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.</dd>
534 <dt><tt><b><a name="linkage_internal">internal</a></b></tt>: </dt>
535 <dd>Similar to private, but the value shows as a local symbol
536 (<tt>STB_LOCAL</tt> in the case of ELF) in the object file. This
537 corresponds to the notion of the '<tt>static</tt>' keyword in C.</dd>
539 <dt><tt><b><a name="linkage_available_externally">available_externally</a></b></tt>: </dt>
540 <dd>Globals with "<tt>available_externally</tt>" linkage are never emitted
541 into the object file corresponding to the LLVM module. They exist to
542 allow inlining and other optimizations to take place given knowledge of
543 the definition of the global, which is known to be somewhere outside the
544 module. Globals with <tt>available_externally</tt> linkage are allowed to
545 be discarded at will, and are otherwise the same as <tt>linkonce_odr</tt>.
546 This linkage type is only allowed on definitions, not declarations.</dd>
548 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
549 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
550 the same name when linkage occurs. This is typically used to implement
551 inline functions, templates, or other code which must be generated in each
552 translation unit that uses it. Unreferenced <tt>linkonce</tt> globals are
553 allowed to be discarded.</dd>
555 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
556 <dd>"<tt>weak</tt>" linkage has the same merging semantics as
557 <tt>linkonce</tt> linkage, except that unreferenced globals with
558 <tt>weak</tt> linkage may not be discarded. This is used for globals that
559 are declared "weak" in C source code.</dd>
561 <dt><tt><b><a name="linkage_common">common</a></b></tt>: </dt>
562 <dd>"<tt>common</tt>" linkage is most similar to "<tt>weak</tt>" linkage, but
563 they are used for tentative definitions in C, such as "<tt>int X;</tt>" at
565 Symbols with "<tt>common</tt>" linkage are merged in the same way as
566 <tt>weak symbols</tt>, and they may not be deleted if unreferenced.
567 Further, <tt>common</tt> symbols may not have an explicit section, and
568 must have a zero initializer. Functions and aliases may not have common
572 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
573 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
574 pointer to array type. When two global variables with appending linkage
575 are linked together, the two global arrays are appended together. This is
576 the LLVM, typesafe, equivalent of having the system linker append together
577 "sections" with identical names when .o files are linked.</dd>
579 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt>: </dt>
580 <dd>The semantics of this linkage follow the ELF object file model: the symbol
581 is weak until linked, if not linked, the symbol becomes null instead of
582 being an undefined reference.</dd>
584 <dt><tt><b><a name="linkage_linkonce">linkonce_odr</a></b></tt>: </dt>
585 <dt><tt><b><a name="linkage_weak">weak_odr</a></b></tt>: </dt>
586 <dd>Some languages allow differing globals to be merged, such as two functions
587 with different semantics. Other languages, such as <tt>C++</tt>, ensure
588 that only equivalent globals are ever merged (the "one definition rule" -
589 "ODR"). Such languages can use the <tt>linkonce_odr</tt>
590 and <tt>weak_odr</tt> linkage types to indicate that the global will only
591 be merged with equivalent globals. These linkage types are otherwise the
592 same as their non-<tt>odr</tt> versions.</dd>
594 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
595 <dd>If none of the above identifiers are used, the global is externally
596 visible, meaning that it participates in linkage and can be used to
597 resolve external symbol references.</dd>
600 <p>The next two types of linkage are targeted for Microsoft Windows platform
601 only. They are designed to support importing (exporting) symbols from (to)
602 DLLs (Dynamic Link Libraries).</p>
605 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt>: </dt>
606 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
607 or variable via a global pointer to a pointer that is set up by the DLL
608 exporting the symbol. On Microsoft Windows targets, the pointer name is
609 formed by combining <code>__imp_</code> and the function or variable
612 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt>: </dt>
613 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
614 pointer to a pointer in a DLL, so that it can be referenced with the
615 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
616 name is formed by combining <code>__imp_</code> and the function or
620 <p>For example, since the "<tt>.LC0</tt>" variable is defined to be internal, if
621 another module defined a "<tt>.LC0</tt>" variable and was linked with this
622 one, one of the two would be renamed, preventing a collision. Since
623 "<tt>main</tt>" and "<tt>puts</tt>" are external (i.e., lacking any linkage
624 declarations), they are accessible outside of the current module.</p>
626 <p>It is illegal for a function <i>declaration</i> to have any linkage type
627 other than "externally visible", <tt>dllimport</tt>
628 or <tt>extern_weak</tt>.</p>
630 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt>, <tt>weak</tt>
631 or <tt>weak_odr</tt> linkages.</p>
635 <!-- ======================================================================= -->
636 <div class="doc_subsection">
637 <a name="callingconv">Calling Conventions</a>
640 <div class="doc_text">
642 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
643 and <a href="#i_invoke">invokes</a> can all have an optional calling
644 convention specified for the call. The calling convention of any pair of
645 dynamic caller/callee must match, or the behavior of the program is
646 undefined. The following calling conventions are supported by LLVM, and more
647 may be added in the future:</p>
650 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
651 <dd>This calling convention (the default if no other calling convention is
652 specified) matches the target C calling conventions. This calling
653 convention supports varargs function calls and tolerates some mismatch in
654 the declared prototype and implemented declaration of the function (as
657 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
658 <dd>This calling convention attempts to make calls as fast as possible
659 (e.g. by passing things in registers). This calling convention allows the
660 target to use whatever tricks it wants to produce fast code for the
661 target, without having to conform to an externally specified ABI
662 (Application Binary Interface). Implementations of this convention should
663 allow arbitrary <a href="CodeGenerator.html#tailcallopt">tail call
664 optimization</a> to be supported. This calling convention does not
665 support varargs and requires the prototype of all callees to exactly match
666 the prototype of the function definition.</dd>
668 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
669 <dd>This calling convention attempts to make code in the caller as efficient
670 as possible under the assumption that the call is not commonly executed.
671 As such, these calls often preserve all registers so that the call does
672 not break any live ranges in the caller side. This calling convention
673 does not support varargs and requires the prototype of all callees to
674 exactly match the prototype of the function definition.</dd>
676 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
677 <dd>Any calling convention may be specified by number, allowing
678 target-specific calling conventions to be used. Target specific calling
679 conventions start at 64.</dd>
682 <p>More calling conventions can be added/defined on an as-needed basis, to
683 support Pascal conventions or any other well-known target-independent
688 <!-- ======================================================================= -->
689 <div class="doc_subsection">
690 <a name="visibility">Visibility Styles</a>
693 <div class="doc_text">
695 <p>All Global Variables and Functions have one of the following visibility
699 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
700 <dd>On targets that use the ELF object file format, default visibility means
701 that the declaration is visible to other modules and, in shared libraries,
702 means that the declared entity may be overridden. On Darwin, default
703 visibility means that the declaration is visible to other modules. Default
704 visibility corresponds to "external linkage" in the language.</dd>
706 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
707 <dd>Two declarations of an object with hidden visibility refer to the same
708 object if they are in the same shared object. Usually, hidden visibility
709 indicates that the symbol will not be placed into the dynamic symbol
710 table, so no other module (executable or shared library) can reference it
713 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
714 <dd>On ELF, protected visibility indicates that the symbol will be placed in
715 the dynamic symbol table, but that references within the defining module
716 will bind to the local symbol. That is, the symbol cannot be overridden by
722 <!-- ======================================================================= -->
723 <div class="doc_subsection">
724 <a name="namedtypes">Named Types</a>
727 <div class="doc_text">
729 <p>LLVM IR allows you to specify name aliases for certain types. This can make
730 it easier to read the IR and make the IR more condensed (particularly when
731 recursive types are involved). An example of a name specification is:</p>
733 <div class="doc_code">
735 %mytype = type { %mytype*, i32 }
739 <p>You may give a name to any <a href="#typesystem">type</a> except
740 "<a href="t_void">void</a>". Type name aliases may be used anywhere a type
741 is expected with the syntax "%mytype".</p>
743 <p>Note that type names are aliases for the structural type that they indicate,
744 and that you can therefore specify multiple names for the same type. This
745 often leads to confusing behavior when dumping out a .ll file. Since LLVM IR
746 uses structural typing, the name is not part of the type. When printing out
747 LLVM IR, the printer will pick <em>one name</em> to render all types of a
748 particular shape. This means that if you have code where two different
749 source types end up having the same LLVM type, that the dumper will sometimes
750 print the "wrong" or unexpected type. This is an important design point and
751 isn't going to change.</p>
755 <!-- ======================================================================= -->
756 <div class="doc_subsection">
757 <a name="globalvars">Global Variables</a>
760 <div class="doc_text">
762 <p>Global variables define regions of memory allocated at compilation time
763 instead of run-time. Global variables may optionally be initialized, may
764 have an explicit section to be placed in, and may have an optional explicit
765 alignment specified. A variable may be defined as "thread_local", which
766 means that it will not be shared by threads (each thread will have a
767 separated copy of the variable). A variable may be defined as a global
768 "constant," which indicates that the contents of the variable
769 will <b>never</b> be modified (enabling better optimization, allowing the
770 global data to be placed in the read-only section of an executable, etc).
771 Note that variables that need runtime initialization cannot be marked
772 "constant" as there is a store to the variable.</p>
774 <p>LLVM explicitly allows <em>declarations</em> of global variables to be marked
775 constant, even if the final definition of the global is not. This capability
776 can be used to enable slightly better optimization of the program, but
777 requires the language definition to guarantee that optimizations based on the
778 'constantness' are valid for the translation units that do not include the
781 <p>As SSA values, global variables define pointer values that are in scope
782 (i.e. they dominate) all basic blocks in the program. Global variables
783 always define a pointer to their "content" type because they describe a
784 region of memory, and all memory objects in LLVM are accessed through
787 <p>A global variable may be declared to reside in a target-specific numbered
788 address space. For targets that support them, address spaces may affect how
789 optimizations are performed and/or what target instructions are used to
790 access the variable. The default address space is zero. The address space
791 qualifier must precede any other attributes.</p>
793 <p>LLVM allows an explicit section to be specified for globals. If the target
794 supports it, it will emit globals to the section specified.</p>
796 <p>An explicit alignment may be specified for a global. If not present, or if
797 the alignment is set to zero, the alignment of the global is set by the
798 target to whatever it feels convenient. If an explicit alignment is
799 specified, the global is forced to have at least that much alignment. All
800 alignments must be a power of 2.</p>
802 <p>For example, the following defines a global in a numbered address space with
803 an initializer, section, and alignment:</p>
805 <div class="doc_code">
807 @G = addrspace(5) constant float 1.0, section "foo", align 4
814 <!-- ======================================================================= -->
815 <div class="doc_subsection">
816 <a name="functionstructure">Functions</a>
819 <div class="doc_text">
821 <p>LLVM function definitions consist of the "<tt>define</tt>" keyord, an
822 optional <a href="#linkage">linkage type</a>, an optional
823 <a href="#visibility">visibility style</a>, an optional
824 <a href="#callingconv">calling convention</a>, a return type, an optional
825 <a href="#paramattrs">parameter attribute</a> for the return type, a function
826 name, a (possibly empty) argument list (each with optional
827 <a href="#paramattrs">parameter attributes</a>), optional
828 <a href="#fnattrs">function attributes</a>, an optional section, an optional
829 alignment, an optional <a href="#gc">garbage collector name</a>, an opening
830 curly brace, a list of basic blocks, and a closing curly brace.</p>
832 <p>LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
833 optional <a href="#linkage">linkage type</a>, an optional
834 <a href="#visibility">visibility style</a>, an optional
835 <a href="#callingconv">calling convention</a>, a return type, an optional
836 <a href="#paramattrs">parameter attribute</a> for the return type, a function
837 name, a possibly empty list of arguments, an optional alignment, and an
838 optional <a href="#gc">garbage collector name</a>.</p>
840 <p>A function definition contains a list of basic blocks, forming the CFG
841 (Control Flow Graph) for the function. Each basic block may optionally start
842 with a label (giving the basic block a symbol table entry), contains a list
843 of instructions, and ends with a <a href="#terminators">terminator</a>
844 instruction (such as a branch or function return).</p>
846 <p>The first basic block in a function is special in two ways: it is immediately
847 executed on entrance to the function, and it is not allowed to have
848 predecessor basic blocks (i.e. there can not be any branches to the entry
849 block of a function). Because the block can have no predecessors, it also
850 cannot have any <a href="#i_phi">PHI nodes</a>.</p>
852 <p>LLVM allows an explicit section to be specified for functions. If the target
853 supports it, it will emit functions to the section specified.</p>
855 <p>An explicit alignment may be specified for a function. If not present, or if
856 the alignment is set to zero, the alignment of the function is set by the
857 target to whatever it feels convenient. If an explicit alignment is
858 specified, the function is forced to have at least that much alignment. All
859 alignments must be a power of 2.</p>
862 <div class="doc_code">
864 define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
865 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
866 <ResultType> @<FunctionName> ([argument list])
867 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
868 [<a href="#gc">gc</a>] { ... }
874 <!-- ======================================================================= -->
875 <div class="doc_subsection">
876 <a name="aliasstructure">Aliases</a>
879 <div class="doc_text">
881 <p>Aliases act as "second name" for the aliasee value (which can be either
882 function, global variable, another alias or bitcast of global value). Aliases
883 may have an optional <a href="#linkage">linkage type</a>, and an
884 optional <a href="#visibility">visibility style</a>.</p>
887 <div class="doc_code">
889 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
895 <!-- ======================================================================= -->
896 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
898 <div class="doc_text">
900 <p>The return type and each parameter of a function type may have a set of
901 <i>parameter attributes</i> associated with them. Parameter attributes are
902 used to communicate additional information about the result or parameters of
903 a function. Parameter attributes are considered to be part of the function,
904 not of the function type, so functions with different parameter attributes
905 can have the same function type.</p>
907 <p>Parameter attributes are simple keywords that follow the type specified. If
908 multiple parameter attributes are needed, they are space separated. For
911 <div class="doc_code">
913 declare i32 @printf(i8* noalias nocapture, ...)
914 declare i32 @atoi(i8 zeroext)
915 declare signext i8 @returns_signed_char()
919 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
920 <tt>readonly</tt>) come immediately after the argument list.</p>
922 <p>Currently, only the following parameter attributes are defined:</p>
925 <dt><tt>zeroext</tt></dt>
926 <dd>This indicates to the code generator that the parameter or return value
927 should be zero-extended to a 32-bit value by the caller (for a parameter)
928 or the callee (for a return value).</dd>
930 <dt><tt>signext</tt></dt>
931 <dd>This indicates to the code generator that the parameter or return value
932 should be sign-extended to a 32-bit value by the caller (for a parameter)
933 or the callee (for a return value).</dd>
935 <dt><tt>inreg</tt></dt>
936 <dd>This indicates that this parameter or return value should be treated in a
937 special target-dependent fashion during while emitting code for a function
938 call or return (usually, by putting it in a register as opposed to memory,
939 though some targets use it to distinguish between two different kinds of
940 registers). Use of this attribute is target-specific.</dd>
942 <dt><tt><a name="byval">byval</a></tt></dt>
943 <dd>This indicates that the pointer parameter should really be passed by value
944 to the function. The attribute implies that a hidden copy of the pointee
945 is made between the caller and the callee, so the callee is unable to
946 modify the value in the callee. This attribute is only valid on LLVM
947 pointer arguments. It is generally used to pass structs and arrays by
948 value, but is also valid on pointers to scalars. The copy is considered
949 to belong to the caller not the callee (for example,
950 <tt><a href="#readonly">readonly</a></tt> functions should not write to
951 <tt>byval</tt> parameters). This is not a valid attribute for return
952 values. The byval attribute also supports specifying an alignment with
953 the align attribute. This has a target-specific effect on the code
954 generator that usually indicates a desired alignment for the synthesized
957 <dt><tt>sret</tt></dt>
958 <dd>This indicates that the pointer parameter specifies the address of a
959 structure that is the return value of the function in the source program.
960 This pointer must be guaranteed by the caller to be valid: loads and
961 stores to the structure may be assumed by the callee to not to trap. This
962 may only be applied to the first parameter. This is not a valid attribute
963 for return values. </dd>
965 <dt><tt>noalias</tt></dt>
966 <dd>This indicates that the pointer does not alias any global or any other
967 parameter. The caller is responsible for ensuring that this is the
968 case. On a function return value, <tt>noalias</tt> additionally indicates
969 that the pointer does not alias any other pointers visible to the
970 caller. For further details, please see the discussion of the NoAlias
972 <a href="http://llvm.org/docs/AliasAnalysis.html#MustMayNo">alias
975 <dt><tt>nocapture</tt></dt>
976 <dd>This indicates that the callee does not make any copies of the pointer
977 that outlive the callee itself. This is not a valid attribute for return
980 <dt><tt>nest</tt></dt>
981 <dd>This indicates that the pointer parameter can be excised using the
982 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
983 attribute for return values.</dd>
988 <!-- ======================================================================= -->
989 <div class="doc_subsection">
990 <a name="gc">Garbage Collector Names</a>
993 <div class="doc_text">
995 <p>Each function may specify a garbage collector name, which is simply a
998 <div class="doc_code">
1000 define void @f() gc "name" { ...
1004 <p>The compiler declares the supported values of <i>name</i>. Specifying a
1005 collector which will cause the compiler to alter its output in order to
1006 support the named garbage collection algorithm.</p>
1010 <!-- ======================================================================= -->
1011 <div class="doc_subsection">
1012 <a name="fnattrs">Function Attributes</a>
1015 <div class="doc_text">
1017 <p>Function attributes are set to communicate additional information about a
1018 function. Function attributes are considered to be part of the function, not
1019 of the function type, so functions with different parameter attributes can
1020 have the same function type.</p>
1022 <p>Function attributes are simple keywords that follow the type specified. If
1023 multiple attributes are needed, they are space separated. For example:</p>
1025 <div class="doc_code">
1027 define void @f() noinline { ... }
1028 define void @f() alwaysinline { ... }
1029 define void @f() alwaysinline optsize { ... }
1030 define void @f() optsize
1035 <dt><tt>alwaysinline</tt></dt>
1036 <dd>This attribute indicates that the inliner should attempt to inline this
1037 function into callers whenever possible, ignoring any active inlining size
1038 threshold for this caller.</dd>
1040 <dt><tt>noinline</tt></dt>
1041 <dd>This attribute indicates that the inliner should never inline this
1042 function in any situation. This attribute may not be used together with
1043 the <tt>alwaysinline</tt> attribute.</dd>
1045 <dt><tt>optsize</tt></dt>
1046 <dd>This attribute suggests that optimization passes and code generator passes
1047 make choices that keep the code size of this function low, and otherwise
1048 do optimizations specifically to reduce code size.</dd>
1050 <dt><tt>noreturn</tt></dt>
1051 <dd>This function attribute indicates that the function never returns
1052 normally. This produces undefined behavior at runtime if the function
1053 ever does dynamically return.</dd>
1055 <dt><tt>nounwind</tt></dt>
1056 <dd>This function attribute indicates that the function never returns with an
1057 unwind or exceptional control flow. If the function does unwind, its
1058 runtime behavior is undefined.</dd>
1060 <dt><tt>readnone</tt></dt>
1061 <dd>This attribute indicates that the function computes its result (or decides
1062 to unwind an exception) based strictly on its arguments, without
1063 dereferencing any pointer arguments or otherwise accessing any mutable
1064 state (e.g. memory, control registers, etc) visible to caller functions.
1065 It does not write through any pointer arguments
1066 (including <tt><a href="#byval">byval</a></tt> arguments) and never
1067 changes any state visible to callers. This means that it cannot unwind
1068 exceptions by calling the <tt>C++</tt> exception throwing methods, but
1069 could use the <tt>unwind</tt> instruction.</dd>
1071 <dt><tt><a name="readonly">readonly</a></tt></dt>
1072 <dd>This attribute indicates that the function does not write through any
1073 pointer arguments (including <tt><a href="#byval">byval</a></tt>
1074 arguments) or otherwise modify any state (e.g. memory, control registers,
1075 etc) visible to caller functions. It may dereference pointer arguments
1076 and read state that may be set in the caller. A readonly function always
1077 returns the same value (or unwinds an exception identically) when called
1078 with the same set of arguments and global state. It cannot unwind an
1079 exception by calling the <tt>C++</tt> exception throwing methods, but may
1080 use the <tt>unwind</tt> instruction.</dd>
1082 <dt><tt><a name="ssp">ssp</a></tt></dt>
1083 <dd>This attribute indicates that the function should emit a stack smashing
1084 protector. It is in the form of a "canary"—a random value placed on
1085 the stack before the local variables that's checked upon return from the
1086 function to see if it has been overwritten. A heuristic is used to
1087 determine if a function needs stack protectors or not.<br>
1089 If a function that has an <tt>ssp</tt> attribute is inlined into a
1090 function that doesn't have an <tt>ssp</tt> attribute, then the resulting
1091 function will have an <tt>ssp</tt> attribute.</dd>
1093 <dt><tt>sspreq</tt></dt>
1094 <dd>This attribute indicates that the function should <em>always</em> emit a
1095 stack smashing protector. This overrides
1096 the <tt><a href="#ssp">ssp</a></tt> function attribute.<br>
1098 If a function that has an <tt>sspreq</tt> attribute is inlined into a
1099 function that doesn't have an <tt>sspreq</tt> attribute or which has
1100 an <tt>ssp</tt> attribute, then the resulting function will have
1101 an <tt>sspreq</tt> attribute.</dd>
1103 <dt><tt>noredzone</tt></dt>
1104 <dd>This attribute indicates that the code generator should not use a red
1105 zone, even if the target-specific ABI normally permits it.</dd>
1107 <dt><tt>noimplicitfloat</tt></dt>
1108 <dd>This attributes disables implicit floating point instructions.</dd>
1110 <dt><tt>naked</tt></dt>
1111 <dd>This attribute disables prologue / epilogue emission for the function.
1112 This can have very system-specific consequences.</dd>
1117 <!-- ======================================================================= -->
1118 <div class="doc_subsection">
1119 <a name="moduleasm">Module-Level Inline Assembly</a>
1122 <div class="doc_text">
1124 <p>Modules may contain "module-level inline asm" blocks, which corresponds to
1125 the GCC "file scope inline asm" blocks. These blocks are internally
1126 concatenated by LLVM and treated as a single unit, but may be separated in
1127 the <tt>.ll</tt> file if desired. The syntax is very simple:</p>
1129 <div class="doc_code">
1131 module asm "inline asm code goes here"
1132 module asm "more can go here"
1136 <p>The strings can contain any character by escaping non-printable characters.
1137 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1140 <p>The inline asm code is simply printed to the machine code .s file when
1141 assembly code is generated.</p>
1145 <!-- ======================================================================= -->
1146 <div class="doc_subsection">
1147 <a name="datalayout">Data Layout</a>
1150 <div class="doc_text">
1152 <p>A module may specify a target specific data layout string that specifies how
1153 data is to be laid out in memory. The syntax for the data layout is
1156 <div class="doc_code">
1158 target datalayout = "<i>layout specification</i>"
1162 <p>The <i>layout specification</i> consists of a list of specifications
1163 separated by the minus sign character ('-'). Each specification starts with
1164 a letter and may include other information after the letter to define some
1165 aspect of the data layout. The specifications accepted are as follows:</p>
1169 <dd>Specifies that the target lays out data in big-endian form. That is, the
1170 bits with the most significance have the lowest address location.</dd>
1173 <dd>Specifies that the target lays out data in little-endian form. That is,
1174 the bits with the least significance have the lowest address
1177 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1178 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1179 <i>preferred</i> alignments. All sizes are in bits. Specifying
1180 the <i>pref</i> alignment is optional. If omitted, the
1181 preceding <tt>:</tt> should be omitted too.</dd>
1183 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1184 <dd>This specifies the alignment for an integer type of a given bit
1185 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1187 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1188 <dd>This specifies the alignment for a vector type of a given bit
1191 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1192 <dd>This specifies the alignment for a floating point type of a given bit
1193 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
1196 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1197 <dd>This specifies the alignment for an aggregate type of a given bit
1200 <dt><tt>s<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1201 <dd>This specifies the alignment for a stack object of a given bit
1205 <p>When constructing the data layout for a given target, LLVM starts with a
1206 default set of specifications which are then (possibly) overriden by the
1207 specifications in the <tt>datalayout</tt> keyword. The default specifications
1208 are given in this list:</p>
1211 <li><tt>E</tt> - big endian</li>
1212 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
1213 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1214 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1215 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1216 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1217 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1218 alignment of 64-bits</li>
1219 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1220 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1221 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1222 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1223 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1224 <li><tt>s0:64:64</tt> - stack objects are 64-bit aligned</li>
1227 <p>When LLVM is determining the alignment for a given type, it uses the
1228 following rules:</p>
1231 <li>If the type sought is an exact match for one of the specifications, that
1232 specification is used.</li>
1234 <li>If no match is found, and the type sought is an integer type, then the
1235 smallest integer type that is larger than the bitwidth of the sought type
1236 is used. If none of the specifications are larger than the bitwidth then
1237 the the largest integer type is used. For example, given the default
1238 specifications above, the i7 type will use the alignment of i8 (next
1239 largest) while both i65 and i256 will use the alignment of i64 (largest
1242 <li>If no match is found, and the type sought is a vector type, then the
1243 largest vector type that is smaller than the sought vector type will be
1244 used as a fall back. This happens because <128 x double> can be
1245 implemented in terms of 64 <2 x double>, for example.</li>
1250 <!-- ======================================================================= -->
1251 <div class="doc_subsection">
1252 <a name="pointeraliasing">Pointer Aliasing Rules</a>
1255 <div class="doc_text">
1257 <p>Any memory access must be done through a pointer value associated
1258 with an address range of the memory access, otherwise the behavior
1259 is undefined. Pointer values are associated with address ranges
1260 according to the following rules:</p>
1263 <li>A pointer value formed from a
1264 <tt><a href="#i_getelementptr">getelementptr</a></tt> instruction
1265 is associated with the addresses associated with the first operand
1266 of the <tt>getelementptr</tt>.</li>
1267 <li>An address of a global variable is associated with the address
1268 range of the variable's storage.</li>
1269 <li>The result value of an allocation instruction is associated with
1270 the address range of the allocated storage.</li>
1271 <li>A null pointer in the default address-space is associated with
1273 <li>A pointer value formed by an
1274 <tt><a href="#i_inttoptr">inttoptr</a></tt> is associated with all
1275 address ranges of all pointer values that contribute (directly or
1276 indirectly) to the computation of the pointer's value.</li>
1277 <li>The result value of a
1278 <tt><a href="#i_bitcast">bitcast</a></tt> is associated with all
1279 addresses associated with the operand of the <tt>bitcast</tt>.</li>
1280 <li>An integer constant other than zero or a pointer value returned
1281 from a function not defined within LLVM may be associated with address
1282 ranges allocated through mechanisms other than those provided by
1283 LLVM. Such ranges shall not overlap with any ranges of addresses
1284 allocated by mechanisms provided by LLVM.</li>
1287 <p>LLVM IR does not associate types with memory. The result type of a
1288 <tt><a href="#i_load">load</a></tt> merely indicates the size and
1289 alignment of the memory from which to load, as well as the
1290 interpretation of the value. The first operand of a
1291 <tt><a href="#i_store">store</a></tt> similarly only indicates the size
1292 and alignment of the store.</p>
1294 <p>Consequently, type-based alias analysis, aka TBAA, aka
1295 <tt>-fstrict-aliasing</tt>, is not applicable to general unadorned
1296 LLVM IR. <a href="#metadata">Metadata</a> may be used to encode
1297 additional information which specialized optimization passes may use
1298 to implement type-based alias analysis.</p>
1302 <!-- *********************************************************************** -->
1303 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
1304 <!-- *********************************************************************** -->
1306 <div class="doc_text">
1308 <p>The LLVM type system is one of the most important features of the
1309 intermediate representation. Being typed enables a number of optimizations
1310 to be performed on the intermediate representation directly, without having
1311 to do extra analyses on the side before the transformation. A strong type
1312 system makes it easier to read the generated code and enables novel analyses
1313 and transformations that are not feasible to perform on normal three address
1314 code representations.</p>
1318 <!-- ======================================================================= -->
1319 <div class="doc_subsection"> <a name="t_classifications">Type
1320 Classifications</a> </div>
1322 <div class="doc_text">
1324 <p>The types fall into a few useful classifications:</p>
1326 <table border="1" cellspacing="0" cellpadding="4">
1328 <tr><th>Classification</th><th>Types</th></tr>
1330 <td><a href="#t_integer">integer</a></td>
1331 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1334 <td><a href="#t_floating">floating point</a></td>
1335 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1338 <td><a name="t_firstclass">first class</a></td>
1339 <td><a href="#t_integer">integer</a>,
1340 <a href="#t_floating">floating point</a>,
1341 <a href="#t_pointer">pointer</a>,
1342 <a href="#t_vector">vector</a>,
1343 <a href="#t_struct">structure</a>,
1344 <a href="#t_array">array</a>,
1345 <a href="#t_label">label</a>,
1346 <a href="#t_metadata">metadata</a>.
1350 <td><a href="#t_primitive">primitive</a></td>
1351 <td><a href="#t_label">label</a>,
1352 <a href="#t_void">void</a>,
1353 <a href="#t_floating">floating point</a>,
1354 <a href="#t_metadata">metadata</a>.</td>
1357 <td><a href="#t_derived">derived</a></td>
1358 <td><a href="#t_integer">integer</a>,
1359 <a href="#t_array">array</a>,
1360 <a href="#t_function">function</a>,
1361 <a href="#t_pointer">pointer</a>,
1362 <a href="#t_struct">structure</a>,
1363 <a href="#t_pstruct">packed structure</a>,
1364 <a href="#t_vector">vector</a>,
1365 <a href="#t_opaque">opaque</a>.
1371 <p>The <a href="#t_firstclass">first class</a> types are perhaps the most
1372 important. Values of these types are the only ones which can be produced by
1373 instructions, passed as arguments, or used as operands to instructions.</p>
1377 <!-- ======================================================================= -->
1378 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
1380 <div class="doc_text">
1382 <p>The primitive types are the fundamental building blocks of the LLVM
1387 <!-- _______________________________________________________________________ -->
1388 <div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1390 <div class="doc_text">
1394 <tr><th>Type</th><th>Description</th></tr>
1395 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1396 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1397 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1398 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1399 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1405 <!-- _______________________________________________________________________ -->
1406 <div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1408 <div class="doc_text">
1411 <p>The void type does not represent any value and has no size.</p>
1420 <!-- _______________________________________________________________________ -->
1421 <div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1423 <div class="doc_text">
1426 <p>The label type represents code labels.</p>
1435 <!-- _______________________________________________________________________ -->
1436 <div class="doc_subsubsection"> <a name="t_metadata">Metadata Type</a> </div>
1438 <div class="doc_text">
1441 <p>The metadata type represents embedded metadata. The only derived type that
1442 may contain metadata is <tt>metadata*</tt> or a function type that returns or
1443 takes metadata typed parameters, but not pointer to metadata types.</p>
1453 <!-- ======================================================================= -->
1454 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1456 <div class="doc_text">
1458 <p>The real power in LLVM comes from the derived types in the system. This is
1459 what allows a programmer to represent arrays, functions, pointers, and other
1460 useful types. Note that these derived types may be recursive: For example,
1461 it is possible to have a two dimensional array.</p>
1465 <!-- _______________________________________________________________________ -->
1466 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1468 <div class="doc_text">
1471 <p>The integer type is a very simple derived type that simply specifies an
1472 arbitrary bit width for the integer type desired. Any bit width from 1 bit to
1473 2^23-1 (about 8 million) can be specified.</p>
1480 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1484 <table class="layout">
1486 <td class="left"><tt>i1</tt></td>
1487 <td class="left">a single-bit integer.</td>
1490 <td class="left"><tt>i32</tt></td>
1491 <td class="left">a 32-bit integer.</td>
1494 <td class="left"><tt>i1942652</tt></td>
1495 <td class="left">a really big integer of over 1 million bits.</td>
1499 <p>Note that the code generator does not yet support large integer types to be
1500 used as function return types. The specific limit on how large a return type
1501 the code generator can currently handle is target-dependent; currently it's
1502 often 64 bits for 32-bit targets and 128 bits for 64-bit targets.</p>
1506 <!-- _______________________________________________________________________ -->
1507 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1509 <div class="doc_text">
1512 <p>The array type is a very simple derived type that arranges elements
1513 sequentially in memory. The array type requires a size (number of elements)
1514 and an underlying data type.</p>
1518 [<# elements> x <elementtype>]
1521 <p>The number of elements is a constant integer value; <tt>elementtype</tt> may
1522 be any type with a size.</p>
1525 <table class="layout">
1527 <td class="left"><tt>[40 x i32]</tt></td>
1528 <td class="left">Array of 40 32-bit integer values.</td>
1531 <td class="left"><tt>[41 x i32]</tt></td>
1532 <td class="left">Array of 41 32-bit integer values.</td>
1535 <td class="left"><tt>[4 x i8]</tt></td>
1536 <td class="left">Array of 4 8-bit integer values.</td>
1539 <p>Here are some examples of multidimensional arrays:</p>
1540 <table class="layout">
1542 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1543 <td class="left">3x4 array of 32-bit integer values.</td>
1546 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1547 <td class="left">12x10 array of single precision floating point values.</td>
1550 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1551 <td class="left">2x3x4 array of 16-bit integer values.</td>
1555 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1556 length array. Normally, accesses past the end of an array are undefined in
1557 LLVM (e.g. it is illegal to access the 5th element of a 3 element array). As
1558 a special case, however, zero length arrays are recognized to be variable
1559 length. This allows implementation of 'pascal style arrays' with the LLVM
1560 type "<tt>{ i32, [0 x float]}</tt>", for example.</p>
1562 <p>Note that the code generator does not yet support large aggregate types to be
1563 used as function return types. The specific limit on how large an aggregate
1564 return type the code generator can currently handle is target-dependent, and
1565 also dependent on the aggregate element types.</p>
1569 <!-- _______________________________________________________________________ -->
1570 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1572 <div class="doc_text">
1575 <p>The function type can be thought of as a function signature. It consists of
1576 a return type and a list of formal parameter types. The return type of a
1577 function type is a scalar type, a void type, or a struct type. If the return
1578 type is a struct type then all struct elements must be of first class types,
1579 and the struct must have at least one element.</p>
1583 <returntype list> (<parameter list>)
1586 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1587 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1588 which indicates that the function takes a variable number of arguments.
1589 Variable argument functions can access their arguments with
1590 the <a href="#int_varargs">variable argument handling intrinsic</a>
1591 functions. '<tt><returntype list></tt>' is a comma-separated list of
1592 <a href="#t_firstclass">first class</a> type specifiers.</p>
1595 <table class="layout">
1597 <td class="left"><tt>i32 (i32)</tt></td>
1598 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1600 </tr><tr class="layout">
1601 <td class="left"><tt>float (i16 signext, i32 *) *
1603 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1604 an <tt>i16</tt> that should be sign extended and a
1605 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1608 </tr><tr class="layout">
1609 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1610 <td class="left">A vararg function that takes at least one
1611 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1612 which returns an integer. This is the signature for <tt>printf</tt> in
1615 </tr><tr class="layout">
1616 <td class="left"><tt>{i32, i32} (i32)</tt></td>
1617 <td class="left">A function taking an <tt>i32</tt>, returning two
1618 <tt>i32</tt> values as an aggregate of type <tt>{ i32, i32 }</tt>
1625 <!-- _______________________________________________________________________ -->
1626 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1628 <div class="doc_text">
1631 <p>The structure type is used to represent a collection of data members together
1632 in memory. The packing of the field types is defined to match the ABI of the
1633 underlying processor. The elements of a structure may be any type that has a
1636 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt> and
1637 '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field with
1638 the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
1642 { <type list> }
1646 <table class="layout">
1648 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1649 <td class="left">A triple of three <tt>i32</tt> values</td>
1650 </tr><tr class="layout">
1651 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1652 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1653 second element is a <a href="#t_pointer">pointer</a> to a
1654 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1655 an <tt>i32</tt>.</td>
1659 <p>Note that the code generator does not yet support large aggregate types to be
1660 used as function return types. The specific limit on how large an aggregate
1661 return type the code generator can currently handle is target-dependent, and
1662 also dependent on the aggregate element types.</p>
1666 <!-- _______________________________________________________________________ -->
1667 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1670 <div class="doc_text">
1673 <p>The packed structure type is used to represent a collection of data members
1674 together in memory. There is no padding between fields. Further, the
1675 alignment of a packed structure is 1 byte. The elements of a packed
1676 structure may be any type that has a size.</p>
1678 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt> and
1679 '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field with
1680 the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
1684 < { <type list> } >
1688 <table class="layout">
1690 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1691 <td class="left">A triple of three <tt>i32</tt> values</td>
1692 </tr><tr class="layout">
1694 <tt>< { float, i32 (i32)* } ></tt></td>
1695 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1696 second element is a <a href="#t_pointer">pointer</a> to a
1697 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1698 an <tt>i32</tt>.</td>
1704 <!-- _______________________________________________________________________ -->
1705 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1707 <div class="doc_text">
1710 <p>As in many languages, the pointer type represents a pointer or reference to
1711 another object, which must live in memory. Pointer types may have an optional
1712 address space attribute defining the target-specific numbered address space
1713 where the pointed-to object resides. The default address space is zero.</p>
1715 <p>Note that LLVM does not permit pointers to void (<tt>void*</tt>) nor does it
1716 permit pointers to labels (<tt>label*</tt>). Use <tt>i8*</tt> instead.</p>
1724 <table class="layout">
1726 <td class="left"><tt>[4 x i32]*</tt></td>
1727 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1728 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1731 <td class="left"><tt>i32 (i32 *) *</tt></td>
1732 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
1733 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1737 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1738 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1739 that resides in address space #5.</td>
1745 <!-- _______________________________________________________________________ -->
1746 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1748 <div class="doc_text">
1751 <p>A vector type is a simple derived type that represents a vector of elements.
1752 Vector types are used when multiple primitive data are operated in parallel
1753 using a single instruction (SIMD). A vector type requires a size (number of
1754 elements) and an underlying primitive data type. Vectors must have a power
1755 of two length (1, 2, 4, 8, 16 ...). Vector types are considered
1756 <a href="#t_firstclass">first class</a>.</p>
1760 < <# elements> x <elementtype> >
1763 <p>The number of elements is a constant integer value; elementtype may be any
1764 integer or floating point type.</p>
1767 <table class="layout">
1769 <td class="left"><tt><4 x i32></tt></td>
1770 <td class="left">Vector of 4 32-bit integer values.</td>
1773 <td class="left"><tt><8 x float></tt></td>
1774 <td class="left">Vector of 8 32-bit floating-point values.</td>
1777 <td class="left"><tt><2 x i64></tt></td>
1778 <td class="left">Vector of 2 64-bit integer values.</td>
1782 <p>Note that the code generator does not yet support large vector types to be
1783 used as function return types. The specific limit on how large a vector
1784 return type codegen can currently handle is target-dependent; currently it's
1785 often a few times longer than a hardware vector register.</p>
1789 <!-- _______________________________________________________________________ -->
1790 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1791 <div class="doc_text">
1794 <p>Opaque types are used to represent unknown types in the system. This
1795 corresponds (for example) to the C notion of a forward declared structure
1796 type. In LLVM, opaque types can eventually be resolved to any type (not just
1797 a structure type).</p>
1805 <table class="layout">
1807 <td class="left"><tt>opaque</tt></td>
1808 <td class="left">An opaque type.</td>
1814 <!-- ======================================================================= -->
1815 <div class="doc_subsection">
1816 <a name="t_uprefs">Type Up-references</a>
1819 <div class="doc_text">
1822 <p>An "up reference" allows you to refer to a lexically enclosing type without
1823 requiring it to have a name. For instance, a structure declaration may
1824 contain a pointer to any of the types it is lexically a member of. Example
1825 of up references (with their equivalent as named type declarations)
1829 { \2 * } %x = type { %x* }
1830 { \2 }* %y = type { %y }*
1834 <p>An up reference is needed by the asmprinter for printing out cyclic types
1835 when there is no declared name for a type in the cycle. Because the
1836 asmprinter does not want to print out an infinite type string, it needs a
1837 syntax to handle recursive types that have no names (all names are optional
1845 <p>The level is the count of the lexical type that is being referred to.</p>
1848 <table class="layout">
1850 <td class="left"><tt>\1*</tt></td>
1851 <td class="left">Self-referential pointer.</td>
1854 <td class="left"><tt>{ { \3*, i8 }, i32 }</tt></td>
1855 <td class="left">Recursive structure where the upref refers to the out-most
1862 <!-- *********************************************************************** -->
1863 <div class="doc_section"> <a name="constants">Constants</a> </div>
1864 <!-- *********************************************************************** -->
1866 <div class="doc_text">
1868 <p>LLVM has several different basic types of constants. This section describes
1869 them all and their syntax.</p>
1873 <!-- ======================================================================= -->
1874 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1876 <div class="doc_text">
1879 <dt><b>Boolean constants</b></dt>
1880 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1881 constants of the <tt><a href="#t_primitive">i1</a></tt> type.</dd>
1883 <dt><b>Integer constants</b></dt>
1884 <dd>Standard integers (such as '4') are constants of
1885 the <a href="#t_integer">integer</a> type. Negative numbers may be used
1886 with integer types.</dd>
1888 <dt><b>Floating point constants</b></dt>
1889 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1890 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1891 notation (see below). The assembler requires the exact decimal value of a
1892 floating-point constant. For example, the assembler accepts 1.25 but
1893 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
1894 constants must have a <a href="#t_floating">floating point</a> type. </dd>
1896 <dt><b>Null pointer constants</b></dt>
1897 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1898 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1901 <p>The one non-intuitive notation for constants is the hexadecimal form of
1902 floating point constants. For example, the form '<tt>double
1903 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than)
1904 '<tt>double 4.5e+15</tt>'. The only time hexadecimal floating point
1905 constants are required (and the only time that they are generated by the
1906 disassembler) is when a floating point constant must be emitted but it cannot
1907 be represented as a decimal floating point number in a reasonable number of
1908 digits. For example, NaN's, infinities, and other special values are
1909 represented in their IEEE hexadecimal format so that assembly and disassembly
1910 do not cause any bits to change in the constants.</p>
1912 <p>When using the hexadecimal form, constants of types float and double are
1913 represented using the 16-digit form shown above (which matches the IEEE754
1914 representation for double); float values must, however, be exactly
1915 representable as IEE754 single precision. Hexadecimal format is always used
1916 for long double, and there are three forms of long double. The 80-bit format
1917 used by x86 is represented as <tt>0xK</tt> followed by 20 hexadecimal digits.
1918 The 128-bit format used by PowerPC (two adjacent doubles) is represented
1919 by <tt>0xM</tt> followed by 32 hexadecimal digits. The IEEE 128-bit format
1920 is represented by <tt>0xL</tt> followed by 32 hexadecimal digits; no
1921 currently supported target uses this format. Long doubles will only work if
1922 they match the long double format on your target. All hexadecimal formats
1923 are big-endian (sign bit at the left).</p>
1927 <!-- ======================================================================= -->
1928 <div class="doc_subsection">
1929 <a name="aggregateconstants"></a> <!-- old anchor -->
1930 <a name="complexconstants">Complex Constants</a>
1933 <div class="doc_text">
1935 <p>Complex constants are a (potentially recursive) combination of simple
1936 constants and smaller complex constants.</p>
1939 <dt><b>Structure constants</b></dt>
1940 <dd>Structure constants are represented with notation similar to structure
1941 type definitions (a comma separated list of elements, surrounded by braces
1942 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
1943 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>".
1944 Structure constants must have <a href="#t_struct">structure type</a>, and
1945 the number and types of elements must match those specified by the
1948 <dt><b>Array constants</b></dt>
1949 <dd>Array constants are represented with notation similar to array type
1950 definitions (a comma separated list of elements, surrounded by square
1951 brackets (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74
1952 ]</tt>". Array constants must have <a href="#t_array">array type</a>, and
1953 the number and types of elements must match those specified by the
1956 <dt><b>Vector constants</b></dt>
1957 <dd>Vector constants are represented with notation similar to vector type
1958 definitions (a comma separated list of elements, surrounded by
1959 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32
1960 42, i32 11, i32 74, i32 100 ></tt>". Vector constants must
1961 have <a href="#t_vector">vector type</a>, and the number and types of
1962 elements must match those specified by the type.</dd>
1964 <dt><b>Zero initialization</b></dt>
1965 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1966 value to zero of <em>any</em> type, including scalar and aggregate types.
1967 This is often used to avoid having to print large zero initializers
1968 (e.g. for large arrays) and is always exactly equivalent to using explicit
1969 zero initializers.</dd>
1971 <dt><b>Metadata node</b></dt>
1972 <dd>A metadata node is a structure-like constant with
1973 <a href="#t_metadata">metadata type</a>. For example: "<tt>metadata !{
1974 i32 0, metadata !"test" }</tt>". Unlike other constants that are meant to
1975 be interpreted as part of the instruction stream, metadata is a place to
1976 attach additional information such as debug info.</dd>
1981 <!-- ======================================================================= -->
1982 <div class="doc_subsection">
1983 <a name="globalconstants">Global Variable and Function Addresses</a>
1986 <div class="doc_text">
1988 <p>The addresses of <a href="#globalvars">global variables</a>
1989 and <a href="#functionstructure">functions</a> are always implicitly valid
1990 (link-time) constants. These constants are explicitly referenced when
1991 the <a href="#identifiers">identifier for the global</a> is used and always
1992 have <a href="#t_pointer">pointer</a> type. For example, the following is a
1993 legal LLVM file:</p>
1995 <div class="doc_code">
1999 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2005 <!-- ======================================================================= -->
2006 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
2007 <div class="doc_text">
2009 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has no
2010 specific value. Undefined values may be of any type and be used anywhere a
2011 constant is permitted.</p>
2013 <p>Undefined values indicate to the compiler that the program is well defined no
2014 matter what value is used, giving the compiler more freedom to optimize.</p>
2018 <!-- ======================================================================= -->
2019 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
2022 <div class="doc_text">
2024 <p>Constant expressions are used to allow expressions involving other constants
2025 to be used as constants. Constant expressions may be of
2026 any <a href="#t_firstclass">first class</a> type and may involve any LLVM
2027 operation that does not have side effects (e.g. load and call are not
2028 supported). The following is the syntax for constant expressions:</p>
2031 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
2032 <dd>Truncate a constant to another type. The bit size of CST must be larger
2033 than the bit size of TYPE. Both types must be integers.</dd>
2035 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
2036 <dd>Zero extend a constant to another type. The bit size of CST must be
2037 smaller or equal to the bit size of TYPE. Both types must be
2040 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
2041 <dd>Sign extend a constant to another type. The bit size of CST must be
2042 smaller or equal to the bit size of TYPE. Both types must be
2045 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
2046 <dd>Truncate a floating point constant to another floating point type. The
2047 size of CST must be larger than the size of TYPE. Both types must be
2048 floating point.</dd>
2050 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
2051 <dd>Floating point extend a constant to another type. The size of CST must be
2052 smaller or equal to the size of TYPE. Both types must be floating
2055 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
2056 <dd>Convert a floating point constant to the corresponding unsigned integer
2057 constant. TYPE must be a scalar or vector integer type. CST must be of
2058 scalar or vector floating point type. Both CST and TYPE must be scalars,
2059 or vectors of the same number of elements. If the value won't fit in the
2060 integer type, the results are undefined.</dd>
2062 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
2063 <dd>Convert a floating point constant to the corresponding signed integer
2064 constant. TYPE must be a scalar or vector integer type. CST must be of
2065 scalar or vector floating point type. Both CST and TYPE must be scalars,
2066 or vectors of the same number of elements. If the value won't fit in the
2067 integer type, the results are undefined.</dd>
2069 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
2070 <dd>Convert an unsigned integer constant to the corresponding floating point
2071 constant. TYPE must be a scalar or vector floating point type. CST must be
2072 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2073 vectors of the same number of elements. If the value won't fit in the
2074 floating point type, the results are undefined.</dd>
2076 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
2077 <dd>Convert a signed integer constant to the corresponding floating point
2078 constant. TYPE must be a scalar or vector floating point type. CST must be
2079 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2080 vectors of the same number of elements. If the value won't fit in the
2081 floating point type, the results are undefined.</dd>
2083 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
2084 <dd>Convert a pointer typed constant to the corresponding integer constant
2085 <tt>TYPE</tt> must be an integer type. <tt>CST</tt> must be of pointer
2086 type. The <tt>CST</tt> value is zero extended, truncated, or unchanged to
2087 make it fit in <tt>TYPE</tt>.</dd>
2089 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
2090 <dd>Convert a integer constant to a pointer constant. TYPE must be a pointer
2091 type. CST must be of integer type. The CST value is zero extended,
2092 truncated, or unchanged to make it fit in a pointer size. This one is
2093 <i>really</i> dangerous!</dd>
2095 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
2096 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
2097 are the same as those for the <a href="#i_bitcast">bitcast
2098 instruction</a>.</dd>
2100 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
2101 <dt><b><tt>getelementptr inbounds ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
2102 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
2103 constants. As with the <a href="#i_getelementptr">getelementptr</a>
2104 instruction, the index list may have zero or more indexes, which are
2105 required to make sense for the type of "CSTPTR".</dd>
2107 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
2108 <dd>Perform the <a href="#i_select">select operation</a> on constants.</dd>
2110 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
2111 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
2113 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
2114 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
2116 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
2117 <dd>Perform the <a href="#i_extractelement">extractelement operation</a> on
2120 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
2121 <dd>Perform the <a href="#i_insertelement">insertelement operation</a> on
2124 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
2125 <dd>Perform the <a href="#i_shufflevector">shufflevector operation</a> on
2128 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
2129 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2130 be any of the <a href="#binaryops">binary</a>
2131 or <a href="#bitwiseops">bitwise binary</a> operations. The constraints
2132 on operands are the same as those for the corresponding instruction
2133 (e.g. no bitwise operations on floating point values are allowed).</dd>
2138 <!-- ======================================================================= -->
2139 <div class="doc_subsection"><a name="metadata">Embedded Metadata</a>
2142 <div class="doc_text">
2144 <p>Embedded metadata provides a way to attach arbitrary data to the instruction
2145 stream without affecting the behaviour of the program. There are two
2146 metadata primitives, strings and nodes. All metadata has the
2147 <tt>metadata</tt> type and is identified in syntax by a preceding exclamation
2148 point ('<tt>!</tt>').</p>
2150 <p>A metadata string is a string surrounded by double quotes. It can contain
2151 any character by escaping non-printable characters with "\xx" where "xx" is
2152 the two digit hex code. For example: "<tt>!"test\00"</tt>".</p>
2154 <p>Metadata nodes are represented with notation similar to structure constants
2155 (a comma separated list of elements, surrounded by braces and preceeded by an
2156 exclamation point). For example: "<tt>!{ metadata !"test\00", i32
2159 <p>A metadata node will attempt to track changes to the values it holds. In the
2160 event that a value is deleted, it will be replaced with a typeless
2161 "<tt>null</tt>", such as "<tt>metadata !{null, i32 10}</tt>".</p>
2163 <p>Optimizations may rely on metadata to provide additional information about
2164 the program that isn't available in the instructions, or that isn't easily
2165 computable. Similarly, the code generator may expect a certain metadata
2166 format to be used to express debugging information.</p>
2170 <!-- *********************************************************************** -->
2171 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
2172 <!-- *********************************************************************** -->
2174 <!-- ======================================================================= -->
2175 <div class="doc_subsection">
2176 <a name="inlineasm">Inline Assembler Expressions</a>
2179 <div class="doc_text">
2181 <p>LLVM supports inline assembler expressions (as opposed
2182 to <a href="#moduleasm"> Module-Level Inline Assembly</a>) through the use of
2183 a special value. This value represents the inline assembler as a string
2184 (containing the instructions to emit), a list of operand constraints (stored
2185 as a string), and a flag that indicates whether or not the inline asm
2186 expression has side effects. An example inline assembler expression is:</p>
2188 <div class="doc_code">
2190 i32 (i32) asm "bswap $0", "=r,r"
2194 <p>Inline assembler expressions may <b>only</b> be used as the callee operand of
2195 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we
2198 <div class="doc_code">
2200 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
2204 <p>Inline asms with side effects not visible in the constraint list must be
2205 marked as having side effects. This is done through the use of the
2206 '<tt>sideeffect</tt>' keyword, like so:</p>
2208 <div class="doc_code">
2210 call void asm sideeffect "eieio", ""()
2214 <p>TODO: The format of the asm and constraints string still need to be
2215 documented here. Constraints on what can be done (e.g. duplication, moving,
2216 etc need to be documented). This is probably best done by reference to
2217 another document that covers inline asm from a holistic perspective.</p>
2222 <!-- *********************************************************************** -->
2223 <div class="doc_section">
2224 <a name="intrinsic_globals">Intrinsic Global Variables</a>
2226 <!-- *********************************************************************** -->
2228 <p>LLVM has a number of "magic" global variables that contain data that affect
2229 code generation or other IR semantics. These are documented here. All globals
2230 of this sort should have a section specified as "<tt>llvm.metadata</tt>". This
2231 section and all globals that start with "<tt>llvm.</tt>" are reserved for use
2234 <!-- ======================================================================= -->
2235 <div class="doc_subsection">
2236 <a name="intg_used">The '<tt>llvm.used</tt>' Global Variable</a>
2239 <div class="doc_text">
2241 <p>The <tt>@llvm.used</tt> global is an array with i8* element type which has <a
2242 href="#linkage_appending">appending linkage</a>. This array contains a list of
2243 pointers to global variables and functions which may optionally have a pointer
2244 cast formed of bitcast or getelementptr. For example, a legal use of it is:</p>
2250 @llvm.used = appending global [2 x i8*] [
2252 i8* bitcast (i32* @Y to i8*)
2253 ], section "llvm.metadata"
2256 <p>If a global variable appears in the <tt>@llvm.used</tt> list, then the
2257 compiler, assembler, and linker are required to treat the symbol as if there is
2258 a reference to the global that it cannot see. For example, if a variable has
2259 internal linkage and no references other than that from the <tt>@llvm.used</tt>
2260 list, it cannot be deleted. This is commonly used to represent references from
2261 inline asms and other things the compiler cannot "see", and corresponds to
2262 "attribute((used))" in GNU C.</p>
2264 <p>On some targets, the code generator must emit a directive to the assembler or
2265 object file to prevent the assembler and linker from molesting the symbol.</p>
2269 <!-- ======================================================================= -->
2270 <div class="doc_subsection">
2271 <a name="intg_compiler_used">The '<tt>llvm.compiler.used</tt>' Global Variable</a>
2274 <div class="doc_text">
2276 <p>The <tt>@llvm.compiler.used</tt> directive is the same as the
2277 <tt>@llvm.used</tt> directive, except that it only prevents the compiler from
2278 touching the symbol. On targets that support it, this allows an intelligent
2279 linker to optimize references to the symbol without being impeded as it would be
2280 by <tt>@llvm.used</tt>.</p>
2282 <p>This is a rare construct that should only be used in rare circumstances, and
2283 should not be exposed to source languages.</p>
2287 <!-- ======================================================================= -->
2288 <div class="doc_subsection">
2289 <a name="intg_global_ctors">The '<tt>llvm.global_ctors</tt>' Global Variable</a>
2292 <div class="doc_text">
2294 <p>TODO: Describe this.</p>
2298 <!-- ======================================================================= -->
2299 <div class="doc_subsection">
2300 <a name="intg_global_dtors">The '<tt>llvm.global_dtors</tt>' Global Variable</a>
2303 <div class="doc_text">
2305 <p>TODO: Describe this.</p>
2310 <!-- *********************************************************************** -->
2311 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
2312 <!-- *********************************************************************** -->
2314 <div class="doc_text">
2316 <p>The LLVM instruction set consists of several different classifications of
2317 instructions: <a href="#terminators">terminator
2318 instructions</a>, <a href="#binaryops">binary instructions</a>,
2319 <a href="#bitwiseops">bitwise binary instructions</a>,
2320 <a href="#memoryops">memory instructions</a>, and
2321 <a href="#otherops">other instructions</a>.</p>
2325 <!-- ======================================================================= -->
2326 <div class="doc_subsection"> <a name="terminators">Terminator
2327 Instructions</a> </div>
2329 <div class="doc_text">
2331 <p>As mentioned <a href="#functionstructure">previously</a>, every basic block
2332 in a program ends with a "Terminator" instruction, which indicates which
2333 block should be executed after the current block is finished. These
2334 terminator instructions typically yield a '<tt>void</tt>' value: they produce
2335 control flow, not values (the one exception being the
2336 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
2338 <p>There are six different terminator instructions: the
2339 '<a href="#i_ret"><tt>ret</tt></a>' instruction, the
2340 '<a href="#i_br"><tt>br</tt></a>' instruction, the
2341 '<a href="#i_switch"><tt>switch</tt></a>' instruction, the
2342 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the
2343 '<a href="#i_unwind"><tt>unwind</tt></a>' instruction, and the
2344 '<a href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
2348 <!-- _______________________________________________________________________ -->
2349 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
2350 Instruction</a> </div>
2352 <div class="doc_text">
2356 ret <type> <value> <i>; Return a value from a non-void function</i>
2357 ret void <i>; Return from void function</i>
2361 <p>The '<tt>ret</tt>' instruction is used to return control flow (and optionally
2362 a value) from a function back to the caller.</p>
2364 <p>There are two forms of the '<tt>ret</tt>' instruction: one that returns a
2365 value and then causes control flow, and one that just causes control flow to
2369 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument, the
2370 return value. The type of the return value must be a
2371 '<a href="#t_firstclass">first class</a>' type.</p>
2373 <p>A function is not <a href="#wellformed">well formed</a> if it it has a
2374 non-void return type and contains a '<tt>ret</tt>' instruction with no return
2375 value or a return value with a type that does not match its type, or if it
2376 has a void return type and contains a '<tt>ret</tt>' instruction with a
2380 <p>When the '<tt>ret</tt>' instruction is executed, control flow returns back to
2381 the calling function's context. If the caller is a
2382 "<a href="#i_call"><tt>call</tt></a>" instruction, execution continues at the
2383 instruction after the call. If the caller was an
2384 "<a href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues at
2385 the beginning of the "normal" destination block. If the instruction returns
2386 a value, that value shall set the call or invoke instruction's return
2391 ret i32 5 <i>; Return an integer value of 5</i>
2392 ret void <i>; Return from a void function</i>
2393 ret { i32, i8 } { i32 4, i8 2 } <i>; Return a struct of values 4 and 2</i>
2396 <p>Note that the code generator does not yet fully support large
2397 return values. The specific sizes that are currently supported are
2398 dependent on the target. For integers, on 32-bit targets the limit
2399 is often 64 bits, and on 64-bit targets the limit is often 128 bits.
2400 For aggregate types, the current limits are dependent on the element
2401 types; for example targets are often limited to 2 total integer
2402 elements and 2 total floating-point elements.</p>
2405 <!-- _______________________________________________________________________ -->
2406 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
2408 <div class="doc_text">
2412 br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
2416 <p>The '<tt>br</tt>' instruction is used to cause control flow to transfer to a
2417 different basic block in the current function. There are two forms of this
2418 instruction, corresponding to a conditional branch and an unconditional
2422 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a single
2423 '<tt>i1</tt>' value and two '<tt>label</tt>' values. The unconditional form
2424 of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>' value as a
2428 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
2429 argument is evaluated. If the value is <tt>true</tt>, control flows to the
2430 '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
2431 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
2436 %cond = <a href="#i_icmp">icmp</a> eq i32 %a, %b
2437 br i1 %cond, label %IfEqual, label %IfUnequal
2439 <a href="#i_ret">ret</a> i32 1
2441 <a href="#i_ret">ret</a> i32 0
2446 <!-- _______________________________________________________________________ -->
2447 <div class="doc_subsubsection">
2448 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
2451 <div class="doc_text">
2455 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
2459 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
2460 several different places. It is a generalization of the '<tt>br</tt>'
2461 instruction, allowing a branch to occur to one of many possible
2465 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
2466 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination,
2467 and an array of pairs of comparison value constants and '<tt>label</tt>'s.
2468 The table is not allowed to contain duplicate constant entries.</p>
2471 <p>The <tt>switch</tt> instruction specifies a table of values and
2472 destinations. When the '<tt>switch</tt>' instruction is executed, this table
2473 is searched for the given value. If the value is found, control flow is
2474 transfered to the corresponding destination; otherwise, control flow is
2475 transfered to the default destination.</p>
2477 <h5>Implementation:</h5>
2478 <p>Depending on properties of the target machine and the particular
2479 <tt>switch</tt> instruction, this instruction may be code generated in
2480 different ways. For example, it could be generated as a series of chained
2481 conditional branches or with a lookup table.</p>
2485 <i>; Emulate a conditional br instruction</i>
2486 %Val = <a href="#i_zext">zext</a> i1 %value to i32
2487 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
2489 <i>; Emulate an unconditional br instruction</i>
2490 switch i32 0, label %dest [ ]
2492 <i>; Implement a jump table:</i>
2493 switch i32 %val, label %otherwise [ i32 0, label %onzero
2495 i32 2, label %ontwo ]
2500 <!-- _______________________________________________________________________ -->
2501 <div class="doc_subsubsection">
2502 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
2505 <div class="doc_text">
2509 <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>]
2510 to label <normal label> unwind label <exception label>
2514 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
2515 function, with the possibility of control flow transfer to either the
2516 '<tt>normal</tt>' label or the '<tt>exception</tt>' label. If the callee
2517 function returns with the "<tt><a href="#i_ret">ret</a></tt>" instruction,
2518 control flow will return to the "normal" label. If the callee (or any
2519 indirect callees) returns with the "<a href="#i_unwind"><tt>unwind</tt></a>"
2520 instruction, control is interrupted and continued at the dynamically nearest
2521 "exception" label.</p>
2524 <p>This instruction requires several arguments:</p>
2527 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
2528 convention</a> the call should use. If none is specified, the call
2529 defaults to using C calling conventions.</li>
2531 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
2532 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
2533 '<tt>inreg</tt>' attributes are valid here.</li>
2535 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
2536 function value being invoked. In most cases, this is a direct function
2537 invocation, but indirect <tt>invoke</tt>s are just as possible, branching
2538 off an arbitrary pointer to function value.</li>
2540 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
2541 function to be invoked. </li>
2543 <li>'<tt>function args</tt>': argument list whose types match the function
2544 signature argument types. If the function signature indicates the
2545 function accepts a variable number of arguments, the extra arguments can
2548 <li>'<tt>normal label</tt>': the label reached when the called function
2549 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
2551 <li>'<tt>exception label</tt>': the label reached when a callee returns with
2552 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
2554 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
2555 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
2556 '<tt>readnone</tt>' attributes are valid here.</li>
2560 <p>This instruction is designed to operate as a standard
2561 '<tt><a href="#i_call">call</a></tt>' instruction in most regards. The
2562 primary difference is that it establishes an association with a label, which
2563 is used by the runtime library to unwind the stack.</p>
2565 <p>This instruction is used in languages with destructors to ensure that proper
2566 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
2567 exception. Additionally, this is important for implementation of
2568 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
2570 <p>For the purposes of the SSA form, the definition of the value returned by the
2571 '<tt>invoke</tt>' instruction is deemed to occur on the edge from the current
2572 block to the "normal" label. If the callee unwinds then no return value is
2577 %retval = invoke i32 @Test(i32 15) to label %Continue
2578 unwind label %TestCleanup <i>; {i32}:retval set</i>
2579 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
2580 unwind label %TestCleanup <i>; {i32}:retval set</i>
2585 <!-- _______________________________________________________________________ -->
2587 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
2588 Instruction</a> </div>
2590 <div class="doc_text">
2598 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
2599 at the first callee in the dynamic call stack which used
2600 an <a href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call.
2601 This is primarily used to implement exception handling.</p>
2604 <p>The '<tt>unwind</tt>' instruction causes execution of the current function to
2605 immediately halt. The dynamic call stack is then searched for the
2606 first <a href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack.
2607 Once found, execution continues at the "exceptional" destination block
2608 specified by the <tt>invoke</tt> instruction. If there is no <tt>invoke</tt>
2609 instruction in the dynamic call chain, undefined behavior results.</p>
2613 <!-- _______________________________________________________________________ -->
2615 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
2616 Instruction</a> </div>
2618 <div class="doc_text">
2626 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
2627 instruction is used to inform the optimizer that a particular portion of the
2628 code is not reachable. This can be used to indicate that the code after a
2629 no-return function cannot be reached, and other facts.</p>
2632 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
2636 <!-- ======================================================================= -->
2637 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
2639 <div class="doc_text">
2641 <p>Binary operators are used to do most of the computation in a program. They
2642 require two operands of the same type, execute an operation on them, and
2643 produce a single value. The operands might represent multiple data, as is
2644 the case with the <a href="#t_vector">vector</a> data type. The result value
2645 has the same type as its operands.</p>
2647 <p>There are several different binary operators:</p>
2651 <!-- _______________________________________________________________________ -->
2652 <div class="doc_subsubsection">
2653 <a name="i_add">'<tt>add</tt>' Instruction</a>
2656 <div class="doc_text">
2660 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2661 <result> = nuw add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2662 <result> = nsw add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2663 <result> = nuw nsw add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2667 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
2670 <p>The two arguments to the '<tt>add</tt>' instruction must
2671 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
2672 integer values. Both arguments must have identical types.</p>
2675 <p>The value produced is the integer sum of the two operands.</p>
2677 <p>If the sum has unsigned overflow, the result returned is the mathematical
2678 result modulo 2<sup>n</sup>, where n is the bit width of the result.</p>
2680 <p>Because LLVM integers use a two's complement representation, this instruction
2681 is appropriate for both signed and unsigned integers.</p>
2683 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
2684 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
2685 <tt>nsw</tt> keywords are present, the result value of the <tt>add</tt>
2686 is undefined if unsigned and/or signed overflow, respectively, occurs.</p>
2690 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
2695 <!-- _______________________________________________________________________ -->
2696 <div class="doc_subsubsection">
2697 <a name="i_fadd">'<tt>fadd</tt>' Instruction</a>
2700 <div class="doc_text">
2704 <result> = fadd <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2708 <p>The '<tt>fadd</tt>' instruction returns the sum of its two operands.</p>
2711 <p>The two arguments to the '<tt>fadd</tt>' instruction must be
2712 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
2713 floating point values. Both arguments must have identical types.</p>
2716 <p>The value produced is the floating point sum of the two operands.</p>
2720 <result> = fadd float 4.0, %var <i>; yields {float}:result = 4.0 + %var</i>
2725 <!-- _______________________________________________________________________ -->
2726 <div class="doc_subsubsection">
2727 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
2730 <div class="doc_text">
2734 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2735 <result> = nuw sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2736 <result> = nsw sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2737 <result> = nuw nsw sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2741 <p>The '<tt>sub</tt>' instruction returns the difference of its two
2744 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
2745 '<tt>neg</tt>' instruction present in most other intermediate
2746 representations.</p>
2749 <p>The two arguments to the '<tt>sub</tt>' instruction must
2750 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
2751 integer values. Both arguments must have identical types.</p>
2754 <p>The value produced is the integer difference of the two operands.</p>
2756 <p>If the difference has unsigned overflow, the result returned is the
2757 mathematical result modulo 2<sup>n</sup>, where n is the bit width of the
2760 <p>Because LLVM integers use a two's complement representation, this instruction
2761 is appropriate for both signed and unsigned integers.</p>
2763 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
2764 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
2765 <tt>nsw</tt> keywords are present, the result value of the <tt>sub</tt>
2766 is undefined if unsigned and/or signed overflow, respectively, occurs.</p>
2770 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
2771 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
2776 <!-- _______________________________________________________________________ -->
2777 <div class="doc_subsubsection">
2778 <a name="i_fsub">'<tt>fsub</tt>' Instruction</a>
2781 <div class="doc_text">
2785 <result> = fsub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2789 <p>The '<tt>fsub</tt>' instruction returns the difference of its two
2792 <p>Note that the '<tt>fsub</tt>' instruction is used to represent the
2793 '<tt>fneg</tt>' instruction present in most other intermediate
2794 representations.</p>
2797 <p>The two arguments to the '<tt>fsub</tt>' instruction must be
2798 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
2799 floating point values. Both arguments must have identical types.</p>
2802 <p>The value produced is the floating point difference of the two operands.</p>
2806 <result> = fsub float 4.0, %var <i>; yields {float}:result = 4.0 - %var</i>
2807 <result> = fsub float -0.0, %val <i>; yields {float}:result = -%var</i>
2812 <!-- _______________________________________________________________________ -->
2813 <div class="doc_subsubsection">
2814 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
2817 <div class="doc_text">
2821 <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2822 <result> = nuw mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2823 <result> = nsw mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2824 <result> = nuw nsw mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2828 <p>The '<tt>mul</tt>' instruction returns the product of its two operands.</p>
2831 <p>The two arguments to the '<tt>mul</tt>' instruction must
2832 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
2833 integer values. Both arguments must have identical types.</p>
2836 <p>The value produced is the integer product of the two operands.</p>
2838 <p>If the result of the multiplication has unsigned overflow, the result
2839 returned is the mathematical result modulo 2<sup>n</sup>, where n is the bit
2840 width of the result.</p>
2842 <p>Because LLVM integers use a two's complement representation, and the result
2843 is the same width as the operands, this instruction returns the correct
2844 result for both signed and unsigned integers. If a full product
2845 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands should
2846 be sign-extended or zero-extended as appropriate to the width of the full
2849 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
2850 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
2851 <tt>nsw</tt> keywords are present, the result value of the <tt>mul</tt>
2852 is undefined if unsigned and/or signed overflow, respectively, occurs.</p>
2856 <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
2861 <!-- _______________________________________________________________________ -->
2862 <div class="doc_subsubsection">
2863 <a name="i_fmul">'<tt>fmul</tt>' Instruction</a>
2866 <div class="doc_text">
2870 <result> = fmul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2874 <p>The '<tt>fmul</tt>' instruction returns the product of its two operands.</p>
2877 <p>The two arguments to the '<tt>fmul</tt>' instruction must be
2878 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
2879 floating point values. Both arguments must have identical types.</p>
2882 <p>The value produced is the floating point product of the two operands.</p>
2886 <result> = fmul float 4.0, %var <i>; yields {float}:result = 4.0 * %var</i>
2891 <!-- _______________________________________________________________________ -->
2892 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
2895 <div class="doc_text">
2899 <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2903 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two operands.</p>
2906 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
2907 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2908 values. Both arguments must have identical types.</p>
2911 <p>The value produced is the unsigned integer quotient of the two operands.</p>
2913 <p>Note that unsigned integer division and signed integer division are distinct
2914 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
2916 <p>Division by zero leads to undefined behavior.</p>
2920 <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2925 <!-- _______________________________________________________________________ -->
2926 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
2929 <div class="doc_text">
2933 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2934 <result> = exact sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2938 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two operands.</p>
2941 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
2942 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2943 values. Both arguments must have identical types.</p>
2946 <p>The value produced is the signed integer quotient of the two operands rounded
2949 <p>Note that signed integer division and unsigned integer division are distinct
2950 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
2952 <p>Division by zero leads to undefined behavior. Overflow also leads to
2953 undefined behavior; this is a rare case, but can occur, for example, by doing
2954 a 32-bit division of -2147483648 by -1.</p>
2956 <p>If the <tt>exact</tt> keyword is present, the result value of the
2957 <tt>sdiv</tt> is undefined if the result would be rounded or if overflow
2962 <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2967 <!-- _______________________________________________________________________ -->
2968 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
2969 Instruction</a> </div>
2971 <div class="doc_text">
2975 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2979 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two operands.</p>
2982 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
2983 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
2984 floating point values. Both arguments must have identical types.</p>
2987 <p>The value produced is the floating point quotient of the two operands.</p>
2991 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
2996 <!-- _______________________________________________________________________ -->
2997 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
3000 <div class="doc_text">
3004 <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3008 <p>The '<tt>urem</tt>' instruction returns the remainder from the unsigned
3009 division of its two arguments.</p>
3012 <p>The two arguments to the '<tt>urem</tt>' instruction must be
3013 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3014 values. Both arguments must have identical types.</p>
3017 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
3018 This instruction always performs an unsigned division to get the
3021 <p>Note that unsigned integer remainder and signed integer remainder are
3022 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
3024 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
3028 <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
3033 <!-- _______________________________________________________________________ -->
3034 <div class="doc_subsubsection">
3035 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
3038 <div class="doc_text">
3042 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3046 <p>The '<tt>srem</tt>' instruction returns the remainder from the signed
3047 division of its two operands. This instruction can also take
3048 <a href="#t_vector">vector</a> versions of the values in which case the
3049 elements must be integers.</p>
3052 <p>The two arguments to the '<tt>srem</tt>' instruction must be
3053 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3054 values. Both arguments must have identical types.</p>
3057 <p>This instruction returns the <i>remainder</i> of a division (where the result
3058 has the same sign as the dividend, <tt>op1</tt>), not the <i>modulo</i>
3059 operator (where the result has the same sign as the divisor, <tt>op2</tt>) of
3060 a value. For more information about the difference,
3061 see <a href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
3062 Math Forum</a>. For a table of how this is implemented in various languages,
3063 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
3064 Wikipedia: modulo operation</a>.</p>
3066 <p>Note that signed integer remainder and unsigned integer remainder are
3067 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
3069 <p>Taking the remainder of a division by zero leads to undefined behavior.
3070 Overflow also leads to undefined behavior; this is a rare case, but can
3071 occur, for example, by taking the remainder of a 32-bit division of
3072 -2147483648 by -1. (The remainder doesn't actually overflow, but this rule
3073 lets srem be implemented using instructions that return both the result of
3074 the division and the remainder.)</p>
3078 <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
3083 <!-- _______________________________________________________________________ -->
3084 <div class="doc_subsubsection">
3085 <a name="i_frem">'<tt>frem</tt>' Instruction</a> </div>
3087 <div class="doc_text">
3091 <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3095 <p>The '<tt>frem</tt>' instruction returns the remainder from the division of
3096 its two operands.</p>
3099 <p>The two arguments to the '<tt>frem</tt>' instruction must be
3100 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3101 floating point values. Both arguments must have identical types.</p>
3104 <p>This instruction returns the <i>remainder</i> of a division. The remainder
3105 has the same sign as the dividend.</p>
3109 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
3114 <!-- ======================================================================= -->
3115 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
3116 Operations</a> </div>
3118 <div class="doc_text">
3120 <p>Bitwise binary operators are used to do various forms of bit-twiddling in a
3121 program. They are generally very efficient instructions and can commonly be
3122 strength reduced from other instructions. They require two operands of the
3123 same type, execute an operation on them, and produce a single value. The
3124 resulting value is the same type as its operands.</p>
3128 <!-- _______________________________________________________________________ -->
3129 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
3130 Instruction</a> </div>
3132 <div class="doc_text">
3136 <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3140 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to the left
3141 a specified number of bits.</p>
3144 <p>Both arguments to the '<tt>shl</tt>' instruction must be the
3145 same <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3146 integer type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3149 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod
3150 2<sup>n</sup>, where <tt>n</tt> is the width of the result. If <tt>op2</tt>
3151 is (statically or dynamically) negative or equal to or larger than the number
3152 of bits in <tt>op1</tt>, the result is undefined. If the arguments are
3153 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
3154 shift amount in <tt>op2</tt>.</p>
3158 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
3159 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
3160 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
3161 <result> = shl i32 1, 32 <i>; undefined</i>
3162 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
3167 <!-- _______________________________________________________________________ -->
3168 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
3169 Instruction</a> </div>
3171 <div class="doc_text">
3175 <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3179 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
3180 operand shifted to the right a specified number of bits with zero fill.</p>
3183 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
3184 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3185 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3188 <p>This instruction always performs a logical shift right operation. The most
3189 significant bits of the result will be filled with zero bits after the shift.
3190 If <tt>op2</tt> is (statically or dynamically) equal to or larger than the
3191 number of bits in <tt>op1</tt>, the result is undefined. If the arguments are
3192 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
3193 shift amount in <tt>op2</tt>.</p>
3197 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
3198 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
3199 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
3200 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
3201 <result> = lshr i32 1, 32 <i>; undefined</i>
3202 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
3207 <!-- _______________________________________________________________________ -->
3208 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
3209 Instruction</a> </div>
3210 <div class="doc_text">
3214 <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3218 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
3219 operand shifted to the right a specified number of bits with sign
3223 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
3224 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3225 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3228 <p>This instruction always performs an arithmetic shift right operation, The
3229 most significant bits of the result will be filled with the sign bit
3230 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
3231 larger than the number of bits in <tt>op1</tt>, the result is undefined. If
3232 the arguments are vectors, each vector element of <tt>op1</tt> is shifted by
3233 the corresponding shift amount in <tt>op2</tt>.</p>
3237 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
3238 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
3239 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
3240 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
3241 <result> = ashr i32 1, 32 <i>; undefined</i>
3242 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
3247 <!-- _______________________________________________________________________ -->
3248 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
3249 Instruction</a> </div>
3251 <div class="doc_text">
3255 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3259 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of its two
3263 <p>The two arguments to the '<tt>and</tt>' instruction must be
3264 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3265 values. Both arguments must have identical types.</p>
3268 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
3270 <table border="1" cellspacing="0" cellpadding="4">
3302 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
3303 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
3304 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
3307 <!-- _______________________________________________________________________ -->
3308 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
3310 <div class="doc_text">
3314 <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3318 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive or of its
3322 <p>The two arguments to the '<tt>or</tt>' instruction must be
3323 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3324 values. Both arguments must have identical types.</p>
3327 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
3329 <table border="1" cellspacing="0" cellpadding="4">
3361 <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
3362 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
3363 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
3368 <!-- _______________________________________________________________________ -->
3369 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
3370 Instruction</a> </div>
3372 <div class="doc_text">
3376 <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3380 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive or of
3381 its two operands. The <tt>xor</tt> is used to implement the "one's
3382 complement" operation, which is the "~" operator in C.</p>
3385 <p>The two arguments to the '<tt>xor</tt>' instruction must be
3386 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3387 values. Both arguments must have identical types.</p>
3390 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
3392 <table border="1" cellspacing="0" cellpadding="4">
3424 <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
3425 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
3426 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
3427 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
3432 <!-- ======================================================================= -->
3433 <div class="doc_subsection">
3434 <a name="vectorops">Vector Operations</a>
3437 <div class="doc_text">
3439 <p>LLVM supports several instructions to represent vector operations in a
3440 target-independent manner. These instructions cover the element-access and
3441 vector-specific operations needed to process vectors effectively. While LLVM
3442 does directly support these vector operations, many sophisticated algorithms
3443 will want to use target-specific intrinsics to take full advantage of a
3444 specific target.</p>
3448 <!-- _______________________________________________________________________ -->
3449 <div class="doc_subsubsection">
3450 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
3453 <div class="doc_text">
3457 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
3461 <p>The '<tt>extractelement</tt>' instruction extracts a single scalar element
3462 from a vector at a specified index.</p>
3466 <p>The first operand of an '<tt>extractelement</tt>' instruction is a value
3467 of <a href="#t_vector">vector</a> type. The second operand is an index
3468 indicating the position from which to extract the element. The index may be
3472 <p>The result is a scalar of the same type as the element type of
3473 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
3474 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
3475 results are undefined.</p>
3479 %result = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
3484 <!-- _______________________________________________________________________ -->
3485 <div class="doc_subsubsection">
3486 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
3489 <div class="doc_text">
3493 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
3497 <p>The '<tt>insertelement</tt>' instruction inserts a scalar element into a
3498 vector at a specified index.</p>
3501 <p>The first operand of an '<tt>insertelement</tt>' instruction is a value
3502 of <a href="#t_vector">vector</a> type. The second operand is a scalar value
3503 whose type must equal the element type of the first operand. The third
3504 operand is an index indicating the position at which to insert the value.
3505 The index may be a variable.</p>
3508 <p>The result is a vector of the same type as <tt>val</tt>. Its element values
3509 are those of <tt>val</tt> except at position <tt>idx</tt>, where it gets the
3510 value <tt>elt</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
3511 results are undefined.</p>
3515 %result = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
3520 <!-- _______________________________________________________________________ -->
3521 <div class="doc_subsubsection">
3522 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
3525 <div class="doc_text">
3529 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
3533 <p>The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
3534 from two input vectors, returning a vector with the same element type as the
3535 input and length that is the same as the shuffle mask.</p>
3538 <p>The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
3539 with types that match each other. The third argument is a shuffle mask whose
3540 element type is always 'i32'. The result of the instruction is a vector
3541 whose length is the same as the shuffle mask and whose element type is the
3542 same as the element type of the first two operands.</p>
3544 <p>The shuffle mask operand is required to be a constant vector with either
3545 constant integer or undef values.</p>
3548 <p>The elements of the two input vectors are numbered from left to right across
3549 both of the vectors. The shuffle mask operand specifies, for each element of
3550 the result vector, which element of the two input vectors the result element
3551 gets. The element selector may be undef (meaning "don't care") and the
3552 second operand may be undef if performing a shuffle from only one vector.</p>
3556 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
3557 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
3558 %result = shufflevector <4 x i32> %v1, <4 x i32> undef,
3559 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
3560 %result = shufflevector <8 x i32> %v1, <8 x i32> undef,
3561 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i>
3562 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
3563 <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>
3568 <!-- ======================================================================= -->
3569 <div class="doc_subsection">
3570 <a name="aggregateops">Aggregate Operations</a>
3573 <div class="doc_text">
3575 <p>LLVM supports several instructions for working with aggregate values.</p>
3579 <!-- _______________________________________________________________________ -->
3580 <div class="doc_subsubsection">
3581 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
3584 <div class="doc_text">
3588 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
3592 <p>The '<tt>extractvalue</tt>' instruction extracts the value of a struct field
3593 or array element from an aggregate value.</p>
3596 <p>The first operand of an '<tt>extractvalue</tt>' instruction is a value
3597 of <a href="#t_struct">struct</a> or <a href="#t_array">array</a> type. The
3598 operands are constant indices to specify which value to extract in a similar
3599 manner as indices in a
3600 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
3603 <p>The result is the value at the position in the aggregate specified by the
3608 %result = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
3613 <!-- _______________________________________________________________________ -->
3614 <div class="doc_subsubsection">
3615 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
3618 <div class="doc_text">
3622 <result> = insertvalue <aggregate type> <val>, <ty> <val>, <idx> <i>; yields <n x <ty>></i>
3626 <p>The '<tt>insertvalue</tt>' instruction inserts a value into a struct field or
3627 array element in an aggregate.</p>
3631 <p>The first operand of an '<tt>insertvalue</tt>' instruction is a value
3632 of <a href="#t_struct">struct</a> or <a href="#t_array">array</a> type. The
3633 second operand is a first-class value to insert. The following operands are
3634 constant indices indicating the position at which to insert the value in a
3635 similar manner as indices in a
3636 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction. The
3637 value to insert must have the same type as the value identified by the
3641 <p>The result is an aggregate of the same type as <tt>val</tt>. Its value is
3642 that of <tt>val</tt> except that the value at the position specified by the
3643 indices is that of <tt>elt</tt>.</p>
3647 %result = insertvalue {i32, float} %agg, i32 1, 0 <i>; yields {i32, float}</i>
3653 <!-- ======================================================================= -->
3654 <div class="doc_subsection">
3655 <a name="memoryops">Memory Access and Addressing Operations</a>
3658 <div class="doc_text">
3660 <p>A key design point of an SSA-based representation is how it represents
3661 memory. In LLVM, no memory locations are in SSA form, which makes things
3662 very simple. This section describes how to read, write, allocate, and free
3667 <!-- _______________________________________________________________________ -->
3668 <div class="doc_subsubsection">
3669 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
3672 <div class="doc_text">
3676 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3680 <p>The '<tt>malloc</tt>' instruction allocates memory from the system heap and
3681 returns a pointer to it. The object is always allocated in the generic
3682 address space (address space zero).</p>
3685 <p>The '<tt>malloc</tt>' instruction allocates
3686 <tt>sizeof(<type>)*NumElements</tt> bytes of memory from the operating
3687 system and returns a pointer of the appropriate type to the program. If
3688 "NumElements" is specified, it is the number of elements allocated, otherwise
3689 "NumElements" is defaulted to be one. If a constant alignment is specified,
3690 the value result of the allocation is guaranteed to be aligned to at least
3691 that boundary. If not specified, or if zero, the target can choose to align
3692 the allocation on any convenient boundary compatible with the type.</p>
3694 <p>'<tt>type</tt>' must be a sized type.</p>
3697 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and a
3698 pointer is returned. The result of a zero byte allocation is undefined. The
3699 result is null if there is insufficient memory available.</p>
3703 %array = malloc [4 x i8] <i>; yields {[%4 x i8]*}:array</i>
3705 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
3706 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
3707 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
3708 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
3709 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
3712 <p>Note that the code generator does not yet respect the alignment value.</p>
3716 <!-- _______________________________________________________________________ -->
3717 <div class="doc_subsubsection">
3718 <a name="i_free">'<tt>free</tt>' Instruction</a>
3721 <div class="doc_text">
3725 free <type> <value> <i>; yields {void}</i>
3729 <p>The '<tt>free</tt>' instruction returns memory back to the unused memory heap
3730 to be reallocated in the future.</p>
3733 <p>'<tt>value</tt>' shall be a pointer value that points to a value that was
3734 allocated with the '<tt><a href="#i_malloc">malloc</a></tt>' instruction.</p>
3737 <p>Access to the memory pointed to by the pointer is no longer defined after
3738 this instruction executes. If the pointer is null, the operation is a
3743 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
3744 free [4 x i8]* %array
3749 <!-- _______________________________________________________________________ -->
3750 <div class="doc_subsubsection">
3751 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
3754 <div class="doc_text">
3758 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3762 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
3763 currently executing function, to be automatically released when this function
3764 returns to its caller. The object is always allocated in the generic address
3765 space (address space zero).</p>
3768 <p>The '<tt>alloca</tt>' instruction
3769 allocates <tt>sizeof(<type>)*NumElements</tt> bytes of memory on the
3770 runtime stack, returning a pointer of the appropriate type to the program.
3771 If "NumElements" is specified, it is the number of elements allocated,
3772 otherwise "NumElements" is defaulted to be one. If a constant alignment is
3773 specified, the value result of the allocation is guaranteed to be aligned to
3774 at least that boundary. If not specified, or if zero, the target can choose
3775 to align the allocation on any convenient boundary compatible with the
3778 <p>'<tt>type</tt>' may be any sized type.</p>
3781 <p>Memory is allocated; a pointer is returned. The operation is undefined if
3782 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
3783 memory is automatically released when the function returns. The
3784 '<tt>alloca</tt>' instruction is commonly used to represent automatic
3785 variables that must have an address available. When the function returns
3786 (either with the <tt><a href="#i_ret">ret</a></tt>
3787 or <tt><a href="#i_unwind">unwind</a></tt> instructions), the memory is
3788 reclaimed. Allocating zero bytes is legal, but the result is undefined.</p>
3792 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
3793 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
3794 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
3795 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
3800 <!-- _______________________________________________________________________ -->
3801 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
3802 Instruction</a> </div>
3804 <div class="doc_text">
3808 <result> = load <ty>* <pointer>[, align <alignment>]
3809 <result> = volatile load <ty>* <pointer>[, align <alignment>]
3813 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
3816 <p>The argument to the '<tt>load</tt>' instruction specifies the memory address
3817 from which to load. The pointer must point to
3818 a <a href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
3819 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify the
3820 number or order of execution of this <tt>load</tt> with other
3821 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
3824 <p>The optional constant "align" argument specifies the alignment of the
3825 operation (that is, the alignment of the memory address). A value of 0 or an
3826 omitted "align" argument means that the operation has the preferential
3827 alignment for the target. It is the responsibility of the code emitter to
3828 ensure that the alignment information is correct. Overestimating the
3829 alignment results in an undefined behavior. Underestimating the alignment may
3830 produce less efficient code. An alignment of 1 is always safe.</p>
3833 <p>The location of memory pointed to is loaded. If the value being loaded is of
3834 scalar type then the number of bytes read does not exceed the minimum number
3835 of bytes needed to hold all bits of the type. For example, loading an
3836 <tt>i24</tt> reads at most three bytes. When loading a value of a type like
3837 <tt>i20</tt> with a size that is not an integral number of bytes, the result
3838 is undefined if the value was not originally written using a store of the
3843 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3844 <a href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
3845 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
3850 <!-- _______________________________________________________________________ -->
3851 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
3852 Instruction</a> </div>
3854 <div class="doc_text">
3858 store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3859 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3863 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
3866 <p>There are two arguments to the '<tt>store</tt>' instruction: a value to store
3867 and an address at which to store it. The type of the
3868 '<tt><pointer></tt>' operand must be a pointer to
3869 the <a href="#t_firstclass">first class</a> type of the
3870 '<tt><value></tt>' operand. If the <tt>store</tt> is marked
3871 as <tt>volatile</tt>, then the optimizer is not allowed to modify the number
3872 or order of execution of this <tt>store</tt> with other
3873 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
3876 <p>The optional constant "align" argument specifies the alignment of the
3877 operation (that is, the alignment of the memory address). A value of 0 or an
3878 omitted "align" argument means that the operation has the preferential
3879 alignment for the target. It is the responsibility of the code emitter to
3880 ensure that the alignment information is correct. Overestimating the
3881 alignment results in an undefined behavior. Underestimating the alignment may
3882 produce less efficient code. An alignment of 1 is always safe.</p>
3885 <p>The contents of memory are updated to contain '<tt><value></tt>' at the
3886 location specified by the '<tt><pointer></tt>' operand. If
3887 '<tt><value></tt>' is of scalar type then the number of bytes written
3888 does not exceed the minimum number of bytes needed to hold all bits of the
3889 type. For example, storing an <tt>i24</tt> writes at most three bytes. When
3890 writing a value of a type like <tt>i20</tt> with a size that is not an
3891 integral number of bytes, it is unspecified what happens to the extra bits
3892 that do not belong to the type, but they will typically be overwritten.</p>
3896 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3897 store i32 3, i32* %ptr <i>; yields {void}</i>
3898 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
3903 <!-- _______________________________________________________________________ -->
3904 <div class="doc_subsubsection">
3905 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
3908 <div class="doc_text">
3912 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
3913 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
3917 <p>The '<tt>getelementptr</tt>' instruction is used to get the address of a
3918 subelement of an aggregate data structure. It performs address calculation
3919 only and does not access memory.</p>
3922 <p>The first argument is always a pointer, and forms the basis of the
3923 calculation. The remaining arguments are indices that indicate which of the
3924 elements of the aggregate object are indexed. The interpretation of each
3925 index is dependent on the type being indexed into. The first index always
3926 indexes the pointer value given as the first argument, the second index
3927 indexes a value of the type pointed to (not necessarily the value directly
3928 pointed to, since the first index can be non-zero), etc. The first type
3929 indexed into must be a pointer value, subsequent types can be arrays, vectors
3930 and structs. Note that subsequent types being indexed into can never be
3931 pointers, since that would require loading the pointer before continuing
3934 <p>The type of each index argument depends on the type it is indexing into.
3935 When indexing into a (optionally packed) structure, only <tt>i32</tt> integer
3936 <b>constants</b> are allowed. When indexing into an array, pointer or
3937 vector, integers of any width are allowed, and they are not required to be
3940 <p>For example, let's consider a C code fragment and how it gets compiled to
3943 <div class="doc_code">
3956 int *foo(struct ST *s) {
3957 return &s[1].Z.B[5][13];
3962 <p>The LLVM code generated by the GCC frontend is:</p>
3964 <div class="doc_code">
3966 %RT = <a href="#namedtypes">type</a> { i8 , [10 x [20 x i32]], i8 }
3967 %ST = <a href="#namedtypes">type</a> { i32, double, %RT }
3969 define i32* @foo(%ST* %s) {
3971 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
3978 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
3979 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
3980 }</tt>' type, a structure. The second index indexes into the third element
3981 of the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
3982 i8 }</tt>' type, another structure. The third index indexes into the second
3983 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
3984 array. The two dimensions of the array are subscripted into, yielding an
3985 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a
3986 pointer to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
3988 <p>Note that it is perfectly legal to index partially through a structure,
3989 returning a pointer to an inner element. Because of this, the LLVM code for
3990 the given testcase is equivalent to:</p>
3993 define i32* @foo(%ST* %s) {
3994 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
3995 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
3996 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
3997 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
3998 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
4003 <p>If the <tt>inbounds</tt> keyword is present, the result value of the
4004 <tt>getelementptr</tt> is undefined if the base pointer is not an
4005 <i>in bounds</i> address of an allocated object, or if any of the addresses
4006 formed by successive addition of the offsets implied by the indices to
4007 the base address are not an <i>in bounds</i> address of that allocated
4009 The <i>in bounds</i> addresses for an allocated object are all the addresses
4010 that point into the object, plus the address one past the end.</p>
4012 <p>If the <tt>inbounds</tt> keyword is not present, the offsets are added to
4013 the base address with silently-wrapping two's complement arithmetic, and
4014 the result value of the <tt>getelementptr</tt> may be outside the object
4015 pointed to by the base pointer. The result value may not necessarily be
4016 used to access memory though, even if it happens to point into allocated
4017 storage. See the <a href="#pointeraliasing">Pointer Aliasing Rules</a>
4018 section for more information.</p>
4020 <p>The getelementptr instruction is often confusing. For some more insight into
4021 how it works, see <a href="GetElementPtr.html">the getelementptr FAQ</a>.</p>
4025 <i>; yields [12 x i8]*:aptr</i>
4026 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
4027 <i>; yields i8*:vptr</i>
4028 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
4029 <i>; yields i8*:eptr</i>
4030 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
4031 <i>; yields i32*:iptr</i>
4032 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
4037 <!-- ======================================================================= -->
4038 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
4041 <div class="doc_text">
4043 <p>The instructions in this category are the conversion instructions (casting)
4044 which all take a single operand and a type. They perform various bit
4045 conversions on the operand.</p>
4049 <!-- _______________________________________________________________________ -->
4050 <div class="doc_subsubsection">
4051 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
4053 <div class="doc_text">
4057 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
4061 <p>The '<tt>trunc</tt>' instruction truncates its operand to the
4062 type <tt>ty2</tt>.</p>
4065 <p>The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
4066 be an <a href="#t_integer">integer</a> type, and a type that specifies the
4067 size and type of the result, which must be
4068 an <a href="#t_integer">integer</a> type. The bit size of <tt>value</tt> must
4069 be larger than the bit size of <tt>ty2</tt>. Equal sized types are not
4073 <p>The '<tt>trunc</tt>' instruction truncates the high order bits
4074 in <tt>value</tt> and converts the remaining bits to <tt>ty2</tt>. Since the
4075 source size must be larger than the destination size, <tt>trunc</tt> cannot
4076 be a <i>no-op cast</i>. It will always truncate bits.</p>
4080 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
4081 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
4082 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
4087 <!-- _______________________________________________________________________ -->
4088 <div class="doc_subsubsection">
4089 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
4091 <div class="doc_text">
4095 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
4099 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
4104 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
4105 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
4106 also be of <a href="#t_integer">integer</a> type. The bit size of the
4107 <tt>value</tt> must be smaller than the bit size of the destination type,
4111 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
4112 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
4114 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
4118 %X = zext i32 257 to i64 <i>; yields i64:257</i>
4119 %Y = zext i1 true to i32 <i>; yields i32:1</i>
4124 <!-- _______________________________________________________________________ -->
4125 <div class="doc_subsubsection">
4126 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
4128 <div class="doc_text">
4132 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
4136 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
4139 <p>The '<tt>sext</tt>' instruction takes a value to cast, which must be of
4140 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
4141 also be of <a href="#t_integer">integer</a> type. The bit size of the
4142 <tt>value</tt> must be smaller than the bit size of the destination type,
4146 <p>The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
4147 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size
4148 of the type <tt>ty2</tt>.</p>
4150 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
4154 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
4155 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
4160 <!-- _______________________________________________________________________ -->
4161 <div class="doc_subsubsection">
4162 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
4165 <div class="doc_text">
4169 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
4173 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
4177 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
4178 point</a> value to cast and a <a href="#t_floating">floating point</a> type
4179 to cast it to. The size of <tt>value</tt> must be larger than the size of
4180 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
4181 <i>no-op cast</i>.</p>
4184 <p>The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
4185 <a href="#t_floating">floating point</a> type to a smaller
4186 <a href="#t_floating">floating point</a> type. If the value cannot fit
4187 within the destination type, <tt>ty2</tt>, then the results are
4192 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
4193 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
4198 <!-- _______________________________________________________________________ -->
4199 <div class="doc_subsubsection">
4200 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
4202 <div class="doc_text">
4206 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
4210 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
4211 floating point value.</p>
4214 <p>The '<tt>fpext</tt>' instruction takes a
4215 <a href="#t_floating">floating point</a> <tt>value</tt> to cast, and
4216 a <a href="#t_floating">floating point</a> type to cast it to. The source
4217 type must be smaller than the destination type.</p>
4220 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
4221 <a href="#t_floating">floating point</a> type to a larger
4222 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
4223 used to make a <i>no-op cast</i> because it always changes bits. Use
4224 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
4228 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
4229 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
4234 <!-- _______________________________________________________________________ -->
4235 <div class="doc_subsubsection">
4236 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
4238 <div class="doc_text">
4242 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
4246 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
4247 unsigned integer equivalent of type <tt>ty2</tt>.</p>
4250 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
4251 scalar or vector <a href="#t_floating">floating point</a> value, and a type
4252 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
4253 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
4254 vector integer type with the same number of elements as <tt>ty</tt></p>
4257 <p>The '<tt>fptoui</tt>' instruction converts its
4258 <a href="#t_floating">floating point</a> operand into the nearest (rounding
4259 towards zero) unsigned integer value. If the value cannot fit
4260 in <tt>ty2</tt>, the results are undefined.</p>
4264 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
4265 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
4266 %X = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
4271 <!-- _______________________________________________________________________ -->
4272 <div class="doc_subsubsection">
4273 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
4275 <div class="doc_text">
4279 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
4283 <p>The '<tt>fptosi</tt>' instruction converts
4284 <a href="#t_floating">floating point</a> <tt>value</tt> to
4285 type <tt>ty2</tt>.</p>
4288 <p>The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
4289 scalar or vector <a href="#t_floating">floating point</a> value, and a type
4290 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
4291 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
4292 vector integer type with the same number of elements as <tt>ty</tt></p>
4295 <p>The '<tt>fptosi</tt>' instruction converts its
4296 <a href="#t_floating">floating point</a> operand into the nearest (rounding
4297 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
4298 the results are undefined.</p>
4302 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
4303 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
4304 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
4309 <!-- _______________________________________________________________________ -->
4310 <div class="doc_subsubsection">
4311 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
4313 <div class="doc_text">
4317 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
4321 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
4322 integer and converts that value to the <tt>ty2</tt> type.</p>
4325 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
4326 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
4327 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
4328 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
4329 floating point type with the same number of elements as <tt>ty</tt></p>
4332 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
4333 integer quantity and converts it to the corresponding floating point
4334 value. If the value cannot fit in the floating point value, the results are
4339 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
4340 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
4345 <!-- _______________________________________________________________________ -->
4346 <div class="doc_subsubsection">
4347 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
4349 <div class="doc_text">
4353 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
4357 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed integer
4358 and converts that value to the <tt>ty2</tt> type.</p>
4361 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
4362 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
4363 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
4364 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
4365 floating point type with the same number of elements as <tt>ty</tt></p>
4368 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed integer
4369 quantity and converts it to the corresponding floating point value. If the
4370 value cannot fit in the floating point value, the results are undefined.</p>
4374 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
4375 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
4380 <!-- _______________________________________________________________________ -->
4381 <div class="doc_subsubsection">
4382 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
4384 <div class="doc_text">
4388 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
4392 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
4393 the integer type <tt>ty2</tt>.</p>
4396 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
4397 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
4398 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.</p>
4401 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
4402 <tt>ty2</tt> by interpreting the pointer value as an integer and either
4403 truncating or zero extending that value to the size of the integer type. If
4404 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
4405 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
4406 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
4411 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
4412 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
4417 <!-- _______________________________________________________________________ -->
4418 <div class="doc_subsubsection">
4419 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
4421 <div class="doc_text">
4425 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
4429 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to a
4430 pointer type, <tt>ty2</tt>.</p>
4433 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
4434 value to cast, and a type to cast it to, which must be a
4435 <a href="#t_pointer">pointer</a> type.</p>
4438 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
4439 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
4440 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
4441 size of a pointer then a truncation is done. If <tt>value</tt> is smaller
4442 than the size of a pointer then a zero extension is done. If they are the
4443 same size, nothing is done (<i>no-op cast</i>).</p>
4447 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
4448 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
4449 %Y = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
4454 <!-- _______________________________________________________________________ -->
4455 <div class="doc_subsubsection">
4456 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
4458 <div class="doc_text">
4462 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
4466 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
4467 <tt>ty2</tt> without changing any bits.</p>
4470 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be a
4471 non-aggregate first class value, and a type to cast it to, which must also be
4472 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes
4473 of <tt>value</tt> and the destination type, <tt>ty2</tt>, must be
4474 identical. If the source type is a pointer, the destination type must also be
4475 a pointer. This instruction supports bitwise conversion of vectors to
4476 integers and to vectors of other types (as long as they have the same
4480 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
4481 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
4482 this conversion. The conversion is done as if the <tt>value</tt> had been
4483 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only
4484 be converted to other pointer types with this instruction. To convert
4485 pointers to other types, use the <a href="#i_inttoptr">inttoptr</a> or
4486 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
4490 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
4491 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
4492 %Z = bitcast <2 x int> %V to i64; <i>; yields i64: %V</i>
4497 <!-- ======================================================================= -->
4498 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
4500 <div class="doc_text">
4502 <p>The instructions in this category are the "miscellaneous" instructions, which
4503 defy better classification.</p>
4507 <!-- _______________________________________________________________________ -->
4508 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
4511 <div class="doc_text">
4515 <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4519 <p>The '<tt>icmp</tt>' instruction returns a boolean value or a vector of
4520 boolean values based on comparison of its two integer, integer vector, or
4521 pointer operands.</p>
4524 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
4525 the condition code indicating the kind of comparison to perform. It is not a
4526 value, just a keyword. The possible condition code are:</p>
4529 <li><tt>eq</tt>: equal</li>
4530 <li><tt>ne</tt>: not equal </li>
4531 <li><tt>ugt</tt>: unsigned greater than</li>
4532 <li><tt>uge</tt>: unsigned greater or equal</li>
4533 <li><tt>ult</tt>: unsigned less than</li>
4534 <li><tt>ule</tt>: unsigned less or equal</li>
4535 <li><tt>sgt</tt>: signed greater than</li>
4536 <li><tt>sge</tt>: signed greater or equal</li>
4537 <li><tt>slt</tt>: signed less than</li>
4538 <li><tt>sle</tt>: signed less or equal</li>
4541 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
4542 <a href="#t_pointer">pointer</a> or integer <a href="#t_vector">vector</a>
4543 typed. They must also be identical types.</p>
4546 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to the
4547 condition code given as <tt>cond</tt>. The comparison performed always yields
4548 either an <a href="#t_primitive"><tt>i1</tt></a> or vector of <tt>i1</tt>
4549 result, as follows:</p>
4552 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
4553 <tt>false</tt> otherwise. No sign interpretation is necessary or
4556 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
4557 <tt>false</tt> otherwise. No sign interpretation is necessary or
4560 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
4561 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4563 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
4564 <tt>true</tt> if <tt>op1</tt> is greater than or equal
4565 to <tt>op2</tt>.</li>
4567 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
4568 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
4570 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
4571 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4573 <li><tt>sgt</tt>: interprets the operands as signed values and yields
4574 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4576 <li><tt>sge</tt>: interprets the operands as signed values and yields
4577 <tt>true</tt> if <tt>op1</tt> is greater than or equal
4578 to <tt>op2</tt>.</li>
4580 <li><tt>slt</tt>: interprets the operands as signed values and yields
4581 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
4583 <li><tt>sle</tt>: interprets the operands as signed values and yields
4584 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4587 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
4588 values are compared as if they were integers.</p>
4590 <p>If the operands are integer vectors, then they are compared element by
4591 element. The result is an <tt>i1</tt> vector with the same number of elements
4592 as the values being compared. Otherwise, the result is an <tt>i1</tt>.</p>
4596 <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
4597 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
4598 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
4599 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
4600 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
4601 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
4604 <p>Note that the code generator does not yet support vector types with
4605 the <tt>icmp</tt> instruction.</p>
4609 <!-- _______________________________________________________________________ -->
4610 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
4613 <div class="doc_text">
4617 <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4621 <p>The '<tt>fcmp</tt>' instruction returns a boolean value or vector of boolean
4622 values based on comparison of its operands.</p>
4624 <p>If the operands are floating point scalars, then the result type is a boolean
4625 (<a href="#t_primitive"><tt>i1</tt></a>).</p>
4627 <p>If the operands are floating point vectors, then the result type is a vector
4628 of boolean with the same number of elements as the operands being
4632 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
4633 the condition code indicating the kind of comparison to perform. It is not a
4634 value, just a keyword. The possible condition code are:</p>
4637 <li><tt>false</tt>: no comparison, always returns false</li>
4638 <li><tt>oeq</tt>: ordered and equal</li>
4639 <li><tt>ogt</tt>: ordered and greater than </li>
4640 <li><tt>oge</tt>: ordered and greater than or equal</li>
4641 <li><tt>olt</tt>: ordered and less than </li>
4642 <li><tt>ole</tt>: ordered and less than or equal</li>
4643 <li><tt>one</tt>: ordered and not equal</li>
4644 <li><tt>ord</tt>: ordered (no nans)</li>
4645 <li><tt>ueq</tt>: unordered or equal</li>
4646 <li><tt>ugt</tt>: unordered or greater than </li>
4647 <li><tt>uge</tt>: unordered or greater than or equal</li>
4648 <li><tt>ult</tt>: unordered or less than </li>
4649 <li><tt>ule</tt>: unordered or less than or equal</li>
4650 <li><tt>une</tt>: unordered or not equal</li>
4651 <li><tt>uno</tt>: unordered (either nans)</li>
4652 <li><tt>true</tt>: no comparison, always returns true</li>
4655 <p><i>Ordered</i> means that neither operand is a QNAN while
4656 <i>unordered</i> means that either operand may be a QNAN.</p>
4658 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be either
4659 a <a href="#t_floating">floating point</a> type or
4660 a <a href="#t_vector">vector</a> of floating point type. They must have
4661 identical types.</p>
4664 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4665 according to the condition code given as <tt>cond</tt>. If the operands are
4666 vectors, then the vectors are compared element by element. Each comparison
4667 performed always yields an <a href="#t_primitive">i1</a> result, as
4671 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
4673 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4674 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
4676 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4677 <tt>op1</tt> is greather than <tt>op2</tt>.</li>
4679 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4680 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4682 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4683 <tt>op1</tt> is less than <tt>op2</tt>.</li>
4685 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4686 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4688 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4689 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
4691 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
4693 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
4694 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
4696 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
4697 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4699 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
4700 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4702 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
4703 <tt>op1</tt> is less than <tt>op2</tt>.</li>
4705 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
4706 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4708 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
4709 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
4711 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
4713 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
4718 <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
4719 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
4720 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
4721 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
4724 <p>Note that the code generator does not yet support vector types with
4725 the <tt>fcmp</tt> instruction.</p>
4729 <!-- _______________________________________________________________________ -->
4730 <div class="doc_subsubsection">
4731 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
4734 <div class="doc_text">
4738 <result> = phi <ty> [ <val0>, <label0>], ...
4742 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in the
4743 SSA graph representing the function.</p>
4746 <p>The type of the incoming values is specified with the first type field. After
4747 this, the '<tt>phi</tt>' instruction takes a list of pairs as arguments, with
4748 one pair for each predecessor basic block of the current block. Only values
4749 of <a href="#t_firstclass">first class</a> type may be used as the value
4750 arguments to the PHI node. Only labels may be used as the label
4753 <p>There must be no non-phi instructions between the start of a basic block and
4754 the PHI instructions: i.e. PHI instructions must be first in a basic
4757 <p>For the purposes of the SSA form, the use of each incoming value is deemed to
4758 occur on the edge from the corresponding predecessor block to the current
4759 block (but after any definition of an '<tt>invoke</tt>' instruction's return
4760 value on the same edge).</p>
4763 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
4764 specified by the pair corresponding to the predecessor basic block that
4765 executed just prior to the current block.</p>
4769 Loop: ; Infinite loop that counts from 0 on up...
4770 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
4771 %nextindvar = add i32 %indvar, 1
4777 <!-- _______________________________________________________________________ -->
4778 <div class="doc_subsubsection">
4779 <a name="i_select">'<tt>select</tt>' Instruction</a>
4782 <div class="doc_text">
4786 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
4788 <i>selty</i> is either i1 or {<N x i1>}
4792 <p>The '<tt>select</tt>' instruction is used to choose one value based on a
4793 condition, without branching.</p>
4797 <p>The '<tt>select</tt>' instruction requires an 'i1' value or a vector of 'i1'
4798 values indicating the condition, and two values of the
4799 same <a href="#t_firstclass">first class</a> type. If the val1/val2 are
4800 vectors and the condition is a scalar, then entire vectors are selected, not
4801 individual elements.</p>
4804 <p>If the condition is an i1 and it evaluates to 1, the instruction returns the
4805 first value argument; otherwise, it returns the second value argument.</p>
4807 <p>If the condition is a vector of i1, then the value arguments must be vectors
4808 of the same size, and the selection is done element by element.</p>
4812 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
4815 <p>Note that the code generator does not yet support conditions
4816 with vector type.</p>
4820 <!-- _______________________________________________________________________ -->
4821 <div class="doc_subsubsection">
4822 <a name="i_call">'<tt>call</tt>' Instruction</a>
4825 <div class="doc_text">
4829 <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>]
4833 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
4836 <p>This instruction requires several arguments:</p>
4839 <li>The optional "tail" marker indicates whether the callee function accesses
4840 any allocas or varargs in the caller. If the "tail" marker is present,
4841 the function call is eligible for tail call optimization. Note that calls
4842 may be marked "tail" even if they do not occur before
4843 a <a href="#i_ret"><tt>ret</tt></a> instruction.</li>
4845 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
4846 convention</a> the call should use. If none is specified, the call
4847 defaults to using C calling conventions.</li>
4849 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
4850 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
4851 '<tt>inreg</tt>' attributes are valid here.</li>
4853 <li>'<tt>ty</tt>': the type of the call instruction itself which is also the
4854 type of the return value. Functions that return no value are marked
4855 <tt><a href="#t_void">void</a></tt>.</li>
4857 <li>'<tt>fnty</tt>': shall be the signature of the pointer to function value
4858 being invoked. The argument types must match the types implied by this
4859 signature. This type can be omitted if the function is not varargs and if
4860 the function type does not return a pointer to a function.</li>
4862 <li>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
4863 be invoked. In most cases, this is a direct function invocation, but
4864 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
4865 to function value.</li>
4867 <li>'<tt>function args</tt>': argument list whose types match the function
4868 signature argument types. All arguments must be of
4869 <a href="#t_firstclass">first class</a> type. If the function signature
4870 indicates the function accepts a variable number of arguments, the extra
4871 arguments can be specified.</li>
4873 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
4874 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
4875 '<tt>readnone</tt>' attributes are valid here.</li>
4879 <p>The '<tt>call</tt>' instruction is used to cause control flow to transfer to
4880 a specified function, with its incoming arguments bound to the specified
4881 values. Upon a '<tt><a href="#i_ret">ret</a></tt>' instruction in the called
4882 function, control flow continues with the instruction after the function
4883 call, and the return value of the function is bound to the result
4888 %retval = call i32 @test(i32 %argc)
4889 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42) <i>; yields i32</i>
4890 %X = tail call i32 @foo() <i>; yields i32</i>
4891 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
4892 call void %foo(i8 97 signext)
4894 %struct.A = type { i32, i8 }
4895 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
4896 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
4897 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
4898 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
4899 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
4904 <!-- _______________________________________________________________________ -->
4905 <div class="doc_subsubsection">
4906 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
4909 <div class="doc_text">
4913 <resultval> = va_arg <va_list*> <arglist>, <argty>
4917 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
4918 the "variable argument" area of a function call. It is used to implement the
4919 <tt>va_arg</tt> macro in C.</p>
4922 <p>This instruction takes a <tt>va_list*</tt> value and the type of the
4923 argument. It returns a value of the specified argument type and increments
4924 the <tt>va_list</tt> to point to the next argument. The actual type
4925 of <tt>va_list</tt> is target specific.</p>
4928 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified type
4929 from the specified <tt>va_list</tt> and causes the <tt>va_list</tt> to point
4930 to the next argument. For more information, see the variable argument
4931 handling <a href="#int_varargs">Intrinsic Functions</a>.</p>
4933 <p>It is legal for this instruction to be called in a function which does not
4934 take a variable number of arguments, for example, the <tt>vfprintf</tt>
4937 <p><tt>va_arg</tt> is an LLVM instruction instead of
4938 an <a href="#intrinsics">intrinsic function</a> because it takes a type as an
4942 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
4944 <p>Note that the code generator does not yet fully support va_arg on many
4945 targets. Also, it does not currently support va_arg with aggregate types on
4950 <!-- *********************************************************************** -->
4951 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
4952 <!-- *********************************************************************** -->
4954 <div class="doc_text">
4956 <p>LLVM supports the notion of an "intrinsic function". These functions have
4957 well known names and semantics and are required to follow certain
4958 restrictions. Overall, these intrinsics represent an extension mechanism for
4959 the LLVM language that does not require changing all of the transformations
4960 in LLVM when adding to the language (or the bitcode reader/writer, the
4961 parser, etc...).</p>
4963 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
4964 prefix is reserved in LLVM for intrinsic names; thus, function names may not
4965 begin with this prefix. Intrinsic functions must always be external
4966 functions: you cannot define the body of intrinsic functions. Intrinsic
4967 functions may only be used in call or invoke instructions: it is illegal to
4968 take the address of an intrinsic function. Additionally, because intrinsic
4969 functions are part of the LLVM language, it is required if any are added that
4970 they be documented here.</p>
4972 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents a
4973 family of functions that perform the same operation but on different data
4974 types. Because LLVM can represent over 8 million different integer types,
4975 overloading is used commonly to allow an intrinsic function to operate on any
4976 integer type. One or more of the argument types or the result type can be
4977 overloaded to accept any integer type. Argument types may also be defined as
4978 exactly matching a previous argument's type or the result type. This allows
4979 an intrinsic function which accepts multiple arguments, but needs all of them
4980 to be of the same type, to only be overloaded with respect to a single
4981 argument or the result.</p>
4983 <p>Overloaded intrinsics will have the names of its overloaded argument types
4984 encoded into its function name, each preceded by a period. Only those types
4985 which are overloaded result in a name suffix. Arguments whose type is matched
4986 against another type do not. For example, the <tt>llvm.ctpop</tt> function
4987 can take an integer of any width and returns an integer of exactly the same
4988 integer width. This leads to a family of functions such as
4989 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29
4990 %val)</tt>. Only one type, the return type, is overloaded, and only one type
4991 suffix is required. Because the argument's type is matched against the return
4992 type, it does not require its own name suffix.</p>
4994 <p>To learn how to add an intrinsic function, please see the
4995 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.</p>
4999 <!-- ======================================================================= -->
5000 <div class="doc_subsection">
5001 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
5004 <div class="doc_text">
5006 <p>Variable argument support is defined in LLVM with
5007 the <a href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
5008 intrinsic functions. These functions are related to the similarly named
5009 macros defined in the <tt><stdarg.h></tt> header file.</p>
5011 <p>All of these functions operate on arguments that use a target-specific value
5012 type "<tt>va_list</tt>". The LLVM assembly language reference manual does
5013 not define what this type is, so all transformations should be prepared to
5014 handle these functions regardless of the type used.</p>
5016 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
5017 instruction and the variable argument handling intrinsic functions are
5020 <div class="doc_code">
5022 define i32 @test(i32 %X, ...) {
5023 ; Initialize variable argument processing
5025 %ap2 = bitcast i8** %ap to i8*
5026 call void @llvm.va_start(i8* %ap2)
5028 ; Read a single integer argument
5029 %tmp = va_arg i8** %ap, i32
5031 ; Demonstrate usage of llvm.va_copy and llvm.va_end
5033 %aq2 = bitcast i8** %aq to i8*
5034 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
5035 call void @llvm.va_end(i8* %aq2)
5037 ; Stop processing of arguments.
5038 call void @llvm.va_end(i8* %ap2)
5042 declare void @llvm.va_start(i8*)
5043 declare void @llvm.va_copy(i8*, i8*)
5044 declare void @llvm.va_end(i8*)
5050 <!-- _______________________________________________________________________ -->
5051 <div class="doc_subsubsection">
5052 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
5056 <div class="doc_text">
5060 declare void %llvm.va_start(i8* <arglist>)
5064 <p>The '<tt>llvm.va_start</tt>' intrinsic initializes <tt>*<arglist></tt>
5065 for subsequent use by <tt><a href="#i_va_arg">va_arg</a></tt>.</p>
5068 <p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
5071 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
5072 macro available in C. In a target-dependent way, it initializes
5073 the <tt>va_list</tt> element to which the argument points, so that the next
5074 call to <tt>va_arg</tt> will produce the first variable argument passed to
5075 the function. Unlike the C <tt>va_start</tt> macro, this intrinsic does not
5076 need to know the last argument of the function as the compiler can figure
5081 <!-- _______________________________________________________________________ -->
5082 <div class="doc_subsubsection">
5083 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
5086 <div class="doc_text">
5090 declare void @llvm.va_end(i8* <arglist>)
5094 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
5095 which has been initialized previously
5096 with <tt><a href="#int_va_start">llvm.va_start</a></tt>
5097 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
5100 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
5103 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
5104 macro available in C. In a target-dependent way, it destroys
5105 the <tt>va_list</tt> element to which the argument points. Calls
5106 to <a href="#int_va_start"><tt>llvm.va_start</tt></a>
5107 and <a href="#int_va_copy"> <tt>llvm.va_copy</tt></a> must be matched exactly
5108 with calls to <tt>llvm.va_end</tt>.</p>
5112 <!-- _______________________________________________________________________ -->
5113 <div class="doc_subsubsection">
5114 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
5117 <div class="doc_text">
5121 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
5125 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
5126 from the source argument list to the destination argument list.</p>
5129 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
5130 The second argument is a pointer to a <tt>va_list</tt> element to copy
5134 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
5135 macro available in C. In a target-dependent way, it copies the
5136 source <tt>va_list</tt> element into the destination <tt>va_list</tt>
5137 element. This intrinsic is necessary because
5138 the <tt><a href="#int_va_start"> llvm.va_start</a></tt> intrinsic may be
5139 arbitrarily complex and require, for example, memory allocation.</p>
5143 <!-- ======================================================================= -->
5144 <div class="doc_subsection">
5145 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
5148 <div class="doc_text">
5150 <p>LLVM support for <a href="GarbageCollection.html">Accurate Garbage
5151 Collection</a> (GC) requires the implementation and generation of these
5152 intrinsics. These intrinsics allow identification of <a href="#int_gcroot">GC
5153 roots on the stack</a>, as well as garbage collector implementations that
5154 require <a href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a>
5155 barriers. Front-ends for type-safe garbage collected languages should generate
5156 these intrinsics to make use of the LLVM garbage collectors. For more details,
5157 see <a href="GarbageCollection.html">Accurate Garbage Collection with
5160 <p>The garbage collection intrinsics only operate on objects in the generic
5161 address space (address space zero).</p>
5165 <!-- _______________________________________________________________________ -->
5166 <div class="doc_subsubsection">
5167 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
5170 <div class="doc_text">
5174 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
5178 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
5179 the code generator, and allows some metadata to be associated with it.</p>
5182 <p>The first argument specifies the address of a stack object that contains the
5183 root pointer. The second pointer (which must be either a constant or a
5184 global value address) contains the meta-data to be associated with the
5188 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
5189 location. At compile-time, the code generator generates information to allow
5190 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
5191 intrinsic may only be used in a function which <a href="#gc">specifies a GC
5196 <!-- _______________________________________________________________________ -->
5197 <div class="doc_subsubsection">
5198 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
5201 <div class="doc_text">
5205 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
5209 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
5210 locations, allowing garbage collector implementations that require read
5214 <p>The second argument is the address to read from, which should be an address
5215 allocated from the garbage collector. The first object is a pointer to the
5216 start of the referenced object, if needed by the language runtime (otherwise
5220 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
5221 instruction, but may be replaced with substantially more complex code by the
5222 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
5223 may only be used in a function which <a href="#gc">specifies a GC
5228 <!-- _______________________________________________________________________ -->
5229 <div class="doc_subsubsection">
5230 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
5233 <div class="doc_text">
5237 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
5241 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
5242 locations, allowing garbage collector implementations that require write
5243 barriers (such as generational or reference counting collectors).</p>
5246 <p>The first argument is the reference to store, the second is the start of the
5247 object to store it to, and the third is the address of the field of Obj to
5248 store to. If the runtime does not require a pointer to the object, Obj may
5252 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
5253 instruction, but may be replaced with substantially more complex code by the
5254 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
5255 may only be used in a function which <a href="#gc">specifies a GC
5260 <!-- ======================================================================= -->
5261 <div class="doc_subsection">
5262 <a name="int_codegen">Code Generator Intrinsics</a>
5265 <div class="doc_text">
5267 <p>These intrinsics are provided by LLVM to expose special features that may
5268 only be implemented with code generator support.</p>
5272 <!-- _______________________________________________________________________ -->
5273 <div class="doc_subsubsection">
5274 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
5277 <div class="doc_text">
5281 declare i8 *@llvm.returnaddress(i32 <level>)
5285 <p>The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
5286 target-specific value indicating the return address of the current function
5287 or one of its callers.</p>
5290 <p>The argument to this intrinsic indicates which function to return the address
5291 for. Zero indicates the calling function, one indicates its caller, etc.
5292 The argument is <b>required</b> to be a constant integer value.</p>
5295 <p>The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer
5296 indicating the return address of the specified call frame, or zero if it
5297 cannot be identified. The value returned by this intrinsic is likely to be
5298 incorrect or 0 for arguments other than zero, so it should only be used for
5299 debugging purposes.</p>
5301 <p>Note that calling this intrinsic does not prevent function inlining or other
5302 aggressive transformations, so the value returned may not be that of the
5303 obvious source-language caller.</p>
5307 <!-- _______________________________________________________________________ -->
5308 <div class="doc_subsubsection">
5309 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
5312 <div class="doc_text">
5316 declare i8 *@llvm.frameaddress(i32 <level>)
5320 <p>The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
5321 target-specific frame pointer value for the specified stack frame.</p>
5324 <p>The argument to this intrinsic indicates which function to return the frame
5325 pointer for. Zero indicates the calling function, one indicates its caller,
5326 etc. The argument is <b>required</b> to be a constant integer value.</p>
5329 <p>The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer
5330 indicating the frame address of the specified call frame, or zero if it
5331 cannot be identified. The value returned by this intrinsic is likely to be
5332 incorrect or 0 for arguments other than zero, so it should only be used for
5333 debugging purposes.</p>
5335 <p>Note that calling this intrinsic does not prevent function inlining or other
5336 aggressive transformations, so the value returned may not be that of the
5337 obvious source-language caller.</p>
5341 <!-- _______________________________________________________________________ -->
5342 <div class="doc_subsubsection">
5343 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
5346 <div class="doc_text">
5350 declare i8 *@llvm.stacksave()
5354 <p>The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state
5355 of the function stack, for use
5356 with <a href="#int_stackrestore"> <tt>llvm.stackrestore</tt></a>. This is
5357 useful for implementing language features like scoped automatic variable
5358 sized arrays in C99.</p>
5361 <p>This intrinsic returns a opaque pointer value that can be passed
5362 to <a href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When
5363 an <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved
5364 from <tt>llvm.stacksave</tt>, it effectively restores the state of the stack
5365 to the state it was in when the <tt>llvm.stacksave</tt> intrinsic executed.
5366 In practice, this pops any <a href="#i_alloca">alloca</a> blocks from the
5367 stack that were allocated after the <tt>llvm.stacksave</tt> was executed.</p>
5371 <!-- _______________________________________________________________________ -->
5372 <div class="doc_subsubsection">
5373 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
5376 <div class="doc_text">
5380 declare void @llvm.stackrestore(i8 * %ptr)
5384 <p>The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
5385 the function stack to the state it was in when the
5386 corresponding <a href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic
5387 executed. This is useful for implementing language features like scoped
5388 automatic variable sized arrays in C99.</p>
5391 <p>See the description
5392 for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.</p>
5396 <!-- _______________________________________________________________________ -->
5397 <div class="doc_subsubsection">
5398 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
5401 <div class="doc_text">
5405 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>)
5409 <p>The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to
5410 insert a prefetch instruction if supported; otherwise, it is a noop.
5411 Prefetches have no effect on the behavior of the program but can change its
5412 performance characteristics.</p>
5415 <p><tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the
5416 specifier determining if the fetch should be for a read (0) or write (1),
5417 and <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
5418 locality, to (3) - extremely local keep in cache. The <tt>rw</tt>
5419 and <tt>locality</tt> arguments must be constant integers.</p>
5422 <p>This intrinsic does not modify the behavior of the program. In particular,
5423 prefetches cannot trap and do not produce a value. On targets that support
5424 this intrinsic, the prefetch can provide hints to the processor cache for
5425 better performance.</p>
5429 <!-- _______________________________________________________________________ -->
5430 <div class="doc_subsubsection">
5431 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
5434 <div class="doc_text">
5438 declare void @llvm.pcmarker(i32 <id>)
5442 <p>The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program
5443 Counter (PC) in a region of code to simulators and other tools. The method
5444 is target specific, but it is expected that the marker will use exported
5445 symbols to transmit the PC of the marker. The marker makes no guarantees
5446 that it will remain with any specific instruction after optimizations. It is
5447 possible that the presence of a marker will inhibit optimizations. The
5448 intended use is to be inserted after optimizations to allow correlations of
5449 simulation runs.</p>
5452 <p><tt>id</tt> is a numerical id identifying the marker.</p>
5455 <p>This intrinsic does not modify the behavior of the program. Backends that do
5456 not support this intrinisic may ignore it.</p>
5460 <!-- _______________________________________________________________________ -->
5461 <div class="doc_subsubsection">
5462 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
5465 <div class="doc_text">
5469 declare i64 @llvm.readcyclecounter( )
5473 <p>The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
5474 counter register (or similar low latency, high accuracy clocks) on those
5475 targets that support it. On X86, it should map to RDTSC. On Alpha, it
5476 should map to RPCC. As the backing counters overflow quickly (on the order
5477 of 9 seconds on alpha), this should only be used for small timings.</p>
5480 <p>When directly supported, reading the cycle counter should not modify any
5481 memory. Implementations are allowed to either return a application specific
5482 value or a system wide value. On backends without support, this is lowered
5483 to a constant 0.</p>
5487 <!-- ======================================================================= -->
5488 <div class="doc_subsection">
5489 <a name="int_libc">Standard C Library Intrinsics</a>
5492 <div class="doc_text">
5494 <p>LLVM provides intrinsics for a few important standard C library functions.
5495 These intrinsics allow source-language front-ends to pass information about
5496 the alignment of the pointer arguments to the code generator, providing
5497 opportunity for more efficient code generation.</p>
5501 <!-- _______________________________________________________________________ -->
5502 <div class="doc_subsubsection">
5503 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
5506 <div class="doc_text">
5509 <p>This is an overloaded intrinsic. You can use <tt>llvm.memcpy</tt> on any
5510 integer bit width. Not all targets support all bit widths however.</p>
5513 declare void @llvm.memcpy.i8(i8 * <dest>, i8 * <src>,
5514 i8 <len>, i32 <align>)
5515 declare void @llvm.memcpy.i16(i8 * <dest>, i8 * <src>,
5516 i16 <len>, i32 <align>)
5517 declare void @llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
5518 i32 <len>, i32 <align>)
5519 declare void @llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
5520 i64 <len>, i32 <align>)
5524 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
5525 source location to the destination location.</p>
5527 <p>Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
5528 intrinsics do not return a value, and takes an extra alignment argument.</p>
5531 <p>The first argument is a pointer to the destination, the second is a pointer
5532 to the source. The third argument is an integer argument specifying the
5533 number of bytes to copy, and the fourth argument is the alignment of the
5534 source and destination locations.</p>
5536 <p>If the call to this intrinisic has an alignment value that is not 0 or 1,
5537 then the caller guarantees that both the source and destination pointers are
5538 aligned to that boundary.</p>
5541 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
5542 source location to the destination location, which are not allowed to
5543 overlap. It copies "len" bytes of memory over. If the argument is known to
5544 be aligned to some boundary, this can be specified as the fourth argument,
5545 otherwise it should be set to 0 or 1.</p>
5549 <!-- _______________________________________________________________________ -->
5550 <div class="doc_subsubsection">
5551 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
5554 <div class="doc_text">
5557 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
5558 width. Not all targets support all bit widths however.</p>
5561 declare void @llvm.memmove.i8(i8 * <dest>, i8 * <src>,
5562 i8 <len>, i32 <align>)
5563 declare void @llvm.memmove.i16(i8 * <dest>, i8 * <src>,
5564 i16 <len>, i32 <align>)
5565 declare void @llvm.memmove.i32(i8 * <dest>, i8 * <src>,
5566 i32 <len>, i32 <align>)
5567 declare void @llvm.memmove.i64(i8 * <dest>, i8 * <src>,
5568 i64 <len>, i32 <align>)
5572 <p>The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the
5573 source location to the destination location. It is similar to the
5574 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to
5577 <p>Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
5578 intrinsics do not return a value, and takes an extra alignment argument.</p>
5581 <p>The first argument is a pointer to the destination, the second is a pointer
5582 to the source. The third argument is an integer argument specifying the
5583 number of bytes to copy, and the fourth argument is the alignment of the
5584 source and destination locations.</p>
5586 <p>If the call to this intrinisic has an alignment value that is not 0 or 1,
5587 then the caller guarantees that the source and destination pointers are
5588 aligned to that boundary.</p>
5591 <p>The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the
5592 source location to the destination location, which may overlap. It copies
5593 "len" bytes of memory over. If the argument is known to be aligned to some
5594 boundary, this can be specified as the fourth argument, otherwise it should
5595 be set to 0 or 1.</p>
5599 <!-- _______________________________________________________________________ -->
5600 <div class="doc_subsubsection">
5601 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
5604 <div class="doc_text">
5607 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
5608 width. Not all targets support all bit widths however.</p>
5611 declare void @llvm.memset.i8(i8 * <dest>, i8 <val>,
5612 i8 <len>, i32 <align>)
5613 declare void @llvm.memset.i16(i8 * <dest>, i8 <val>,
5614 i16 <len>, i32 <align>)
5615 declare void @llvm.memset.i32(i8 * <dest>, i8 <val>,
5616 i32 <len>, i32 <align>)
5617 declare void @llvm.memset.i64(i8 * <dest>, i8 <val>,
5618 i64 <len>, i32 <align>)
5622 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a
5623 particular byte value.</p>
5625 <p>Note that, unlike the standard libc function, the <tt>llvm.memset</tt>
5626 intrinsic does not return a value, and takes an extra alignment argument.</p>
5629 <p>The first argument is a pointer to the destination to fill, the second is the
5630 byte value to fill it with, the third argument is an integer argument
5631 specifying the number of bytes to fill, and the fourth argument is the known
5632 alignment of destination location.</p>
5634 <p>If the call to this intrinisic has an alignment value that is not 0 or 1,
5635 then the caller guarantees that the destination pointer is aligned to that
5639 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting
5640 at the destination location. If the argument is known to be aligned to some
5641 boundary, this can be specified as the fourth argument, otherwise it should
5642 be set to 0 or 1.</p>
5646 <!-- _______________________________________________________________________ -->
5647 <div class="doc_subsubsection">
5648 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
5651 <div class="doc_text">
5654 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
5655 floating point or vector of floating point type. Not all targets support all
5659 declare float @llvm.sqrt.f32(float %Val)
5660 declare double @llvm.sqrt.f64(double %Val)
5661 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
5662 declare fp128 @llvm.sqrt.f128(fp128 %Val)
5663 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
5667 <p>The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
5668 returning the same value as the libm '<tt>sqrt</tt>' functions would.
5669 Unlike <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined
5670 behavior for negative numbers other than -0.0 (which allows for better
5671 optimization, because there is no need to worry about errno being
5672 set). <tt>llvm.sqrt(-0.0)</tt> is defined to return -0.0 like IEEE sqrt.</p>
5675 <p>The argument and return value are floating point numbers of the same
5679 <p>This function returns the sqrt of the specified operand if it is a
5680 nonnegative floating point number.</p>
5684 <!-- _______________________________________________________________________ -->
5685 <div class="doc_subsubsection">
5686 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
5689 <div class="doc_text">
5692 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
5693 floating point or vector of floating point type. Not all targets support all
5697 declare float @llvm.powi.f32(float %Val, i32 %power)
5698 declare double @llvm.powi.f64(double %Val, i32 %power)
5699 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
5700 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
5701 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
5705 <p>The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
5706 specified (positive or negative) power. The order of evaluation of
5707 multiplications is not defined. When a vector of floating point type is
5708 used, the second argument remains a scalar integer value.</p>
5711 <p>The second argument is an integer power, and the first is a value to raise to
5715 <p>This function returns the first value raised to the second power with an
5716 unspecified sequence of rounding operations.</p>
5720 <!-- _______________________________________________________________________ -->
5721 <div class="doc_subsubsection">
5722 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
5725 <div class="doc_text">
5728 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
5729 floating point or vector of floating point type. Not all targets support all
5733 declare float @llvm.sin.f32(float %Val)
5734 declare double @llvm.sin.f64(double %Val)
5735 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
5736 declare fp128 @llvm.sin.f128(fp128 %Val)
5737 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
5741 <p>The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.</p>
5744 <p>The argument and return value are floating point numbers of the same
5748 <p>This function returns the sine of the specified operand, returning the same
5749 values as the libm <tt>sin</tt> functions would, and handles error conditions
5750 in the same way.</p>
5754 <!-- _______________________________________________________________________ -->
5755 <div class="doc_subsubsection">
5756 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
5759 <div class="doc_text">
5762 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
5763 floating point or vector of floating point type. Not all targets support all
5767 declare float @llvm.cos.f32(float %Val)
5768 declare double @llvm.cos.f64(double %Val)
5769 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
5770 declare fp128 @llvm.cos.f128(fp128 %Val)
5771 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
5775 <p>The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.</p>
5778 <p>The argument and return value are floating point numbers of the same
5782 <p>This function returns the cosine of the specified operand, returning the same
5783 values as the libm <tt>cos</tt> functions would, and handles error conditions
5784 in the same way.</p>
5788 <!-- _______________________________________________________________________ -->
5789 <div class="doc_subsubsection">
5790 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
5793 <div class="doc_text">
5796 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
5797 floating point or vector of floating point type. Not all targets support all
5801 declare float @llvm.pow.f32(float %Val, float %Power)
5802 declare double @llvm.pow.f64(double %Val, double %Power)
5803 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
5804 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
5805 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
5809 <p>The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
5810 specified (positive or negative) power.</p>
5813 <p>The second argument is a floating point power, and the first is a value to
5814 raise to that power.</p>
5817 <p>This function returns the first value raised to the second power, returning
5818 the same values as the libm <tt>pow</tt> functions would, and handles error
5819 conditions in the same way.</p>
5823 <!-- ======================================================================= -->
5824 <div class="doc_subsection">
5825 <a name="int_manip">Bit Manipulation Intrinsics</a>
5828 <div class="doc_text">
5830 <p>LLVM provides intrinsics for a few important bit manipulation operations.
5831 These allow efficient code generation for some algorithms.</p>
5835 <!-- _______________________________________________________________________ -->
5836 <div class="doc_subsubsection">
5837 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
5840 <div class="doc_text">
5843 <p>This is an overloaded intrinsic function. You can use bswap on any integer
5844 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
5847 declare i16 @llvm.bswap.i16(i16 <id>)
5848 declare i32 @llvm.bswap.i32(i32 <id>)
5849 declare i64 @llvm.bswap.i64(i64 <id>)
5853 <p>The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
5854 values with an even number of bytes (positive multiple of 16 bits). These
5855 are useful for performing operations on data that is not in the target's
5856 native byte order.</p>
5859 <p>The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
5860 and low byte of the input i16 swapped. Similarly,
5861 the <tt>llvm.bswap.i32</tt> intrinsic returns an i32 value that has the four
5862 bytes of the input i32 swapped, so that if the input bytes are numbered 0, 1,
5863 2, 3 then the returned i32 will have its bytes in 3, 2, 1, 0 order.
5864 The <tt>llvm.bswap.i48</tt>, <tt>llvm.bswap.i64</tt> and other intrinsics
5865 extend this concept to additional even-byte lengths (6 bytes, 8 bytes and
5866 more, respectively).</p>
5870 <!-- _______________________________________________________________________ -->
5871 <div class="doc_subsubsection">
5872 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
5875 <div class="doc_text">
5878 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
5879 width. Not all targets support all bit widths however.</p>
5882 declare i8 @llvm.ctpop.i8(i8 <src>)
5883 declare i16 @llvm.ctpop.i16(i16 <src>)
5884 declare i32 @llvm.ctpop.i32(i32 <src>)
5885 declare i64 @llvm.ctpop.i64(i64 <src>)
5886 declare i256 @llvm.ctpop.i256(i256 <src>)
5890 <p>The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set
5894 <p>The only argument is the value to be counted. The argument may be of any
5895 integer type. The return type must match the argument type.</p>
5898 <p>The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.</p>
5902 <!-- _______________________________________________________________________ -->
5903 <div class="doc_subsubsection">
5904 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
5907 <div class="doc_text">
5910 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
5911 integer bit width. Not all targets support all bit widths however.</p>
5914 declare i8 @llvm.ctlz.i8 (i8 <src>)
5915 declare i16 @llvm.ctlz.i16(i16 <src>)
5916 declare i32 @llvm.ctlz.i32(i32 <src>)
5917 declare i64 @llvm.ctlz.i64(i64 <src>)
5918 declare i256 @llvm.ctlz.i256(i256 <src>)
5922 <p>The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
5923 leading zeros in a variable.</p>
5926 <p>The only argument is the value to be counted. The argument may be of any
5927 integer type. The return type must match the argument type.</p>
5930 <p>The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant)
5931 zeros in a variable. If the src == 0 then the result is the size in bits of
5932 the type of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.</p>
5936 <!-- _______________________________________________________________________ -->
5937 <div class="doc_subsubsection">
5938 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
5941 <div class="doc_text">
5944 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
5945 integer bit width. Not all targets support all bit widths however.</p>
5948 declare i8 @llvm.cttz.i8 (i8 <src>)
5949 declare i16 @llvm.cttz.i16(i16 <src>)
5950 declare i32 @llvm.cttz.i32(i32 <src>)
5951 declare i64 @llvm.cttz.i64(i64 <src>)
5952 declare i256 @llvm.cttz.i256(i256 <src>)
5956 <p>The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
5960 <p>The only argument is the value to be counted. The argument may be of any
5961 integer type. The return type must match the argument type.</p>
5964 <p>The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant)
5965 zeros in a variable. If the src == 0 then the result is the size in bits of
5966 the type of src. For example, <tt>llvm.cttz(2) = 1</tt>.</p>
5970 <!-- ======================================================================= -->
5971 <div class="doc_subsection">
5972 <a name="int_overflow">Arithmetic with Overflow Intrinsics</a>
5975 <div class="doc_text">
5977 <p>LLVM provides intrinsics for some arithmetic with overflow operations.</p>
5981 <!-- _______________________________________________________________________ -->
5982 <div class="doc_subsubsection">
5983 <a name="int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt>' Intrinsics</a>
5986 <div class="doc_text">
5989 <p>This is an overloaded intrinsic. You can use <tt>llvm.sadd.with.overflow</tt>
5990 on any integer bit width.</p>
5993 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
5994 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
5995 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
5999 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
6000 a signed addition of the two arguments, and indicate whether an overflow
6001 occurred during the signed summation.</p>
6004 <p>The arguments (%a and %b) and the first element of the result structure may
6005 be of integer types of any bit width, but they must have the same bit
6006 width. The second element of the result structure must be of
6007 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6008 undergo signed addition.</p>
6011 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
6012 a signed addition of the two variables. They return a structure — the
6013 first element of which is the signed summation, and the second element of
6014 which is a bit specifying if the signed summation resulted in an
6019 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6020 %sum = extractvalue {i32, i1} %res, 0
6021 %obit = extractvalue {i32, i1} %res, 1
6022 br i1 %obit, label %overflow, label %normal
6027 <!-- _______________________________________________________________________ -->
6028 <div class="doc_subsubsection">
6029 <a name="int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt>' Intrinsics</a>
6032 <div class="doc_text">
6035 <p>This is an overloaded intrinsic. You can use <tt>llvm.uadd.with.overflow</tt>
6036 on any integer bit width.</p>
6039 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
6040 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6041 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
6045 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
6046 an unsigned addition of the two arguments, and indicate whether a carry
6047 occurred during the unsigned summation.</p>
6050 <p>The arguments (%a and %b) and the first element of the result structure may
6051 be of integer types of any bit width, but they must have the same bit
6052 width. The second element of the result structure must be of
6053 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6054 undergo unsigned addition.</p>
6057 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
6058 an unsigned addition of the two arguments. They return a structure —
6059 the first element of which is the sum, and the second element of which is a
6060 bit specifying if the unsigned summation resulted in a carry.</p>
6064 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6065 %sum = extractvalue {i32, i1} %res, 0
6066 %obit = extractvalue {i32, i1} %res, 1
6067 br i1 %obit, label %carry, label %normal
6072 <!-- _______________________________________________________________________ -->
6073 <div class="doc_subsubsection">
6074 <a name="int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt>' Intrinsics</a>
6077 <div class="doc_text">
6080 <p>This is an overloaded intrinsic. You can use <tt>llvm.ssub.with.overflow</tt>
6081 on any integer bit width.</p>
6084 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
6085 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6086 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
6090 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
6091 a signed subtraction of the two arguments, and indicate whether an overflow
6092 occurred during the signed subtraction.</p>
6095 <p>The arguments (%a and %b) and the first element of the result structure may
6096 be of integer types of any bit width, but they must have the same bit
6097 width. The second element of the result structure must be of
6098 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6099 undergo signed subtraction.</p>
6102 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
6103 a signed subtraction of the two arguments. They return a structure —
6104 the first element of which is the subtraction, and the second element of
6105 which is a bit specifying if the signed subtraction resulted in an
6110 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6111 %sum = extractvalue {i32, i1} %res, 0
6112 %obit = extractvalue {i32, i1} %res, 1
6113 br i1 %obit, label %overflow, label %normal
6118 <!-- _______________________________________________________________________ -->
6119 <div class="doc_subsubsection">
6120 <a name="int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt>' Intrinsics</a>
6123 <div class="doc_text">
6126 <p>This is an overloaded intrinsic. You can use <tt>llvm.usub.with.overflow</tt>
6127 on any integer bit width.</p>
6130 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
6131 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6132 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
6136 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
6137 an unsigned subtraction of the two arguments, and indicate whether an
6138 overflow occurred during the unsigned subtraction.</p>
6141 <p>The arguments (%a and %b) and the first element of the result structure may
6142 be of integer types of any bit width, but they must have the same bit
6143 width. The second element of the result structure must be of
6144 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6145 undergo unsigned subtraction.</p>
6148 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
6149 an unsigned subtraction of the two arguments. They return a structure —
6150 the first element of which is the subtraction, and the second element of
6151 which is a bit specifying if the unsigned subtraction resulted in an
6156 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6157 %sum = extractvalue {i32, i1} %res, 0
6158 %obit = extractvalue {i32, i1} %res, 1
6159 br i1 %obit, label %overflow, label %normal
6164 <!-- _______________________________________________________________________ -->
6165 <div class="doc_subsubsection">
6166 <a name="int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt>' Intrinsics</a>
6169 <div class="doc_text">
6172 <p>This is an overloaded intrinsic. You can use <tt>llvm.smul.with.overflow</tt>
6173 on any integer bit width.</p>
6176 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
6177 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6178 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
6183 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
6184 a signed multiplication of the two arguments, and indicate whether an
6185 overflow occurred during the signed multiplication.</p>
6188 <p>The arguments (%a and %b) and the first element of the result structure may
6189 be of integer types of any bit width, but they must have the same bit
6190 width. The second element of the result structure must be of
6191 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6192 undergo signed multiplication.</p>
6195 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
6196 a signed multiplication of the two arguments. They return a structure —
6197 the first element of which is the multiplication, and the second element of
6198 which is a bit specifying if the signed multiplication resulted in an
6203 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6204 %sum = extractvalue {i32, i1} %res, 0
6205 %obit = extractvalue {i32, i1} %res, 1
6206 br i1 %obit, label %overflow, label %normal
6211 <!-- _______________________________________________________________________ -->
6212 <div class="doc_subsubsection">
6213 <a name="int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt>' Intrinsics</a>
6216 <div class="doc_text">
6219 <p>This is an overloaded intrinsic. You can use <tt>llvm.umul.with.overflow</tt>
6220 on any integer bit width.</p>
6223 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
6224 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6225 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
6229 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
6230 a unsigned multiplication of the two arguments, and indicate whether an
6231 overflow occurred during the unsigned multiplication.</p>
6234 <p>The arguments (%a and %b) and the first element of the result structure may
6235 be of integer types of any bit width, but they must have the same bit
6236 width. The second element of the result structure must be of
6237 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6238 undergo unsigned multiplication.</p>
6241 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
6242 an unsigned multiplication of the two arguments. They return a structure
6243 — the first element of which is the multiplication, and the second
6244 element of which is a bit specifying if the unsigned multiplication resulted
6249 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6250 %sum = extractvalue {i32, i1} %res, 0
6251 %obit = extractvalue {i32, i1} %res, 1
6252 br i1 %obit, label %overflow, label %normal
6257 <!-- ======================================================================= -->
6258 <div class="doc_subsection">
6259 <a name="int_debugger">Debugger Intrinsics</a>
6262 <div class="doc_text">
6264 <p>The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt>
6265 prefix), are described in
6266 the <a href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source
6267 Level Debugging</a> document.</p>
6271 <!-- ======================================================================= -->
6272 <div class="doc_subsection">
6273 <a name="int_eh">Exception Handling Intrinsics</a>
6276 <div class="doc_text">
6278 <p>The LLVM exception handling intrinsics (which all start with
6279 <tt>llvm.eh.</tt> prefix), are described in
6280 the <a href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
6281 Handling</a> document.</p>
6285 <!-- ======================================================================= -->
6286 <div class="doc_subsection">
6287 <a name="int_trampoline">Trampoline Intrinsic</a>
6290 <div class="doc_text">
6292 <p>This intrinsic makes it possible to excise one parameter, marked with
6293 the <tt>nest</tt> attribute, from a function. The result is a callable
6294 function pointer lacking the nest parameter - the caller does not need to
6295 provide a value for it. Instead, the value to use is stored in advance in a
6296 "trampoline", a block of memory usually allocated on the stack, which also
6297 contains code to splice the nest value into the argument list. This is used
6298 to implement the GCC nested function address extension.</p>
6300 <p>For example, if the function is
6301 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
6302 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as
6305 <div class="doc_code">
6307 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
6308 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
6309 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
6310 %fp = bitcast i8* %p to i32 (i32, i32)*
6314 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
6315 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
6319 <!-- _______________________________________________________________________ -->
6320 <div class="doc_subsubsection">
6321 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
6324 <div class="doc_text">
6328 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
6332 <p>This fills the memory pointed to by <tt>tramp</tt> with code and returns a
6333 function pointer suitable for executing it.</p>
6336 <p>The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
6337 pointers. The <tt>tramp</tt> argument must point to a sufficiently large and
6338 sufficiently aligned block of memory; this memory is written to by the
6339 intrinsic. Note that the size and the alignment are target-specific - LLVM
6340 currently provides no portable way of determining them, so a front-end that
6341 generates this intrinsic needs to have some target-specific knowledge.
6342 The <tt>func</tt> argument must hold a function bitcast to
6343 an <tt>i8*</tt>.</p>
6346 <p>The block of memory pointed to by <tt>tramp</tt> is filled with target
6347 dependent code, turning it into a function. A pointer to this function is
6348 returned, but needs to be bitcast to an <a href="#int_trampoline">appropriate
6349 function pointer type</a> before being called. The new function's signature
6350 is the same as that of <tt>func</tt> with any arguments marked with
6351 the <tt>nest</tt> attribute removed. At most one such <tt>nest</tt> argument
6352 is allowed, and it must be of pointer type. Calling the new function is
6353 equivalent to calling <tt>func</tt> with the same argument list, but
6354 with <tt>nval</tt> used for the missing <tt>nest</tt> argument. If, after
6355 calling <tt>llvm.init.trampoline</tt>, the memory pointed to
6356 by <tt>tramp</tt> is modified, then the effect of any later call to the
6357 returned function pointer is undefined.</p>
6361 <!-- ======================================================================= -->
6362 <div class="doc_subsection">
6363 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
6366 <div class="doc_text">
6368 <p>These intrinsic functions expand the "universal IR" of LLVM to represent
6369 hardware constructs for atomic operations and memory synchronization. This
6370 provides an interface to the hardware, not an interface to the programmer. It
6371 is aimed at a low enough level to allow any programming models or APIs
6372 (Application Programming Interfaces) which need atomic behaviors to map
6373 cleanly onto it. It is also modeled primarily on hardware behavior. Just as
6374 hardware provides a "universal IR" for source languages, it also provides a
6375 starting point for developing a "universal" atomic operation and
6376 synchronization IR.</p>
6378 <p>These do <em>not</em> form an API such as high-level threading libraries,
6379 software transaction memory systems, atomic primitives, and intrinsic
6380 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
6381 application libraries. The hardware interface provided by LLVM should allow
6382 a clean implementation of all of these APIs and parallel programming models.
6383 No one model or paradigm should be selected above others unless the hardware
6384 itself ubiquitously does so.</p>
6388 <!-- _______________________________________________________________________ -->
6389 <div class="doc_subsubsection">
6390 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
6392 <div class="doc_text">
6395 declare void @llvm.memory.barrier( i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>, i1 <device> )
6399 <p>The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
6400 specific pairs of memory access types.</p>
6403 <p>The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
6404 The first four arguments enables a specific barrier as listed below. The
6405 fith argument specifies that the barrier applies to io or device or uncached
6409 <li><tt>ll</tt>: load-load barrier</li>
6410 <li><tt>ls</tt>: load-store barrier</li>
6411 <li><tt>sl</tt>: store-load barrier</li>
6412 <li><tt>ss</tt>: store-store barrier</li>
6413 <li><tt>device</tt>: barrier applies to device and uncached memory also.</li>
6417 <p>This intrinsic causes the system to enforce some ordering constraints upon
6418 the loads and stores of the program. This barrier does not
6419 indicate <em>when</em> any events will occur, it only enforces
6420 an <em>order</em> in which they occur. For any of the specified pairs of load
6421 and store operations (f.ex. load-load, or store-load), all of the first
6422 operations preceding the barrier will complete before any of the second
6423 operations succeeding the barrier begin. Specifically the semantics for each
6424 pairing is as follows:</p>
6427 <li><tt>ll</tt>: All loads before the barrier must complete before any load
6428 after the barrier begins.</li>
6429 <li><tt>ls</tt>: All loads before the barrier must complete before any
6430 store after the barrier begins.</li>
6431 <li><tt>ss</tt>: All stores before the barrier must complete before any
6432 store after the barrier begins.</li>
6433 <li><tt>sl</tt>: All stores before the barrier must complete before any
6434 load after the barrier begins.</li>
6437 <p>These semantics are applied with a logical "and" behavior when more than one
6438 is enabled in a single memory barrier intrinsic.</p>
6440 <p>Backends may implement stronger barriers than those requested when they do
6441 not support as fine grained a barrier as requested. Some architectures do
6442 not need all types of barriers and on such architectures, these become
6450 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
6451 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
6452 <i>; guarantee the above finishes</i>
6453 store i32 8, %ptr <i>; before this begins</i>
6458 <!-- _______________________________________________________________________ -->
6459 <div class="doc_subsubsection">
6460 <a name="int_atomic_cmp_swap">'<tt>llvm.atomic.cmp.swap.*</tt>' Intrinsic</a>
6463 <div class="doc_text">
6466 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.cmp.swap</tt> on
6467 any integer bit width and for different address spaces. Not all targets
6468 support all bit widths however.</p>
6471 declare i8 @llvm.atomic.cmp.swap.i8.p0i8( i8* <ptr>, i8 <cmp>, i8 <val> )
6472 declare i16 @llvm.atomic.cmp.swap.i16.p0i16( i16* <ptr>, i16 <cmp>, i16 <val> )
6473 declare i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* <ptr>, i32 <cmp>, i32 <val> )
6474 declare i64 @llvm.atomic.cmp.swap.i64.p0i64( i64* <ptr>, i64 <cmp>, i64 <val> )
6478 <p>This loads a value in memory and compares it to a given value. If they are
6479 equal, it stores a new value into the memory.</p>
6482 <p>The <tt>llvm.atomic.cmp.swap</tt> intrinsic takes three arguments. The result
6483 as well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
6484 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
6485 this integer type. While any bit width integer may be used, targets may only
6486 lower representations they support in hardware.</p>
6489 <p>This entire intrinsic must be executed atomically. It first loads the value
6490 in memory pointed to by <tt>ptr</tt> and compares it with the
6491 value <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the
6492 memory. The loaded value is yielded in all cases. This provides the
6493 equivalent of an atomic compare-and-swap operation within the SSA
6501 %val1 = add i32 4, 4
6502 %result1 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 4, %val1 )
6503 <i>; yields {i32}:result1 = 4</i>
6504 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
6505 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
6507 %val2 = add i32 1, 1
6508 %result2 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 5, %val2 )
6509 <i>; yields {i32}:result2 = 8</i>
6510 %stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
6512 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
6517 <!-- _______________________________________________________________________ -->
6518 <div class="doc_subsubsection">
6519 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
6521 <div class="doc_text">
6524 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
6525 integer bit width. Not all targets support all bit widths however.</p>
6528 declare i8 @llvm.atomic.swap.i8.p0i8( i8* <ptr>, i8 <val> )
6529 declare i16 @llvm.atomic.swap.i16.p0i16( i16* <ptr>, i16 <val> )
6530 declare i32 @llvm.atomic.swap.i32.p0i32( i32* <ptr>, i32 <val> )
6531 declare i64 @llvm.atomic.swap.i64.p0i64( i64* <ptr>, i64 <val> )
6535 <p>This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
6536 the value from memory. It then stores the value in <tt>val</tt> in the memory
6537 at <tt>ptr</tt>.</p>
6540 <p>The <tt>llvm.atomic.swap</tt> intrinsic takes two arguments. Both
6541 the <tt>val</tt> argument and the result must be integers of the same bit
6542 width. The first argument, <tt>ptr</tt>, must be a pointer to a value of this
6543 integer type. The targets may only lower integer representations they
6547 <p>This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
6548 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
6549 equivalent of an atomic swap operation within the SSA framework.</p>
6556 %val1 = add i32 4, 4
6557 %result1 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val1 )
6558 <i>; yields {i32}:result1 = 4</i>
6559 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
6560 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
6562 %val2 = add i32 1, 1
6563 %result2 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val2 )
6564 <i>; yields {i32}:result2 = 8</i>
6566 %stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
6567 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
6572 <!-- _______________________________________________________________________ -->
6573 <div class="doc_subsubsection">
6574 <a name="int_atomic_load_add">'<tt>llvm.atomic.load.add.*</tt>' Intrinsic</a>
6578 <div class="doc_text">
6581 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.add</tt> on
6582 any integer bit width. Not all targets support all bit widths however.</p>
6585 declare i8 @llvm.atomic.load.add.i8..p0i8( i8* <ptr>, i8 <delta> )
6586 declare i16 @llvm.atomic.load.add.i16..p0i16( i16* <ptr>, i16 <delta> )
6587 declare i32 @llvm.atomic.load.add.i32..p0i32( i32* <ptr>, i32 <delta> )
6588 declare i64 @llvm.atomic.load.add.i64..p0i64( i64* <ptr>, i64 <delta> )
6592 <p>This intrinsic adds <tt>delta</tt> to the value stored in memory
6593 at <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.</p>
6596 <p>The intrinsic takes two arguments, the first a pointer to an integer value
6597 and the second an integer value. The result is also an integer value. These
6598 integer types can have any bit width, but they must all have the same bit
6599 width. The targets may only lower integer representations they support.</p>
6602 <p>This intrinsic does a series of operations atomically. It first loads the
6603 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
6604 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.</p>
6610 %result1 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 4 )
6611 <i>; yields {i32}:result1 = 4</i>
6612 %result2 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 2 )
6613 <i>; yields {i32}:result2 = 8</i>
6614 %result3 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 5 )
6615 <i>; yields {i32}:result3 = 10</i>
6616 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
6621 <!-- _______________________________________________________________________ -->
6622 <div class="doc_subsubsection">
6623 <a name="int_atomic_load_sub">'<tt>llvm.atomic.load.sub.*</tt>' Intrinsic</a>
6627 <div class="doc_text">
6630 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.sub</tt> on
6631 any integer bit width and for different address spaces. Not all targets
6632 support all bit widths however.</p>
6635 declare i8 @llvm.atomic.load.sub.i8.p0i32( i8* <ptr>, i8 <delta> )
6636 declare i16 @llvm.atomic.load.sub.i16.p0i32( i16* <ptr>, i16 <delta> )
6637 declare i32 @llvm.atomic.load.sub.i32.p0i32( i32* <ptr>, i32 <delta> )
6638 declare i64 @llvm.atomic.load.sub.i64.p0i32( i64* <ptr>, i64 <delta> )
6642 <p>This intrinsic subtracts <tt>delta</tt> to the value stored in memory at
6643 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.</p>
6646 <p>The intrinsic takes two arguments, the first a pointer to an integer value
6647 and the second an integer value. The result is also an integer value. These
6648 integer types can have any bit width, but they must all have the same bit
6649 width. The targets may only lower integer representations they support.</p>
6652 <p>This intrinsic does a series of operations atomically. It first loads the
6653 value stored at <tt>ptr</tt>. It then subtracts <tt>delta</tt>, stores the
6654 result to <tt>ptr</tt>. It yields the original value stored
6655 at <tt>ptr</tt>.</p>
6661 %result1 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 4 )
6662 <i>; yields {i32}:result1 = 8</i>
6663 %result2 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 2 )
6664 <i>; yields {i32}:result2 = 4</i>
6665 %result3 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 5 )
6666 <i>; yields {i32}:result3 = 2</i>
6667 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = -3</i>
6672 <!-- _______________________________________________________________________ -->
6673 <div class="doc_subsubsection">
6674 <a name="int_atomic_load_and">'<tt>llvm.atomic.load.and.*</tt>' Intrinsic</a><br>
6675 <a name="int_atomic_load_nand">'<tt>llvm.atomic.load.nand.*</tt>' Intrinsic</a><br>
6676 <a name="int_atomic_load_or">'<tt>llvm.atomic.load.or.*</tt>' Intrinsic</a><br>
6677 <a name="int_atomic_load_xor">'<tt>llvm.atomic.load.xor.*</tt>' Intrinsic</a><br>
6680 <div class="doc_text">
6683 <p>These are overloaded intrinsics. You can
6684 use <tt>llvm.atomic.load_and</tt>, <tt>llvm.atomic.load_nand</tt>,
6685 <tt>llvm.atomic.load_or</tt>, and <tt>llvm.atomic.load_xor</tt> on any integer
6686 bit width and for different address spaces. Not all targets support all bit
6690 declare i8 @llvm.atomic.load.and.i8.p0i8( i8* <ptr>, i8 <delta> )
6691 declare i16 @llvm.atomic.load.and.i16.p0i16( i16* <ptr>, i16 <delta> )
6692 declare i32 @llvm.atomic.load.and.i32.p0i32( i32* <ptr>, i32 <delta> )
6693 declare i64 @llvm.atomic.load.and.i64.p0i64( i64* <ptr>, i64 <delta> )
6697 declare i8 @llvm.atomic.load.or.i8.p0i8( i8* <ptr>, i8 <delta> )
6698 declare i16 @llvm.atomic.load.or.i16.p0i16( i16* <ptr>, i16 <delta> )
6699 declare i32 @llvm.atomic.load.or.i32.p0i32( i32* <ptr>, i32 <delta> )
6700 declare i64 @llvm.atomic.load.or.i64.p0i64( i64* <ptr>, i64 <delta> )
6704 declare i8 @llvm.atomic.load.nand.i8.p0i32( i8* <ptr>, i8 <delta> )
6705 declare i16 @llvm.atomic.load.nand.i16.p0i32( i16* <ptr>, i16 <delta> )
6706 declare i32 @llvm.atomic.load.nand.i32.p0i32( i32* <ptr>, i32 <delta> )
6707 declare i64 @llvm.atomic.load.nand.i64.p0i32( i64* <ptr>, i64 <delta> )
6711 declare i8 @llvm.atomic.load.xor.i8.p0i32( i8* <ptr>, i8 <delta> )
6712 declare i16 @llvm.atomic.load.xor.i16.p0i32( i16* <ptr>, i16 <delta> )
6713 declare i32 @llvm.atomic.load.xor.i32.p0i32( i32* <ptr>, i32 <delta> )
6714 declare i64 @llvm.atomic.load.xor.i64.p0i32( i64* <ptr>, i64 <delta> )
6718 <p>These intrinsics bitwise the operation (and, nand, or, xor) <tt>delta</tt> to
6719 the value stored in memory at <tt>ptr</tt>. It yields the original value
6720 at <tt>ptr</tt>.</p>
6723 <p>These intrinsics take two arguments, the first a pointer to an integer value
6724 and the second an integer value. The result is also an integer value. These
6725 integer types can have any bit width, but they must all have the same bit
6726 width. The targets may only lower integer representations they support.</p>
6729 <p>These intrinsics does a series of operations atomically. They first load the
6730 value stored at <tt>ptr</tt>. They then do the bitwise
6731 operation <tt>delta</tt>, store the result to <tt>ptr</tt>. They yield the
6732 original value stored at <tt>ptr</tt>.</p>
6737 store i32 0x0F0F, %ptr
6738 %result0 = call i32 @llvm.atomic.load.nand.i32.p0i32( i32* %ptr, i32 0xFF )
6739 <i>; yields {i32}:result0 = 0x0F0F</i>
6740 %result1 = call i32 @llvm.atomic.load.and.i32.p0i32( i32* %ptr, i32 0xFF )
6741 <i>; yields {i32}:result1 = 0xFFFFFFF0</i>
6742 %result2 = call i32 @llvm.atomic.load.or.i32.p0i32( i32* %ptr, i32 0F )
6743 <i>; yields {i32}:result2 = 0xF0</i>
6744 %result3 = call i32 @llvm.atomic.load.xor.i32.p0i32( i32* %ptr, i32 0F )
6745 <i>; yields {i32}:result3 = FF</i>
6746 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = F0</i>
6751 <!-- _______________________________________________________________________ -->
6752 <div class="doc_subsubsection">
6753 <a name="int_atomic_load_max">'<tt>llvm.atomic.load.max.*</tt>' Intrinsic</a><br>
6754 <a name="int_atomic_load_min">'<tt>llvm.atomic.load.min.*</tt>' Intrinsic</a><br>
6755 <a name="int_atomic_load_umax">'<tt>llvm.atomic.load.umax.*</tt>' Intrinsic</a><br>
6756 <a name="int_atomic_load_umin">'<tt>llvm.atomic.load.umin.*</tt>' Intrinsic</a><br>
6759 <div class="doc_text">
6762 <p>These are overloaded intrinsics. You can use <tt>llvm.atomic.load_max</tt>,
6763 <tt>llvm.atomic.load_min</tt>, <tt>llvm.atomic.load_umax</tt>, and
6764 <tt>llvm.atomic.load_umin</tt> on any integer bit width and for different
6765 address spaces. Not all targets support all bit widths however.</p>
6768 declare i8 @llvm.atomic.load.max.i8.p0i8( i8* <ptr>, i8 <delta> )
6769 declare i16 @llvm.atomic.load.max.i16.p0i16( i16* <ptr>, i16 <delta> )
6770 declare i32 @llvm.atomic.load.max.i32.p0i32( i32* <ptr>, i32 <delta> )
6771 declare i64 @llvm.atomic.load.max.i64.p0i64( i64* <ptr>, i64 <delta> )
6775 declare i8 @llvm.atomic.load.min.i8.p0i8( i8* <ptr>, i8 <delta> )
6776 declare i16 @llvm.atomic.load.min.i16.p0i16( i16* <ptr>, i16 <delta> )
6777 declare i32 @llvm.atomic.load.min.i32..p0i32( i32* <ptr>, i32 <delta> )
6778 declare i64 @llvm.atomic.load.min.i64..p0i64( i64* <ptr>, i64 <delta> )
6782 declare i8 @llvm.atomic.load.umax.i8.p0i8( i8* <ptr>, i8 <delta> )
6783 declare i16 @llvm.atomic.load.umax.i16.p0i16( i16* <ptr>, i16 <delta> )
6784 declare i32 @llvm.atomic.load.umax.i32.p0i32( i32* <ptr>, i32 <delta> )
6785 declare i64 @llvm.atomic.load.umax.i64.p0i64( i64* <ptr>, i64 <delta> )
6789 declare i8 @llvm.atomic.load.umin.i8..p0i8( i8* <ptr>, i8 <delta> )
6790 declare i16 @llvm.atomic.load.umin.i16.p0i16( i16* <ptr>, i16 <delta> )
6791 declare i32 @llvm.atomic.load.umin.i32..p0i32( i32* <ptr>, i32 <delta> )
6792 declare i64 @llvm.atomic.load.umin.i64..p0i64( i64* <ptr>, i64 <delta> )
6796 <p>These intrinsics takes the signed or unsigned minimum or maximum of
6797 <tt>delta</tt> and the value stored in memory at <tt>ptr</tt>. It yields the
6798 original value at <tt>ptr</tt>.</p>
6801 <p>These intrinsics take two arguments, the first a pointer to an integer value
6802 and the second an integer value. The result is also an integer value. These
6803 integer types can have any bit width, but they must all have the same bit
6804 width. The targets may only lower integer representations they support.</p>
6807 <p>These intrinsics does a series of operations atomically. They first load the
6808 value stored at <tt>ptr</tt>. They then do the signed or unsigned min or
6809 max <tt>delta</tt> and the value, store the result to <tt>ptr</tt>. They
6810 yield the original value stored at <tt>ptr</tt>.</p>
6816 %result0 = call i32 @llvm.atomic.load.min.i32.p0i32( i32* %ptr, i32 -2 )
6817 <i>; yields {i32}:result0 = 7</i>
6818 %result1 = call i32 @llvm.atomic.load.max.i32.p0i32( i32* %ptr, i32 8 )
6819 <i>; yields {i32}:result1 = -2</i>
6820 %result2 = call i32 @llvm.atomic.load.umin.i32.p0i32( i32* %ptr, i32 10 )
6821 <i>; yields {i32}:result2 = 8</i>
6822 %result3 = call i32 @llvm.atomic.load.umax.i32.p0i32( i32* %ptr, i32 30 )
6823 <i>; yields {i32}:result3 = 8</i>
6824 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 30</i>
6829 <!-- ======================================================================= -->
6830 <div class="doc_subsection">
6831 <a name="int_general">General Intrinsics</a>
6834 <div class="doc_text">
6836 <p>This class of intrinsics is designed to be generic and has no specific
6841 <!-- _______________________________________________________________________ -->
6842 <div class="doc_subsubsection">
6843 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
6846 <div class="doc_text">
6850 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
6854 <p>The '<tt>llvm.var.annotation</tt>' intrinsic.</p>
6857 <p>The first argument is a pointer to a value, the second is a pointer to a
6858 global string, the third is a pointer to a global string which is the source
6859 file name, and the last argument is the line number.</p>
6862 <p>This intrinsic allows annotation of local variables with arbitrary strings.
6863 This can be useful for special purpose optimizations that want to look for
6864 these annotations. These have no other defined use, they are ignored by code
6865 generation and optimization.</p>
6869 <!-- _______________________________________________________________________ -->
6870 <div class="doc_subsubsection">
6871 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
6874 <div class="doc_text">
6877 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
6878 any integer bit width.</p>
6881 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
6882 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
6883 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
6884 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
6885 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
6889 <p>The '<tt>llvm.annotation</tt>' intrinsic.</p>
6892 <p>The first argument is an integer value (result of some expression), the
6893 second is a pointer to a global string, the third is a pointer to a global
6894 string which is the source file name, and the last argument is the line
6895 number. It returns the value of the first argument.</p>
6898 <p>This intrinsic allows annotations to be put on arbitrary expressions with
6899 arbitrary strings. This can be useful for special purpose optimizations that
6900 want to look for these annotations. These have no other defined use, they
6901 are ignored by code generation and optimization.</p>
6905 <!-- _______________________________________________________________________ -->
6906 <div class="doc_subsubsection">
6907 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
6910 <div class="doc_text">
6914 declare void @llvm.trap()
6918 <p>The '<tt>llvm.trap</tt>' intrinsic.</p>
6924 <p>This intrinsics is lowered to the target dependent trap instruction. If the
6925 target does not have a trap instruction, this intrinsic will be lowered to
6926 the call of the <tt>abort()</tt> function.</p>
6930 <!-- _______________________________________________________________________ -->
6931 <div class="doc_subsubsection">
6932 <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a>
6935 <div class="doc_text">
6939 declare void @llvm.stackprotector( i8* <guard>, i8** <slot> )
6943 <p>The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and
6944 stores it onto the stack at <tt>slot</tt>. The stack slot is adjusted to
6945 ensure that it is placed on the stack before local variables.</p>
6948 <p>The <tt>llvm.stackprotector</tt> intrinsic requires two pointer
6949 arguments. The first argument is the value loaded from the stack
6950 guard <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt>
6951 that has enough space to hold the value of the guard.</p>
6954 <p>This intrinsic causes the prologue/epilogue inserter to force the position of
6955 the <tt>AllocaInst</tt> stack slot to be before local variables on the
6956 stack. This is to ensure that if a local variable on the stack is
6957 overwritten, it will destroy the value of the guard. When the function exits,
6958 the guard on the stack is checked against the original guard. If they're
6959 different, then the program aborts by calling the <tt>__stack_chk_fail()</tt>
6964 <!-- *********************************************************************** -->
6967 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
6968 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
6969 <a href="http://validator.w3.org/check/referer"><img
6970 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
6972 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
6973 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
6974 Last modified: $Date$