element type as the
3680 input and length that is the same as the shuffle mask.</p>
3683 <p>The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
3684 with types that match each other. The third argument is a shuffle mask whose
3685 element type is always 'i32'. The result of the instruction is a vector
3686 whose length is the same as the shuffle mask and whose element type is the
3687 same as the element type of the first two operands.</p>
3689 <p>The shuffle mask operand is required to be a constant vector with either
3690 constant integer or undef values.</p>
3693 <p>The elements of the two input vectors are numbered from left to right across
3694 both of the vectors. The shuffle mask operand specifies, for each element of
3695 the result vector, which element of the two input vectors the result element
3696 gets. The element selector may be undef (meaning "don't care") and the
3697 second operand may be undef if performing a shuffle from only one vector.</p>
3701 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
3702 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
3703 %result = shufflevector <4 x i32> %v1, <4 x i32> undef,
3704 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
3705 %result = shufflevector <8 x i32> %v1, <8 x i32> undef,
3706 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i>
3707 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
3708 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > <i>; yields <8 x i32></i>
3713 <!-- ======================================================================= -->
3714 <div class="doc_subsection">
3715 <a name="aggregateops">Aggregate Operations</a>
3718 <div class="doc_text">
3720 <p>LLVM supports several instructions for working with aggregate values.</p>
3724 <!-- _______________________________________________________________________ -->
3725 <div class="doc_subsubsection">
3726 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
3729 <div class="doc_text">
3733 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
3737 <p>The '<tt>extractvalue</tt>' instruction extracts the value of a struct field
3738 or array element from an aggregate value.</p>
3741 <p>The first operand of an '<tt>extractvalue</tt>' instruction is a value
3742 of <a href="#t_struct">struct</a> or <a href="#t_array">array</a> type. The
3743 operands are constant indices to specify which value to extract in a similar
3744 manner as indices in a
3745 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
3748 <p>The result is the value at the position in the aggregate specified by the
3753 %result = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
3758 <!-- _______________________________________________________________________ -->
3759 <div class="doc_subsubsection">
3760 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
3763 <div class="doc_text">
3767 <result> = insertvalue <aggregate type> <val>, <ty> <val>, <idx> <i>; yields <n x <ty>></i>
3771 <p>The '<tt>insertvalue</tt>' instruction inserts a value into a struct field or
3772 array element in an aggregate.</p>
3776 <p>The first operand of an '<tt>insertvalue</tt>' instruction is a value
3777 of <a href="#t_struct">struct</a> or <a href="#t_array">array</a> type. The
3778 second operand is a first-class value to insert. The following operands are
3779 constant indices indicating the position at which to insert the value in a
3780 similar manner as indices in a
3781 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction. The
3782 value to insert must have the same type as the value identified by the
3786 <p>The result is an aggregate of the same type as <tt>val</tt>. Its value is
3787 that of <tt>val</tt> except that the value at the position specified by the
3788 indices is that of <tt>elt</tt>.</p>
3792 %result = insertvalue {i32, float} %agg, i32 1, 0 <i>; yields {i32, float}</i>
3798 <!-- ======================================================================= -->
3799 <div class="doc_subsection">
3800 <a name="memoryops">Memory Access and Addressing Operations</a>
3803 <div class="doc_text">
3805 <p>A key design point of an SSA-based representation is how it represents
3806 memory. In LLVM, no memory locations are in SSA form, which makes things
3807 very simple. This section describes how to read, write, allocate, and free
3812 <!-- _______________________________________________________________________ -->
3813 <div class="doc_subsubsection">
3814 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
3817 <div class="doc_text">
3821 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3825 <p>The '<tt>malloc</tt>' instruction allocates memory from the system heap and
3826 returns a pointer to it. The object is always allocated in the generic
3827 address space (address space zero).</p>
3830 <p>The '<tt>malloc</tt>' instruction allocates
3831 <tt>sizeof(<type>)*NumElements</tt> bytes of memory from the operating
3832 system and returns a pointer of the appropriate type to the program. If
3833 "NumElements" is specified, it is the number of elements allocated, otherwise
3834 "NumElements" is defaulted to be one. If a constant alignment is specified,
3835 the value result of the allocation is guaranteed to be aligned to at least
3836 that boundary. If not specified, or if zero, the target can choose to align
3837 the allocation on any convenient boundary compatible with the type.</p>
3839 <p>'<tt>type</tt>' must be a sized type.</p>
3842 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and a
3843 pointer is returned. The result of a zero byte allocation is undefined. The
3844 result is null if there is insufficient memory available.</p>
3848 %array = malloc [4 x i8] <i>; yields {[%4 x i8]*}:array</i>
3850 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
3851 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
3852 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
3853 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
3854 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
3857 <p>Note that the code generator does not yet respect the alignment value.</p>
3861 <!-- _______________________________________________________________________ -->
3862 <div class="doc_subsubsection">
3863 <a name="i_free">'<tt>free</tt>' Instruction</a>
3866 <div class="doc_text">
3870 free <type> <value> <i>; yields {void}</i>
3874 <p>The '<tt>free</tt>' instruction returns memory back to the unused memory heap
3875 to be reallocated in the future.</p>
3878 <p>'<tt>value</tt>' shall be a pointer value that points to a value that was
3879 allocated with the '<tt><a href="#i_malloc">malloc</a></tt>' instruction.</p>
3882 <p>Access to the memory pointed to by the pointer is no longer defined after
3883 this instruction executes. If the pointer is null, the operation is a
3888 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
3889 free [4 x i8]* %array
3894 <!-- _______________________________________________________________________ -->
3895 <div class="doc_subsubsection">
3896 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
3899 <div class="doc_text">
3903 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3907 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
3908 currently executing function, to be automatically released when this function
3909 returns to its caller. The object is always allocated in the generic address
3910 space (address space zero).</p>
3913 <p>The '<tt>alloca</tt>' instruction
3914 allocates <tt>sizeof(<type>)*NumElements</tt> bytes of memory on the
3915 runtime stack, returning a pointer of the appropriate type to the program.
3916 If "NumElements" is specified, it is the number of elements allocated,
3917 otherwise "NumElements" is defaulted to be one. If a constant alignment is
3918 specified, the value result of the allocation is guaranteed to be aligned to
3919 at least that boundary. If not specified, or if zero, the target can choose
3920 to align the allocation on any convenient boundary compatible with the
3923 <p>'<tt>type</tt>' may be any sized type.</p>
3926 <p>Memory is allocated; a pointer is returned. The operation is undefined if
3927 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
3928 memory is automatically released when the function returns. The
3929 '<tt>alloca</tt>' instruction is commonly used to represent automatic
3930 variables that must have an address available. When the function returns
3931 (either with the <tt><a href="#i_ret">ret</a></tt>
3932 or <tt><a href="#i_unwind">unwind</a></tt> instructions), the memory is
3933 reclaimed. Allocating zero bytes is legal, but the result is undefined.</p>
3937 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
3938 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
3939 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
3940 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
3945 <!-- _______________________________________________________________________ -->
3946 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
3947 Instruction</a> </div>
3949 <div class="doc_text">
3953 <result> = load <ty>* <pointer>[, align <alignment>]
3954 <result> = volatile load <ty>* <pointer>[, align <alignment>]
3958 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
3961 <p>The argument to the '<tt>load</tt>' instruction specifies the memory address
3962 from which to load. The pointer must point to
3963 a <a href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
3964 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify the
3965 number or order of execution of this <tt>load</tt> with other
3966 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
3969 <p>The optional constant "align" argument specifies the alignment of the
3970 operation (that is, the alignment of the memory address). A value of 0 or an
3971 omitted "align" argument means that the operation has the preferential
3972 alignment for the target. It is the responsibility of the code emitter to
3973 ensure that the alignment information is correct. Overestimating the
3974 alignment results in an undefined behavior. Underestimating the alignment may
3975 produce less efficient code. An alignment of 1 is always safe.</p>
3978 <p>The location of memory pointed to is loaded. If the value being loaded is of
3979 scalar type then the number of bytes read does not exceed the minimum number
3980 of bytes needed to hold all bits of the type. For example, loading an
3981 <tt>i24</tt> reads at most three bytes. When loading a value of a type like
3982 <tt>i20</tt> with a size that is not an integral number of bytes, the result
3983 is undefined if the value was not originally written using a store of the
3988 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3989 <a href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
3990 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
3995 <!-- _______________________________________________________________________ -->
3996 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
3997 Instruction</a> </div>
3999 <div class="doc_text">
4003 store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
4004 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
4008 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
4011 <p>There are two arguments to the '<tt>store</tt>' instruction: a value to store
4012 and an address at which to store it. The type of the
4013 '<tt><pointer></tt>' operand must be a pointer to
4014 the <a href="#t_firstclass">first class</a> type of the
4015 '<tt><value></tt>' operand. If the <tt>store</tt> is marked
4016 as <tt>volatile</tt>, then the optimizer is not allowed to modify the number
4017 or order of execution of this <tt>store</tt> with other
4018 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
4021 <p>The optional constant "align" argument specifies the alignment of the
4022 operation (that is, the alignment of the memory address). A value of 0 or an
4023 omitted "align" argument means that the operation has the preferential
4024 alignment for the target. It is the responsibility of the code emitter to
4025 ensure that the alignment information is correct. Overestimating the
4026 alignment results in an undefined behavior. Underestimating the alignment may
4027 produce less efficient code. An alignment of 1 is always safe.</p>
4030 <p>The contents of memory are updated to contain '<tt><value></tt>' at the
4031 location specified by the '<tt><pointer></tt>' operand. If
4032 '<tt><value></tt>' is of scalar type then the number of bytes written
4033 does not exceed the minimum number of bytes needed to hold all bits of the
4034 type. For example, storing an <tt>i24</tt> writes at most three bytes. When
4035 writing a value of a type like <tt>i20</tt> with a size that is not an
4036 integral number of bytes, it is unspecified what happens to the extra bits
4037 that do not belong to the type, but they will typically be overwritten.</p>
4041 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
4042 store i32 3, i32* %ptr <i>; yields {void}</i>
4043 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
4048 <!-- _______________________________________________________________________ -->
4049 <div class="doc_subsubsection">
4050 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
4053 <div class="doc_text">
4057 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
4058 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
4062 <p>The '<tt>getelementptr</tt>' instruction is used to get the address of a
4063 subelement of an aggregate data structure. It performs address calculation
4064 only and does not access memory.</p>
4067 <p>The first argument is always a pointer, and forms the basis of the
4068 calculation. The remaining arguments are indices that indicate which of the
4069 elements of the aggregate object are indexed. The interpretation of each
4070 index is dependent on the type being indexed into. The first index always
4071 indexes the pointer value given as the first argument, the second index
4072 indexes a value of the type pointed to (not necessarily the value directly
4073 pointed to, since the first index can be non-zero), etc. The first type
4074 indexed into must be a pointer value, subsequent types can be arrays, vectors
4075 and structs. Note that subsequent types being indexed into can never be
4076 pointers, since that would require loading the pointer before continuing
4079 <p>The type of each index argument depends on the type it is indexing into.
4080 When indexing into a (optionally packed) structure, only <tt>i32</tt> integer
4081 <b>constants</b> are allowed. When indexing into an array, pointer or
4082 vector, integers of any width are allowed, and they are not required to be
4085 <p>For example, let's consider a C code fragment and how it gets compiled to
4088 <div class="doc_code">
4101 int *foo(struct ST *s) {
4102 return &s[1].Z.B[5][13];
4107 <p>The LLVM code generated by the GCC frontend is:</p>
4109 <div class="doc_code">
4111 %RT = <a href="#namedtypes">type</a> { i8 , [10 x [20 x i32]], i8 }
4112 %ST = <a href="#namedtypes">type</a> { i32, double, %RT }
4114 define i32* @foo(%ST* %s) {
4116 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
4123 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
4124 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
4125 }</tt>' type, a structure. The second index indexes into the third element
4126 of the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
4127 i8 }</tt>' type, another structure. The third index indexes into the second
4128 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
4129 array. The two dimensions of the array are subscripted into, yielding an
4130 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a
4131 pointer to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
4133 <p>Note that it is perfectly legal to index partially through a structure,
4134 returning a pointer to an inner element. Because of this, the LLVM code for
4135 the given testcase is equivalent to:</p>
4138 define i32* @foo(%ST* %s) {
4139 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
4140 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
4141 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
4142 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
4143 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
4148 <p>If the <tt>inbounds</tt> keyword is present, the result value of the
4149 <tt>getelementptr</tt> is undefined if the base pointer is not an
4150 <i>in bounds</i> address of an allocated object, or if any of the addresses
4151 that would be formed by successive addition of the offsets implied by the
4152 indices to the base address with infinitely precise arithmetic are not an
4153 <i>in bounds</i> address of that allocated object.
4154 The <i>in bounds</i> addresses for an allocated object are all the addresses
4155 that point into the object, plus the address one byte past the end.</p>
4157 <p>If the <tt>inbounds</tt> keyword is not present, the offsets are added to
4158 the base address with silently-wrapping two's complement arithmetic, and
4159 the result value of the <tt>getelementptr</tt> may be outside the object
4160 pointed to by the base pointer. The result value may not necessarily be
4161 used to access memory though, even if it happens to point into allocated
4162 storage. See the <a href="#pointeraliasing">Pointer Aliasing Rules</a>
4163 section for more information.</p>
4165 <p>The getelementptr instruction is often confusing. For some more insight into
4166 how it works, see <a href="GetElementPtr.html">the getelementptr FAQ</a>.</p>
4170 <i>; yields [12 x i8]*:aptr</i>
4171 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
4172 <i>; yields i8*:vptr</i>
4173 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
4174 <i>; yields i8*:eptr</i>
4175 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
4176 <i>; yields i32*:iptr</i>
4177 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
4182 <!-- ======================================================================= -->
4183 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
4186 <div class="doc_text">
4188 <p>The instructions in this category are the conversion instructions (casting)
4189 which all take a single operand and a type. They perform various bit
4190 conversions on the operand.</p>
4194 <!-- _______________________________________________________________________ -->
4195 <div class="doc_subsubsection">
4196 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
4198 <div class="doc_text">
4202 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
4206 <p>The '<tt>trunc</tt>' instruction truncates its operand to the
4207 type <tt>ty2</tt>.</p>
4210 <p>The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
4211 be an <a href="#t_integer">integer</a> type, and a type that specifies the
4212 size and type of the result, which must be
4213 an <a href="#t_integer">integer</a> type. The bit size of <tt>value</tt> must
4214 be larger than the bit size of <tt>ty2</tt>. Equal sized types are not
4218 <p>The '<tt>trunc</tt>' instruction truncates the high order bits
4219 in <tt>value</tt> and converts the remaining bits to <tt>ty2</tt>. Since the
4220 source size must be larger than the destination size, <tt>trunc</tt> cannot
4221 be a <i>no-op cast</i>. It will always truncate bits.</p>
4225 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
4226 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
4227 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
4232 <!-- _______________________________________________________________________ -->
4233 <div class="doc_subsubsection">
4234 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
4236 <div class="doc_text">
4240 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
4244 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
4249 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
4250 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
4251 also be of <a href="#t_integer">integer</a> type. The bit size of the
4252 <tt>value</tt> must be smaller than the bit size of the destination type,
4256 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
4257 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
4259 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
4263 %X = zext i32 257 to i64 <i>; yields i64:257</i>
4264 %Y = zext i1 true to i32 <i>; yields i32:1</i>
4269 <!-- _______________________________________________________________________ -->
4270 <div class="doc_subsubsection">
4271 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
4273 <div class="doc_text">
4277 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
4281 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
4284 <p>The '<tt>sext</tt>' instruction takes a value to cast, which must be of
4285 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
4286 also be of <a href="#t_integer">integer</a> type. The bit size of the
4287 <tt>value</tt> must be smaller than the bit size of the destination type,
4291 <p>The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
4292 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size
4293 of the type <tt>ty2</tt>.</p>
4295 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
4299 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
4300 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
4305 <!-- _______________________________________________________________________ -->
4306 <div class="doc_subsubsection">
4307 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
4310 <div class="doc_text">
4314 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
4318 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
4322 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
4323 point</a> value to cast and a <a href="#t_floating">floating point</a> type
4324 to cast it to. The size of <tt>value</tt> must be larger than the size of
4325 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
4326 <i>no-op cast</i>.</p>
4329 <p>The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
4330 <a href="#t_floating">floating point</a> type to a smaller
4331 <a href="#t_floating">floating point</a> type. If the value cannot fit
4332 within the destination type, <tt>ty2</tt>, then the results are
4337 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
4338 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
4343 <!-- _______________________________________________________________________ -->
4344 <div class="doc_subsubsection">
4345 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
4347 <div class="doc_text">
4351 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
4355 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
4356 floating point value.</p>
4359 <p>The '<tt>fpext</tt>' instruction takes a
4360 <a href="#t_floating">floating point</a> <tt>value</tt> to cast, and
4361 a <a href="#t_floating">floating point</a> type to cast it to. The source
4362 type must be smaller than the destination type.</p>
4365 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
4366 <a href="#t_floating">floating point</a> type to a larger
4367 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
4368 used to make a <i>no-op cast</i> because it always changes bits. Use
4369 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
4373 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
4374 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
4379 <!-- _______________________________________________________________________ -->
4380 <div class="doc_subsubsection">
4381 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
4383 <div class="doc_text">
4387 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
4391 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
4392 unsigned integer equivalent of type <tt>ty2</tt>.</p>
4395 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
4396 scalar or vector <a href="#t_floating">floating point</a> value, and a type
4397 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
4398 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
4399 vector integer type with the same number of elements as <tt>ty</tt></p>
4402 <p>The '<tt>fptoui</tt>' instruction converts its
4403 <a href="#t_floating">floating point</a> operand into the nearest (rounding
4404 towards zero) unsigned integer value. If the value cannot fit
4405 in <tt>ty2</tt>, the results are undefined.</p>
4409 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
4410 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
4411 %X = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
4416 <!-- _______________________________________________________________________ -->
4417 <div class="doc_subsubsection">
4418 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
4420 <div class="doc_text">
4424 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
4428 <p>The '<tt>fptosi</tt>' instruction converts
4429 <a href="#t_floating">floating point</a> <tt>value</tt> to
4430 type <tt>ty2</tt>.</p>
4433 <p>The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
4434 scalar or vector <a href="#t_floating">floating point</a> value, and a type
4435 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
4436 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
4437 vector integer type with the same number of elements as <tt>ty</tt></p>
4440 <p>The '<tt>fptosi</tt>' instruction converts its
4441 <a href="#t_floating">floating point</a> operand into the nearest (rounding
4442 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
4443 the results are undefined.</p>
4447 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
4448 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
4449 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
4454 <!-- _______________________________________________________________________ -->
4455 <div class="doc_subsubsection">
4456 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
4458 <div class="doc_text">
4462 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
4466 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
4467 integer and converts that value to the <tt>ty2</tt> type.</p>
4470 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
4471 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
4472 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
4473 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
4474 floating point type with the same number of elements as <tt>ty</tt></p>
4477 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
4478 integer quantity and converts it to the corresponding floating point
4479 value. If the value cannot fit in the floating point value, the results are
4484 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
4485 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
4490 <!-- _______________________________________________________________________ -->
4491 <div class="doc_subsubsection">
4492 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
4494 <div class="doc_text">
4498 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
4502 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed integer
4503 and converts that value to the <tt>ty2</tt> type.</p>
4506 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
4507 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
4508 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
4509 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
4510 floating point type with the same number of elements as <tt>ty</tt></p>
4513 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed integer
4514 quantity and converts it to the corresponding floating point value. If the
4515 value cannot fit in the floating point value, the results are undefined.</p>
4519 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
4520 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
4525 <!-- _______________________________________________________________________ -->
4526 <div class="doc_subsubsection">
4527 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
4529 <div class="doc_text">
4533 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
4537 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
4538 the integer type <tt>ty2</tt>.</p>
4541 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
4542 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
4543 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.</p>
4546 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
4547 <tt>ty2</tt> by interpreting the pointer value as an integer and either
4548 truncating or zero extending that value to the size of the integer type. If
4549 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
4550 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
4551 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
4556 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
4557 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
4562 <!-- _______________________________________________________________________ -->
4563 <div class="doc_subsubsection">
4564 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
4566 <div class="doc_text">
4570 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
4574 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to a
4575 pointer type, <tt>ty2</tt>.</p>
4578 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
4579 value to cast, and a type to cast it to, which must be a
4580 <a href="#t_pointer">pointer</a> type.</p>
4583 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
4584 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
4585 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
4586 size of a pointer then a truncation is done. If <tt>value</tt> is smaller
4587 than the size of a pointer then a zero extension is done. If they are the
4588 same size, nothing is done (<i>no-op cast</i>).</p>
4592 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
4593 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
4594 %Y = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
4599 <!-- _______________________________________________________________________ -->
4600 <div class="doc_subsubsection">
4601 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
4603 <div class="doc_text">
4607 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
4611 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
4612 <tt>ty2</tt> without changing any bits.</p>
4615 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be a
4616 non-aggregate first class value, and a type to cast it to, which must also be
4617 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes
4618 of <tt>value</tt> and the destination type, <tt>ty2</tt>, must be
4619 identical. If the source type is a pointer, the destination type must also be
4620 a pointer. This instruction supports bitwise conversion of vectors to
4621 integers and to vectors of other types (as long as they have the same
4625 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
4626 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
4627 this conversion. The conversion is done as if the <tt>value</tt> had been
4628 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only
4629 be converted to other pointer types with this instruction. To convert
4630 pointers to other types, use the <a href="#i_inttoptr">inttoptr</a> or
4631 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
4635 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
4636 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
4637 %Z = bitcast <2 x int> %V to i64; <i>; yields i64: %V</i>
4642 <!-- ======================================================================= -->
4643 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
4645 <div class="doc_text">
4647 <p>The instructions in this category are the "miscellaneous" instructions, which
4648 defy better classification.</p>
4652 <!-- _______________________________________________________________________ -->
4653 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
4656 <div class="doc_text">
4660 <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4664 <p>The '<tt>icmp</tt>' instruction returns a boolean value or a vector of
4665 boolean values based on comparison of its two integer, integer vector, or
4666 pointer operands.</p>
4669 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
4670 the condition code indicating the kind of comparison to perform. It is not a
4671 value, just a keyword. The possible condition code are:</p>
4674 <li><tt>eq</tt>: equal</li>
4675 <li><tt>ne</tt>: not equal </li>
4676 <li><tt>ugt</tt>: unsigned greater than</li>
4677 <li><tt>uge</tt>: unsigned greater or equal</li>
4678 <li><tt>ult</tt>: unsigned less than</li>
4679 <li><tt>ule</tt>: unsigned less or equal</li>
4680 <li><tt>sgt</tt>: signed greater than</li>
4681 <li><tt>sge</tt>: signed greater or equal</li>
4682 <li><tt>slt</tt>: signed less than</li>
4683 <li><tt>sle</tt>: signed less or equal</li>
4686 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
4687 <a href="#t_pointer">pointer</a> or integer <a href="#t_vector">vector</a>
4688 typed. They must also be identical types.</p>
4691 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to the
4692 condition code given as <tt>cond</tt>. The comparison performed always yields
4693 either an <a href="#t_primitive"><tt>i1</tt></a> or vector of <tt>i1</tt>
4694 result, as follows:</p>
4697 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
4698 <tt>false</tt> otherwise. No sign interpretation is necessary or
4701 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
4702 <tt>false</tt> otherwise. No sign interpretation is necessary or
4705 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
4706 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4708 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
4709 <tt>true</tt> if <tt>op1</tt> is greater than or equal
4710 to <tt>op2</tt>.</li>
4712 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
4713 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
4715 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
4716 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4718 <li><tt>sgt</tt>: interprets the operands as signed values and yields
4719 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4721 <li><tt>sge</tt>: interprets the operands as signed values and yields
4722 <tt>true</tt> if <tt>op1</tt> is greater than or equal
4723 to <tt>op2</tt>.</li>
4725 <li><tt>slt</tt>: interprets the operands as signed values and yields
4726 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
4728 <li><tt>sle</tt>: interprets the operands as signed values and yields
4729 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4732 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
4733 values are compared as if they were integers.</p>
4735 <p>If the operands are integer vectors, then they are compared element by
4736 element. The result is an <tt>i1</tt> vector with the same number of elements
4737 as the values being compared. Otherwise, the result is an <tt>i1</tt>.</p>
4741 <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
4742 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
4743 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
4744 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
4745 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
4746 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
4749 <p>Note that the code generator does not yet support vector types with
4750 the <tt>icmp</tt> instruction.</p>
4754 <!-- _______________________________________________________________________ -->
4755 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
4758 <div class="doc_text">
4762 <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4766 <p>The '<tt>fcmp</tt>' instruction returns a boolean value or vector of boolean
4767 values based on comparison of its operands.</p>
4769 <p>If the operands are floating point scalars, then the result type is a boolean
4770 (<a href="#t_primitive"><tt>i1</tt></a>).</p>
4772 <p>If the operands are floating point vectors, then the result type is a vector
4773 of boolean with the same number of elements as the operands being
4777 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
4778 the condition code indicating the kind of comparison to perform. It is not a
4779 value, just a keyword. The possible condition code are:</p>
4782 <li><tt>false</tt>: no comparison, always returns false</li>
4783 <li><tt>oeq</tt>: ordered and equal</li>
4784 <li><tt>ogt</tt>: ordered and greater than </li>
4785 <li><tt>oge</tt>: ordered and greater than or equal</li>
4786 <li><tt>olt</tt>: ordered and less than </li>
4787 <li><tt>ole</tt>: ordered and less than or equal</li>
4788 <li><tt>one</tt>: ordered and not equal</li>
4789 <li><tt>ord</tt>: ordered (no nans)</li>
4790 <li><tt>ueq</tt>: unordered or equal</li>
4791 <li><tt>ugt</tt>: unordered or greater than </li>
4792 <li><tt>uge</tt>: unordered or greater than or equal</li>
4793 <li><tt>ult</tt>: unordered or less than </li>
4794 <li><tt>ule</tt>: unordered or less than or equal</li>
4795 <li><tt>une</tt>: unordered or not equal</li>
4796 <li><tt>uno</tt>: unordered (either nans)</li>
4797 <li><tt>true</tt>: no comparison, always returns true</li>
4800 <p><i>Ordered</i> means that neither operand is a QNAN while
4801 <i>unordered</i> means that either operand may be a QNAN.</p>
4803 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be either
4804 a <a href="#t_floating">floating point</a> type or
4805 a <a href="#t_vector">vector</a> of floating point type. They must have
4806 identical types.</p>
4809 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4810 according to the condition code given as <tt>cond</tt>. If the operands are
4811 vectors, then the vectors are compared element by element. Each comparison
4812 performed always yields an <a href="#t_primitive">i1</a> result, as
4816 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
4818 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4819 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
4821 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4822 <tt>op1</tt> is greather than <tt>op2</tt>.</li>
4824 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4825 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4827 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4828 <tt>op1</tt> is less than <tt>op2</tt>.</li>
4830 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4831 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4833 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4834 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
4836 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
4838 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
4839 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
4841 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
4842 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4844 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
4845 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4847 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
4848 <tt>op1</tt> is less than <tt>op2</tt>.</li>
4850 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
4851 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4853 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
4854 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
4856 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
4858 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
4863 <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
4864 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
4865 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
4866 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
4869 <p>Note that the code generator does not yet support vector types with
4870 the <tt>fcmp</tt> instruction.</p>
4874 <!-- _______________________________________________________________________ -->
4875 <div class="doc_subsubsection">
4876 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
4879 <div class="doc_text">
4883 <result> = phi <ty> [ <val0>, <label0>], ...
4887 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in the
4888 SSA graph representing the function.</p>
4891 <p>The type of the incoming values is specified with the first type field. After
4892 this, the '<tt>phi</tt>' instruction takes a list of pairs as arguments, with
4893 one pair for each predecessor basic block of the current block. Only values
4894 of <a href="#t_firstclass">first class</a> type may be used as the value
4895 arguments to the PHI node. Only labels may be used as the label
4898 <p>There must be no non-phi instructions between the start of a basic block and
4899 the PHI instructions: i.e. PHI instructions must be first in a basic
4902 <p>For the purposes of the SSA form, the use of each incoming value is deemed to
4903 occur on the edge from the corresponding predecessor block to the current
4904 block (but after any definition of an '<tt>invoke</tt>' instruction's return
4905 value on the same edge).</p>
4908 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
4909 specified by the pair corresponding to the predecessor basic block that
4910 executed just prior to the current block.</p>
4914 Loop: ; Infinite loop that counts from 0 on up...
4915 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
4916 %nextindvar = add i32 %indvar, 1
4922 <!-- _______________________________________________________________________ -->
4923 <div class="doc_subsubsection">
4924 <a name="i_select">'<tt>select</tt>' Instruction</a>
4927 <div class="doc_text">
4931 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
4933 <i>selty</i> is either i1 or {<N x i1>}
4937 <p>The '<tt>select</tt>' instruction is used to choose one value based on a
4938 condition, without branching.</p>
4942 <p>The '<tt>select</tt>' instruction requires an 'i1' value or a vector of 'i1'
4943 values indicating the condition, and two values of the
4944 same <a href="#t_firstclass">first class</a> type. If the val1/val2 are
4945 vectors and the condition is a scalar, then entire vectors are selected, not
4946 individual elements.</p>
4949 <p>If the condition is an i1 and it evaluates to 1, the instruction returns the
4950 first value argument; otherwise, it returns the second value argument.</p>
4952 <p>If the condition is a vector of i1, then the value arguments must be vectors
4953 of the same size, and the selection is done element by element.</p>
4957 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
4960 <p>Note that the code generator does not yet support conditions
4961 with vector type.</p>
4965 <!-- _______________________________________________________________________ -->
4966 <div class="doc_subsubsection">
4967 <a name="i_call">'<tt>call</tt>' Instruction</a>
4970 <div class="doc_text">
4974 <result> = [tail] call [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>] <ty> [<fnty>*] <fnptrval>(<function args>) [<a href="#fnattrs">fn attrs</a>]
4978 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
4981 <p>This instruction requires several arguments:</p>
4984 <li>The optional "tail" marker indicates whether the callee function accesses
4985 any allocas or varargs in the caller. If the "tail" marker is present,
4986 the function call is eligible for tail call optimization. Note that calls
4987 may be marked "tail" even if they do not occur before
4988 a <a href="#i_ret"><tt>ret</tt></a> instruction.</li>
4990 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
4991 convention</a> the call should use. If none is specified, the call
4992 defaults to using C calling conventions.</li>
4994 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
4995 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
4996 '<tt>inreg</tt>' attributes are valid here.</li>
4998 <li>'<tt>ty</tt>': the type of the call instruction itself which is also the
4999 type of the return value. Functions that return no value are marked
5000 <tt><a href="#t_void">void</a></tt>.</li>
5002 <li>'<tt>fnty</tt>': shall be the signature of the pointer to function value
5003 being invoked. The argument types must match the types implied by this
5004 signature. This type can be omitted if the function is not varargs and if
5005 the function type does not return a pointer to a function.</li>
5007 <li>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
5008 be invoked. In most cases, this is a direct function invocation, but
5009 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
5010 to function value.</li>
5012 <li>'<tt>function args</tt>': argument list whose types match the function
5013 signature argument types. All arguments must be of
5014 <a href="#t_firstclass">first class</a> type. If the function signature
5015 indicates the function accepts a variable number of arguments, the extra
5016 arguments can be specified.</li>
5018 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
5019 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
5020 '<tt>readnone</tt>' attributes are valid here.</li>
5024 <p>The '<tt>call</tt>' instruction is used to cause control flow to transfer to
5025 a specified function, with its incoming arguments bound to the specified
5026 values. Upon a '<tt><a href="#i_ret">ret</a></tt>' instruction in the called
5027 function, control flow continues with the instruction after the function
5028 call, and the return value of the function is bound to the result
5033 %retval = call i32 @test(i32 %argc)
5034 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42) <i>; yields i32</i>
5035 %X = tail call i32 @foo() <i>; yields i32</i>
5036 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
5037 call void %foo(i8 97 signext)
5039 %struct.A = type { i32, i8 }
5040 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
5041 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
5042 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
5043 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
5044 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
5049 <!-- _______________________________________________________________________ -->
5050 <div class="doc_subsubsection">
5051 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
5054 <div class="doc_text">
5058 <resultval> = va_arg <va_list*> <arglist>, <argty>
5062 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
5063 the "variable argument" area of a function call. It is used to implement the
5064 <tt>va_arg</tt> macro in C.</p>
5067 <p>This instruction takes a <tt>va_list*</tt> value and the type of the
5068 argument. It returns a value of the specified argument type and increments
5069 the <tt>va_list</tt> to point to the next argument. The actual type
5070 of <tt>va_list</tt> is target specific.</p>
5073 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified type
5074 from the specified <tt>va_list</tt> and causes the <tt>va_list</tt> to point
5075 to the next argument. For more information, see the variable argument
5076 handling <a href="#int_varargs">Intrinsic Functions</a>.</p>
5078 <p>It is legal for this instruction to be called in a function which does not
5079 take a variable number of arguments, for example, the <tt>vfprintf</tt>
5082 <p><tt>va_arg</tt> is an LLVM instruction instead of
5083 an <a href="#intrinsics">intrinsic function</a> because it takes a type as an
5087 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
5089 <p>Note that the code generator does not yet fully support va_arg on many
5090 targets. Also, it does not currently support va_arg with aggregate types on
5095 <!-- *********************************************************************** -->
5096 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
5097 <!-- *********************************************************************** -->
5099 <div class="doc_text">
5101 <p>LLVM supports the notion of an "intrinsic function". These functions have
5102 well known names and semantics and are required to follow certain
5103 restrictions. Overall, these intrinsics represent an extension mechanism for
5104 the LLVM language that does not require changing all of the transformations
5105 in LLVM when adding to the language (or the bitcode reader/writer, the
5106 parser, etc...).</p>
5108 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
5109 prefix is reserved in LLVM for intrinsic names; thus, function names may not
5110 begin with this prefix. Intrinsic functions must always be external
5111 functions: you cannot define the body of intrinsic functions. Intrinsic
5112 functions may only be used in call or invoke instructions: it is illegal to
5113 take the address of an intrinsic function. Additionally, because intrinsic
5114 functions are part of the LLVM language, it is required if any are added that
5115 they be documented here.</p>
5117 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents a
5118 family of functions that perform the same operation but on different data
5119 types. Because LLVM can represent over 8 million different integer types,
5120 overloading is used commonly to allow an intrinsic function to operate on any
5121 integer type. One or more of the argument types or the result type can be
5122 overloaded to accept any integer type. Argument types may also be defined as
5123 exactly matching a previous argument's type or the result type. This allows
5124 an intrinsic function which accepts multiple arguments, but needs all of them
5125 to be of the same type, to only be overloaded with respect to a single
5126 argument or the result.</p>
5128 <p>Overloaded intrinsics will have the names of its overloaded argument types
5129 encoded into its function name, each preceded by a period. Only those types
5130 which are overloaded result in a name suffix. Arguments whose type is matched
5131 against another type do not. For example, the <tt>llvm.ctpop</tt> function
5132 can take an integer of any width and returns an integer of exactly the same
5133 integer width. This leads to a family of functions such as
5134 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29
5135 %val)</tt>. Only one type, the return type, is overloaded, and only one type
5136 suffix is required. Because the argument's type is matched against the return
5137 type, it does not require its own name suffix.</p>
5139 <p>To learn how to add an intrinsic function, please see the
5140 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.</p>
5144 <!-- ======================================================================= -->
5145 <div class="doc_subsection">
5146 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
5149 <div class="doc_text">
5151 <p>Variable argument support is defined in LLVM with
5152 the <a href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
5153 intrinsic functions. These functions are related to the similarly named
5154 macros defined in the <tt><stdarg.h></tt> header file.</p>
5156 <p>All of these functions operate on arguments that use a target-specific value
5157 type "<tt>va_list</tt>". The LLVM assembly language reference manual does
5158 not define what this type is, so all transformations should be prepared to
5159 handle these functions regardless of the type used.</p>
5161 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
5162 instruction and the variable argument handling intrinsic functions are
5165 <div class="doc_code">
5167 define i32 @test(i32 %X, ...) {
5168 ; Initialize variable argument processing
5170 %ap2 = bitcast i8** %ap to i8*
5171 call void @llvm.va_start(i8* %ap2)
5173 ; Read a single integer argument
5174 %tmp = va_arg i8** %ap, i32
5176 ; Demonstrate usage of llvm.va_copy and llvm.va_end
5178 %aq2 = bitcast i8** %aq to i8*
5179 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
5180 call void @llvm.va_end(i8* %aq2)
5182 ; Stop processing of arguments.
5183 call void @llvm.va_end(i8* %ap2)
5187 declare void @llvm.va_start(i8*)
5188 declare void @llvm.va_copy(i8*, i8*)
5189 declare void @llvm.va_end(i8*)
5195 <!-- _______________________________________________________________________ -->
5196 <div class="doc_subsubsection">
5197 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
5201 <div class="doc_text">
5205 declare void %llvm.va_start(i8* <arglist>)
5209 <p>The '<tt>llvm.va_start</tt>' intrinsic initializes <tt>*<arglist></tt>
5210 for subsequent use by <tt><a href="#i_va_arg">va_arg</a></tt>.</p>
5213 <p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
5216 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
5217 macro available in C. In a target-dependent way, it initializes
5218 the <tt>va_list</tt> element to which the argument points, so that the next
5219 call to <tt>va_arg</tt> will produce the first variable argument passed to
5220 the function. Unlike the C <tt>va_start</tt> macro, this intrinsic does not
5221 need to know the last argument of the function as the compiler can figure
5226 <!-- _______________________________________________________________________ -->
5227 <div class="doc_subsubsection">
5228 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
5231 <div class="doc_text">
5235 declare void @llvm.va_end(i8* <arglist>)
5239 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
5240 which has been initialized previously
5241 with <tt><a href="#int_va_start">llvm.va_start</a></tt>
5242 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
5245 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
5248 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
5249 macro available in C. In a target-dependent way, it destroys
5250 the <tt>va_list</tt> element to which the argument points. Calls
5251 to <a href="#int_va_start"><tt>llvm.va_start</tt></a>
5252 and <a href="#int_va_copy"> <tt>llvm.va_copy</tt></a> must be matched exactly
5253 with calls to <tt>llvm.va_end</tt>.</p>
5257 <!-- _______________________________________________________________________ -->
5258 <div class="doc_subsubsection">
5259 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
5262 <div class="doc_text">
5266 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
5270 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
5271 from the source argument list to the destination argument list.</p>
5274 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
5275 The second argument is a pointer to a <tt>va_list</tt> element to copy
5279 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
5280 macro available in C. In a target-dependent way, it copies the
5281 source <tt>va_list</tt> element into the destination <tt>va_list</tt>
5282 element. This intrinsic is necessary because
5283 the <tt><a href="#int_va_start"> llvm.va_start</a></tt> intrinsic may be
5284 arbitrarily complex and require, for example, memory allocation.</p>
5288 <!-- ======================================================================= -->
5289 <div class="doc_subsection">
5290 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
5293 <div class="doc_text">
5295 <p>LLVM support for <a href="GarbageCollection.html">Accurate Garbage
5296 Collection</a> (GC) requires the implementation and generation of these
5297 intrinsics. These intrinsics allow identification of <a href="#int_gcroot">GC
5298 roots on the stack</a>, as well as garbage collector implementations that
5299 require <a href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a>
5300 barriers. Front-ends for type-safe garbage collected languages should generate
5301 these intrinsics to make use of the LLVM garbage collectors. For more details,
5302 see <a href="GarbageCollection.html">Accurate Garbage Collection with
5305 <p>The garbage collection intrinsics only operate on objects in the generic
5306 address space (address space zero).</p>
5310 <!-- _______________________________________________________________________ -->
5311 <div class="doc_subsubsection">
5312 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
5315 <div class="doc_text">
5319 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
5323 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
5324 the code generator, and allows some metadata to be associated with it.</p>
5327 <p>The first argument specifies the address of a stack object that contains the
5328 root pointer. The second pointer (which must be either a constant or a
5329 global value address) contains the meta-data to be associated with the
5333 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
5334 location. At compile-time, the code generator generates information to allow
5335 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
5336 intrinsic may only be used in a function which <a href="#gc">specifies a GC
5341 <!-- _______________________________________________________________________ -->
5342 <div class="doc_subsubsection">
5343 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
5346 <div class="doc_text">
5350 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
5354 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
5355 locations, allowing garbage collector implementations that require read
5359 <p>The second argument is the address to read from, which should be an address
5360 allocated from the garbage collector. The first object is a pointer to the
5361 start of the referenced object, if needed by the language runtime (otherwise
5365 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
5366 instruction, but may be replaced with substantially more complex code by the
5367 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
5368 may only be used in a function which <a href="#gc">specifies a GC
5373 <!-- _______________________________________________________________________ -->
5374 <div class="doc_subsubsection">
5375 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
5378 <div class="doc_text">
5382 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
5386 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
5387 locations, allowing garbage collector implementations that require write
5388 barriers (such as generational or reference counting collectors).</p>
5391 <p>The first argument is the reference to store, the second is the start of the
5392 object to store it to, and the third is the address of the field of Obj to
5393 store to. If the runtime does not require a pointer to the object, Obj may
5397 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
5398 instruction, but may be replaced with substantially more complex code by the
5399 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
5400 may only be used in a function which <a href="#gc">specifies a GC
5405 <!-- ======================================================================= -->
5406 <div class="doc_subsection">
5407 <a name="int_codegen">Code Generator Intrinsics</a>
5410 <div class="doc_text">
5412 <p>These intrinsics are provided by LLVM to expose special features that may
5413 only be implemented with code generator support.</p>
5417 <!-- _______________________________________________________________________ -->
5418 <div class="doc_subsubsection">
5419 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
5422 <div class="doc_text">
5426 declare i8 *@llvm.returnaddress(i32 <level>)
5430 <p>The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
5431 target-specific value indicating the return address of the current function
5432 or one of its callers.</p>
5435 <p>The argument to this intrinsic indicates which function to return the address
5436 for. Zero indicates the calling function, one indicates its caller, etc.
5437 The argument is <b>required</b> to be a constant integer value.</p>
5440 <p>The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer
5441 indicating the return address of the specified call frame, or zero if it
5442 cannot be identified. The value returned by this intrinsic is likely to be
5443 incorrect or 0 for arguments other than zero, so it should only be used for
5444 debugging purposes.</p>
5446 <p>Note that calling this intrinsic does not prevent function inlining or other
5447 aggressive transformations, so the value returned may not be that of the
5448 obvious source-language caller.</p>
5452 <!-- _______________________________________________________________________ -->
5453 <div class="doc_subsubsection">
5454 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
5457 <div class="doc_text">
5461 declare i8 *@llvm.frameaddress(i32 <level>)
5465 <p>The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
5466 target-specific frame pointer value for the specified stack frame.</p>
5469 <p>The argument to this intrinsic indicates which function to return the frame
5470 pointer for. Zero indicates the calling function, one indicates its caller,
5471 etc. The argument is <b>required</b> to be a constant integer value.</p>
5474 <p>The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer
5475 indicating the frame address of the specified call frame, or zero if it
5476 cannot be identified. The value returned by this intrinsic is likely to be
5477 incorrect or 0 for arguments other than zero, so it should only be used for
5478 debugging purposes.</p>
5480 <p>Note that calling this intrinsic does not prevent function inlining or other
5481 aggressive transformations, so the value returned may not be that of the
5482 obvious source-language caller.</p>
5486 <!-- _______________________________________________________________________ -->
5487 <div class="doc_subsubsection">
5488 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
5491 <div class="doc_text">
5495 declare i8 *@llvm.stacksave()
5499 <p>The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state
5500 of the function stack, for use
5501 with <a href="#int_stackrestore"> <tt>llvm.stackrestore</tt></a>. This is
5502 useful for implementing language features like scoped automatic variable
5503 sized arrays in C99.</p>
5506 <p>This intrinsic returns a opaque pointer value that can be passed
5507 to <a href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When
5508 an <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved
5509 from <tt>llvm.stacksave</tt>, it effectively restores the state of the stack
5510 to the state it was in when the <tt>llvm.stacksave</tt> intrinsic executed.
5511 In practice, this pops any <a href="#i_alloca">alloca</a> blocks from the
5512 stack that were allocated after the <tt>llvm.stacksave</tt> was executed.</p>
5516 <!-- _______________________________________________________________________ -->
5517 <div class="doc_subsubsection">
5518 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
5521 <div class="doc_text">
5525 declare void @llvm.stackrestore(i8 * %ptr)
5529 <p>The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
5530 the function stack to the state it was in when the
5531 corresponding <a href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic
5532 executed. This is useful for implementing language features like scoped
5533 automatic variable sized arrays in C99.</p>
5536 <p>See the description
5537 for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.</p>
5541 <!-- _______________________________________________________________________ -->
5542 <div class="doc_subsubsection">
5543 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
5546 <div class="doc_text">
5550 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>)
5554 <p>The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to
5555 insert a prefetch instruction if supported; otherwise, it is a noop.
5556 Prefetches have no effect on the behavior of the program but can change its
5557 performance characteristics.</p>
5560 <p><tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the
5561 specifier determining if the fetch should be for a read (0) or write (1),
5562 and <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
5563 locality, to (3) - extremely local keep in cache. The <tt>rw</tt>
5564 and <tt>locality</tt> arguments must be constant integers.</p>
5567 <p>This intrinsic does not modify the behavior of the program. In particular,
5568 prefetches cannot trap and do not produce a value. On targets that support
5569 this intrinsic, the prefetch can provide hints to the processor cache for
5570 better performance.</p>
5574 <!-- _______________________________________________________________________ -->
5575 <div class="doc_subsubsection">
5576 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
5579 <div class="doc_text">
5583 declare void @llvm.pcmarker(i32 <id>)
5587 <p>The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program
5588 Counter (PC) in a region of code to simulators and other tools. The method
5589 is target specific, but it is expected that the marker will use exported
5590 symbols to transmit the PC of the marker. The marker makes no guarantees
5591 that it will remain with any specific instruction after optimizations. It is
5592 possible that the presence of a marker will inhibit optimizations. The
5593 intended use is to be inserted after optimizations to allow correlations of
5594 simulation runs.</p>
5597 <p><tt>id</tt> is a numerical id identifying the marker.</p>
5600 <p>This intrinsic does not modify the behavior of the program. Backends that do
5601 not support this intrinisic may ignore it.</p>
5605 <!-- _______________________________________________________________________ -->
5606 <div class="doc_subsubsection">
5607 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
5610 <div class="doc_text">
5614 declare i64 @llvm.readcyclecounter( )
5618 <p>The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
5619 counter register (or similar low latency, high accuracy clocks) on those
5620 targets that support it. On X86, it should map to RDTSC. On Alpha, it
5621 should map to RPCC. As the backing counters overflow quickly (on the order
5622 of 9 seconds on alpha), this should only be used for small timings.</p>
5625 <p>When directly supported, reading the cycle counter should not modify any
5626 memory. Implementations are allowed to either return a application specific
5627 value or a system wide value. On backends without support, this is lowered
5628 to a constant 0.</p>
5632 <!-- ======================================================================= -->
5633 <div class="doc_subsection">
5634 <a name="int_libc">Standard C Library Intrinsics</a>
5637 <div class="doc_text">
5639 <p>LLVM provides intrinsics for a few important standard C library functions.
5640 These intrinsics allow source-language front-ends to pass information about
5641 the alignment of the pointer arguments to the code generator, providing
5642 opportunity for more efficient code generation.</p>
5646 <!-- _______________________________________________________________________ -->
5647 <div class="doc_subsubsection">
5648 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
5651 <div class="doc_text">
5654 <p>This is an overloaded intrinsic. You can use <tt>llvm.memcpy</tt> on any
5655 integer bit width. Not all targets support all bit widths however.</p>
5658 declare void @llvm.memcpy.i8(i8 * <dest>, i8 * <src>,
5659 i8 <len>, i32 <align>)
5660 declare void @llvm.memcpy.i16(i8 * <dest>, i8 * <src>,
5661 i16 <len>, i32 <align>)
5662 declare void @llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
5663 i32 <len>, i32 <align>)
5664 declare void @llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
5665 i64 <len>, i32 <align>)
5669 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
5670 source location to the destination location.</p>
5672 <p>Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
5673 intrinsics do not return a value, and takes an extra alignment argument.</p>
5676 <p>The first argument is a pointer to the destination, the second is a pointer
5677 to the source. The third argument is an integer argument specifying the
5678 number of bytes to copy, and the fourth argument is the alignment of the
5679 source and destination locations.</p>
5681 <p>If the call to this intrinisic has an alignment value that is not 0 or 1,
5682 then the caller guarantees that both the source and destination pointers are
5683 aligned to that boundary.</p>
5686 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
5687 source location to the destination location, which are not allowed to
5688 overlap. It copies "len" bytes of memory over. If the argument is known to
5689 be aligned to some boundary, this can be specified as the fourth argument,
5690 otherwise it should be set to 0 or 1.</p>
5694 <!-- _______________________________________________________________________ -->
5695 <div class="doc_subsubsection">
5696 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
5699 <div class="doc_text">
5702 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
5703 width. Not all targets support all bit widths however.</p>
5706 declare void @llvm.memmove.i8(i8 * <dest>, i8 * <src>,
5707 i8 <len>, i32 <align>)
5708 declare void @llvm.memmove.i16(i8 * <dest>, i8 * <src>,
5709 i16 <len>, i32 <align>)
5710 declare void @llvm.memmove.i32(i8 * <dest>, i8 * <src>,
5711 i32 <len>, i32 <align>)
5712 declare void @llvm.memmove.i64(i8 * <dest>, i8 * <src>,
5713 i64 <len>, i32 <align>)
5717 <p>The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the
5718 source location to the destination location. It is similar to the
5719 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to
5722 <p>Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
5723 intrinsics do not return a value, and takes an extra alignment argument.</p>
5726 <p>The first argument is a pointer to the destination, the second is a pointer
5727 to the source. The third argument is an integer argument specifying the
5728 number of bytes to copy, and the fourth argument is the alignment of the
5729 source and destination locations.</p>
5731 <p>If the call to this intrinisic has an alignment value that is not 0 or 1,
5732 then the caller guarantees that the source and destination pointers are
5733 aligned to that boundary.</p>
5736 <p>The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the
5737 source location to the destination location, which may overlap. It copies
5738 "len" bytes of memory over. If the argument is known to be aligned to some
5739 boundary, this can be specified as the fourth argument, otherwise it should
5740 be set to 0 or 1.</p>
5744 <!-- _______________________________________________________________________ -->
5745 <div class="doc_subsubsection">
5746 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
5749 <div class="doc_text">
5752 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
5753 width. Not all targets support all bit widths however.</p>
5756 declare void @llvm.memset.i8(i8 * <dest>, i8 <val>,
5757 i8 <len>, i32 <align>)
5758 declare void @llvm.memset.i16(i8 * <dest>, i8 <val>,
5759 i16 <len>, i32 <align>)
5760 declare void @llvm.memset.i32(i8 * <dest>, i8 <val>,
5761 i32 <len>, i32 <align>)
5762 declare void @llvm.memset.i64(i8 * <dest>, i8 <val>,
5763 i64 <len>, i32 <align>)
5767 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a
5768 particular byte value.</p>
5770 <p>Note that, unlike the standard libc function, the <tt>llvm.memset</tt>
5771 intrinsic does not return a value, and takes an extra alignment argument.</p>
5774 <p>The first argument is a pointer to the destination to fill, the second is the
5775 byte value to fill it with, the third argument is an integer argument
5776 specifying the number of bytes to fill, and the fourth argument is the known
5777 alignment of destination location.</p>
5779 <p>If the call to this intrinisic has an alignment value that is not 0 or 1,
5780 then the caller guarantees that the destination pointer is aligned to that
5784 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting
5785 at the destination location. If the argument is known to be aligned to some
5786 boundary, this can be specified as the fourth argument, otherwise it should
5787 be set to 0 or 1.</p>
5791 <!-- _______________________________________________________________________ -->
5792 <div class="doc_subsubsection">
5793 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
5796 <div class="doc_text">
5799 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
5800 floating point or vector of floating point type. Not all targets support all
5804 declare float @llvm.sqrt.f32(float %Val)
5805 declare double @llvm.sqrt.f64(double %Val)
5806 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
5807 declare fp128 @llvm.sqrt.f128(fp128 %Val)
5808 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
5812 <p>The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
5813 returning the same value as the libm '<tt>sqrt</tt>' functions would.
5814 Unlike <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined
5815 behavior for negative numbers other than -0.0 (which allows for better
5816 optimization, because there is no need to worry about errno being
5817 set). <tt>llvm.sqrt(-0.0)</tt> is defined to return -0.0 like IEEE sqrt.</p>
5820 <p>The argument and return value are floating point numbers of the same
5824 <p>This function returns the sqrt of the specified operand if it is a
5825 nonnegative floating point number.</p>
5829 <!-- _______________________________________________________________________ -->
5830 <div class="doc_subsubsection">
5831 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
5834 <div class="doc_text">
5837 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
5838 floating point or vector of floating point type. Not all targets support all
5842 declare float @llvm.powi.f32(float %Val, i32 %power)
5843 declare double @llvm.powi.f64(double %Val, i32 %power)
5844 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
5845 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
5846 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
5850 <p>The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
5851 specified (positive or negative) power. The order of evaluation of
5852 multiplications is not defined. When a vector of floating point type is
5853 used, the second argument remains a scalar integer value.</p>
5856 <p>The second argument is an integer power, and the first is a value to raise to
5860 <p>This function returns the first value raised to the second power with an
5861 unspecified sequence of rounding operations.</p>
5865 <!-- _______________________________________________________________________ -->
5866 <div class="doc_subsubsection">
5867 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
5870 <div class="doc_text">
5873 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
5874 floating point or vector of floating point type. Not all targets support all
5878 declare float @llvm.sin.f32(float %Val)
5879 declare double @llvm.sin.f64(double %Val)
5880 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
5881 declare fp128 @llvm.sin.f128(fp128 %Val)
5882 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
5886 <p>The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.</p>
5889 <p>The argument and return value are floating point numbers of the same
5893 <p>This function returns the sine of the specified operand, returning the same
5894 values as the libm <tt>sin</tt> functions would, and handles error conditions
5895 in the same way.</p>
5899 <!-- _______________________________________________________________________ -->
5900 <div class="doc_subsubsection">
5901 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
5904 <div class="doc_text">
5907 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
5908 floating point or vector of floating point type. Not all targets support all
5912 declare float @llvm.cos.f32(float %Val)
5913 declare double @llvm.cos.f64(double %Val)
5914 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
5915 declare fp128 @llvm.cos.f128(fp128 %Val)
5916 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
5920 <p>The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.</p>
5923 <p>The argument and return value are floating point numbers of the same
5927 <p>This function returns the cosine of the specified operand, returning the same
5928 values as the libm <tt>cos</tt> functions would, and handles error conditions
5929 in the same way.</p>
5933 <!-- _______________________________________________________________________ -->
5934 <div class="doc_subsubsection">
5935 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
5938 <div class="doc_text">
5941 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
5942 floating point or vector of floating point type. Not all targets support all
5946 declare float @llvm.pow.f32(float %Val, float %Power)
5947 declare double @llvm.pow.f64(double %Val, double %Power)
5948 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
5949 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
5950 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
5954 <p>The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
5955 specified (positive or negative) power.</p>
5958 <p>The second argument is a floating point power, and the first is a value to
5959 raise to that power.</p>
5962 <p>This function returns the first value raised to the second power, returning
5963 the same values as the libm <tt>pow</tt> functions would, and handles error
5964 conditions in the same way.</p>
5968 <!-- ======================================================================= -->
5969 <div class="doc_subsection">
5970 <a name="int_manip">Bit Manipulation Intrinsics</a>
5973 <div class="doc_text">
5975 <p>LLVM provides intrinsics for a few important bit manipulation operations.
5976 These allow efficient code generation for some algorithms.</p>
5980 <!-- _______________________________________________________________________ -->
5981 <div class="doc_subsubsection">
5982 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
5985 <div class="doc_text">
5988 <p>This is an overloaded intrinsic function. You can use bswap on any integer
5989 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
5992 declare i16 @llvm.bswap.i16(i16 <id>)
5993 declare i32 @llvm.bswap.i32(i32 <id>)
5994 declare i64 @llvm.bswap.i64(i64 <id>)
5998 <p>The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
5999 values with an even number of bytes (positive multiple of 16 bits). These
6000 are useful for performing operations on data that is not in the target's
6001 native byte order.</p>
6004 <p>The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
6005 and low byte of the input i16 swapped. Similarly,
6006 the <tt>llvm.bswap.i32</tt> intrinsic returns an i32 value that has the four
6007 bytes of the input i32 swapped, so that if the input bytes are numbered 0, 1,
6008 2, 3 then the returned i32 will have its bytes in 3, 2, 1, 0 order.
6009 The <tt>llvm.bswap.i48</tt>, <tt>llvm.bswap.i64</tt> and other intrinsics
6010 extend this concept to additional even-byte lengths (6 bytes, 8 bytes and
6011 more, respectively).</p>
6015 <!-- _______________________________________________________________________ -->
6016 <div class="doc_subsubsection">
6017 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
6020 <div class="doc_text">
6023 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
6024 width. Not all targets support all bit widths however.</p>
6027 declare i8 @llvm.ctpop.i8(i8 <src>)
6028 declare i16 @llvm.ctpop.i16(i16 <src>)
6029 declare i32 @llvm.ctpop.i32(i32 <src>)
6030 declare i64 @llvm.ctpop.i64(i64 <src>)
6031 declare i256 @llvm.ctpop.i256(i256 <src>)
6035 <p>The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set
6039 <p>The only argument is the value to be counted. The argument may be of any
6040 integer type. The return type must match the argument type.</p>
6043 <p>The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.</p>
6047 <!-- _______________________________________________________________________ -->
6048 <div class="doc_subsubsection">
6049 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
6052 <div class="doc_text">
6055 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
6056 integer bit width. Not all targets support all bit widths however.</p>
6059 declare i8 @llvm.ctlz.i8 (i8 <src>)
6060 declare i16 @llvm.ctlz.i16(i16 <src>)
6061 declare i32 @llvm.ctlz.i32(i32 <src>)
6062 declare i64 @llvm.ctlz.i64(i64 <src>)
6063 declare i256 @llvm.ctlz.i256(i256 <src>)
6067 <p>The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
6068 leading zeros in a variable.</p>
6071 <p>The only argument is the value to be counted. The argument may be of any
6072 integer type. The return type must match the argument type.</p>
6075 <p>The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant)
6076 zeros in a variable. If the src == 0 then the result is the size in bits of
6077 the type of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.</p>
6081 <!-- _______________________________________________________________________ -->
6082 <div class="doc_subsubsection">
6083 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
6086 <div class="doc_text">
6089 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
6090 integer bit width. Not all targets support all bit widths however.</p>
6093 declare i8 @llvm.cttz.i8 (i8 <src>)
6094 declare i16 @llvm.cttz.i16(i16 <src>)
6095 declare i32 @llvm.cttz.i32(i32 <src>)
6096 declare i64 @llvm.cttz.i64(i64 <src>)
6097 declare i256 @llvm.cttz.i256(i256 <src>)
6101 <p>The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
6105 <p>The only argument is the value to be counted. The argument may be of any
6106 integer type. The return type must match the argument type.</p>
6109 <p>The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant)
6110 zeros in a variable. If the src == 0 then the result is the size in bits of
6111 the type of src. For example, <tt>llvm.cttz(2) = 1</tt>.</p>
6115 <!-- ======================================================================= -->
6116 <div class="doc_subsection">
6117 <a name="int_overflow">Arithmetic with Overflow Intrinsics</a>
6120 <div class="doc_text">
6122 <p>LLVM provides intrinsics for some arithmetic with overflow operations.</p>
6126 <!-- _______________________________________________________________________ -->
6127 <div class="doc_subsubsection">
6128 <a name="int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt>' Intrinsics</a>
6131 <div class="doc_text">
6134 <p>This is an overloaded intrinsic. You can use <tt>llvm.sadd.with.overflow</tt>
6135 on any integer bit width.</p>
6138 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
6139 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6140 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
6144 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
6145 a signed addition of the two arguments, and indicate whether an overflow
6146 occurred during the signed summation.</p>
6149 <p>The arguments (%a and %b) and the first element of the result structure may
6150 be of integer types of any bit width, but they must have the same bit
6151 width. The second element of the result structure must be of
6152 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6153 undergo signed addition.</p>
6156 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
6157 a signed addition of the two variables. They return a structure — the
6158 first element of which is the signed summation, and the second element of
6159 which is a bit specifying if the signed summation resulted in an
6164 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6165 %sum = extractvalue {i32, i1} %res, 0
6166 %obit = extractvalue {i32, i1} %res, 1
6167 br i1 %obit, label %overflow, label %normal
6172 <!-- _______________________________________________________________________ -->
6173 <div class="doc_subsubsection">
6174 <a name="int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt>' Intrinsics</a>
6177 <div class="doc_text">
6180 <p>This is an overloaded intrinsic. You can use <tt>llvm.uadd.with.overflow</tt>
6181 on any integer bit width.</p>
6184 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
6185 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6186 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
6190 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
6191 an unsigned addition of the two arguments, and indicate whether a carry
6192 occurred during the unsigned summation.</p>
6195 <p>The arguments (%a and %b) and the first element of the result structure may
6196 be of integer types of any bit width, but they must have the same bit
6197 width. The second element of the result structure must be of
6198 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6199 undergo unsigned addition.</p>
6202 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
6203 an unsigned addition of the two arguments. They return a structure —
6204 the first element of which is the sum, and the second element of which is a
6205 bit specifying if the unsigned summation resulted in a carry.</p>
6209 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6210 %sum = extractvalue {i32, i1} %res, 0
6211 %obit = extractvalue {i32, i1} %res, 1
6212 br i1 %obit, label %carry, label %normal
6217 <!-- _______________________________________________________________________ -->
6218 <div class="doc_subsubsection">
6219 <a name="int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt>' Intrinsics</a>
6222 <div class="doc_text">
6225 <p>This is an overloaded intrinsic. You can use <tt>llvm.ssub.with.overflow</tt>
6226 on any integer bit width.</p>
6229 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
6230 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6231 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
6235 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
6236 a signed subtraction of the two arguments, and indicate whether an overflow
6237 occurred during the signed subtraction.</p>
6240 <p>The arguments (%a and %b) and the first element of the result structure may
6241 be of integer types of any bit width, but they must have the same bit
6242 width. The second element of the result structure must be of
6243 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6244 undergo signed subtraction.</p>
6247 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
6248 a signed subtraction of the two arguments. They return a structure —
6249 the first element of which is the subtraction, and the second element of
6250 which is a bit specifying if the signed subtraction resulted in an
6255 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6256 %sum = extractvalue {i32, i1} %res, 0
6257 %obit = extractvalue {i32, i1} %res, 1
6258 br i1 %obit, label %overflow, label %normal
6263 <!-- _______________________________________________________________________ -->
6264 <div class="doc_subsubsection">
6265 <a name="int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt>' Intrinsics</a>
6268 <div class="doc_text">
6271 <p>This is an overloaded intrinsic. You can use <tt>llvm.usub.with.overflow</tt>
6272 on any integer bit width.</p>
6275 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
6276 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6277 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
6281 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
6282 an unsigned subtraction of the two arguments, and indicate whether an
6283 overflow occurred during the unsigned subtraction.</p>
6286 <p>The arguments (%a and %b) and the first element of the result structure may
6287 be of integer types of any bit width, but they must have the same bit
6288 width. The second element of the result structure must be of
6289 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6290 undergo unsigned subtraction.</p>
6293 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
6294 an unsigned subtraction of the two arguments. They return a structure —
6295 the first element of which is the subtraction, and the second element of
6296 which is a bit specifying if the unsigned subtraction resulted in an
6301 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6302 %sum = extractvalue {i32, i1} %res, 0
6303 %obit = extractvalue {i32, i1} %res, 1
6304 br i1 %obit, label %overflow, label %normal
6309 <!-- _______________________________________________________________________ -->
6310 <div class="doc_subsubsection">
6311 <a name="int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt>' Intrinsics</a>
6314 <div class="doc_text">
6317 <p>This is an overloaded intrinsic. You can use <tt>llvm.smul.with.overflow</tt>
6318 on any integer bit width.</p>
6321 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
6322 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6323 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
6328 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
6329 a signed multiplication of the two arguments, and indicate whether an
6330 overflow occurred during the signed multiplication.</p>
6333 <p>The arguments (%a and %b) and the first element of the result structure may
6334 be of integer types of any bit width, but they must have the same bit
6335 width. The second element of the result structure must be of
6336 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6337 undergo signed multiplication.</p>
6340 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
6341 a signed multiplication of the two arguments. They return a structure —
6342 the first element of which is the multiplication, and the second element of
6343 which is a bit specifying if the signed multiplication resulted in an
6348 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6349 %sum = extractvalue {i32, i1} %res, 0
6350 %obit = extractvalue {i32, i1} %res, 1
6351 br i1 %obit, label %overflow, label %normal
6356 <!-- _______________________________________________________________________ -->
6357 <div class="doc_subsubsection">
6358 <a name="int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt>' Intrinsics</a>
6361 <div class="doc_text">
6364 <p>This is an overloaded intrinsic. You can use <tt>llvm.umul.with.overflow</tt>
6365 on any integer bit width.</p>
6368 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
6369 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6370 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
6374 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
6375 a unsigned multiplication of the two arguments, and indicate whether an
6376 overflow occurred during the unsigned multiplication.</p>
6379 <p>The arguments (%a and %b) and the first element of the result structure may
6380 be of integer types of any bit width, but they must have the same bit
6381 width. The second element of the result structure must be of
6382 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6383 undergo unsigned multiplication.</p>
6386 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
6387 an unsigned multiplication of the two arguments. They return a structure
6388 — the first element of which is the multiplication, and the second
6389 element of which is a bit specifying if the unsigned multiplication resulted
6394 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6395 %sum = extractvalue {i32, i1} %res, 0
6396 %obit = extractvalue {i32, i1} %res, 1
6397 br i1 %obit, label %overflow, label %normal
6402 <!-- ======================================================================= -->
6403 <div class="doc_subsection">
6404 <a name="int_debugger">Debugger Intrinsics</a>
6407 <div class="doc_text">
6409 <p>The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt>
6410 prefix), are described in
6411 the <a href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source
6412 Level Debugging</a> document.</p>
6416 <!-- ======================================================================= -->
6417 <div class="doc_subsection">
6418 <a name="int_eh">Exception Handling Intrinsics</a>
6421 <div class="doc_text">
6423 <p>The LLVM exception handling intrinsics (which all start with
6424 <tt>llvm.eh.</tt> prefix), are described in
6425 the <a href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
6426 Handling</a> document.</p>
6430 <!-- ======================================================================= -->
6431 <div class="doc_subsection">
6432 <a name="int_trampoline">Trampoline Intrinsic</a>
6435 <div class="doc_text">
6437 <p>This intrinsic makes it possible to excise one parameter, marked with
6438 the <tt>nest</tt> attribute, from a function. The result is a callable
6439 function pointer lacking the nest parameter - the caller does not need to
6440 provide a value for it. Instead, the value to use is stored in advance in a
6441 "trampoline", a block of memory usually allocated on the stack, which also
6442 contains code to splice the nest value into the argument list. This is used
6443 to implement the GCC nested function address extension.</p>
6445 <p>For example, if the function is
6446 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
6447 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as
6450 <div class="doc_code">
6452 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
6453 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
6454 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
6455 %fp = bitcast i8* %p to i32 (i32, i32)*
6459 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
6460 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
6464 <!-- _______________________________________________________________________ -->
6465 <div class="doc_subsubsection">
6466 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
6469 <div class="doc_text">
6473 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
6477 <p>This fills the memory pointed to by <tt>tramp</tt> with code and returns a
6478 function pointer suitable for executing it.</p>
6481 <p>The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
6482 pointers. The <tt>tramp</tt> argument must point to a sufficiently large and
6483 sufficiently aligned block of memory; this memory is written to by the
6484 intrinsic. Note that the size and the alignment are target-specific - LLVM
6485 currently provides no portable way of determining them, so a front-end that
6486 generates this intrinsic needs to have some target-specific knowledge.
6487 The <tt>func</tt> argument must hold a function bitcast to
6488 an <tt>i8*</tt>.</p>
6491 <p>The block of memory pointed to by <tt>tramp</tt> is filled with target
6492 dependent code, turning it into a function. A pointer to this function is
6493 returned, but needs to be bitcast to an <a href="#int_trampoline">appropriate
6494 function pointer type</a> before being called. The new function's signature
6495 is the same as that of <tt>func</tt> with any arguments marked with
6496 the <tt>nest</tt> attribute removed. At most one such <tt>nest</tt> argument
6497 is allowed, and it must be of pointer type. Calling the new function is
6498 equivalent to calling <tt>func</tt> with the same argument list, but
6499 with <tt>nval</tt> used for the missing <tt>nest</tt> argument. If, after
6500 calling <tt>llvm.init.trampoline</tt>, the memory pointed to
6501 by <tt>tramp</tt> is modified, then the effect of any later call to the
6502 returned function pointer is undefined.</p>
6506 <!-- ======================================================================= -->
6507 <div class="doc_subsection">
6508 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
6511 <div class="doc_text">
6513 <p>These intrinsic functions expand the "universal IR" of LLVM to represent
6514 hardware constructs for atomic operations and memory synchronization. This
6515 provides an interface to the hardware, not an interface to the programmer. It
6516 is aimed at a low enough level to allow any programming models or APIs
6517 (Application Programming Interfaces) which need atomic behaviors to map
6518 cleanly onto it. It is also modeled primarily on hardware behavior. Just as
6519 hardware provides a "universal IR" for source languages, it also provides a
6520 starting point for developing a "universal" atomic operation and
6521 synchronization IR.</p>
6523 <p>These do <em>not</em> form an API such as high-level threading libraries,
6524 software transaction memory systems, atomic primitives, and intrinsic
6525 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
6526 application libraries. The hardware interface provided by LLVM should allow
6527 a clean implementation of all of these APIs and parallel programming models.
6528 No one model or paradigm should be selected above others unless the hardware
6529 itself ubiquitously does so.</p>
6533 <!-- _______________________________________________________________________ -->
6534 <div class="doc_subsubsection">
6535 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
6537 <div class="doc_text">
6540 declare void @llvm.memory.barrier( i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>, i1 <device> )
6544 <p>The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
6545 specific pairs of memory access types.</p>
6548 <p>The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
6549 The first four arguments enables a specific barrier as listed below. The
6550 fith argument specifies that the barrier applies to io or device or uncached
6554 <li><tt>ll</tt>: load-load barrier</li>
6555 <li><tt>ls</tt>: load-store barrier</li>
6556 <li><tt>sl</tt>: store-load barrier</li>
6557 <li><tt>ss</tt>: store-store barrier</li>
6558 <li><tt>device</tt>: barrier applies to device and uncached memory also.</li>
6562 <p>This intrinsic causes the system to enforce some ordering constraints upon
6563 the loads and stores of the program. This barrier does not
6564 indicate <em>when</em> any events will occur, it only enforces
6565 an <em>order</em> in which they occur. For any of the specified pairs of load
6566 and store operations (f.ex. load-load, or store-load), all of the first
6567 operations preceding the barrier will complete before any of the second
6568 operations succeeding the barrier begin. Specifically the semantics for each
6569 pairing is as follows:</p>
6572 <li><tt>ll</tt>: All loads before the barrier must complete before any load
6573 after the barrier begins.</li>
6574 <li><tt>ls</tt>: All loads before the barrier must complete before any
6575 store after the barrier begins.</li>
6576 <li><tt>ss</tt>: All stores before the barrier must complete before any
6577 store after the barrier begins.</li>
6578 <li><tt>sl</tt>: All stores before the barrier must complete before any
6579 load after the barrier begins.</li>
6582 <p>These semantics are applied with a logical "and" behavior when more than one
6583 is enabled in a single memory barrier intrinsic.</p>
6585 <p>Backends may implement stronger barriers than those requested when they do
6586 not support as fine grained a barrier as requested. Some architectures do
6587 not need all types of barriers and on such architectures, these become
6595 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
6596 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
6597 <i>; guarantee the above finishes</i>
6598 store i32 8, %ptr <i>; before this begins</i>
6603 <!-- _______________________________________________________________________ -->
6604 <div class="doc_subsubsection">
6605 <a name="int_atomic_cmp_swap">'<tt>llvm.atomic.cmp.swap.*</tt>' Intrinsic</a>
6608 <div class="doc_text">
6611 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.cmp.swap</tt> on
6612 any integer bit width and for different address spaces. Not all targets
6613 support all bit widths however.</p>
6616 declare i8 @llvm.atomic.cmp.swap.i8.p0i8( i8* <ptr>, i8 <cmp>, i8 <val> )
6617 declare i16 @llvm.atomic.cmp.swap.i16.p0i16( i16* <ptr>, i16 <cmp>, i16 <val> )
6618 declare i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* <ptr>, i32 <cmp>, i32 <val> )
6619 declare i64 @llvm.atomic.cmp.swap.i64.p0i64( i64* <ptr>, i64 <cmp>, i64 <val> )
6623 <p>This loads a value in memory and compares it to a given value. If they are
6624 equal, it stores a new value into the memory.</p>
6627 <p>The <tt>llvm.atomic.cmp.swap</tt> intrinsic takes three arguments. The result
6628 as well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
6629 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
6630 this integer type. While any bit width integer may be used, targets may only
6631 lower representations they support in hardware.</p>
6634 <p>This entire intrinsic must be executed atomically. It first loads the value
6635 in memory pointed to by <tt>ptr</tt> and compares it with the
6636 value <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the
6637 memory. The loaded value is yielded in all cases. This provides the
6638 equivalent of an atomic compare-and-swap operation within the SSA
6646 %val1 = add i32 4, 4
6647 %result1 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 4, %val1 )
6648 <i>; yields {i32}:result1 = 4</i>
6649 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
6650 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
6652 %val2 = add i32 1, 1
6653 %result2 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 5, %val2 )
6654 <i>; yields {i32}:result2 = 8</i>
6655 %stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
6657 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
6662 <!-- _______________________________________________________________________ -->
6663 <div class="doc_subsubsection">
6664 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
6666 <div class="doc_text">
6669 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
6670 integer bit width. Not all targets support all bit widths however.</p>
6673 declare i8 @llvm.atomic.swap.i8.p0i8( i8* <ptr>, i8 <val> )
6674 declare i16 @llvm.atomic.swap.i16.p0i16( i16* <ptr>, i16 <val> )
6675 declare i32 @llvm.atomic.swap.i32.p0i32( i32* <ptr>, i32 <val> )
6676 declare i64 @llvm.atomic.swap.i64.p0i64( i64* <ptr>, i64 <val> )
6680 <p>This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
6681 the value from memory. It then stores the value in <tt>val</tt> in the memory
6682 at <tt>ptr</tt>.</p>
6685 <p>The <tt>llvm.atomic.swap</tt> intrinsic takes two arguments. Both
6686 the <tt>val</tt> argument and the result must be integers of the same bit
6687 width. The first argument, <tt>ptr</tt>, must be a pointer to a value of this
6688 integer type. The targets may only lower integer representations they
6692 <p>This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
6693 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
6694 equivalent of an atomic swap operation within the SSA framework.</p>
6701 %val1 = add i32 4, 4
6702 %result1 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val1 )
6703 <i>; yields {i32}:result1 = 4</i>
6704 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
6705 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
6707 %val2 = add i32 1, 1
6708 %result2 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val2 )
6709 <i>; yields {i32}:result2 = 8</i>
6711 %stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
6712 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
6717 <!-- _______________________________________________________________________ -->
6718 <div class="doc_subsubsection">
6719 <a name="int_atomic_load_add">'<tt>llvm.atomic.load.add.*</tt>' Intrinsic</a>
6723 <div class="doc_text">
6726 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.add</tt> on
6727 any integer bit width. Not all targets support all bit widths however.</p>
6730 declare i8 @llvm.atomic.load.add.i8..p0i8( i8* <ptr>, i8 <delta> )
6731 declare i16 @llvm.atomic.load.add.i16..p0i16( i16* <ptr>, i16 <delta> )
6732 declare i32 @llvm.atomic.load.add.i32..p0i32( i32* <ptr>, i32 <delta> )
6733 declare i64 @llvm.atomic.load.add.i64..p0i64( i64* <ptr>, i64 <delta> )
6737 <p>This intrinsic adds <tt>delta</tt> to the value stored in memory
6738 at <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.</p>
6741 <p>The intrinsic takes two arguments, the first a pointer to an integer value
6742 and the second an integer value. The result is also an integer value. These
6743 integer types can have any bit width, but they must all have the same bit
6744 width. The targets may only lower integer representations they support.</p>
6747 <p>This intrinsic does a series of operations atomically. It first loads the
6748 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
6749 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.</p>
6755 %result1 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 4 )
6756 <i>; yields {i32}:result1 = 4</i>
6757 %result2 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 2 )
6758 <i>; yields {i32}:result2 = 8</i>
6759 %result3 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 5 )
6760 <i>; yields {i32}:result3 = 10</i>
6761 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
6766 <!-- _______________________________________________________________________ -->
6767 <div class="doc_subsubsection">
6768 <a name="int_atomic_load_sub">'<tt>llvm.atomic.load.sub.*</tt>' Intrinsic</a>
6772 <div class="doc_text">
6775 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.sub</tt> on
6776 any integer bit width and for different address spaces. Not all targets
6777 support all bit widths however.</p>
6780 declare i8 @llvm.atomic.load.sub.i8.p0i32( i8* <ptr>, i8 <delta> )
6781 declare i16 @llvm.atomic.load.sub.i16.p0i32( i16* <ptr>, i16 <delta> )
6782 declare i32 @llvm.atomic.load.sub.i32.p0i32( i32* <ptr>, i32 <delta> )
6783 declare i64 @llvm.atomic.load.sub.i64.p0i32( i64* <ptr>, i64 <delta> )
6787 <p>This intrinsic subtracts <tt>delta</tt> to the value stored in memory at
6788 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.</p>
6791 <p>The intrinsic takes two arguments, the first a pointer to an integer value
6792 and the second an integer value. The result is also an integer value. These
6793 integer types can have any bit width, but they must all have the same bit
6794 width. The targets may only lower integer representations they support.</p>
6797 <p>This intrinsic does a series of operations atomically. It first loads the
6798 value stored at <tt>ptr</tt>. It then subtracts <tt>delta</tt>, stores the
6799 result to <tt>ptr</tt>. It yields the original value stored
6800 at <tt>ptr</tt>.</p>
6806 %result1 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 4 )
6807 <i>; yields {i32}:result1 = 8</i>
6808 %result2 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 2 )
6809 <i>; yields {i32}:result2 = 4</i>
6810 %result3 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 5 )
6811 <i>; yields {i32}:result3 = 2</i>
6812 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = -3</i>
6817 <!-- _______________________________________________________________________ -->
6818 <div class="doc_subsubsection">
6819 <a name="int_atomic_load_and">'<tt>llvm.atomic.load.and.*</tt>' Intrinsic</a><br>
6820 <a name="int_atomic_load_nand">'<tt>llvm.atomic.load.nand.*</tt>' Intrinsic</a><br>
6821 <a name="int_atomic_load_or">'<tt>llvm.atomic.load.or.*</tt>' Intrinsic</a><br>
6822 <a name="int_atomic_load_xor">'<tt>llvm.atomic.load.xor.*</tt>' Intrinsic</a><br>
6825 <div class="doc_text">
6828 <p>These are overloaded intrinsics. You can
6829 use <tt>llvm.atomic.load_and</tt>, <tt>llvm.atomic.load_nand</tt>,
6830 <tt>llvm.atomic.load_or</tt>, and <tt>llvm.atomic.load_xor</tt> on any integer
6831 bit width and for different address spaces. Not all targets support all bit
6835 declare i8 @llvm.atomic.load.and.i8.p0i8( i8* <ptr>, i8 <delta> )
6836 declare i16 @llvm.atomic.load.and.i16.p0i16( i16* <ptr>, i16 <delta> )
6837 declare i32 @llvm.atomic.load.and.i32.p0i32( i32* <ptr>, i32 <delta> )
6838 declare i64 @llvm.atomic.load.and.i64.p0i64( i64* <ptr>, i64 <delta> )
6842 declare i8 @llvm.atomic.load.or.i8.p0i8( i8* <ptr>, i8 <delta> )
6843 declare i16 @llvm.atomic.load.or.i16.p0i16( i16* <ptr>, i16 <delta> )
6844 declare i32 @llvm.atomic.load.or.i32.p0i32( i32* <ptr>, i32 <delta> )
6845 declare i64 @llvm.atomic.load.or.i64.p0i64( i64* <ptr>, i64 <delta> )
6849 declare i8 @llvm.atomic.load.nand.i8.p0i32( i8* <ptr>, i8 <delta> )
6850 declare i16 @llvm.atomic.load.nand.i16.p0i32( i16* <ptr>, i16 <delta> )
6851 declare i32 @llvm.atomic.load.nand.i32.p0i32( i32* <ptr>, i32 <delta> )
6852 declare i64 @llvm.atomic.load.nand.i64.p0i32( i64* <ptr>, i64 <delta> )
6856 declare i8 @llvm.atomic.load.xor.i8.p0i32( i8* <ptr>, i8 <delta> )
6857 declare i16 @llvm.atomic.load.xor.i16.p0i32( i16* <ptr>, i16 <delta> )
6858 declare i32 @llvm.atomic.load.xor.i32.p0i32( i32* <ptr>, i32 <delta> )
6859 declare i64 @llvm.atomic.load.xor.i64.p0i32( i64* <ptr>, i64 <delta> )
6863 <p>These intrinsics bitwise the operation (and, nand, or, xor) <tt>delta</tt> to
6864 the value stored in memory at <tt>ptr</tt>. It yields the original value
6865 at <tt>ptr</tt>.</p>
6868 <p>These intrinsics take two arguments, the first a pointer to an integer value
6869 and the second an integer value. The result is also an integer value. These
6870 integer types can have any bit width, but they must all have the same bit
6871 width. The targets may only lower integer representations they support.</p>
6874 <p>These intrinsics does a series of operations atomically. They first load the
6875 value stored at <tt>ptr</tt>. They then do the bitwise
6876 operation <tt>delta</tt>, store the result to <tt>ptr</tt>. They yield the
6877 original value stored at <tt>ptr</tt>.</p>
6882 store i32 0x0F0F, %ptr
6883 %result0 = call i32 @llvm.atomic.load.nand.i32.p0i32( i32* %ptr, i32 0xFF )
6884 <i>; yields {i32}:result0 = 0x0F0F</i>
6885 %result1 = call i32 @llvm.atomic.load.and.i32.p0i32( i32* %ptr, i32 0xFF )
6886 <i>; yields {i32}:result1 = 0xFFFFFFF0</i>
6887 %result2 = call i32 @llvm.atomic.load.or.i32.p0i32( i32* %ptr, i32 0F )
6888 <i>; yields {i32}:result2 = 0xF0</i>
6889 %result3 = call i32 @llvm.atomic.load.xor.i32.p0i32( i32* %ptr, i32 0F )
6890 <i>; yields {i32}:result3 = FF</i>
6891 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = F0</i>
6896 <!-- _______________________________________________________________________ -->
6897 <div class="doc_subsubsection">
6898 <a name="int_atomic_load_max">'<tt>llvm.atomic.load.max.*</tt>' Intrinsic</a><br>
6899 <a name="int_atomic_load_min">'<tt>llvm.atomic.load.min.*</tt>' Intrinsic</a><br>
6900 <a name="int_atomic_load_umax">'<tt>llvm.atomic.load.umax.*</tt>' Intrinsic</a><br>
6901 <a name="int_atomic_load_umin">'<tt>llvm.atomic.load.umin.*</tt>' Intrinsic</a><br>
6904 <div class="doc_text">
6907 <p>These are overloaded intrinsics. You can use <tt>llvm.atomic.load_max</tt>,
6908 <tt>llvm.atomic.load_min</tt>, <tt>llvm.atomic.load_umax</tt>, and
6909 <tt>llvm.atomic.load_umin</tt> on any integer bit width and for different
6910 address spaces. Not all targets support all bit widths however.</p>
6913 declare i8 @llvm.atomic.load.max.i8.p0i8( i8* <ptr>, i8 <delta> )
6914 declare i16 @llvm.atomic.load.max.i16.p0i16( i16* <ptr>, i16 <delta> )
6915 declare i32 @llvm.atomic.load.max.i32.p0i32( i32* <ptr>, i32 <delta> )
6916 declare i64 @llvm.atomic.load.max.i64.p0i64( i64* <ptr>, i64 <delta> )
6920 declare i8 @llvm.atomic.load.min.i8.p0i8( i8* <ptr>, i8 <delta> )
6921 declare i16 @llvm.atomic.load.min.i16.p0i16( i16* <ptr>, i16 <delta> )
6922 declare i32 @llvm.atomic.load.min.i32..p0i32( i32* <ptr>, i32 <delta> )
6923 declare i64 @llvm.atomic.load.min.i64..p0i64( i64* <ptr>, i64 <delta> )
6927 declare i8 @llvm.atomic.load.umax.i8.p0i8( i8* <ptr>, i8 <delta> )
6928 declare i16 @llvm.atomic.load.umax.i16.p0i16( i16* <ptr>, i16 <delta> )
6929 declare i32 @llvm.atomic.load.umax.i32.p0i32( i32* <ptr>, i32 <delta> )
6930 declare i64 @llvm.atomic.load.umax.i64.p0i64( i64* <ptr>, i64 <delta> )
6934 declare i8 @llvm.atomic.load.umin.i8..p0i8( i8* <ptr>, i8 <delta> )
6935 declare i16 @llvm.atomic.load.umin.i16.p0i16( i16* <ptr>, i16 <delta> )
6936 declare i32 @llvm.atomic.load.umin.i32..p0i32( i32* <ptr>, i32 <delta> )
6937 declare i64 @llvm.atomic.load.umin.i64..p0i64( i64* <ptr>, i64 <delta> )
6941 <p>These intrinsics takes the signed or unsigned minimum or maximum of
6942 <tt>delta</tt> and the value stored in memory at <tt>ptr</tt>. It yields the
6943 original value at <tt>ptr</tt>.</p>
6946 <p>These intrinsics take two arguments, the first a pointer to an integer value
6947 and the second an integer value. The result is also an integer value. These
6948 integer types can have any bit width, but they must all have the same bit
6949 width. The targets may only lower integer representations they support.</p>
6952 <p>These intrinsics does a series of operations atomically. They first load the
6953 value stored at <tt>ptr</tt>. They then do the signed or unsigned min or
6954 max <tt>delta</tt> and the value, store the result to <tt>ptr</tt>. They
6955 yield the original value stored at <tt>ptr</tt>.</p>
6961 %result0 = call i32 @llvm.atomic.load.min.i32.p0i32( i32* %ptr, i32 -2 )
6962 <i>; yields {i32}:result0 = 7</i>
6963 %result1 = call i32 @llvm.atomic.load.max.i32.p0i32( i32* %ptr, i32 8 )
6964 <i>; yields {i32}:result1 = -2</i>
6965 %result2 = call i32 @llvm.atomic.load.umin.i32.p0i32( i32* %ptr, i32 10 )
6966 <i>; yields {i32}:result2 = 8</i>
6967 %result3 = call i32 @llvm.atomic.load.umax.i32.p0i32( i32* %ptr, i32 30 )
6968 <i>; yields {i32}:result3 = 8</i>
6969 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 30</i>
6974 <!-- ======================================================================= -->
6975 <div class="doc_subsection">
6976 <a name="int_general">General Intrinsics</a>
6979 <div class="doc_text">
6981 <p>This class of intrinsics is designed to be generic and has no specific
6986 <!-- _______________________________________________________________________ -->
6987 <div class="doc_subsubsection">
6988 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
6991 <div class="doc_text">
6995 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
6999 <p>The '<tt>llvm.var.annotation</tt>' intrinsic.</p>
7002 <p>The first argument is a pointer to a value, the second is a pointer to a
7003 global string, the third is a pointer to a global string which is the source
7004 file name, and the last argument is the line number.</p>
7007 <p>This intrinsic allows annotation of local variables with arbitrary strings.
7008 This can be useful for special purpose optimizations that want to look for
7009 these annotations. These have no other defined use, they are ignored by code
7010 generation and optimization.</p>
7014 <!-- _______________________________________________________________________ -->
7015 <div class="doc_subsubsection">
7016 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
7019 <div class="doc_text">
7022 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
7023 any integer bit width.</p>
7026 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
7027 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
7028 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
7029 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
7030 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
7034 <p>The '<tt>llvm.annotation</tt>' intrinsic.</p>
7037 <p>The first argument is an integer value (result of some expression), the
7038 second is a pointer to a global string, the third is a pointer to a global
7039 string which is the source file name, and the last argument is the line
7040 number. It returns the value of the first argument.</p>
7043 <p>This intrinsic allows annotations to be put on arbitrary expressions with
7044 arbitrary strings. This can be useful for special purpose optimizations that
7045 want to look for these annotations. These have no other defined use, they
7046 are ignored by code generation and optimization.</p>
7050 <!-- _______________________________________________________________________ -->
7051 <div class="doc_subsubsection">
7052 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
7055 <div class="doc_text">
7059 declare void @llvm.trap()
7063 <p>The '<tt>llvm.trap</tt>' intrinsic.</p>
7069 <p>This intrinsics is lowered to the target dependent trap instruction. If the
7070 target does not have a trap instruction, this intrinsic will be lowered to
7071 the call of the <tt>abort()</tt> function.</p>
7075 <!-- _______________________________________________________________________ -->
7076 <div class="doc_subsubsection">
7077 <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a>
7080 <div class="doc_text">
7084 declare void @llvm.stackprotector( i8* <guard>, i8** <slot> )
7088 <p>The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and
7089 stores it onto the stack at <tt>slot</tt>. The stack slot is adjusted to
7090 ensure that it is placed on the stack before local variables.</p>
7093 <p>The <tt>llvm.stackprotector</tt> intrinsic requires two pointer
7094 arguments. The first argument is the value loaded from the stack
7095 guard <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt>
7096 that has enough space to hold the value of the guard.</p>
7099 <p>This intrinsic causes the prologue/epilogue inserter to force the position of
7100 the <tt>AllocaInst</tt> stack slot to be before local variables on the
7101 stack. This is to ensure that if a local variable on the stack is
7102 overwritten, it will destroy the value of the guard. When the function exits,
7103 the guard on the stack is checked against the original guard. If they're
7104 different, then the program aborts by calling the <tt>__stack_chk_fail()</tt>
7109 <!-- *********************************************************************** -->
7112 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
7113 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
7114 <a href="http://validator.w3.org/check/referer"><img
7115 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
7117 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
7118 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
7119 Last modified: $Date$