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_odr">'<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_integer">Integer Type</a></li>
60 <li><a href="#t_floating">Floating Point Types</a></li>
61 <li><a href="#t_void">Void Type</a></li>
62 <li><a href="#t_label">Label Type</a></li>
63 <li><a href="#t_metadata">Metadata Type</a></li>
66 <li><a href="#t_derived">Derived Types</a>
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_alloca">'<tt>alloca</tt>' Instruction</a></li>
160 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
161 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
162 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
165 <li><a href="#convertops">Conversion Operations</a>
167 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
168 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
169 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
170 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
171 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
172 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
173 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
174 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
175 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
176 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
177 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
178 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
181 <li><a href="#otherops">Other Operations</a>
183 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
184 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
185 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
186 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
187 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
188 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
193 <li><a href="#intrinsics">Intrinsic Functions</a>
195 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
197 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
198 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
199 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
202 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
204 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
205 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
206 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
209 <li><a href="#int_codegen">Code Generator Intrinsics</a>
211 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
212 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
213 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
214 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
215 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
216 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
217 <li><a href="#int_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
220 <li><a href="#int_libc">Standard C Library Intrinsics</a>
222 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
223 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
224 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
225 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
226 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
227 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
228 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
229 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
232 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
234 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
235 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
236 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
237 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
240 <li><a href="#int_overflow">Arithmetic with Overflow Intrinsics</a>
242 <li><a href="#int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt> Intrinsics</a></li>
243 <li><a href="#int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt> Intrinsics</a></li>
244 <li><a href="#int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt> Intrinsics</a></li>
245 <li><a href="#int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt> Intrinsics</a></li>
246 <li><a href="#int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt> Intrinsics</a></li>
247 <li><a href="#int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt> Intrinsics</a></li>
250 <li><a href="#int_debugger">Debugger intrinsics</a></li>
251 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
252 <li><a href="#int_trampoline">Trampoline Intrinsic</a>
254 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
257 <li><a href="#int_atomics">Atomic intrinsics</a>
259 <li><a href="#int_memory_barrier"><tt>llvm.memory_barrier</tt></a></li>
260 <li><a href="#int_atomic_cmp_swap"><tt>llvm.atomic.cmp.swap</tt></a></li>
261 <li><a href="#int_atomic_swap"><tt>llvm.atomic.swap</tt></a></li>
262 <li><a href="#int_atomic_load_add"><tt>llvm.atomic.load.add</tt></a></li>
263 <li><a href="#int_atomic_load_sub"><tt>llvm.atomic.load.sub</tt></a></li>
264 <li><a href="#int_atomic_load_and"><tt>llvm.atomic.load.and</tt></a></li>
265 <li><a href="#int_atomic_load_nand"><tt>llvm.atomic.load.nand</tt></a></li>
266 <li><a href="#int_atomic_load_or"><tt>llvm.atomic.load.or</tt></a></li>
267 <li><a href="#int_atomic_load_xor"><tt>llvm.atomic.load.xor</tt></a></li>
268 <li><a href="#int_atomic_load_max"><tt>llvm.atomic.load.max</tt></a></li>
269 <li><a href="#int_atomic_load_min"><tt>llvm.atomic.load.min</tt></a></li>
270 <li><a href="#int_atomic_load_umax"><tt>llvm.atomic.load.umax</tt></a></li>
271 <li><a href="#int_atomic_load_umin"><tt>llvm.atomic.load.umin</tt></a></li>
274 <li><a href="#int_memorymarkers">Memory Use Markers</a>
276 <li><a href="#int_lifetime_start"><tt>llvm.lifetime.start</tt></a></li>
277 <li><a href="#int_lifetime_end"><tt>llvm.lifetime.end</tt></a></li>
278 <li><a href="#int_invariant_start"><tt>llvm.invariant.start</tt></a></li>
279 <li><a href="#int_invariant_end"><tt>llvm.invariant.end</tt></a></li>
282 <li><a href="#int_general">General intrinsics</a>
284 <li><a href="#int_var_annotation">
285 '<tt>llvm.var.annotation</tt>' Intrinsic</a></li>
286 <li><a href="#int_annotation">
287 '<tt>llvm.annotation.*</tt>' Intrinsic</a></li>
288 <li><a href="#int_trap">
289 '<tt>llvm.trap</tt>' Intrinsic</a></li>
290 <li><a href="#int_stackprotector">
291 '<tt>llvm.stackprotector</tt>' Intrinsic</a></li>
298 <div class="doc_author">
299 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
300 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
303 <!-- *********************************************************************** -->
304 <div class="doc_section"> <a name="abstract">Abstract </a></div>
305 <!-- *********************************************************************** -->
307 <div class="doc_text">
309 <p>This document is a reference manual for the LLVM assembly language. LLVM is
310 a Static Single Assignment (SSA) based representation that provides type
311 safety, low-level operations, flexibility, and the capability of representing
312 'all' high-level languages cleanly. It is the common code representation
313 used throughout all phases of the LLVM compilation strategy.</p>
317 <!-- *********************************************************************** -->
318 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
319 <!-- *********************************************************************** -->
321 <div class="doc_text">
323 <p>The LLVM code representation is designed to be used in three different forms:
324 as an in-memory compiler IR, as an on-disk bitcode representation (suitable
325 for fast loading by a Just-In-Time compiler), and as a human readable
326 assembly language representation. This allows LLVM to provide a powerful
327 intermediate representation for efficient compiler transformations and
328 analysis, while providing a natural means to debug and visualize the
329 transformations. The three different forms of LLVM are all equivalent. This
330 document describes the human readable representation and notation.</p>
332 <p>The LLVM representation aims to be light-weight and low-level while being
333 expressive, typed, and extensible at the same time. It aims to be a
334 "universal IR" of sorts, by being at a low enough level that high-level ideas
335 may be cleanly mapped to it (similar to how microprocessors are "universal
336 IR's", allowing many source languages to be mapped to them). By providing
337 type information, LLVM can be used as the target of optimizations: for
338 example, through pointer analysis, it can be proven that a C automatic
339 variable is never accessed outside of the current function... allowing it to
340 be promoted to a simple SSA value instead of a memory location.</p>
344 <!-- _______________________________________________________________________ -->
345 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
347 <div class="doc_text">
349 <p>It is important to note that this document describes 'well formed' LLVM
350 assembly language. There is a difference between what the parser accepts and
351 what is considered 'well formed'. For example, the following instruction is
352 syntactically okay, but not well formed:</p>
354 <div class="doc_code">
356 %x = <a href="#i_add">add</a> i32 1, %x
360 <p>...because the definition of <tt>%x</tt> does not dominate all of its
361 uses. The LLVM infrastructure provides a verification pass that may be used
362 to verify that an LLVM module is well formed. This pass is automatically run
363 by the parser after parsing input assembly and by the optimizer before it
364 outputs bitcode. The violations pointed out by the verifier pass indicate
365 bugs in transformation passes or input to the parser.</p>
369 <!-- Describe the typesetting conventions here. -->
371 <!-- *********************************************************************** -->
372 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
373 <!-- *********************************************************************** -->
375 <div class="doc_text">
377 <p>LLVM identifiers come in two basic types: global and local. Global
378 identifiers (functions, global variables) begin with the <tt>'@'</tt>
379 character. Local identifiers (register names, types) begin with
380 the <tt>'%'</tt> character. Additionally, there are three different formats
381 for identifiers, for different purposes:</p>
384 <li>Named values are represented as a string of characters with their prefix.
385 For example, <tt>%foo</tt>, <tt>@DivisionByZero</tt>,
386 <tt>%a.really.long.identifier</tt>. The actual regular expression used is
387 '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'. Identifiers which require
388 other characters in their names can be surrounded with quotes. Special
389 characters may be escaped using <tt>"\xx"</tt> where <tt>xx</tt> is the
390 ASCII code for the character in hexadecimal. In this way, any character
391 can be used in a name value, even quotes themselves.</li>
393 <li>Unnamed values are represented as an unsigned numeric value with their
394 prefix. For example, <tt>%12</tt>, <tt>@2</tt>, <tt>%44</tt>.</li>
396 <li>Constants, which are described in a <a href="#constants">section about
397 constants</a>, below.</li>
400 <p>LLVM requires that values start with a prefix for two reasons: Compilers
401 don't need to worry about name clashes with reserved words, and the set of
402 reserved words may be expanded in the future without penalty. Additionally,
403 unnamed identifiers allow a compiler to quickly come up with a temporary
404 variable without having to avoid symbol table conflicts.</p>
406 <p>Reserved words in LLVM are very similar to reserved words in other
407 languages. There are keywords for different opcodes
408 ('<tt><a href="#i_add">add</a></tt>',
409 '<tt><a href="#i_bitcast">bitcast</a></tt>',
410 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names
411 ('<tt><a href="#t_void">void</a></tt>',
412 '<tt><a href="#t_primitive">i32</a></tt>', etc...), and others. These
413 reserved words cannot conflict with variable names, because none of them
414 start with a prefix character (<tt>'%'</tt> or <tt>'@'</tt>).</p>
416 <p>Here is an example of LLVM code to multiply the integer variable
417 '<tt>%X</tt>' by 8:</p>
421 <div class="doc_code">
423 %result = <a href="#i_mul">mul</a> i32 %X, 8
427 <p>After strength reduction:</p>
429 <div class="doc_code">
431 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
435 <p>And the hard way:</p>
437 <div class="doc_code">
439 <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
440 <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
441 %result = <a href="#i_add">add</a> i32 %1, %1
445 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several important
446 lexical features of LLVM:</p>
449 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
452 <li>Unnamed temporaries are created when the result of a computation is not
453 assigned to a named value.</li>
455 <li>Unnamed temporaries are numbered sequentially</li>
458 <p>...and it also shows a convention that we follow in this document. When
459 demonstrating instructions, we will follow an instruction with a comment that
460 defines the type and name of value produced. Comments are shown in italic
465 <!-- *********************************************************************** -->
466 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
467 <!-- *********************************************************************** -->
469 <!-- ======================================================================= -->
470 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
473 <div class="doc_text">
475 <p>LLVM programs are composed of "Module"s, each of which is a translation unit
476 of the input programs. Each module consists of functions, global variables,
477 and symbol table entries. Modules may be combined together with the LLVM
478 linker, which merges function (and global variable) definitions, resolves
479 forward declarations, and merges symbol table entries. Here is an example of
480 the "hello world" module:</p>
482 <div class="doc_code">
483 <pre><i>; Declare the string constant as a global constant...</i>
484 <a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a
485 href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
487 <i>; External declaration of the puts function</i>
488 <a href="#functionstructure">declare</a> i32 @puts(i8 *) <i>; i32(i8 *)* </i>
490 <i>; Definition of main function</i>
491 define i32 @main() { <i>; i32()* </i>
492 <i>; Convert [13 x i8]* to i8 *...</i>
494 href="#i_getelementptr">getelementptr</a> [13 x i8]* @.LC0, i64 0, i64 0 <i>; i8 *</i>
496 <i>; Call puts function to write out the string to stdout...</i>
498 href="#i_call">call</a> i32 @puts(i8 * %cast210) <i>; i32</i>
500 href="#i_ret">ret</a> i32 0<br>}<br>
504 <p>This example is made up of a <a href="#globalvars">global variable</a> named
505 "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>" function, and
506 a <a href="#functionstructure">function definition</a> for
509 <p>In general, a module is made up of a list of global values, where both
510 functions and global variables are global values. Global values are
511 represented by a pointer to a memory location (in this case, a pointer to an
512 array of char, and a pointer to a function), and have one of the
513 following <a href="#linkage">linkage types</a>.</p>
517 <!-- ======================================================================= -->
518 <div class="doc_subsection">
519 <a name="linkage">Linkage Types</a>
522 <div class="doc_text">
524 <p>All Global Variables and Functions have one of the following types of
528 <dt><tt><b><a name="linkage_private">private</a></b></tt>: </dt>
529 <dd>Global values with private linkage are only directly accessible by objects
530 in the current module. In particular, linking code into a module with an
531 private global value may cause the private to be renamed as necessary to
532 avoid collisions. Because the symbol is private to the module, all
533 references can be updated. This doesn't show up in any symbol table in the
536 <dt><tt><b><a name="linkage_linker_private">linker_private</a></b></tt>: </dt>
537 <dd>Similar to private, but the symbol is passed through the assembler and
538 removed by the linker after evaluation. Note that (unlike private
539 symbols) linker_private symbols are subject to coalescing by the linker:
540 weak symbols get merged and redefinitions are rejected. However, unlike
541 normal strong symbols, they are removed by the linker from the final
542 linked image (executable or dynamic library).</dd>
544 <dt><tt><b><a name="linkage_internal">internal</a></b></tt>: </dt>
545 <dd>Similar to private, but the value shows as a local symbol
546 (<tt>STB_LOCAL</tt> in the case of ELF) in the object file. This
547 corresponds to the notion of the '<tt>static</tt>' keyword in C.</dd>
549 <dt><tt><b><a name="linkage_available_externally">available_externally</a></b></tt>: </dt>
550 <dd>Globals with "<tt>available_externally</tt>" linkage are never emitted
551 into the object file corresponding to the LLVM module. They exist to
552 allow inlining and other optimizations to take place given knowledge of
553 the definition of the global, which is known to be somewhere outside the
554 module. Globals with <tt>available_externally</tt> linkage are allowed to
555 be discarded at will, and are otherwise the same as <tt>linkonce_odr</tt>.
556 This linkage type is only allowed on definitions, not declarations.</dd>
558 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
559 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
560 the same name when linkage occurs. This is typically used to implement
561 inline functions, templates, or other code which must be generated in each
562 translation unit that uses it. Unreferenced <tt>linkonce</tt> globals are
563 allowed to be discarded.</dd>
565 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
566 <dd>"<tt>weak</tt>" linkage has the same merging semantics as
567 <tt>linkonce</tt> linkage, except that unreferenced globals with
568 <tt>weak</tt> linkage may not be discarded. This is used for globals that
569 are declared "weak" in C source code.</dd>
571 <dt><tt><b><a name="linkage_common">common</a></b></tt>: </dt>
572 <dd>"<tt>common</tt>" linkage is most similar to "<tt>weak</tt>" linkage, but
573 they are used for tentative definitions in C, such as "<tt>int X;</tt>" at
575 Symbols with "<tt>common</tt>" linkage are merged in the same way as
576 <tt>weak symbols</tt>, and they may not be deleted if unreferenced.
577 <tt>common</tt> symbols may not have an explicit section,
578 must have a zero initializer, and may not be marked '<a
579 href="#globalvars"><tt>constant</tt></a>'. Functions and aliases may not
580 have common linkage.</dd>
583 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
584 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
585 pointer to array type. When two global variables with appending linkage
586 are linked together, the two global arrays are appended together. This is
587 the LLVM, typesafe, equivalent of having the system linker append together
588 "sections" with identical names when .o files are linked.</dd>
590 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt>: </dt>
591 <dd>The semantics of this linkage follow the ELF object file model: the symbol
592 is weak until linked, if not linked, the symbol becomes null instead of
593 being an undefined reference.</dd>
595 <dt><tt><b><a name="linkage_linkonce_odr">linkonce_odr</a></b></tt>: </dt>
596 <dt><tt><b><a name="linkage_weak_odr">weak_odr</a></b></tt>: </dt>
597 <dd>Some languages allow differing globals to be merged, such as two functions
598 with different semantics. Other languages, such as <tt>C++</tt>, ensure
599 that only equivalent globals are ever merged (the "one definition rule" -
600 "ODR"). Such languages can use the <tt>linkonce_odr</tt>
601 and <tt>weak_odr</tt> linkage types to indicate that the global will only
602 be merged with equivalent globals. These linkage types are otherwise the
603 same as their non-<tt>odr</tt> versions.</dd>
605 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
606 <dd>If none of the above identifiers are used, the global is externally
607 visible, meaning that it participates in linkage and can be used to
608 resolve external symbol references.</dd>
611 <p>The next two types of linkage are targeted for Microsoft Windows platform
612 only. They are designed to support importing (exporting) symbols from (to)
613 DLLs (Dynamic Link Libraries).</p>
616 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt>: </dt>
617 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
618 or variable via a global pointer to a pointer that is set up by the DLL
619 exporting the symbol. On Microsoft Windows targets, the pointer name is
620 formed by combining <code>__imp_</code> and the function or variable
623 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt>: </dt>
624 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
625 pointer to a pointer in a DLL, so that it can be referenced with the
626 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
627 name is formed by combining <code>__imp_</code> and the function or
631 <p>For example, since the "<tt>.LC0</tt>" variable is defined to be internal, if
632 another module defined a "<tt>.LC0</tt>" variable and was linked with this
633 one, one of the two would be renamed, preventing a collision. Since
634 "<tt>main</tt>" and "<tt>puts</tt>" are external (i.e., lacking any linkage
635 declarations), they are accessible outside of the current module.</p>
637 <p>It is illegal for a function <i>declaration</i> to have any linkage type
638 other than "externally visible", <tt>dllimport</tt>
639 or <tt>extern_weak</tt>.</p>
641 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt>, <tt>weak</tt>
642 or <tt>weak_odr</tt> linkages.</p>
646 <!-- ======================================================================= -->
647 <div class="doc_subsection">
648 <a name="callingconv">Calling Conventions</a>
651 <div class="doc_text">
653 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
654 and <a href="#i_invoke">invokes</a> can all have an optional calling
655 convention specified for the call. The calling convention of any pair of
656 dynamic caller/callee must match, or the behavior of the program is
657 undefined. The following calling conventions are supported by LLVM, and more
658 may be added in the future:</p>
661 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
662 <dd>This calling convention (the default if no other calling convention is
663 specified) matches the target C calling conventions. This calling
664 convention supports varargs function calls and tolerates some mismatch in
665 the declared prototype and implemented declaration of the function (as
668 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
669 <dd>This calling convention attempts to make calls as fast as possible
670 (e.g. by passing things in registers). This calling convention allows the
671 target to use whatever tricks it wants to produce fast code for the
672 target, without having to conform to an externally specified ABI
673 (Application Binary Interface). Implementations of this convention should
674 allow arbitrary <a href="CodeGenerator.html#tailcallopt">tail call
675 optimization</a> to be supported. This calling convention does not
676 support varargs and requires the prototype of all callees to exactly match
677 the prototype of the function definition.</dd>
679 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
680 <dd>This calling convention attempts to make code in the caller as efficient
681 as possible under the assumption that the call is not commonly executed.
682 As such, these calls often preserve all registers so that the call does
683 not break any live ranges in the caller side. This calling convention
684 does not support varargs and requires the prototype of all callees to
685 exactly match the prototype of the function definition.</dd>
687 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
688 <dd>Any calling convention may be specified by number, allowing
689 target-specific calling conventions to be used. Target specific calling
690 conventions start at 64.</dd>
693 <p>More calling conventions can be added/defined on an as-needed basis, to
694 support Pascal conventions or any other well-known target-independent
699 <!-- ======================================================================= -->
700 <div class="doc_subsection">
701 <a name="visibility">Visibility Styles</a>
704 <div class="doc_text">
706 <p>All Global Variables and Functions have one of the following visibility
710 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
711 <dd>On targets that use the ELF object file format, default visibility means
712 that the declaration is visible to other modules and, in shared libraries,
713 means that the declared entity may be overridden. On Darwin, default
714 visibility means that the declaration is visible to other modules. Default
715 visibility corresponds to "external linkage" in the language.</dd>
717 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
718 <dd>Two declarations of an object with hidden visibility refer to the same
719 object if they are in the same shared object. Usually, hidden visibility
720 indicates that the symbol will not be placed into the dynamic symbol
721 table, so no other module (executable or shared library) can reference it
724 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
725 <dd>On ELF, protected visibility indicates that the symbol will be placed in
726 the dynamic symbol table, but that references within the defining module
727 will bind to the local symbol. That is, the symbol cannot be overridden by
733 <!-- ======================================================================= -->
734 <div class="doc_subsection">
735 <a name="namedtypes">Named Types</a>
738 <div class="doc_text">
740 <p>LLVM IR allows you to specify name aliases for certain types. This can make
741 it easier to read the IR and make the IR more condensed (particularly when
742 recursive types are involved). An example of a name specification is:</p>
744 <div class="doc_code">
746 %mytype = type { %mytype*, i32 }
750 <p>You may give a name to any <a href="#typesystem">type</a> except
751 "<a href="t_void">void</a>". Type name aliases may be used anywhere a type
752 is expected with the syntax "%mytype".</p>
754 <p>Note that type names are aliases for the structural type that they indicate,
755 and that you can therefore specify multiple names for the same type. This
756 often leads to confusing behavior when dumping out a .ll file. Since LLVM IR
757 uses structural typing, the name is not part of the type. When printing out
758 LLVM IR, the printer will pick <em>one name</em> to render all types of a
759 particular shape. This means that if you have code where two different
760 source types end up having the same LLVM type, that the dumper will sometimes
761 print the "wrong" or unexpected type. This is an important design point and
762 isn't going to change.</p>
766 <!-- ======================================================================= -->
767 <div class="doc_subsection">
768 <a name="globalvars">Global Variables</a>
771 <div class="doc_text">
773 <p>Global variables define regions of memory allocated at compilation time
774 instead of run-time. Global variables may optionally be initialized, may
775 have an explicit section to be placed in, and may have an optional explicit
776 alignment specified. A variable may be defined as "thread_local", which
777 means that it will not be shared by threads (each thread will have a
778 separated copy of the variable). A variable may be defined as a global
779 "constant," which indicates that the contents of the variable
780 will <b>never</b> be modified (enabling better optimization, allowing the
781 global data to be placed in the read-only section of an executable, etc).
782 Note that variables that need runtime initialization cannot be marked
783 "constant" as there is a store to the variable.</p>
785 <p>LLVM explicitly allows <em>declarations</em> of global variables to be marked
786 constant, even if the final definition of the global is not. This capability
787 can be used to enable slightly better optimization of the program, but
788 requires the language definition to guarantee that optimizations based on the
789 'constantness' are valid for the translation units that do not include the
792 <p>As SSA values, global variables define pointer values that are in scope
793 (i.e. they dominate) all basic blocks in the program. Global variables
794 always define a pointer to their "content" type because they describe a
795 region of memory, and all memory objects in LLVM are accessed through
798 <p>A global variable may be declared to reside in a target-specific numbered
799 address space. For targets that support them, address spaces may affect how
800 optimizations are performed and/or what target instructions are used to
801 access the variable. The default address space is zero. The address space
802 qualifier must precede any other attributes.</p>
804 <p>LLVM allows an explicit section to be specified for globals. If the target
805 supports it, it will emit globals to the section specified.</p>
807 <p>An explicit alignment may be specified for a global. If not present, or if
808 the alignment is set to zero, the alignment of the global is set by the
809 target to whatever it feels convenient. If an explicit alignment is
810 specified, the global is forced to have at least that much alignment. All
811 alignments must be a power of 2.</p>
813 <p>For example, the following defines a global in a numbered address space with
814 an initializer, section, and alignment:</p>
816 <div class="doc_code">
818 @G = addrspace(5) constant float 1.0, section "foo", align 4
825 <!-- ======================================================================= -->
826 <div class="doc_subsection">
827 <a name="functionstructure">Functions</a>
830 <div class="doc_text">
832 <p>LLVM function definitions consist of the "<tt>define</tt>" keyord, 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) argument list (each with optional
838 <a href="#paramattrs">parameter attributes</a>), optional
839 <a href="#fnattrs">function attributes</a>, an optional section, an optional
840 alignment, an optional <a href="#gc">garbage collector name</a>, an opening
841 curly brace, a list of basic blocks, and a closing curly brace.</p>
843 <p>LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
844 optional <a href="#linkage">linkage type</a>, an optional
845 <a href="#visibility">visibility style</a>, an optional
846 <a href="#callingconv">calling convention</a>, a return type, an optional
847 <a href="#paramattrs">parameter attribute</a> for the return type, a function
848 name, a possibly empty list of arguments, an optional alignment, and an
849 optional <a href="#gc">garbage collector name</a>.</p>
851 <p>A function definition contains a list of basic blocks, forming the CFG
852 (Control Flow Graph) for the function. Each basic block may optionally start
853 with a label (giving the basic block a symbol table entry), contains a list
854 of instructions, and ends with a <a href="#terminators">terminator</a>
855 instruction (such as a branch or function return).</p>
857 <p>The first basic block in a function is special in two ways: it is immediately
858 executed on entrance to the function, and it is not allowed to have
859 predecessor basic blocks (i.e. there can not be any branches to the entry
860 block of a function). Because the block can have no predecessors, it also
861 cannot have any <a href="#i_phi">PHI nodes</a>.</p>
863 <p>LLVM allows an explicit section to be specified for functions. If the target
864 supports it, it will emit functions to the section specified.</p>
866 <p>An explicit alignment may be specified for a function. If not present, or if
867 the alignment is set to zero, the alignment of the function is set by the
868 target to whatever it feels convenient. If an explicit alignment is
869 specified, the function is forced to have at least that much alignment. All
870 alignments must be a power of 2.</p>
873 <div class="doc_code">
875 define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
876 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
877 <ResultType> @<FunctionName> ([argument list])
878 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
879 [<a href="#gc">gc</a>] { ... }
885 <!-- ======================================================================= -->
886 <div class="doc_subsection">
887 <a name="aliasstructure">Aliases</a>
890 <div class="doc_text">
892 <p>Aliases act as "second name" for the aliasee value (which can be either
893 function, global variable, another alias or bitcast of global value). Aliases
894 may have an optional <a href="#linkage">linkage type</a>, and an
895 optional <a href="#visibility">visibility style</a>.</p>
898 <div class="doc_code">
900 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
906 <!-- ======================================================================= -->
907 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
909 <div class="doc_text">
911 <p>The return type and each parameter of a function type may have a set of
912 <i>parameter attributes</i> associated with them. Parameter attributes are
913 used to communicate additional information about the result or parameters of
914 a function. Parameter attributes are considered to be part of the function,
915 not of the function type, so functions with different parameter attributes
916 can have the same function type.</p>
918 <p>Parameter attributes are simple keywords that follow the type specified. If
919 multiple parameter attributes are needed, they are space separated. For
922 <div class="doc_code">
924 declare i32 @printf(i8* noalias nocapture, ...)
925 declare i32 @atoi(i8 zeroext)
926 declare signext i8 @returns_signed_char()
930 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
931 <tt>readonly</tt>) come immediately after the argument list.</p>
933 <p>Currently, only the following parameter attributes are defined:</p>
936 <dt><tt>zeroext</tt></dt>
937 <dd>This indicates to the code generator that the parameter or return value
938 should be zero-extended to a 32-bit value by the caller (for a parameter)
939 or the callee (for a return value).</dd>
941 <dt><tt>signext</tt></dt>
942 <dd>This indicates to the code generator that the parameter or return value
943 should be sign-extended to a 32-bit value by the caller (for a parameter)
944 or the callee (for a return value).</dd>
946 <dt><tt>inreg</tt></dt>
947 <dd>This indicates that this parameter or return value should be treated in a
948 special target-dependent fashion during while emitting code for a function
949 call or return (usually, by putting it in a register as opposed to memory,
950 though some targets use it to distinguish between two different kinds of
951 registers). Use of this attribute is target-specific.</dd>
953 <dt><tt><a name="byval">byval</a></tt></dt>
954 <dd>This indicates that the pointer parameter should really be passed by value
955 to the function. The attribute implies that a hidden copy of the pointee
956 is made between the caller and the callee, so the callee is unable to
957 modify the value in the callee. This attribute is only valid on LLVM
958 pointer arguments. It is generally used to pass structs and arrays by
959 value, but is also valid on pointers to scalars. The copy is considered
960 to belong to the caller not the callee (for example,
961 <tt><a href="#readonly">readonly</a></tt> functions should not write to
962 <tt>byval</tt> parameters). This is not a valid attribute for return
963 values. The byval attribute also supports specifying an alignment with
964 the align attribute. This has a target-specific effect on the code
965 generator that usually indicates a desired alignment for the synthesized
968 <dt><tt>sret</tt></dt>
969 <dd>This indicates that the pointer parameter specifies the address of a
970 structure that is the return value of the function in the source program.
971 This pointer must be guaranteed by the caller to be valid: loads and
972 stores to the structure may be assumed by the callee to not to trap. This
973 may only be applied to the first parameter. This is not a valid attribute
974 for return values. </dd>
976 <dt><tt>noalias</tt></dt>
977 <dd>This indicates that the pointer does not alias any global or any other
978 parameter. The caller is responsible for ensuring that this is the
979 case. On a function return value, <tt>noalias</tt> additionally indicates
980 that the pointer does not alias any other pointers visible to the
981 caller. For further details, please see the discussion of the NoAlias
983 <a href="http://llvm.org/docs/AliasAnalysis.html#MustMayNo">alias
986 <dt><tt>nocapture</tt></dt>
987 <dd>This indicates that the callee does not make any copies of the pointer
988 that outlive the callee itself. This is not a valid attribute for return
991 <dt><tt>nest</tt></dt>
992 <dd>This indicates that the pointer parameter can be excised using the
993 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
994 attribute for return values.</dd>
999 <!-- ======================================================================= -->
1000 <div class="doc_subsection">
1001 <a name="gc">Garbage Collector Names</a>
1004 <div class="doc_text">
1006 <p>Each function may specify a garbage collector name, which is simply a
1009 <div class="doc_code">
1011 define void @f() gc "name" { ...
1015 <p>The compiler declares the supported values of <i>name</i>. Specifying a
1016 collector which will cause the compiler to alter its output in order to
1017 support the named garbage collection algorithm.</p>
1021 <!-- ======================================================================= -->
1022 <div class="doc_subsection">
1023 <a name="fnattrs">Function Attributes</a>
1026 <div class="doc_text">
1028 <p>Function attributes are set to communicate additional information about a
1029 function. Function attributes are considered to be part of the function, not
1030 of the function type, so functions with different parameter attributes can
1031 have the same function type.</p>
1033 <p>Function attributes are simple keywords that follow the type specified. If
1034 multiple attributes are needed, they are space separated. For example:</p>
1036 <div class="doc_code">
1038 define void @f() noinline { ... }
1039 define void @f() alwaysinline { ... }
1040 define void @f() alwaysinline optsize { ... }
1041 define void @f() optsize
1046 <dt><tt>alwaysinline</tt></dt>
1047 <dd>This attribute indicates that the inliner should attempt to inline this
1048 function into callers whenever possible, ignoring any active inlining size
1049 threshold for this caller.</dd>
1051 <dt><tt>inlinehint</tt></dt>
1052 <dd>This attribute indicates that the source code contained a hint that inlining
1053 this function is desirable (such as the "inline" keyword in C/C++). It
1054 is just a hint; it imposes no requirements on the inliner.</dd>
1056 <dt><tt>noinline</tt></dt>
1057 <dd>This attribute indicates that the inliner should never inline this
1058 function in any situation. This attribute may not be used together with
1059 the <tt>alwaysinline</tt> attribute.</dd>
1061 <dt><tt>optsize</tt></dt>
1062 <dd>This attribute suggests that optimization passes and code generator passes
1063 make choices that keep the code size of this function low, and otherwise
1064 do optimizations specifically to reduce code size.</dd>
1066 <dt><tt>noreturn</tt></dt>
1067 <dd>This function attribute indicates that the function never returns
1068 normally. This produces undefined behavior at runtime if the function
1069 ever does dynamically return.</dd>
1071 <dt><tt>nounwind</tt></dt>
1072 <dd>This function attribute indicates that the function never returns with an
1073 unwind or exceptional control flow. If the function does unwind, its
1074 runtime behavior is undefined.</dd>
1076 <dt><tt>readnone</tt></dt>
1077 <dd>This attribute indicates that the function computes its result (or decides
1078 to unwind an exception) based strictly on its arguments, without
1079 dereferencing any pointer arguments or otherwise accessing any mutable
1080 state (e.g. memory, control registers, etc) visible to caller functions.
1081 It does not write through any pointer arguments
1082 (including <tt><a href="#byval">byval</a></tt> arguments) and never
1083 changes any state visible to callers. This means that it cannot unwind
1084 exceptions by calling the <tt>C++</tt> exception throwing methods, but
1085 could use the <tt>unwind</tt> instruction.</dd>
1087 <dt><tt><a name="readonly">readonly</a></tt></dt>
1088 <dd>This attribute indicates that the function does not write through any
1089 pointer arguments (including <tt><a href="#byval">byval</a></tt>
1090 arguments) or otherwise modify any state (e.g. memory, control registers,
1091 etc) visible to caller functions. It may dereference pointer arguments
1092 and read state that may be set in the caller. A readonly function always
1093 returns the same value (or unwinds an exception identically) when called
1094 with the same set of arguments and global state. It cannot unwind an
1095 exception by calling the <tt>C++</tt> exception throwing methods, but may
1096 use the <tt>unwind</tt> instruction.</dd>
1098 <dt><tt><a name="ssp">ssp</a></tt></dt>
1099 <dd>This attribute indicates that the function should emit a stack smashing
1100 protector. It is in the form of a "canary"—a random value placed on
1101 the stack before the local variables that's checked upon return from the
1102 function to see if it has been overwritten. A heuristic is used to
1103 determine if a function needs stack protectors or not.<br>
1105 If a function that has an <tt>ssp</tt> attribute is inlined into a
1106 function that doesn't have an <tt>ssp</tt> attribute, then the resulting
1107 function will have an <tt>ssp</tt> attribute.</dd>
1109 <dt><tt>sspreq</tt></dt>
1110 <dd>This attribute indicates that the function should <em>always</em> emit a
1111 stack smashing protector. This overrides
1112 the <tt><a href="#ssp">ssp</a></tt> function attribute.<br>
1114 If a function that has an <tt>sspreq</tt> attribute is inlined into a
1115 function that doesn't have an <tt>sspreq</tt> attribute or which has
1116 an <tt>ssp</tt> attribute, then the resulting function will have
1117 an <tt>sspreq</tt> attribute.</dd>
1119 <dt><tt>noredzone</tt></dt>
1120 <dd>This attribute indicates that the code generator should not use a red
1121 zone, even if the target-specific ABI normally permits it.</dd>
1123 <dt><tt>noimplicitfloat</tt></dt>
1124 <dd>This attributes disables implicit floating point instructions.</dd>
1126 <dt><tt>naked</tt></dt>
1127 <dd>This attribute disables prologue / epilogue emission for the function.
1128 This can have very system-specific consequences.</dd>
1133 <!-- ======================================================================= -->
1134 <div class="doc_subsection">
1135 <a name="moduleasm">Module-Level Inline Assembly</a>
1138 <div class="doc_text">
1140 <p>Modules may contain "module-level inline asm" blocks, which corresponds to
1141 the GCC "file scope inline asm" blocks. These blocks are internally
1142 concatenated by LLVM and treated as a single unit, but may be separated in
1143 the <tt>.ll</tt> file if desired. The syntax is very simple:</p>
1145 <div class="doc_code">
1147 module asm "inline asm code goes here"
1148 module asm "more can go here"
1152 <p>The strings can contain any character by escaping non-printable characters.
1153 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1156 <p>The inline asm code is simply printed to the machine code .s file when
1157 assembly code is generated.</p>
1161 <!-- ======================================================================= -->
1162 <div class="doc_subsection">
1163 <a name="datalayout">Data Layout</a>
1166 <div class="doc_text">
1168 <p>A module may specify a target specific data layout string that specifies how
1169 data is to be laid out in memory. The syntax for the data layout is
1172 <div class="doc_code">
1174 target datalayout = "<i>layout specification</i>"
1178 <p>The <i>layout specification</i> consists of a list of specifications
1179 separated by the minus sign character ('-'). Each specification starts with
1180 a letter and may include other information after the letter to define some
1181 aspect of the data layout. The specifications accepted are as follows:</p>
1185 <dd>Specifies that the target lays out data in big-endian form. That is, the
1186 bits with the most significance have the lowest address location.</dd>
1189 <dd>Specifies that the target lays out data in little-endian form. That is,
1190 the bits with the least significance have the lowest address
1193 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1194 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1195 <i>preferred</i> alignments. All sizes are in bits. Specifying
1196 the <i>pref</i> alignment is optional. If omitted, the
1197 preceding <tt>:</tt> should be omitted too.</dd>
1199 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1200 <dd>This specifies the alignment for an integer type of a given bit
1201 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1203 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1204 <dd>This specifies the alignment for a vector type of a given bit
1207 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1208 <dd>This specifies the alignment for a floating point type of a given bit
1209 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
1212 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1213 <dd>This specifies the alignment for an aggregate type of a given bit
1216 <dt><tt>s<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1217 <dd>This specifies the alignment for a stack object of a given bit
1221 <p>When constructing the data layout for a given target, LLVM starts with a
1222 default set of specifications which are then (possibly) overriden by the
1223 specifications in the <tt>datalayout</tt> keyword. The default specifications
1224 are given in this list:</p>
1227 <li><tt>E</tt> - big endian</li>
1228 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
1229 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1230 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1231 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1232 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1233 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1234 alignment of 64-bits</li>
1235 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1236 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1237 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1238 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1239 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1240 <li><tt>s0:64:64</tt> - stack objects are 64-bit aligned</li>
1243 <p>When LLVM is determining the alignment for a given type, it uses the
1244 following rules:</p>
1247 <li>If the type sought is an exact match for one of the specifications, that
1248 specification is used.</li>
1250 <li>If no match is found, and the type sought is an integer type, then the
1251 smallest integer type that is larger than the bitwidth of the sought type
1252 is used. If none of the specifications are larger than the bitwidth then
1253 the the largest integer type is used. For example, given the default
1254 specifications above, the i7 type will use the alignment of i8 (next
1255 largest) while both i65 and i256 will use the alignment of i64 (largest
1258 <li>If no match is found, and the type sought is a vector type, then the
1259 largest vector type that is smaller than the sought vector type will be
1260 used as a fall back. This happens because <128 x double> can be
1261 implemented in terms of 64 <2 x double>, for example.</li>
1266 <!-- ======================================================================= -->
1267 <div class="doc_subsection">
1268 <a name="pointeraliasing">Pointer Aliasing Rules</a>
1271 <div class="doc_text">
1273 <p>Any memory access must be done through a pointer value associated
1274 with an address range of the memory access, otherwise the behavior
1275 is undefined. Pointer values are associated with address ranges
1276 according to the following rules:</p>
1279 <li>A pointer value formed from a
1280 <tt><a href="#i_getelementptr">getelementptr</a></tt> instruction
1281 is associated with the addresses associated with the first operand
1282 of the <tt>getelementptr</tt>.</li>
1283 <li>An address of a global variable is associated with the address
1284 range of the variable's storage.</li>
1285 <li>The result value of an allocation instruction is associated with
1286 the address range of the allocated storage.</li>
1287 <li>A null pointer in the default address-space is associated with
1289 <li>A pointer value formed by an
1290 <tt><a href="#i_inttoptr">inttoptr</a></tt> is associated with all
1291 address ranges of all pointer values that contribute (directly or
1292 indirectly) to the computation of the pointer's value.</li>
1293 <li>The result value of a
1294 <tt><a href="#i_bitcast">bitcast</a></tt> is associated with all
1295 addresses associated with the operand of the <tt>bitcast</tt>.</li>
1296 <li>An integer constant other than zero or a pointer value returned
1297 from a function not defined within LLVM may be associated with address
1298 ranges allocated through mechanisms other than those provided by
1299 LLVM. Such ranges shall not overlap with any ranges of addresses
1300 allocated by mechanisms provided by LLVM.</li>
1303 <p>LLVM IR does not associate types with memory. The result type of a
1304 <tt><a href="#i_load">load</a></tt> merely indicates the size and
1305 alignment of the memory from which to load, as well as the
1306 interpretation of the value. The first operand of a
1307 <tt><a href="#i_store">store</a></tt> similarly only indicates the size
1308 and alignment of the store.</p>
1310 <p>Consequently, type-based alias analysis, aka TBAA, aka
1311 <tt>-fstrict-aliasing</tt>, is not applicable to general unadorned
1312 LLVM IR. <a href="#metadata">Metadata</a> may be used to encode
1313 additional information which specialized optimization passes may use
1314 to implement type-based alias analysis.</p>
1318 <!-- *********************************************************************** -->
1319 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
1320 <!-- *********************************************************************** -->
1322 <div class="doc_text">
1324 <p>The LLVM type system is one of the most important features of the
1325 intermediate representation. Being typed enables a number of optimizations
1326 to be performed on the intermediate representation directly, without having
1327 to do extra analyses on the side before the transformation. A strong type
1328 system makes it easier to read the generated code and enables novel analyses
1329 and transformations that are not feasible to perform on normal three address
1330 code representations.</p>
1334 <!-- ======================================================================= -->
1335 <div class="doc_subsection"> <a name="t_classifications">Type
1336 Classifications</a> </div>
1338 <div class="doc_text">
1340 <p>The types fall into a few useful classifications:</p>
1342 <table border="1" cellspacing="0" cellpadding="4">
1344 <tr><th>Classification</th><th>Types</th></tr>
1346 <td><a href="#t_integer">integer</a></td>
1347 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1350 <td><a href="#t_floating">floating point</a></td>
1351 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1354 <td><a name="t_firstclass">first class</a></td>
1355 <td><a href="#t_integer">integer</a>,
1356 <a href="#t_floating">floating point</a>,
1357 <a href="#t_pointer">pointer</a>,
1358 <a href="#t_vector">vector</a>,
1359 <a href="#t_struct">structure</a>,
1360 <a href="#t_array">array</a>,
1361 <a href="#t_label">label</a>,
1362 <a href="#t_metadata">metadata</a>.
1366 <td><a href="#t_primitive">primitive</a></td>
1367 <td><a href="#t_label">label</a>,
1368 <a href="#t_void">void</a>,
1369 <a href="#t_floating">floating point</a>,
1370 <a href="#t_metadata">metadata</a>.</td>
1373 <td><a href="#t_derived">derived</a></td>
1374 <td><a href="#t_integer">integer</a>,
1375 <a href="#t_array">array</a>,
1376 <a href="#t_function">function</a>,
1377 <a href="#t_pointer">pointer</a>,
1378 <a href="#t_struct">structure</a>,
1379 <a href="#t_pstruct">packed structure</a>,
1380 <a href="#t_vector">vector</a>,
1381 <a href="#t_opaque">opaque</a>.
1387 <p>The <a href="#t_firstclass">first class</a> types are perhaps the most
1388 important. Values of these types are the only ones which can be produced by
1393 <!-- ======================================================================= -->
1394 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
1396 <div class="doc_text">
1398 <p>The primitive types are the fundamental building blocks of the LLVM
1403 <!-- _______________________________________________________________________ -->
1404 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1406 <div class="doc_text">
1409 <p>The integer type is a very simple type that simply specifies an arbitrary
1410 bit width for the integer type desired. Any bit width from 1 bit to
1411 2<sup>23</sup>-1 (about 8 million) can be specified.</p>
1418 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1422 <table class="layout">
1424 <td class="left"><tt>i1</tt></td>
1425 <td class="left">a single-bit integer.</td>
1428 <td class="left"><tt>i32</tt></td>
1429 <td class="left">a 32-bit integer.</td>
1432 <td class="left"><tt>i1942652</tt></td>
1433 <td class="left">a really big integer of over 1 million bits.</td>
1437 <p>Note that the code generator does not yet support large integer types to be
1438 used as function return types. The specific limit on how large a return type
1439 the code generator can currently handle is target-dependent; currently it's
1440 often 64 bits for 32-bit targets and 128 bits for 64-bit targets.</p>
1444 <!-- _______________________________________________________________________ -->
1445 <div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1447 <div class="doc_text">
1451 <tr><th>Type</th><th>Description</th></tr>
1452 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1453 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1454 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1455 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1456 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1462 <!-- _______________________________________________________________________ -->
1463 <div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1465 <div class="doc_text">
1468 <p>The void type does not represent any value and has no size.</p>
1477 <!-- _______________________________________________________________________ -->
1478 <div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1480 <div class="doc_text">
1483 <p>The label type represents code labels.</p>
1492 <!-- _______________________________________________________________________ -->
1493 <div class="doc_subsubsection"> <a name="t_metadata">Metadata Type</a> </div>
1495 <div class="doc_text">
1498 <p>The metadata type represents embedded metadata. No derived types may be
1499 created from metadata except for <a href="#t_function">function</a>
1510 <!-- ======================================================================= -->
1511 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1513 <div class="doc_text">
1515 <p>The real power in LLVM comes from the derived types in the system. This is
1516 what allows a programmer to represent arrays, functions, pointers, and other
1517 useful types. Each of these types contain one or more element types which
1518 may be a primitive type, or another derived type. For example, it is
1519 possible to have a two dimensional array, using an array as the element type
1520 of another array.</p>
1524 <!-- _______________________________________________________________________ -->
1525 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1527 <div class="doc_text">
1530 <p>The array type is a very simple derived type that arranges elements
1531 sequentially in memory. The array type requires a size (number of elements)
1532 and an underlying data type.</p>
1536 [<# elements> x <elementtype>]
1539 <p>The number of elements is a constant integer value; <tt>elementtype</tt> may
1540 be any type with a size.</p>
1543 <table class="layout">
1545 <td class="left"><tt>[40 x i32]</tt></td>
1546 <td class="left">Array of 40 32-bit integer values.</td>
1549 <td class="left"><tt>[41 x i32]</tt></td>
1550 <td class="left">Array of 41 32-bit integer values.</td>
1553 <td class="left"><tt>[4 x i8]</tt></td>
1554 <td class="left">Array of 4 8-bit integer values.</td>
1557 <p>Here are some examples of multidimensional arrays:</p>
1558 <table class="layout">
1560 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1561 <td class="left">3x4 array of 32-bit integer values.</td>
1564 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1565 <td class="left">12x10 array of single precision floating point values.</td>
1568 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1569 <td class="left">2x3x4 array of 16-bit integer values.</td>
1573 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1574 length array. Normally, accesses past the end of an array are undefined in
1575 LLVM (e.g. it is illegal to access the 5th element of a 3 element array). As
1576 a special case, however, zero length arrays are recognized to be variable
1577 length. This allows implementation of 'pascal style arrays' with the LLVM
1578 type "<tt>{ i32, [0 x float]}</tt>", for example.</p>
1580 <p>Note that the code generator does not yet support large aggregate types to be
1581 used as function return types. The specific limit on how large an aggregate
1582 return type the code generator can currently handle is target-dependent, and
1583 also dependent on the aggregate element types.</p>
1587 <!-- _______________________________________________________________________ -->
1588 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1590 <div class="doc_text">
1593 <p>The function type can be thought of as a function signature. It consists of
1594 a return type and a list of formal parameter types. The return type of a
1595 function type is a scalar type, a void type, or a struct type. If the return
1596 type is a struct type then all struct elements must be of first class types,
1597 and the struct must have at least one element.</p>
1601 <returntype> (<parameter list>)
1604 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1605 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1606 which indicates that the function takes a variable number of arguments.
1607 Variable argument functions can access their arguments with
1608 the <a href="#int_varargs">variable argument handling intrinsic</a>
1609 functions. '<tt><returntype></tt>' is a any type except
1610 <a href="#t_label">label</a>.</p>
1613 <table class="layout">
1615 <td class="left"><tt>i32 (i32)</tt></td>
1616 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1618 </tr><tr class="layout">
1619 <td class="left"><tt>float (i16 signext, i32 *) *
1621 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1622 an <tt>i16</tt> that should be sign extended and a
1623 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1626 </tr><tr class="layout">
1627 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1628 <td class="left">A vararg function that takes at least one
1629 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1630 which returns an integer. This is the signature for <tt>printf</tt> in
1633 </tr><tr class="layout">
1634 <td class="left"><tt>{i32, i32} (i32)</tt></td>
1635 <td class="left">A function taking an <tt>i32</tt>, returning a
1636 <a href="#t_struct">structure</a> containing two <tt>i32</tt> values
1643 <!-- _______________________________________________________________________ -->
1644 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1646 <div class="doc_text">
1649 <p>The structure type is used to represent a collection of data members together
1650 in memory. The packing of the field types is defined to match the ABI of the
1651 underlying processor. The elements of a structure may be any type that has a
1654 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt> and
1655 '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field with
1656 the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
1660 { <type list> }
1664 <table class="layout">
1666 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1667 <td class="left">A triple of three <tt>i32</tt> values</td>
1668 </tr><tr class="layout">
1669 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1670 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1671 second element is a <a href="#t_pointer">pointer</a> to a
1672 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1673 an <tt>i32</tt>.</td>
1677 <p>Note that the code generator does not yet support large aggregate types to be
1678 used as function return types. The specific limit on how large an aggregate
1679 return type the code generator can currently handle is target-dependent, and
1680 also dependent on the aggregate element types.</p>
1684 <!-- _______________________________________________________________________ -->
1685 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1688 <div class="doc_text">
1691 <p>The packed structure type is used to represent a collection of data members
1692 together in memory. There is no padding between fields. Further, the
1693 alignment of a packed structure is 1 byte. The elements of a packed
1694 structure may be any type that has a size.</p>
1696 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt> and
1697 '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a field with
1698 the '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
1702 < { <type list> } >
1706 <table class="layout">
1708 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1709 <td class="left">A triple of three <tt>i32</tt> values</td>
1710 </tr><tr class="layout">
1712 <tt>< { float, i32 (i32)* } ></tt></td>
1713 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1714 second element is a <a href="#t_pointer">pointer</a> to a
1715 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1716 an <tt>i32</tt>.</td>
1722 <!-- _______________________________________________________________________ -->
1723 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1725 <div class="doc_text">
1728 <p>As in many languages, the pointer type represents a pointer or reference to
1729 another object, which must live in memory. Pointer types may have an optional
1730 address space attribute defining the target-specific numbered address space
1731 where the pointed-to object resides. The default address space is zero.</p>
1733 <p>Note that LLVM does not permit pointers to void (<tt>void*</tt>) nor does it
1734 permit pointers to labels (<tt>label*</tt>). Use <tt>i8*</tt> instead.</p>
1742 <table class="layout">
1744 <td class="left"><tt>[4 x i32]*</tt></td>
1745 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1746 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1749 <td class="left"><tt>i32 (i32 *) *</tt></td>
1750 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
1751 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1755 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1756 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1757 that resides in address space #5.</td>
1763 <!-- _______________________________________________________________________ -->
1764 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1766 <div class="doc_text">
1769 <p>A vector type is a simple derived type that represents a vector of elements.
1770 Vector types are used when multiple primitive data are operated in parallel
1771 using a single instruction (SIMD). A vector type requires a size (number of
1772 elements) and an underlying primitive data type. Vectors must have a power
1773 of two length (1, 2, 4, 8, 16 ...). Vector types are considered
1774 <a href="#t_firstclass">first class</a>.</p>
1778 < <# elements> x <elementtype> >
1781 <p>The number of elements is a constant integer value; elementtype may be any
1782 integer or floating point type.</p>
1785 <table class="layout">
1787 <td class="left"><tt><4 x i32></tt></td>
1788 <td class="left">Vector of 4 32-bit integer values.</td>
1791 <td class="left"><tt><8 x float></tt></td>
1792 <td class="left">Vector of 8 32-bit floating-point values.</td>
1795 <td class="left"><tt><2 x i64></tt></td>
1796 <td class="left">Vector of 2 64-bit integer values.</td>
1800 <p>Note that the code generator does not yet support large vector types to be
1801 used as function return types. The specific limit on how large a vector
1802 return type codegen can currently handle is target-dependent; currently it's
1803 often a few times longer than a hardware vector register.</p>
1807 <!-- _______________________________________________________________________ -->
1808 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1809 <div class="doc_text">
1812 <p>Opaque types are used to represent unknown types in the system. This
1813 corresponds (for example) to the C notion of a forward declared structure
1814 type. In LLVM, opaque types can eventually be resolved to any type (not just
1815 a structure type).</p>
1823 <table class="layout">
1825 <td class="left"><tt>opaque</tt></td>
1826 <td class="left">An opaque type.</td>
1832 <!-- ======================================================================= -->
1833 <div class="doc_subsection">
1834 <a name="t_uprefs">Type Up-references</a>
1837 <div class="doc_text">
1840 <p>An "up reference" allows you to refer to a lexically enclosing type without
1841 requiring it to have a name. For instance, a structure declaration may
1842 contain a pointer to any of the types it is lexically a member of. Example
1843 of up references (with their equivalent as named type declarations)
1847 { \2 * } %x = type { %x* }
1848 { \2 }* %y = type { %y }*
1852 <p>An up reference is needed by the asmprinter for printing out cyclic types
1853 when there is no declared name for a type in the cycle. Because the
1854 asmprinter does not want to print out an infinite type string, it needs a
1855 syntax to handle recursive types that have no names (all names are optional
1863 <p>The level is the count of the lexical type that is being referred to.</p>
1866 <table class="layout">
1868 <td class="left"><tt>\1*</tt></td>
1869 <td class="left">Self-referential pointer.</td>
1872 <td class="left"><tt>{ { \3*, i8 }, i32 }</tt></td>
1873 <td class="left">Recursive structure where the upref refers to the out-most
1880 <!-- *********************************************************************** -->
1881 <div class="doc_section"> <a name="constants">Constants</a> </div>
1882 <!-- *********************************************************************** -->
1884 <div class="doc_text">
1886 <p>LLVM has several different basic types of constants. This section describes
1887 them all and their syntax.</p>
1891 <!-- ======================================================================= -->
1892 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1894 <div class="doc_text">
1897 <dt><b>Boolean constants</b></dt>
1898 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1899 constants of the <tt><a href="#t_integer">i1</a></tt> type.</dd>
1901 <dt><b>Integer constants</b></dt>
1902 <dd>Standard integers (such as '4') are constants of
1903 the <a href="#t_integer">integer</a> type. Negative numbers may be used
1904 with integer types.</dd>
1906 <dt><b>Floating point constants</b></dt>
1907 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1908 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1909 notation (see below). The assembler requires the exact decimal value of a
1910 floating-point constant. For example, the assembler accepts 1.25 but
1911 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
1912 constants must have a <a href="#t_floating">floating point</a> type. </dd>
1914 <dt><b>Null pointer constants</b></dt>
1915 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1916 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1919 <p>The one non-intuitive notation for constants is the hexadecimal form of
1920 floating point constants. For example, the form '<tt>double
1921 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than)
1922 '<tt>double 4.5e+15</tt>'. The only time hexadecimal floating point
1923 constants are required (and the only time that they are generated by the
1924 disassembler) is when a floating point constant must be emitted but it cannot
1925 be represented as a decimal floating point number in a reasonable number of
1926 digits. For example, NaN's, infinities, and other special values are
1927 represented in their IEEE hexadecimal format so that assembly and disassembly
1928 do not cause any bits to change in the constants.</p>
1930 <p>When using the hexadecimal form, constants of types float and double are
1931 represented using the 16-digit form shown above (which matches the IEEE754
1932 representation for double); float values must, however, be exactly
1933 representable as IEE754 single precision. Hexadecimal format is always used
1934 for long double, and there are three forms of long double. The 80-bit format
1935 used by x86 is represented as <tt>0xK</tt> followed by 20 hexadecimal digits.
1936 The 128-bit format used by PowerPC (two adjacent doubles) is represented
1937 by <tt>0xM</tt> followed by 32 hexadecimal digits. The IEEE 128-bit format
1938 is represented by <tt>0xL</tt> followed by 32 hexadecimal digits; no
1939 currently supported target uses this format. Long doubles will only work if
1940 they match the long double format on your target. All hexadecimal formats
1941 are big-endian (sign bit at the left).</p>
1945 <!-- ======================================================================= -->
1946 <div class="doc_subsection">
1947 <a name="aggregateconstants"></a> <!-- old anchor -->
1948 <a name="complexconstants">Complex Constants</a>
1951 <div class="doc_text">
1953 <p>Complex constants are a (potentially recursive) combination of simple
1954 constants and smaller complex constants.</p>
1957 <dt><b>Structure constants</b></dt>
1958 <dd>Structure constants are represented with notation similar to structure
1959 type definitions (a comma separated list of elements, surrounded by braces
1960 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
1961 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>".
1962 Structure constants must have <a href="#t_struct">structure type</a>, and
1963 the number and types of elements must match those specified by the
1966 <dt><b>Array constants</b></dt>
1967 <dd>Array constants are represented with notation similar to array type
1968 definitions (a comma separated list of elements, surrounded by square
1969 brackets (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74
1970 ]</tt>". Array constants must have <a href="#t_array">array type</a>, and
1971 the number and types of elements must match those specified by the
1974 <dt><b>Vector constants</b></dt>
1975 <dd>Vector constants are represented with notation similar to vector type
1976 definitions (a comma separated list of elements, surrounded by
1977 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32
1978 42, i32 11, i32 74, i32 100 ></tt>". Vector constants must
1979 have <a href="#t_vector">vector type</a>, and the number and types of
1980 elements must match those specified by the type.</dd>
1982 <dt><b>Zero initialization</b></dt>
1983 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1984 value to zero of <em>any</em> type, including scalar and aggregate types.
1985 This is often used to avoid having to print large zero initializers
1986 (e.g. for large arrays) and is always exactly equivalent to using explicit
1987 zero initializers.</dd>
1989 <dt><b>Metadata node</b></dt>
1990 <dd>A metadata node is a structure-like constant with
1991 <a href="#t_metadata">metadata type</a>. For example: "<tt>metadata !{
1992 i32 0, metadata !"test" }</tt>". Unlike other constants that are meant to
1993 be interpreted as part of the instruction stream, metadata is a place to
1994 attach additional information such as debug info.</dd>
1999 <!-- ======================================================================= -->
2000 <div class="doc_subsection">
2001 <a name="globalconstants">Global Variable and Function Addresses</a>
2004 <div class="doc_text">
2006 <p>The addresses of <a href="#globalvars">global variables</a>
2007 and <a href="#functionstructure">functions</a> are always implicitly valid
2008 (link-time) constants. These constants are explicitly referenced when
2009 the <a href="#identifiers">identifier for the global</a> is used and always
2010 have <a href="#t_pointer">pointer</a> type. For example, the following is a
2011 legal LLVM file:</p>
2013 <div class="doc_code">
2017 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
2023 <!-- ======================================================================= -->
2024 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
2025 <div class="doc_text">
2027 <p>The string '<tt>undef</tt>' can be used anywhere a constant is expected, and
2028 indicates that the user of the value may receive an unspecified bit-pattern.
2029 Undefined values may be of any type (other than label or void) and be used
2030 anywhere a constant is permitted.</p>
2032 <p>Undefined values are useful because they indicate to the compiler that the
2033 program is well defined no matter what value is used. This gives the
2034 compiler more freedom to optimize. Here are some examples of (potentially
2035 surprising) transformations that are valid (in pseudo IR):</p>
2038 <div class="doc_code">
2050 <p>This is safe because all of the output bits are affected by the undef bits.
2051 Any output bit can have a zero or one depending on the input bits.</p>
2053 <div class="doc_code">
2066 <p>These logical operations have bits that are not always affected by the input.
2067 For example, if "%X" has a zero bit, then the output of the 'and' operation will
2068 always be a zero, no matter what the corresponding bit from the undef is. As
2069 such, it is unsafe to optimize or assume that the result of the and is undef.
2070 However, it is safe to assume that all bits of the undef could be 0, and
2071 optimize the and to 0. Likewise, it is safe to assume that all the bits of
2072 the undef operand to the or could be set, allowing the or to be folded to
2075 <div class="doc_code">
2077 %A = select undef, %X, %Y
2078 %B = select undef, 42, %Y
2079 %C = select %X, %Y, undef
2091 <p>This set of examples show that undefined select (and conditional branch)
2092 conditions can go "either way" but they have to come from one of the two
2093 operands. In the %A example, if %X and %Y were both known to have a clear low
2094 bit, then %A would have to have a cleared low bit. However, in the %C example,
2095 the optimizer is allowed to assume that the undef operand could be the same as
2096 %Y, allowing the whole select to be eliminated.</p>
2099 <div class="doc_code">
2101 %A = xor undef, undef
2120 <p>This example points out that two undef operands are not necessarily the same.
2121 This can be surprising to people (and also matches C semantics) where they
2122 assume that "X^X" is always zero, even if X is undef. This isn't true for a
2123 number of reasons, but the short answer is that an undef "variable" can
2124 arbitrarily change its value over its "live range". This is true because the
2125 "variable" doesn't actually <em>have a live range</em>. Instead, the value is
2126 logically read from arbitrary registers that happen to be around when needed,
2127 so the value is not necessarily consistent over time. In fact, %A and %C need
2128 to have the same semantics or the core LLVM "replace all uses with" concept
2131 <div class="doc_code">
2141 <p>These examples show the crucial difference between an <em>undefined
2142 value</em> and <em>undefined behavior</em>. An undefined value (like undef) is
2143 allowed to have an arbitrary bit-pattern. This means that the %A operation
2144 can be constant folded to undef because the undef could be an SNaN, and fdiv is
2145 not (currently) defined on SNaN's. However, in the second example, we can make
2146 a more aggressive assumption: because the undef is allowed to be an arbitrary
2147 value, we are allowed to assume that it could be zero. Since a divide by zero
2148 has <em>undefined behavior</em>, we are allowed to assume that the operation
2149 does not execute at all. This allows us to delete the divide and all code after
2150 it: since the undefined operation "can't happen", the optimizer can assume that
2151 it occurs in dead code.
2154 <div class="doc_code">
2156 a: store undef -> %X
2157 b: store %X -> undef
2164 <p>These examples reiterate the fdiv example: a store "of" an undefined value
2165 can be assumed to not have any effect: we can assume that the value is
2166 overwritten with bits that happen to match what was already there. However, a
2167 store "to" an undefined location could clobber arbitrary memory, therefore, it
2168 has undefined behavior.</p>
2172 <!-- ======================================================================= -->
2173 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
2176 <div class="doc_text">
2178 <p>Constant expressions are used to allow expressions involving other constants
2179 to be used as constants. Constant expressions may be of
2180 any <a href="#t_firstclass">first class</a> type and may involve any LLVM
2181 operation that does not have side effects (e.g. load and call are not
2182 supported). The following is the syntax for constant expressions:</p>
2185 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
2186 <dd>Truncate a constant to another type. The bit size of CST must be larger
2187 than the bit size of TYPE. Both types must be integers.</dd>
2189 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
2190 <dd>Zero extend a constant to another type. The bit size of CST must be
2191 smaller or equal to the bit size of TYPE. Both types must be
2194 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
2195 <dd>Sign extend a constant to another type. The bit size of CST must be
2196 smaller or equal to the bit size of TYPE. Both types must be
2199 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
2200 <dd>Truncate a floating point constant to another floating point type. The
2201 size of CST must be larger than the size of TYPE. Both types must be
2202 floating point.</dd>
2204 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
2205 <dd>Floating point extend a constant to another type. The size of CST must be
2206 smaller or equal to the size of TYPE. Both types must be floating
2209 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
2210 <dd>Convert a floating point constant to the corresponding unsigned integer
2211 constant. TYPE must be a scalar or vector integer type. CST must be of
2212 scalar or vector floating point type. Both CST and TYPE must be scalars,
2213 or vectors of the same number of elements. If the value won't fit in the
2214 integer type, the results are undefined.</dd>
2216 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
2217 <dd>Convert a floating point constant to the corresponding signed integer
2218 constant. TYPE must be a scalar or vector integer type. CST must be of
2219 scalar or vector floating point type. Both CST and TYPE must be scalars,
2220 or vectors of the same number of elements. If the value won't fit in the
2221 integer type, the results are undefined.</dd>
2223 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
2224 <dd>Convert an unsigned integer constant to the corresponding floating point
2225 constant. TYPE must be a scalar or vector floating point type. CST must be
2226 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2227 vectors of the same number of elements. If the value won't fit in the
2228 floating point type, the results are undefined.</dd>
2230 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
2231 <dd>Convert a signed integer constant to the corresponding floating point
2232 constant. TYPE must be a scalar or vector floating point type. CST must be
2233 of scalar or vector integer type. Both CST and TYPE must be scalars, or
2234 vectors of the same number of elements. If the value won't fit in the
2235 floating point type, the results are undefined.</dd>
2237 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
2238 <dd>Convert a pointer typed constant to the corresponding integer constant
2239 <tt>TYPE</tt> must be an integer type. <tt>CST</tt> must be of pointer
2240 type. The <tt>CST</tt> value is zero extended, truncated, or unchanged to
2241 make it fit in <tt>TYPE</tt>.</dd>
2243 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
2244 <dd>Convert a integer constant to a pointer constant. TYPE must be a pointer
2245 type. CST must be of integer type. The CST value is zero extended,
2246 truncated, or unchanged to make it fit in a pointer size. This one is
2247 <i>really</i> dangerous!</dd>
2249 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
2250 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
2251 are the same as those for the <a href="#i_bitcast">bitcast
2252 instruction</a>.</dd>
2254 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
2255 <dt><b><tt>getelementptr inbounds ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
2256 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
2257 constants. As with the <a href="#i_getelementptr">getelementptr</a>
2258 instruction, the index list may have zero or more indexes, which are
2259 required to make sense for the type of "CSTPTR".</dd>
2261 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
2262 <dd>Perform the <a href="#i_select">select operation</a> on constants.</dd>
2264 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
2265 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
2267 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
2268 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
2270 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
2271 <dd>Perform the <a href="#i_extractelement">extractelement operation</a> on
2274 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
2275 <dd>Perform the <a href="#i_insertelement">insertelement operation</a> on
2278 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
2279 <dd>Perform the <a href="#i_shufflevector">shufflevector operation</a> on
2282 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
2283 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2284 be any of the <a href="#binaryops">binary</a>
2285 or <a href="#bitwiseops">bitwise binary</a> operations. The constraints
2286 on operands are the same as those for the corresponding instruction
2287 (e.g. no bitwise operations on floating point values are allowed).</dd>
2292 <!-- ======================================================================= -->
2293 <div class="doc_subsection"><a name="metadata">Embedded Metadata</a>
2296 <div class="doc_text">
2298 <p>Embedded metadata provides a way to attach arbitrary data to the instruction
2299 stream without affecting the behaviour of the program. There are two
2300 metadata primitives, strings and nodes. All metadata has the
2301 <tt>metadata</tt> type and is identified in syntax by a preceding exclamation
2302 point ('<tt>!</tt>').</p>
2304 <p>A metadata string is a string surrounded by double quotes. It can contain
2305 any character by escaping non-printable characters with "\xx" where "xx" is
2306 the two digit hex code. For example: "<tt>!"test\00"</tt>".</p>
2308 <p>Metadata nodes are represented with notation similar to structure constants
2309 (a comma separated list of elements, surrounded by braces and preceded by an
2310 exclamation point). For example: "<tt>!{ metadata !"test\00", i32
2313 <p>A metadata node will attempt to track changes to the values it holds. In the
2314 event that a value is deleted, it will be replaced with a typeless
2315 "<tt>null</tt>", such as "<tt>metadata !{null, i32 10}</tt>".</p>
2317 <p>Optimizations may rely on metadata to provide additional information about
2318 the program that isn't available in the instructions, or that isn't easily
2319 computable. Similarly, the code generator may expect a certain metadata
2320 format to be used to express debugging information.</p>
2324 <!-- *********************************************************************** -->
2325 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
2326 <!-- *********************************************************************** -->
2328 <!-- ======================================================================= -->
2329 <div class="doc_subsection">
2330 <a name="inlineasm">Inline Assembler Expressions</a>
2333 <div class="doc_text">
2335 <p>LLVM supports inline assembler expressions (as opposed
2336 to <a href="#moduleasm"> Module-Level Inline Assembly</a>) through the use of
2337 a special value. This value represents the inline assembler as a string
2338 (containing the instructions to emit), a list of operand constraints (stored
2339 as a string), a flag that indicates whether or not the inline asm
2340 expression has side effects, and a flag indicating whether the function
2341 containing the asm needs to align its stack conservatively. An example
2342 inline assembler expression is:</p>
2344 <div class="doc_code">
2346 i32 (i32) asm "bswap $0", "=r,r"
2350 <p>Inline assembler expressions may <b>only</b> be used as the callee operand of
2351 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we
2354 <div class="doc_code">
2356 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
2360 <p>Inline asms with side effects not visible in the constraint list must be
2361 marked as having side effects. This is done through the use of the
2362 '<tt>sideeffect</tt>' keyword, like so:</p>
2364 <div class="doc_code">
2366 call void asm sideeffect "eieio", ""()
2370 <p>In some cases inline asms will contain code that will not work unless the
2371 stack is aligned in some way, such as calls or SSE instructions on x86,
2372 yet will not contain code that does that alignment within the asm.
2373 The compiler should make conservative assumptions about what the asm might
2374 contain and should generate its usual stack alignment code in the prologue
2375 if the '<tt>alignstack</tt>' keyword is present:</p>
2377 <div class="doc_code">
2379 call void asm alignstack "eieio", ""()
2383 <p>If both keywords appear the '<tt>sideeffect</tt>' keyword must come
2386 <p>TODO: The format of the asm and constraints string still need to be
2387 documented here. Constraints on what can be done (e.g. duplication, moving,
2388 etc need to be documented). This is probably best done by reference to
2389 another document that covers inline asm from a holistic perspective.</p>
2394 <!-- *********************************************************************** -->
2395 <div class="doc_section">
2396 <a name="intrinsic_globals">Intrinsic Global Variables</a>
2398 <!-- *********************************************************************** -->
2400 <p>LLVM has a number of "magic" global variables that contain data that affect
2401 code generation or other IR semantics. These are documented here. All globals
2402 of this sort should have a section specified as "<tt>llvm.metadata</tt>". This
2403 section and all globals that start with "<tt>llvm.</tt>" are reserved for use
2406 <!-- ======================================================================= -->
2407 <div class="doc_subsection">
2408 <a name="intg_used">The '<tt>llvm.used</tt>' Global Variable</a>
2411 <div class="doc_text">
2413 <p>The <tt>@llvm.used</tt> global is an array with i8* element type which has <a
2414 href="#linkage_appending">appending linkage</a>. This array contains a list of
2415 pointers to global variables and functions which may optionally have a pointer
2416 cast formed of bitcast or getelementptr. For example, a legal use of it is:</p>
2422 @llvm.used = appending global [2 x i8*] [
2424 i8* bitcast (i32* @Y to i8*)
2425 ], section "llvm.metadata"
2428 <p>If a global variable appears in the <tt>@llvm.used</tt> list, then the
2429 compiler, assembler, and linker are required to treat the symbol as if there is
2430 a reference to the global that it cannot see. For example, if a variable has
2431 internal linkage and no references other than that from the <tt>@llvm.used</tt>
2432 list, it cannot be deleted. This is commonly used to represent references from
2433 inline asms and other things the compiler cannot "see", and corresponds to
2434 "attribute((used))" in GNU C.</p>
2436 <p>On some targets, the code generator must emit a directive to the assembler or
2437 object file to prevent the assembler and linker from molesting the symbol.</p>
2441 <!-- ======================================================================= -->
2442 <div class="doc_subsection">
2443 <a name="intg_compiler_used">The '<tt>llvm.compiler.used</tt>' Global Variable</a>
2446 <div class="doc_text">
2448 <p>The <tt>@llvm.compiler.used</tt> directive is the same as the
2449 <tt>@llvm.used</tt> directive, except that it only prevents the compiler from
2450 touching the symbol. On targets that support it, this allows an intelligent
2451 linker to optimize references to the symbol without being impeded as it would be
2452 by <tt>@llvm.used</tt>.</p>
2454 <p>This is a rare construct that should only be used in rare circumstances, and
2455 should not be exposed to source languages.</p>
2459 <!-- ======================================================================= -->
2460 <div class="doc_subsection">
2461 <a name="intg_global_ctors">The '<tt>llvm.global_ctors</tt>' Global Variable</a>
2464 <div class="doc_text">
2466 <p>TODO: Describe this.</p>
2470 <!-- ======================================================================= -->
2471 <div class="doc_subsection">
2472 <a name="intg_global_dtors">The '<tt>llvm.global_dtors</tt>' Global Variable</a>
2475 <div class="doc_text">
2477 <p>TODO: Describe this.</p>
2482 <!-- *********************************************************************** -->
2483 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
2484 <!-- *********************************************************************** -->
2486 <div class="doc_text">
2488 <p>The LLVM instruction set consists of several different classifications of
2489 instructions: <a href="#terminators">terminator
2490 instructions</a>, <a href="#binaryops">binary instructions</a>,
2491 <a href="#bitwiseops">bitwise binary instructions</a>,
2492 <a href="#memoryops">memory instructions</a>, and
2493 <a href="#otherops">other instructions</a>.</p>
2497 <!-- ======================================================================= -->
2498 <div class="doc_subsection"> <a name="terminators">Terminator
2499 Instructions</a> </div>
2501 <div class="doc_text">
2503 <p>As mentioned <a href="#functionstructure">previously</a>, every basic block
2504 in a program ends with a "Terminator" instruction, which indicates which
2505 block should be executed after the current block is finished. These
2506 terminator instructions typically yield a '<tt>void</tt>' value: they produce
2507 control flow, not values (the one exception being the
2508 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
2510 <p>There are six different terminator instructions: the
2511 '<a href="#i_ret"><tt>ret</tt></a>' instruction, the
2512 '<a href="#i_br"><tt>br</tt></a>' instruction, the
2513 '<a href="#i_switch"><tt>switch</tt></a>' instruction, the
2514 '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the
2515 '<a href="#i_unwind"><tt>unwind</tt></a>' instruction, and the
2516 '<a href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
2520 <!-- _______________________________________________________________________ -->
2521 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
2522 Instruction</a> </div>
2524 <div class="doc_text">
2528 ret <type> <value> <i>; Return a value from a non-void function</i>
2529 ret void <i>; Return from void function</i>
2533 <p>The '<tt>ret</tt>' instruction is used to return control flow (and optionally
2534 a value) from a function back to the caller.</p>
2536 <p>There are two forms of the '<tt>ret</tt>' instruction: one that returns a
2537 value and then causes control flow, and one that just causes control flow to
2541 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument, the
2542 return value. The type of the return value must be a
2543 '<a href="#t_firstclass">first class</a>' type.</p>
2545 <p>A function is not <a href="#wellformed">well formed</a> if it it has a
2546 non-void return type and contains a '<tt>ret</tt>' instruction with no return
2547 value or a return value with a type that does not match its type, or if it
2548 has a void return type and contains a '<tt>ret</tt>' instruction with a
2552 <p>When the '<tt>ret</tt>' instruction is executed, control flow returns back to
2553 the calling function's context. If the caller is a
2554 "<a href="#i_call"><tt>call</tt></a>" instruction, execution continues at the
2555 instruction after the call. If the caller was an
2556 "<a href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues at
2557 the beginning of the "normal" destination block. If the instruction returns
2558 a value, that value shall set the call or invoke instruction's return
2563 ret i32 5 <i>; Return an integer value of 5</i>
2564 ret void <i>; Return from a void function</i>
2565 ret { i32, i8 } { i32 4, i8 2 } <i>; Return a struct of values 4 and 2</i>
2568 <p>Note that the code generator does not yet fully support large
2569 return values. The specific sizes that are currently supported are
2570 dependent on the target. For integers, on 32-bit targets the limit
2571 is often 64 bits, and on 64-bit targets the limit is often 128 bits.
2572 For aggregate types, the current limits are dependent on the element
2573 types; for example targets are often limited to 2 total integer
2574 elements and 2 total floating-point elements.</p>
2577 <!-- _______________________________________________________________________ -->
2578 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
2580 <div class="doc_text">
2584 br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
2588 <p>The '<tt>br</tt>' instruction is used to cause control flow to transfer to a
2589 different basic block in the current function. There are two forms of this
2590 instruction, corresponding to a conditional branch and an unconditional
2594 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a single
2595 '<tt>i1</tt>' value and two '<tt>label</tt>' values. The unconditional form
2596 of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>' value as a
2600 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
2601 argument is evaluated. If the value is <tt>true</tt>, control flows to the
2602 '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
2603 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
2608 %cond = <a href="#i_icmp">icmp</a> eq i32 %a, %b
2609 br i1 %cond, label %IfEqual, label %IfUnequal
2611 <a href="#i_ret">ret</a> i32 1
2613 <a href="#i_ret">ret</a> i32 0
2618 <!-- _______________________________________________________________________ -->
2619 <div class="doc_subsubsection">
2620 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
2623 <div class="doc_text">
2627 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
2631 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
2632 several different places. It is a generalization of the '<tt>br</tt>'
2633 instruction, allowing a branch to occur to one of many possible
2637 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
2638 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination,
2639 and an array of pairs of comparison value constants and '<tt>label</tt>'s.
2640 The table is not allowed to contain duplicate constant entries.</p>
2643 <p>The <tt>switch</tt> instruction specifies a table of values and
2644 destinations. When the '<tt>switch</tt>' instruction is executed, this table
2645 is searched for the given value. If the value is found, control flow is
2646 transferred to the corresponding destination; otherwise, control flow is
2647 transferred to the default destination.</p>
2649 <h5>Implementation:</h5>
2650 <p>Depending on properties of the target machine and the particular
2651 <tt>switch</tt> instruction, this instruction may be code generated in
2652 different ways. For example, it could be generated as a series of chained
2653 conditional branches or with a lookup table.</p>
2657 <i>; Emulate a conditional br instruction</i>
2658 %Val = <a href="#i_zext">zext</a> i1 %value to i32
2659 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
2661 <i>; Emulate an unconditional br instruction</i>
2662 switch i32 0, label %dest [ ]
2664 <i>; Implement a jump table:</i>
2665 switch i32 %val, label %otherwise [ i32 0, label %onzero
2667 i32 2, label %ontwo ]
2672 <!-- _______________________________________________________________________ -->
2673 <div class="doc_subsubsection">
2674 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
2677 <div class="doc_text">
2681 <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>]
2682 to label <normal label> unwind label <exception label>
2686 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
2687 function, with the possibility of control flow transfer to either the
2688 '<tt>normal</tt>' label or the '<tt>exception</tt>' label. If the callee
2689 function returns with the "<tt><a href="#i_ret">ret</a></tt>" instruction,
2690 control flow will return to the "normal" label. If the callee (or any
2691 indirect callees) returns with the "<a href="#i_unwind"><tt>unwind</tt></a>"
2692 instruction, control is interrupted and continued at the dynamically nearest
2693 "exception" label.</p>
2696 <p>This instruction requires several arguments:</p>
2699 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
2700 convention</a> the call should use. If none is specified, the call
2701 defaults to using C calling conventions.</li>
2703 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
2704 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
2705 '<tt>inreg</tt>' attributes are valid here.</li>
2707 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
2708 function value being invoked. In most cases, this is a direct function
2709 invocation, but indirect <tt>invoke</tt>s are just as possible, branching
2710 off an arbitrary pointer to function value.</li>
2712 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
2713 function to be invoked. </li>
2715 <li>'<tt>function args</tt>': argument list whose types match the function
2716 signature argument types. If the function signature indicates the
2717 function accepts a variable number of arguments, the extra arguments can
2720 <li>'<tt>normal label</tt>': the label reached when the called function
2721 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
2723 <li>'<tt>exception label</tt>': the label reached when a callee returns with
2724 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
2726 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
2727 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
2728 '<tt>readnone</tt>' attributes are valid here.</li>
2732 <p>This instruction is designed to operate as a standard
2733 '<tt><a href="#i_call">call</a></tt>' instruction in most regards. The
2734 primary difference is that it establishes an association with a label, which
2735 is used by the runtime library to unwind the stack.</p>
2737 <p>This instruction is used in languages with destructors to ensure that proper
2738 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
2739 exception. Additionally, this is important for implementation of
2740 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
2742 <p>For the purposes of the SSA form, the definition of the value returned by the
2743 '<tt>invoke</tt>' instruction is deemed to occur on the edge from the current
2744 block to the "normal" label. If the callee unwinds then no return value is
2749 %retval = invoke i32 @Test(i32 15) to label %Continue
2750 unwind label %TestCleanup <i>; {i32}:retval set</i>
2751 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
2752 unwind label %TestCleanup <i>; {i32}:retval set</i>
2757 <!-- _______________________________________________________________________ -->
2759 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
2760 Instruction</a> </div>
2762 <div class="doc_text">
2770 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
2771 at the first callee in the dynamic call stack which used
2772 an <a href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call.
2773 This is primarily used to implement exception handling.</p>
2776 <p>The '<tt>unwind</tt>' instruction causes execution of the current function to
2777 immediately halt. The dynamic call stack is then searched for the
2778 first <a href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack.
2779 Once found, execution continues at the "exceptional" destination block
2780 specified by the <tt>invoke</tt> instruction. If there is no <tt>invoke</tt>
2781 instruction in the dynamic call chain, undefined behavior results.</p>
2785 <!-- _______________________________________________________________________ -->
2787 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
2788 Instruction</a> </div>
2790 <div class="doc_text">
2798 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
2799 instruction is used to inform the optimizer that a particular portion of the
2800 code is not reachable. This can be used to indicate that the code after a
2801 no-return function cannot be reached, and other facts.</p>
2804 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
2808 <!-- ======================================================================= -->
2809 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
2811 <div class="doc_text">
2813 <p>Binary operators are used to do most of the computation in a program. They
2814 require two operands of the same type, execute an operation on them, and
2815 produce a single value. The operands might represent multiple data, as is
2816 the case with the <a href="#t_vector">vector</a> data type. The result value
2817 has the same type as its operands.</p>
2819 <p>There are several different binary operators:</p>
2823 <!-- _______________________________________________________________________ -->
2824 <div class="doc_subsubsection">
2825 <a name="i_add">'<tt>add</tt>' Instruction</a>
2828 <div class="doc_text">
2832 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2833 <result> = add nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2834 <result> = add nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2835 <result> = add nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2839 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
2842 <p>The two arguments to the '<tt>add</tt>' instruction must
2843 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
2844 integer values. Both arguments must have identical types.</p>
2847 <p>The value produced is the integer sum of the two operands.</p>
2849 <p>If the sum has unsigned overflow, the result returned is the mathematical
2850 result modulo 2<sup>n</sup>, where n is the bit width of the result.</p>
2852 <p>Because LLVM integers use a two's complement representation, this instruction
2853 is appropriate for both signed and unsigned integers.</p>
2855 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
2856 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
2857 <tt>nsw</tt> keywords are present, the result value of the <tt>add</tt>
2858 is undefined if unsigned and/or signed overflow, respectively, occurs.</p>
2862 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
2867 <!-- _______________________________________________________________________ -->
2868 <div class="doc_subsubsection">
2869 <a name="i_fadd">'<tt>fadd</tt>' Instruction</a>
2872 <div class="doc_text">
2876 <result> = fadd <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2880 <p>The '<tt>fadd</tt>' instruction returns the sum of its two operands.</p>
2883 <p>The two arguments to the '<tt>fadd</tt>' instruction must be
2884 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
2885 floating point values. Both arguments must have identical types.</p>
2888 <p>The value produced is the floating point sum of the two operands.</p>
2892 <result> = fadd float 4.0, %var <i>; yields {float}:result = 4.0 + %var</i>
2897 <!-- _______________________________________________________________________ -->
2898 <div class="doc_subsubsection">
2899 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
2902 <div class="doc_text">
2906 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2907 <result> = sub nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2908 <result> = sub nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2909 <result> = sub nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2913 <p>The '<tt>sub</tt>' instruction returns the difference of its two
2916 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
2917 '<tt>neg</tt>' instruction present in most other intermediate
2918 representations.</p>
2921 <p>The two arguments to the '<tt>sub</tt>' instruction must
2922 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
2923 integer values. Both arguments must have identical types.</p>
2926 <p>The value produced is the integer difference of the two operands.</p>
2928 <p>If the difference has unsigned overflow, the result returned is the
2929 mathematical result modulo 2<sup>n</sup>, where n is the bit width of the
2932 <p>Because LLVM integers use a two's complement representation, this instruction
2933 is appropriate for both signed and unsigned integers.</p>
2935 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
2936 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
2937 <tt>nsw</tt> keywords are present, the result value of the <tt>sub</tt>
2938 is undefined if unsigned and/or signed overflow, respectively, occurs.</p>
2942 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
2943 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
2948 <!-- _______________________________________________________________________ -->
2949 <div class="doc_subsubsection">
2950 <a name="i_fsub">'<tt>fsub</tt>' Instruction</a>
2953 <div class="doc_text">
2957 <result> = fsub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2961 <p>The '<tt>fsub</tt>' instruction returns the difference of its two
2964 <p>Note that the '<tt>fsub</tt>' instruction is used to represent the
2965 '<tt>fneg</tt>' instruction present in most other intermediate
2966 representations.</p>
2969 <p>The two arguments to the '<tt>fsub</tt>' instruction must be
2970 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
2971 floating point values. Both arguments must have identical types.</p>
2974 <p>The value produced is the floating point difference of the two operands.</p>
2978 <result> = fsub float 4.0, %var <i>; yields {float}:result = 4.0 - %var</i>
2979 <result> = fsub float -0.0, %val <i>; yields {float}:result = -%var</i>
2984 <!-- _______________________________________________________________________ -->
2985 <div class="doc_subsubsection">
2986 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
2989 <div class="doc_text">
2993 <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2994 <result> = mul nuw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2995 <result> = mul nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2996 <result> = mul nuw nsw <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3000 <p>The '<tt>mul</tt>' instruction returns the product of its two operands.</p>
3003 <p>The two arguments to the '<tt>mul</tt>' instruction must
3004 be <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3005 integer values. Both arguments must have identical types.</p>
3008 <p>The value produced is the integer product of the two operands.</p>
3010 <p>If the result of the multiplication has unsigned overflow, the result
3011 returned is the mathematical result modulo 2<sup>n</sup>, where n is the bit
3012 width of the result.</p>
3014 <p>Because LLVM integers use a two's complement representation, and the result
3015 is the same width as the operands, this instruction returns the correct
3016 result for both signed and unsigned integers. If a full product
3017 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands should
3018 be sign-extended or zero-extended as appropriate to the width of the full
3021 <p><tt>nuw</tt> and <tt>nsw</tt> stand for "No Unsigned Wrap"
3022 and "No Signed Wrap", respectively. If the <tt>nuw</tt> and/or
3023 <tt>nsw</tt> keywords are present, the result value of the <tt>mul</tt>
3024 is undefined if unsigned and/or signed overflow, respectively, occurs.</p>
3028 <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
3033 <!-- _______________________________________________________________________ -->
3034 <div class="doc_subsubsection">
3035 <a name="i_fmul">'<tt>fmul</tt>' Instruction</a>
3038 <div class="doc_text">
3042 <result> = fmul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3046 <p>The '<tt>fmul</tt>' instruction returns the product of its two operands.</p>
3049 <p>The two arguments to the '<tt>fmul</tt>' instruction must be
3050 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3051 floating point values. Both arguments must have identical types.</p>
3054 <p>The value produced is the floating point product of the two operands.</p>
3058 <result> = fmul float 4.0, %var <i>; yields {float}:result = 4.0 * %var</i>
3063 <!-- _______________________________________________________________________ -->
3064 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
3067 <div class="doc_text">
3071 <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3075 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two operands.</p>
3078 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
3079 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3080 values. Both arguments must have identical types.</p>
3083 <p>The value produced is the unsigned integer quotient of the two operands.</p>
3085 <p>Note that unsigned integer division and signed integer division are distinct
3086 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
3088 <p>Division by zero leads to undefined behavior.</p>
3092 <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
3097 <!-- _______________________________________________________________________ -->
3098 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
3101 <div class="doc_text">
3105 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3106 <result> = sdiv exact <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3110 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two operands.</p>
3113 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
3114 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3115 values. Both arguments must have identical types.</p>
3118 <p>The value produced is the signed integer quotient of the two operands rounded
3121 <p>Note that signed integer division and unsigned integer division are distinct
3122 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
3124 <p>Division by zero leads to undefined behavior. Overflow also leads to
3125 undefined behavior; this is a rare case, but can occur, for example, by doing
3126 a 32-bit division of -2147483648 by -1.</p>
3128 <p>If the <tt>exact</tt> keyword is present, the result value of the
3129 <tt>sdiv</tt> is undefined if the result would be rounded or if overflow
3134 <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
3139 <!-- _______________________________________________________________________ -->
3140 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
3141 Instruction</a> </div>
3143 <div class="doc_text">
3147 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3151 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two operands.</p>
3154 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
3155 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3156 floating point values. Both arguments must have identical types.</p>
3159 <p>The value produced is the floating point quotient of the two operands.</p>
3163 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
3168 <!-- _______________________________________________________________________ -->
3169 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
3172 <div class="doc_text">
3176 <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3180 <p>The '<tt>urem</tt>' instruction returns the remainder from the unsigned
3181 division of its two arguments.</p>
3184 <p>The two arguments to the '<tt>urem</tt>' instruction must be
3185 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3186 values. Both arguments must have identical types.</p>
3189 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
3190 This instruction always performs an unsigned division to get the
3193 <p>Note that unsigned integer remainder and signed integer remainder are
3194 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
3196 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
3200 <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
3205 <!-- _______________________________________________________________________ -->
3206 <div class="doc_subsubsection">
3207 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
3210 <div class="doc_text">
3214 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3218 <p>The '<tt>srem</tt>' instruction returns the remainder from the signed
3219 division of its two operands. This instruction can also take
3220 <a href="#t_vector">vector</a> versions of the values in which case the
3221 elements must be integers.</p>
3224 <p>The two arguments to the '<tt>srem</tt>' instruction must be
3225 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3226 values. Both arguments must have identical types.</p>
3229 <p>This instruction returns the <i>remainder</i> of a division (where the result
3230 has the same sign as the dividend, <tt>op1</tt>), not the <i>modulo</i>
3231 operator (where the result has the same sign as the divisor, <tt>op2</tt>) of
3232 a value. For more information about the difference,
3233 see <a href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
3234 Math Forum</a>. For a table of how this is implemented in various languages,
3235 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
3236 Wikipedia: modulo operation</a>.</p>
3238 <p>Note that signed integer remainder and unsigned integer remainder are
3239 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
3241 <p>Taking the remainder of a division by zero leads to undefined behavior.
3242 Overflow also leads to undefined behavior; this is a rare case, but can
3243 occur, for example, by taking the remainder of a 32-bit division of
3244 -2147483648 by -1. (The remainder doesn't actually overflow, but this rule
3245 lets srem be implemented using instructions that return both the result of
3246 the division and the remainder.)</p>
3250 <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
3255 <!-- _______________________________________________________________________ -->
3256 <div class="doc_subsubsection">
3257 <a name="i_frem">'<tt>frem</tt>' Instruction</a> </div>
3259 <div class="doc_text">
3263 <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3267 <p>The '<tt>frem</tt>' instruction returns the remainder from the division of
3268 its two operands.</p>
3271 <p>The two arguments to the '<tt>frem</tt>' instruction must be
3272 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
3273 floating point values. Both arguments must have identical types.</p>
3276 <p>This instruction returns the <i>remainder</i> of a division. The remainder
3277 has the same sign as the dividend.</p>
3281 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
3286 <!-- ======================================================================= -->
3287 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
3288 Operations</a> </div>
3290 <div class="doc_text">
3292 <p>Bitwise binary operators are used to do various forms of bit-twiddling in a
3293 program. They are generally very efficient instructions and can commonly be
3294 strength reduced from other instructions. They require two operands of the
3295 same type, execute an operation on them, and produce a single value. The
3296 resulting value is the same type as its operands.</p>
3300 <!-- _______________________________________________________________________ -->
3301 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
3302 Instruction</a> </div>
3304 <div class="doc_text">
3308 <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3312 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to the left
3313 a specified number of bits.</p>
3316 <p>Both arguments to the '<tt>shl</tt>' instruction must be the
3317 same <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
3318 integer type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3321 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod
3322 2<sup>n</sup>, where <tt>n</tt> is the width of the result. If <tt>op2</tt>
3323 is (statically or dynamically) negative or equal to or larger than the number
3324 of bits in <tt>op1</tt>, the result is undefined. If the arguments are
3325 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
3326 shift amount in <tt>op2</tt>.</p>
3330 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
3331 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
3332 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
3333 <result> = shl i32 1, 32 <i>; undefined</i>
3334 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
3339 <!-- _______________________________________________________________________ -->
3340 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
3341 Instruction</a> </div>
3343 <div class="doc_text">
3347 <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3351 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
3352 operand shifted to the right a specified number of bits with zero fill.</p>
3355 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
3356 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3357 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3360 <p>This instruction always performs a logical shift right operation. The most
3361 significant bits of the result will be filled with zero bits after the shift.
3362 If <tt>op2</tt> is (statically or dynamically) equal to or larger than the
3363 number of bits in <tt>op1</tt>, the result is undefined. If the arguments are
3364 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding
3365 shift amount in <tt>op2</tt>.</p>
3369 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
3370 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
3371 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
3372 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
3373 <result> = lshr i32 1, 32 <i>; undefined</i>
3374 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
3379 <!-- _______________________________________________________________________ -->
3380 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
3381 Instruction</a> </div>
3382 <div class="doc_text">
3386 <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3390 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
3391 operand shifted to the right a specified number of bits with sign
3395 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
3396 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3397 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3400 <p>This instruction always performs an arithmetic shift right operation, The
3401 most significant bits of the result will be filled with the sign bit
3402 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
3403 larger than the number of bits in <tt>op1</tt>, the result is undefined. If
3404 the arguments are vectors, each vector element of <tt>op1</tt> is shifted by
3405 the corresponding shift amount in <tt>op2</tt>.</p>
3409 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
3410 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
3411 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
3412 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
3413 <result> = ashr i32 1, 32 <i>; undefined</i>
3414 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
3419 <!-- _______________________________________________________________________ -->
3420 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
3421 Instruction</a> </div>
3423 <div class="doc_text">
3427 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3431 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of its two
3435 <p>The two arguments to the '<tt>and</tt>' instruction must be
3436 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3437 values. Both arguments must have identical types.</p>
3440 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
3442 <table border="1" cellspacing="0" cellpadding="4">
3474 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
3475 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
3476 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
3479 <!-- _______________________________________________________________________ -->
3480 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
3482 <div class="doc_text">
3486 <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3490 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive or of its
3494 <p>The two arguments to the '<tt>or</tt>' instruction must be
3495 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3496 values. Both arguments must have identical types.</p>
3499 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
3501 <table border="1" cellspacing="0" cellpadding="4">
3533 <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
3534 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
3535 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
3540 <!-- _______________________________________________________________________ -->
3541 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
3542 Instruction</a> </div>
3544 <div class="doc_text">
3548 <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3552 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive or of
3553 its two operands. The <tt>xor</tt> is used to implement the "one's
3554 complement" operation, which is the "~" operator in C.</p>
3557 <p>The two arguments to the '<tt>xor</tt>' instruction must be
3558 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3559 values. Both arguments must have identical types.</p>
3562 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
3564 <table border="1" cellspacing="0" cellpadding="4">
3596 <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
3597 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
3598 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
3599 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
3604 <!-- ======================================================================= -->
3605 <div class="doc_subsection">
3606 <a name="vectorops">Vector Operations</a>
3609 <div class="doc_text">
3611 <p>LLVM supports several instructions to represent vector operations in a
3612 target-independent manner. These instructions cover the element-access and
3613 vector-specific operations needed to process vectors effectively. While LLVM
3614 does directly support these vector operations, many sophisticated algorithms
3615 will want to use target-specific intrinsics to take full advantage of a
3616 specific target.</p>
3620 <!-- _______________________________________________________________________ -->
3621 <div class="doc_subsubsection">
3622 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
3625 <div class="doc_text">
3629 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
3633 <p>The '<tt>extractelement</tt>' instruction extracts a single scalar element
3634 from a vector at a specified index.</p>
3638 <p>The first operand of an '<tt>extractelement</tt>' instruction is a value
3639 of <a href="#t_vector">vector</a> type. The second operand is an index
3640 indicating the position from which to extract the element. The index may be
3644 <p>The result is a scalar of the same type as the element type of
3645 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
3646 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
3647 results are undefined.</p>
3651 %result = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
3656 <!-- _______________________________________________________________________ -->
3657 <div class="doc_subsubsection">
3658 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
3661 <div class="doc_text">
3665 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
3669 <p>The '<tt>insertelement</tt>' instruction inserts a scalar element into a
3670 vector at a specified index.</p>
3673 <p>The first operand of an '<tt>insertelement</tt>' instruction is a value
3674 of <a href="#t_vector">vector</a> type. The second operand is a scalar value
3675 whose type must equal the element type of the first operand. The third
3676 operand is an index indicating the position at which to insert the value.
3677 The index may be a variable.</p>
3680 <p>The result is a vector of the same type as <tt>val</tt>. Its element values
3681 are those of <tt>val</tt> except at position <tt>idx</tt>, where it gets the
3682 value <tt>elt</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
3683 results are undefined.</p>
3687 %result = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
3692 <!-- _______________________________________________________________________ -->
3693 <div class="doc_subsubsection">
3694 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
3697 <div class="doc_text">
3701 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
3705 <p>The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
3706 from two input vectors, returning a vector with the same element type as the
3707 input and length that is the same as the shuffle mask.</p>
3710 <p>The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
3711 with types that match each other. The third argument is a shuffle mask whose
3712 element type is always 'i32'. The result of the instruction is a vector
3713 whose length is the same as the shuffle mask and whose element type is the
3714 same as the element type of the first two operands.</p>
3716 <p>The shuffle mask operand is required to be a constant vector with either
3717 constant integer or undef values.</p>
3720 <p>The elements of the two input vectors are numbered from left to right across
3721 both of the vectors. The shuffle mask operand specifies, for each element of
3722 the result vector, which element of the two input vectors the result element
3723 gets. The element selector may be undef (meaning "don't care") and the
3724 second operand may be undef if performing a shuffle from only one vector.</p>
3728 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
3729 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
3730 %result = shufflevector <4 x i32> %v1, <4 x i32> undef,
3731 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
3732 %result = shufflevector <8 x i32> %v1, <8 x i32> undef,
3733 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i>
3734 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
3735 <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>
3740 <!-- ======================================================================= -->
3741 <div class="doc_subsection">
3742 <a name="aggregateops">Aggregate Operations</a>
3745 <div class="doc_text">
3747 <p>LLVM supports several instructions for working with aggregate values.</p>
3751 <!-- _______________________________________________________________________ -->
3752 <div class="doc_subsubsection">
3753 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
3756 <div class="doc_text">
3760 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
3764 <p>The '<tt>extractvalue</tt>' instruction extracts the value of a struct field
3765 or array element from an aggregate value.</p>
3768 <p>The first operand of an '<tt>extractvalue</tt>' instruction is a value
3769 of <a href="#t_struct">struct</a> or <a href="#t_array">array</a> type. The
3770 operands are constant indices to specify which value to extract in a similar
3771 manner as indices in a
3772 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.</p>
3775 <p>The result is the value at the position in the aggregate specified by the
3780 %result = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
3785 <!-- _______________________________________________________________________ -->
3786 <div class="doc_subsubsection">
3787 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
3790 <div class="doc_text">
3794 <result> = insertvalue <aggregate type> <val>, <ty> <val>, <idx> <i>; yields <n x <ty>></i>
3798 <p>The '<tt>insertvalue</tt>' instruction inserts a value into a struct field or
3799 array element in an aggregate.</p>
3803 <p>The first operand of an '<tt>insertvalue</tt>' instruction is a value
3804 of <a href="#t_struct">struct</a> or <a href="#t_array">array</a> type. The
3805 second operand is a first-class value to insert. The following operands are
3806 constant indices indicating the position at which to insert the value in a
3807 similar manner as indices in a
3808 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction. The
3809 value to insert must have the same type as the value identified by the
3813 <p>The result is an aggregate of the same type as <tt>val</tt>. Its value is
3814 that of <tt>val</tt> except that the value at the position specified by the
3815 indices is that of <tt>elt</tt>.</p>
3819 %result = insertvalue {i32, float} %agg, i32 1, 0 <i>; yields {i32, float}</i>
3825 <!-- ======================================================================= -->
3826 <div class="doc_subsection">
3827 <a name="memoryops">Memory Access and Addressing Operations</a>
3830 <div class="doc_text">
3832 <p>A key design point of an SSA-based representation is how it represents
3833 memory. In LLVM, no memory locations are in SSA form, which makes things
3834 very simple. This section describes how to read, write, and allocate
3839 <!-- _______________________________________________________________________ -->
3840 <div class="doc_subsubsection">
3841 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
3844 <div class="doc_text">
3848 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3852 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
3853 currently executing function, to be automatically released when this function
3854 returns to its caller. The object is always allocated in the generic address
3855 space (address space zero).</p>
3858 <p>The '<tt>alloca</tt>' instruction
3859 allocates <tt>sizeof(<type>)*NumElements</tt> bytes of memory on the
3860 runtime stack, returning a pointer of the appropriate type to the program.
3861 If "NumElements" is specified, it is the number of elements allocated,
3862 otherwise "NumElements" is defaulted to be one. If a constant alignment is
3863 specified, the value result of the allocation is guaranteed to be aligned to
3864 at least that boundary. If not specified, or if zero, the target can choose
3865 to align the allocation on any convenient boundary compatible with the
3868 <p>'<tt>type</tt>' may be any sized type.</p>
3871 <p>Memory is allocated; a pointer is returned. The operation is undefined if
3872 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
3873 memory is automatically released when the function returns. The
3874 '<tt>alloca</tt>' instruction is commonly used to represent automatic
3875 variables that must have an address available. When the function returns
3876 (either with the <tt><a href="#i_ret">ret</a></tt>
3877 or <tt><a href="#i_unwind">unwind</a></tt> instructions), the memory is
3878 reclaimed. Allocating zero bytes is legal, but the result is undefined.</p>
3882 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
3883 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
3884 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
3885 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
3890 <!-- _______________________________________________________________________ -->
3891 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
3892 Instruction</a> </div>
3894 <div class="doc_text">
3898 <result> = load <ty>* <pointer>[, align <alignment>]
3899 <result> = volatile load <ty>* <pointer>[, align <alignment>]
3903 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
3906 <p>The argument to the '<tt>load</tt>' instruction specifies the memory address
3907 from which to load. The pointer must point to
3908 a <a href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
3909 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify the
3910 number or order of execution of this <tt>load</tt> with other
3911 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
3914 <p>The optional constant "align" argument specifies the alignment of the
3915 operation (that is, the alignment of the memory address). A value of 0 or an
3916 omitted "align" argument means that the operation has the preferential
3917 alignment for the target. It is the responsibility of the code emitter to
3918 ensure that the alignment information is correct. Overestimating the
3919 alignment results in an undefined behavior. Underestimating the alignment may
3920 produce less efficient code. An alignment of 1 is always safe.</p>
3923 <p>The location of memory pointed to is loaded. If the value being loaded is of
3924 scalar type then the number of bytes read does not exceed the minimum number
3925 of bytes needed to hold all bits of the type. For example, loading an
3926 <tt>i24</tt> reads at most three bytes. When loading a value of a type like
3927 <tt>i20</tt> with a size that is not an integral number of bytes, the result
3928 is undefined if the value was not originally written using a store of the
3933 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3934 <a href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
3935 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
3940 <!-- _______________________________________________________________________ -->
3941 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
3942 Instruction</a> </div>
3944 <div class="doc_text">
3948 store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3949 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3953 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
3956 <p>There are two arguments to the '<tt>store</tt>' instruction: a value to store
3957 and an address at which to store it. The type of the
3958 '<tt><pointer></tt>' operand must be a pointer to
3959 the <a href="#t_firstclass">first class</a> type of the
3960 '<tt><value></tt>' operand. If the <tt>store</tt> is marked
3961 as <tt>volatile</tt>, then the optimizer is not allowed to modify the number
3962 or order of execution of this <tt>store</tt> with other
3963 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
3966 <p>The optional constant "align" argument specifies the alignment of the
3967 operation (that is, the alignment of the memory address). A value of 0 or an
3968 omitted "align" argument means that the operation has the preferential
3969 alignment for the target. It is the responsibility of the code emitter to
3970 ensure that the alignment information is correct. Overestimating the
3971 alignment results in an undefined behavior. Underestimating the alignment may
3972 produce less efficient code. An alignment of 1 is always safe.</p>
3975 <p>The contents of memory are updated to contain '<tt><value></tt>' at the
3976 location specified by the '<tt><pointer></tt>' operand. If
3977 '<tt><value></tt>' is of scalar type then the number of bytes written
3978 does not exceed the minimum number of bytes needed to hold all bits of the
3979 type. For example, storing an <tt>i24</tt> writes at most three bytes. When
3980 writing a value of a type like <tt>i20</tt> with a size that is not an
3981 integral number of bytes, it is unspecified what happens to the extra bits
3982 that do not belong to the type, but they will typically be overwritten.</p>
3986 %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3987 store i32 3, i32* %ptr <i>; yields {void}</i>
3988 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
3993 <!-- _______________________________________________________________________ -->
3994 <div class="doc_subsubsection">
3995 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
3998 <div class="doc_text">
4002 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
4003 <result> = getelementptr inbounds <pty>* <ptrval>{, <ty> <idx>}*
4007 <p>The '<tt>getelementptr</tt>' instruction is used to get the address of a
4008 subelement of an aggregate data structure. It performs address calculation
4009 only and does not access memory.</p>
4012 <p>The first argument is always a pointer, and forms the basis of the
4013 calculation. The remaining arguments are indices that indicate which of the
4014 elements of the aggregate object are indexed. The interpretation of each
4015 index is dependent on the type being indexed into. The first index always
4016 indexes the pointer value given as the first argument, the second index
4017 indexes a value of the type pointed to (not necessarily the value directly
4018 pointed to, since the first index can be non-zero), etc. The first type
4019 indexed into must be a pointer value, subsequent types can be arrays, vectors
4020 and structs. Note that subsequent types being indexed into can never be
4021 pointers, since that would require loading the pointer before continuing
4024 <p>The type of each index argument depends on the type it is indexing into.
4025 When indexing into a (optionally packed) structure, only <tt>i32</tt> integer
4026 <b>constants</b> are allowed. When indexing into an array, pointer or
4027 vector, integers of any width are allowed, and they are not required to be
4030 <p>For example, let's consider a C code fragment and how it gets compiled to
4033 <div class="doc_code">
4046 int *foo(struct ST *s) {
4047 return &s[1].Z.B[5][13];
4052 <p>The LLVM code generated by the GCC frontend is:</p>
4054 <div class="doc_code">
4056 %RT = <a href="#namedtypes">type</a> { i8 , [10 x [20 x i32]], i8 }
4057 %ST = <a href="#namedtypes">type</a> { i32, double, %RT }
4059 define i32* @foo(%ST* %s) {
4061 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
4068 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
4069 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
4070 }</tt>' type, a structure. The second index indexes into the third element
4071 of the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
4072 i8 }</tt>' type, another structure. The third index indexes into the second
4073 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
4074 array. The two dimensions of the array are subscripted into, yielding an
4075 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a
4076 pointer to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
4078 <p>Note that it is perfectly legal to index partially through a structure,
4079 returning a pointer to an inner element. Because of this, the LLVM code for
4080 the given testcase is equivalent to:</p>
4083 define i32* @foo(%ST* %s) {
4084 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
4085 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
4086 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
4087 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
4088 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
4093 <p>If the <tt>inbounds</tt> keyword is present, the result value of the
4094 <tt>getelementptr</tt> is undefined if the base pointer is not an
4095 <i>in bounds</i> address of an allocated object, or if any of the addresses
4096 that would be formed by successive addition of the offsets implied by the
4097 indices to the base address with infinitely precise arithmetic are not an
4098 <i>in bounds</i> address of that allocated object.
4099 The <i>in bounds</i> addresses for an allocated object are all the addresses
4100 that point into the object, plus the address one byte past the end.</p>
4102 <p>If the <tt>inbounds</tt> keyword is not present, the offsets are added to
4103 the base address with silently-wrapping two's complement arithmetic, and
4104 the result value of the <tt>getelementptr</tt> may be outside the object
4105 pointed to by the base pointer. The result value may not necessarily be
4106 used to access memory though, even if it happens to point into allocated
4107 storage. See the <a href="#pointeraliasing">Pointer Aliasing Rules</a>
4108 section for more information.</p>
4110 <p>The getelementptr instruction is often confusing. For some more insight into
4111 how it works, see <a href="GetElementPtr.html">the getelementptr FAQ</a>.</p>
4115 <i>; yields [12 x i8]*:aptr</i>
4116 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
4117 <i>; yields i8*:vptr</i>
4118 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
4119 <i>; yields i8*:eptr</i>
4120 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
4121 <i>; yields i32*:iptr</i>
4122 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
4127 <!-- ======================================================================= -->
4128 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
4131 <div class="doc_text">
4133 <p>The instructions in this category are the conversion instructions (casting)
4134 which all take a single operand and a type. They perform various bit
4135 conversions on the operand.</p>
4139 <!-- _______________________________________________________________________ -->
4140 <div class="doc_subsubsection">
4141 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
4143 <div class="doc_text">
4147 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
4151 <p>The '<tt>trunc</tt>' instruction truncates its operand to the
4152 type <tt>ty2</tt>.</p>
4155 <p>The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
4156 be an <a href="#t_integer">integer</a> type, and a type that specifies the
4157 size and type of the result, which must be
4158 an <a href="#t_integer">integer</a> type. The bit size of <tt>value</tt> must
4159 be larger than the bit size of <tt>ty2</tt>. Equal sized types are not
4163 <p>The '<tt>trunc</tt>' instruction truncates the high order bits
4164 in <tt>value</tt> and converts the remaining bits to <tt>ty2</tt>. Since the
4165 source size must be larger than the destination size, <tt>trunc</tt> cannot
4166 be a <i>no-op cast</i>. It will always truncate bits.</p>
4170 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
4171 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
4172 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
4177 <!-- _______________________________________________________________________ -->
4178 <div class="doc_subsubsection">
4179 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
4181 <div class="doc_text">
4185 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
4189 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
4194 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
4195 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
4196 also be of <a href="#t_integer">integer</a> type. The bit size of the
4197 <tt>value</tt> must be smaller than the bit size of the destination type,
4201 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
4202 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
4204 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
4208 %X = zext i32 257 to i64 <i>; yields i64:257</i>
4209 %Y = zext i1 true to i32 <i>; yields i32:1</i>
4214 <!-- _______________________________________________________________________ -->
4215 <div class="doc_subsubsection">
4216 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
4218 <div class="doc_text">
4222 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
4226 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
4229 <p>The '<tt>sext</tt>' instruction takes a value to cast, which must be of
4230 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
4231 also be of <a href="#t_integer">integer</a> type. The bit size of the
4232 <tt>value</tt> must be smaller than the bit size of the destination type,
4236 <p>The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
4237 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size
4238 of the type <tt>ty2</tt>.</p>
4240 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
4244 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
4245 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
4250 <!-- _______________________________________________________________________ -->
4251 <div class="doc_subsubsection">
4252 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
4255 <div class="doc_text">
4259 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
4263 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
4267 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
4268 point</a> value to cast and a <a href="#t_floating">floating point</a> type
4269 to cast it to. The size of <tt>value</tt> must be larger than the size of
4270 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
4271 <i>no-op cast</i>.</p>
4274 <p>The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
4275 <a href="#t_floating">floating point</a> type to a smaller
4276 <a href="#t_floating">floating point</a> type. If the value cannot fit
4277 within the destination type, <tt>ty2</tt>, then the results are
4282 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
4283 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
4288 <!-- _______________________________________________________________________ -->
4289 <div class="doc_subsubsection">
4290 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
4292 <div class="doc_text">
4296 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
4300 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
4301 floating point value.</p>
4304 <p>The '<tt>fpext</tt>' instruction takes a
4305 <a href="#t_floating">floating point</a> <tt>value</tt> to cast, and
4306 a <a href="#t_floating">floating point</a> type to cast it to. The source
4307 type must be smaller than the destination type.</p>
4310 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
4311 <a href="#t_floating">floating point</a> type to a larger
4312 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
4313 used to make a <i>no-op cast</i> because it always changes bits. Use
4314 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
4318 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
4319 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
4324 <!-- _______________________________________________________________________ -->
4325 <div class="doc_subsubsection">
4326 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
4328 <div class="doc_text">
4332 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
4336 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
4337 unsigned integer equivalent of type <tt>ty2</tt>.</p>
4340 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
4341 scalar or vector <a href="#t_floating">floating point</a> value, and a type
4342 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
4343 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
4344 vector integer type with the same number of elements as <tt>ty</tt></p>
4347 <p>The '<tt>fptoui</tt>' instruction converts its
4348 <a href="#t_floating">floating point</a> operand into the nearest (rounding
4349 towards zero) unsigned integer value. If the value cannot fit
4350 in <tt>ty2</tt>, the results are undefined.</p>
4354 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
4355 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
4356 %X = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
4361 <!-- _______________________________________________________________________ -->
4362 <div class="doc_subsubsection">
4363 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
4365 <div class="doc_text">
4369 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
4373 <p>The '<tt>fptosi</tt>' instruction converts
4374 <a href="#t_floating">floating point</a> <tt>value</tt> to
4375 type <tt>ty2</tt>.</p>
4378 <p>The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
4379 scalar or vector <a href="#t_floating">floating point</a> value, and a type
4380 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
4381 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
4382 vector integer type with the same number of elements as <tt>ty</tt></p>
4385 <p>The '<tt>fptosi</tt>' instruction converts its
4386 <a href="#t_floating">floating point</a> operand into the nearest (rounding
4387 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
4388 the results are undefined.</p>
4392 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
4393 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
4394 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
4399 <!-- _______________________________________________________________________ -->
4400 <div class="doc_subsubsection">
4401 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
4403 <div class="doc_text">
4407 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
4411 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
4412 integer and converts that value to the <tt>ty2</tt> type.</p>
4415 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
4416 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
4417 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
4418 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
4419 floating point type with the same number of elements as <tt>ty</tt></p>
4422 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
4423 integer quantity and converts it to the corresponding floating point
4424 value. If the value cannot fit in the floating point value, the results are
4429 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
4430 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
4435 <!-- _______________________________________________________________________ -->
4436 <div class="doc_subsubsection">
4437 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
4439 <div class="doc_text">
4443 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
4447 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed integer
4448 and converts that value to the <tt>ty2</tt> type.</p>
4451 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
4452 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast
4453 it to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
4454 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
4455 floating point type with the same number of elements as <tt>ty</tt></p>
4458 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed integer
4459 quantity and converts it to the corresponding floating point value. If the
4460 value cannot fit in the floating point value, the results are undefined.</p>
4464 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
4465 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
4470 <!-- _______________________________________________________________________ -->
4471 <div class="doc_subsubsection">
4472 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
4474 <div class="doc_text">
4478 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
4482 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
4483 the integer type <tt>ty2</tt>.</p>
4486 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
4487 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
4488 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.</p>
4491 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
4492 <tt>ty2</tt> by interpreting the pointer value as an integer and either
4493 truncating or zero extending that value to the size of the integer type. If
4494 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
4495 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
4496 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
4501 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
4502 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
4507 <!-- _______________________________________________________________________ -->
4508 <div class="doc_subsubsection">
4509 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
4511 <div class="doc_text">
4515 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
4519 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to a
4520 pointer type, <tt>ty2</tt>.</p>
4523 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
4524 value to cast, and a type to cast it to, which must be a
4525 <a href="#t_pointer">pointer</a> type.</p>
4528 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
4529 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
4530 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
4531 size of a pointer then a truncation is done. If <tt>value</tt> is smaller
4532 than the size of a pointer then a zero extension is done. If they are the
4533 same size, nothing is done (<i>no-op cast</i>).</p>
4537 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
4538 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
4539 %Y = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
4544 <!-- _______________________________________________________________________ -->
4545 <div class="doc_subsubsection">
4546 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
4548 <div class="doc_text">
4552 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
4556 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
4557 <tt>ty2</tt> without changing any bits.</p>
4560 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be a
4561 non-aggregate first class value, and a type to cast it to, which must also be
4562 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes
4563 of <tt>value</tt> and the destination type, <tt>ty2</tt>, must be
4564 identical. If the source type is a pointer, the destination type must also be
4565 a pointer. This instruction supports bitwise conversion of vectors to
4566 integers and to vectors of other types (as long as they have the same
4570 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
4571 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
4572 this conversion. The conversion is done as if the <tt>value</tt> had been
4573 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only
4574 be converted to other pointer types with this instruction. To convert
4575 pointers to other types, use the <a href="#i_inttoptr">inttoptr</a> or
4576 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
4580 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
4581 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
4582 %Z = bitcast <2 x int> %V to i64; <i>; yields i64: %V</i>
4587 <!-- ======================================================================= -->
4588 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
4590 <div class="doc_text">
4592 <p>The instructions in this category are the "miscellaneous" instructions, which
4593 defy better classification.</p>
4597 <!-- _______________________________________________________________________ -->
4598 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
4601 <div class="doc_text">
4605 <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4609 <p>The '<tt>icmp</tt>' instruction returns a boolean value or a vector of
4610 boolean values based on comparison of its two integer, integer vector, or
4611 pointer operands.</p>
4614 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
4615 the condition code indicating the kind of comparison to perform. It is not a
4616 value, just a keyword. The possible condition code are:</p>
4619 <li><tt>eq</tt>: equal</li>
4620 <li><tt>ne</tt>: not equal </li>
4621 <li><tt>ugt</tt>: unsigned greater than</li>
4622 <li><tt>uge</tt>: unsigned greater or equal</li>
4623 <li><tt>ult</tt>: unsigned less than</li>
4624 <li><tt>ule</tt>: unsigned less or equal</li>
4625 <li><tt>sgt</tt>: signed greater than</li>
4626 <li><tt>sge</tt>: signed greater or equal</li>
4627 <li><tt>slt</tt>: signed less than</li>
4628 <li><tt>sle</tt>: signed less or equal</li>
4631 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
4632 <a href="#t_pointer">pointer</a> or integer <a href="#t_vector">vector</a>
4633 typed. They must also be identical types.</p>
4636 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to the
4637 condition code given as <tt>cond</tt>. The comparison performed always yields
4638 either an <a href="#t_integer"><tt>i1</tt></a> or vector of <tt>i1</tt>
4639 result, as follows:</p>
4642 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
4643 <tt>false</tt> otherwise. No sign interpretation is necessary or
4646 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
4647 <tt>false</tt> otherwise. No sign interpretation is necessary or
4650 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
4651 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4653 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
4654 <tt>true</tt> if <tt>op1</tt> is greater than or equal
4655 to <tt>op2</tt>.</li>
4657 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
4658 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
4660 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
4661 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4663 <li><tt>sgt</tt>: interprets the operands as signed values and yields
4664 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4666 <li><tt>sge</tt>: interprets the operands as signed values and yields
4667 <tt>true</tt> if <tt>op1</tt> is greater than or equal
4668 to <tt>op2</tt>.</li>
4670 <li><tt>slt</tt>: interprets the operands as signed values and yields
4671 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
4673 <li><tt>sle</tt>: interprets the operands as signed values and yields
4674 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4677 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
4678 values are compared as if they were integers.</p>
4680 <p>If the operands are integer vectors, then they are compared element by
4681 element. The result is an <tt>i1</tt> vector with the same number of elements
4682 as the values being compared. Otherwise, the result is an <tt>i1</tt>.</p>
4686 <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
4687 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
4688 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
4689 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
4690 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
4691 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
4694 <p>Note that the code generator does not yet support vector types with
4695 the <tt>icmp</tt> instruction.</p>
4699 <!-- _______________________________________________________________________ -->
4700 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
4703 <div class="doc_text">
4707 <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4711 <p>The '<tt>fcmp</tt>' instruction returns a boolean value or vector of boolean
4712 values based on comparison of its operands.</p>
4714 <p>If the operands are floating point scalars, then the result type is a boolean
4715 (<a href="#t_integer"><tt>i1</tt></a>).</p>
4717 <p>If the operands are floating point vectors, then the result type is a vector
4718 of boolean with the same number of elements as the operands being
4722 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
4723 the condition code indicating the kind of comparison to perform. It is not a
4724 value, just a keyword. The possible condition code are:</p>
4727 <li><tt>false</tt>: no comparison, always returns false</li>
4728 <li><tt>oeq</tt>: ordered and equal</li>
4729 <li><tt>ogt</tt>: ordered and greater than </li>
4730 <li><tt>oge</tt>: ordered and greater than or equal</li>
4731 <li><tt>olt</tt>: ordered and less than </li>
4732 <li><tt>ole</tt>: ordered and less than or equal</li>
4733 <li><tt>one</tt>: ordered and not equal</li>
4734 <li><tt>ord</tt>: ordered (no nans)</li>
4735 <li><tt>ueq</tt>: unordered or equal</li>
4736 <li><tt>ugt</tt>: unordered or greater than </li>
4737 <li><tt>uge</tt>: unordered or greater than or equal</li>
4738 <li><tt>ult</tt>: unordered or less than </li>
4739 <li><tt>ule</tt>: unordered or less than or equal</li>
4740 <li><tt>une</tt>: unordered or not equal</li>
4741 <li><tt>uno</tt>: unordered (either nans)</li>
4742 <li><tt>true</tt>: no comparison, always returns true</li>
4745 <p><i>Ordered</i> means that neither operand is a QNAN while
4746 <i>unordered</i> means that either operand may be a QNAN.</p>
4748 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be either
4749 a <a href="#t_floating">floating point</a> type or
4750 a <a href="#t_vector">vector</a> of floating point type. They must have
4751 identical types.</p>
4754 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4755 according to the condition code given as <tt>cond</tt>. If the operands are
4756 vectors, then the vectors are compared element by element. Each comparison
4757 performed always yields an <a href="#t_integer">i1</a> result, as
4761 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
4763 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4764 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
4766 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4767 <tt>op1</tt> is greather than <tt>op2</tt>.</li>
4769 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4770 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4772 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4773 <tt>op1</tt> is less than <tt>op2</tt>.</li>
4775 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4776 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4778 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4779 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
4781 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
4783 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
4784 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
4786 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
4787 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4789 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
4790 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4792 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
4793 <tt>op1</tt> is less than <tt>op2</tt>.</li>
4795 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
4796 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4798 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
4799 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
4801 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
4803 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
4808 <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
4809 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
4810 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
4811 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
4814 <p>Note that the code generator does not yet support vector types with
4815 the <tt>fcmp</tt> instruction.</p>
4819 <!-- _______________________________________________________________________ -->
4820 <div class="doc_subsubsection">
4821 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
4824 <div class="doc_text">
4828 <result> = phi <ty> [ <val0>, <label0>], ...
4832 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in the
4833 SSA graph representing the function.</p>
4836 <p>The type of the incoming values is specified with the first type field. After
4837 this, the '<tt>phi</tt>' instruction takes a list of pairs as arguments, with
4838 one pair for each predecessor basic block of the current block. Only values
4839 of <a href="#t_firstclass">first class</a> type may be used as the value
4840 arguments to the PHI node. Only labels may be used as the label
4843 <p>There must be no non-phi instructions between the start of a basic block and
4844 the PHI instructions: i.e. PHI instructions must be first in a basic
4847 <p>For the purposes of the SSA form, the use of each incoming value is deemed to
4848 occur on the edge from the corresponding predecessor block to the current
4849 block (but after any definition of an '<tt>invoke</tt>' instruction's return
4850 value on the same edge).</p>
4853 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
4854 specified by the pair corresponding to the predecessor basic block that
4855 executed just prior to the current block.</p>
4859 Loop: ; Infinite loop that counts from 0 on up...
4860 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
4861 %nextindvar = add i32 %indvar, 1
4867 <!-- _______________________________________________________________________ -->
4868 <div class="doc_subsubsection">
4869 <a name="i_select">'<tt>select</tt>' Instruction</a>
4872 <div class="doc_text">
4876 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
4878 <i>selty</i> is either i1 or {<N x i1>}
4882 <p>The '<tt>select</tt>' instruction is used to choose one value based on a
4883 condition, without branching.</p>
4887 <p>The '<tt>select</tt>' instruction requires an 'i1' value or a vector of 'i1'
4888 values indicating the condition, and two values of the
4889 same <a href="#t_firstclass">first class</a> type. If the val1/val2 are
4890 vectors and the condition is a scalar, then entire vectors are selected, not
4891 individual elements.</p>
4894 <p>If the condition is an i1 and it evaluates to 1, the instruction returns the
4895 first value argument; otherwise, it returns the second value argument.</p>
4897 <p>If the condition is a vector of i1, then the value arguments must be vectors
4898 of the same size, and the selection is done element by element.</p>
4902 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
4905 <p>Note that the code generator does not yet support conditions
4906 with vector type.</p>
4910 <!-- _______________________________________________________________________ -->
4911 <div class="doc_subsubsection">
4912 <a name="i_call">'<tt>call</tt>' Instruction</a>
4915 <div class="doc_text">
4919 <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>]
4923 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
4926 <p>This instruction requires several arguments:</p>
4929 <li>The optional "tail" marker indicates whether the callee function accesses
4930 any allocas or varargs in the caller. If the "tail" marker is present,
4931 the function call is eligible for tail call optimization. Note that calls
4932 may be marked "tail" even if they do not occur before
4933 a <a href="#i_ret"><tt>ret</tt></a> instruction.</li>
4935 <li>The optional "cconv" marker indicates which <a href="#callingconv">calling
4936 convention</a> the call should use. If none is specified, the call
4937 defaults to using C calling conventions.</li>
4939 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
4940 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>', and
4941 '<tt>inreg</tt>' attributes are valid here.</li>
4943 <li>'<tt>ty</tt>': the type of the call instruction itself which is also the
4944 type of the return value. Functions that return no value are marked
4945 <tt><a href="#t_void">void</a></tt>.</li>
4947 <li>'<tt>fnty</tt>': shall be the signature of the pointer to function value
4948 being invoked. The argument types must match the types implied by this
4949 signature. This type can be omitted if the function is not varargs and if
4950 the function type does not return a pointer to a function.</li>
4952 <li>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
4953 be invoked. In most cases, this is a direct function invocation, but
4954 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
4955 to function value.</li>
4957 <li>'<tt>function args</tt>': argument list whose types match the function
4958 signature argument types. All arguments must be of
4959 <a href="#t_firstclass">first class</a> type. If the function signature
4960 indicates the function accepts a variable number of arguments, the extra
4961 arguments can be specified.</li>
4963 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
4964 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
4965 '<tt>readnone</tt>' attributes are valid here.</li>
4969 <p>The '<tt>call</tt>' instruction is used to cause control flow to transfer to
4970 a specified function, with its incoming arguments bound to the specified
4971 values. Upon a '<tt><a href="#i_ret">ret</a></tt>' instruction in the called
4972 function, control flow continues with the instruction after the function
4973 call, and the return value of the function is bound to the result
4978 %retval = call i32 @test(i32 %argc)
4979 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42) <i>; yields i32</i>
4980 %X = tail call i32 @foo() <i>; yields i32</i>
4981 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
4982 call void %foo(i8 97 signext)
4984 %struct.A = type { i32, i8 }
4985 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
4986 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
4987 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
4988 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
4989 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
4992 <p>llvm treats calls to some functions with names and arguments that match the
4993 standard C99 library as being the C99 library functions, and may perform
4994 optimizations or generate code for them under that assumption. This is
4995 something we'd like to change in the future to provide better support for
4996 freestanding environments and non-C-based langauges.</p>
5000 <!-- _______________________________________________________________________ -->
5001 <div class="doc_subsubsection">
5002 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
5005 <div class="doc_text">
5009 <resultval> = va_arg <va_list*> <arglist>, <argty>
5013 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
5014 the "variable argument" area of a function call. It is used to implement the
5015 <tt>va_arg</tt> macro in C.</p>
5018 <p>This instruction takes a <tt>va_list*</tt> value and the type of the
5019 argument. It returns a value of the specified argument type and increments
5020 the <tt>va_list</tt> to point to the next argument. The actual type
5021 of <tt>va_list</tt> is target specific.</p>
5024 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified type
5025 from the specified <tt>va_list</tt> and causes the <tt>va_list</tt> to point
5026 to the next argument. For more information, see the variable argument
5027 handling <a href="#int_varargs">Intrinsic Functions</a>.</p>
5029 <p>It is legal for this instruction to be called in a function which does not
5030 take a variable number of arguments, for example, the <tt>vfprintf</tt>
5033 <p><tt>va_arg</tt> is an LLVM instruction instead of
5034 an <a href="#intrinsics">intrinsic function</a> because it takes a type as an
5038 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
5040 <p>Note that the code generator does not yet fully support va_arg on many
5041 targets. Also, it does not currently support va_arg with aggregate types on
5046 <!-- *********************************************************************** -->
5047 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
5048 <!-- *********************************************************************** -->
5050 <div class="doc_text">
5052 <p>LLVM supports the notion of an "intrinsic function". These functions have
5053 well known names and semantics and are required to follow certain
5054 restrictions. Overall, these intrinsics represent an extension mechanism for
5055 the LLVM language that does not require changing all of the transformations
5056 in LLVM when adding to the language (or the bitcode reader/writer, the
5057 parser, etc...).</p>
5059 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
5060 prefix is reserved in LLVM for intrinsic names; thus, function names may not
5061 begin with this prefix. Intrinsic functions must always be external
5062 functions: you cannot define the body of intrinsic functions. Intrinsic
5063 functions may only be used in call or invoke instructions: it is illegal to
5064 take the address of an intrinsic function. Additionally, because intrinsic
5065 functions are part of the LLVM language, it is required if any are added that
5066 they be documented here.</p>
5068 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents a
5069 family of functions that perform the same operation but on different data
5070 types. Because LLVM can represent over 8 million different integer types,
5071 overloading is used commonly to allow an intrinsic function to operate on any
5072 integer type. One or more of the argument types or the result type can be
5073 overloaded to accept any integer type. Argument types may also be defined as
5074 exactly matching a previous argument's type or the result type. This allows
5075 an intrinsic function which accepts multiple arguments, but needs all of them
5076 to be of the same type, to only be overloaded with respect to a single
5077 argument or the result.</p>
5079 <p>Overloaded intrinsics will have the names of its overloaded argument types
5080 encoded into its function name, each preceded by a period. Only those types
5081 which are overloaded result in a name suffix. Arguments whose type is matched
5082 against another type do not. For example, the <tt>llvm.ctpop</tt> function
5083 can take an integer of any width and returns an integer of exactly the same
5084 integer width. This leads to a family of functions such as
5085 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29
5086 %val)</tt>. Only one type, the return type, is overloaded, and only one type
5087 suffix is required. Because the argument's type is matched against the return
5088 type, it does not require its own name suffix.</p>
5090 <p>To learn how to add an intrinsic function, please see the
5091 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.</p>
5095 <!-- ======================================================================= -->
5096 <div class="doc_subsection">
5097 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
5100 <div class="doc_text">
5102 <p>Variable argument support is defined in LLVM with
5103 the <a href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
5104 intrinsic functions. These functions are related to the similarly named
5105 macros defined in the <tt><stdarg.h></tt> header file.</p>
5107 <p>All of these functions operate on arguments that use a target-specific value
5108 type "<tt>va_list</tt>". The LLVM assembly language reference manual does
5109 not define what this type is, so all transformations should be prepared to
5110 handle these functions regardless of the type used.</p>
5112 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
5113 instruction and the variable argument handling intrinsic functions are
5116 <div class="doc_code">
5118 define i32 @test(i32 %X, ...) {
5119 ; Initialize variable argument processing
5121 %ap2 = bitcast i8** %ap to i8*
5122 call void @llvm.va_start(i8* %ap2)
5124 ; Read a single integer argument
5125 %tmp = va_arg i8** %ap, i32
5127 ; Demonstrate usage of llvm.va_copy and llvm.va_end
5129 %aq2 = bitcast i8** %aq to i8*
5130 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
5131 call void @llvm.va_end(i8* %aq2)
5133 ; Stop processing of arguments.
5134 call void @llvm.va_end(i8* %ap2)
5138 declare void @llvm.va_start(i8*)
5139 declare void @llvm.va_copy(i8*, i8*)
5140 declare void @llvm.va_end(i8*)
5146 <!-- _______________________________________________________________________ -->
5147 <div class="doc_subsubsection">
5148 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
5152 <div class="doc_text">
5156 declare void %llvm.va_start(i8* <arglist>)
5160 <p>The '<tt>llvm.va_start</tt>' intrinsic initializes <tt>*<arglist></tt>
5161 for subsequent use by <tt><a href="#i_va_arg">va_arg</a></tt>.</p>
5164 <p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
5167 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
5168 macro available in C. In a target-dependent way, it initializes
5169 the <tt>va_list</tt> element to which the argument points, so that the next
5170 call to <tt>va_arg</tt> will produce the first variable argument passed to
5171 the function. Unlike the C <tt>va_start</tt> macro, this intrinsic does not
5172 need to know the last argument of the function as the compiler can figure
5177 <!-- _______________________________________________________________________ -->
5178 <div class="doc_subsubsection">
5179 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
5182 <div class="doc_text">
5186 declare void @llvm.va_end(i8* <arglist>)
5190 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
5191 which has been initialized previously
5192 with <tt><a href="#int_va_start">llvm.va_start</a></tt>
5193 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
5196 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
5199 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
5200 macro available in C. In a target-dependent way, it destroys
5201 the <tt>va_list</tt> element to which the argument points. Calls
5202 to <a href="#int_va_start"><tt>llvm.va_start</tt></a>
5203 and <a href="#int_va_copy"> <tt>llvm.va_copy</tt></a> must be matched exactly
5204 with calls to <tt>llvm.va_end</tt>.</p>
5208 <!-- _______________________________________________________________________ -->
5209 <div class="doc_subsubsection">
5210 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
5213 <div class="doc_text">
5217 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
5221 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
5222 from the source argument list to the destination argument list.</p>
5225 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
5226 The second argument is a pointer to a <tt>va_list</tt> element to copy
5230 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
5231 macro available in C. In a target-dependent way, it copies the
5232 source <tt>va_list</tt> element into the destination <tt>va_list</tt>
5233 element. This intrinsic is necessary because
5234 the <tt><a href="#int_va_start"> llvm.va_start</a></tt> intrinsic may be
5235 arbitrarily complex and require, for example, memory allocation.</p>
5239 <!-- ======================================================================= -->
5240 <div class="doc_subsection">
5241 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
5244 <div class="doc_text">
5246 <p>LLVM support for <a href="GarbageCollection.html">Accurate Garbage
5247 Collection</a> (GC) requires the implementation and generation of these
5248 intrinsics. These intrinsics allow identification of <a href="#int_gcroot">GC
5249 roots on the stack</a>, as well as garbage collector implementations that
5250 require <a href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a>
5251 barriers. Front-ends for type-safe garbage collected languages should generate
5252 these intrinsics to make use of the LLVM garbage collectors. For more details,
5253 see <a href="GarbageCollection.html">Accurate Garbage Collection with
5256 <p>The garbage collection intrinsics only operate on objects in the generic
5257 address space (address space zero).</p>
5261 <!-- _______________________________________________________________________ -->
5262 <div class="doc_subsubsection">
5263 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
5266 <div class="doc_text">
5270 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
5274 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
5275 the code generator, and allows some metadata to be associated with it.</p>
5278 <p>The first argument specifies the address of a stack object that contains the
5279 root pointer. The second pointer (which must be either a constant or a
5280 global value address) contains the meta-data to be associated with the
5284 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
5285 location. At compile-time, the code generator generates information to allow
5286 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
5287 intrinsic may only be used in a function which <a href="#gc">specifies a GC
5292 <!-- _______________________________________________________________________ -->
5293 <div class="doc_subsubsection">
5294 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
5297 <div class="doc_text">
5301 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
5305 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
5306 locations, allowing garbage collector implementations that require read
5310 <p>The second argument is the address to read from, which should be an address
5311 allocated from the garbage collector. The first object is a pointer to the
5312 start of the referenced object, if needed by the language runtime (otherwise
5316 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
5317 instruction, but may be replaced with substantially more complex code by the
5318 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
5319 may only be used in a function which <a href="#gc">specifies a GC
5324 <!-- _______________________________________________________________________ -->
5325 <div class="doc_subsubsection">
5326 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
5329 <div class="doc_text">
5333 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
5337 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
5338 locations, allowing garbage collector implementations that require write
5339 barriers (such as generational or reference counting collectors).</p>
5342 <p>The first argument is the reference to store, the second is the start of the
5343 object to store it to, and the third is the address of the field of Obj to
5344 store to. If the runtime does not require a pointer to the object, Obj may
5348 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
5349 instruction, but may be replaced with substantially more complex code by the
5350 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
5351 may only be used in a function which <a href="#gc">specifies a GC
5356 <!-- ======================================================================= -->
5357 <div class="doc_subsection">
5358 <a name="int_codegen">Code Generator Intrinsics</a>
5361 <div class="doc_text">
5363 <p>These intrinsics are provided by LLVM to expose special features that may
5364 only be implemented with code generator support.</p>
5368 <!-- _______________________________________________________________________ -->
5369 <div class="doc_subsubsection">
5370 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
5373 <div class="doc_text">
5377 declare i8 *@llvm.returnaddress(i32 <level>)
5381 <p>The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
5382 target-specific value indicating the return address of the current function
5383 or one of its callers.</p>
5386 <p>The argument to this intrinsic indicates which function to return the address
5387 for. Zero indicates the calling function, one indicates its caller, etc.
5388 The argument is <b>required</b> to be a constant integer value.</p>
5391 <p>The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer
5392 indicating the return address of the specified call frame, or zero if it
5393 cannot be identified. The value returned by this intrinsic is likely to be
5394 incorrect or 0 for arguments other than zero, so it should only be used for
5395 debugging purposes.</p>
5397 <p>Note that calling this intrinsic does not prevent function inlining or other
5398 aggressive transformations, so the value returned may not be that of the
5399 obvious source-language caller.</p>
5403 <!-- _______________________________________________________________________ -->
5404 <div class="doc_subsubsection">
5405 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
5408 <div class="doc_text">
5412 declare i8 *@llvm.frameaddress(i32 <level>)
5416 <p>The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
5417 target-specific frame pointer value for the specified stack frame.</p>
5420 <p>The argument to this intrinsic indicates which function to return the frame
5421 pointer for. Zero indicates the calling function, one indicates its caller,
5422 etc. The argument is <b>required</b> to be a constant integer value.</p>
5425 <p>The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer
5426 indicating the frame address of the specified call frame, or zero if it
5427 cannot be identified. The value returned by this intrinsic is likely to be
5428 incorrect or 0 for arguments other than zero, so it should only be used for
5429 debugging purposes.</p>
5431 <p>Note that calling this intrinsic does not prevent function inlining or other
5432 aggressive transformations, so the value returned may not be that of the
5433 obvious source-language caller.</p>
5437 <!-- _______________________________________________________________________ -->
5438 <div class="doc_subsubsection">
5439 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
5442 <div class="doc_text">
5446 declare i8 *@llvm.stacksave()
5450 <p>The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state
5451 of the function stack, for use
5452 with <a href="#int_stackrestore"> <tt>llvm.stackrestore</tt></a>. This is
5453 useful for implementing language features like scoped automatic variable
5454 sized arrays in C99.</p>
5457 <p>This intrinsic returns a opaque pointer value that can be passed
5458 to <a href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When
5459 an <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved
5460 from <tt>llvm.stacksave</tt>, it effectively restores the state of the stack
5461 to the state it was in when the <tt>llvm.stacksave</tt> intrinsic executed.
5462 In practice, this pops any <a href="#i_alloca">alloca</a> blocks from the
5463 stack that were allocated after the <tt>llvm.stacksave</tt> was executed.</p>
5467 <!-- _______________________________________________________________________ -->
5468 <div class="doc_subsubsection">
5469 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
5472 <div class="doc_text">
5476 declare void @llvm.stackrestore(i8 * %ptr)
5480 <p>The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
5481 the function stack to the state it was in when the
5482 corresponding <a href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic
5483 executed. This is useful for implementing language features like scoped
5484 automatic variable sized arrays in C99.</p>
5487 <p>See the description
5488 for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.</p>
5492 <!-- _______________________________________________________________________ -->
5493 <div class="doc_subsubsection">
5494 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
5497 <div class="doc_text">
5501 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>)
5505 <p>The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to
5506 insert a prefetch instruction if supported; otherwise, it is a noop.
5507 Prefetches have no effect on the behavior of the program but can change its
5508 performance characteristics.</p>
5511 <p><tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the
5512 specifier determining if the fetch should be for a read (0) or write (1),
5513 and <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
5514 locality, to (3) - extremely local keep in cache. The <tt>rw</tt>
5515 and <tt>locality</tt> arguments must be constant integers.</p>
5518 <p>This intrinsic does not modify the behavior of the program. In particular,
5519 prefetches cannot trap and do not produce a value. On targets that support
5520 this intrinsic, the prefetch can provide hints to the processor cache for
5521 better performance.</p>
5525 <!-- _______________________________________________________________________ -->
5526 <div class="doc_subsubsection">
5527 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
5530 <div class="doc_text">
5534 declare void @llvm.pcmarker(i32 <id>)
5538 <p>The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program
5539 Counter (PC) in a region of code to simulators and other tools. The method
5540 is target specific, but it is expected that the marker will use exported
5541 symbols to transmit the PC of the marker. The marker makes no guarantees
5542 that it will remain with any specific instruction after optimizations. It is
5543 possible that the presence of a marker will inhibit optimizations. The
5544 intended use is to be inserted after optimizations to allow correlations of
5545 simulation runs.</p>
5548 <p><tt>id</tt> is a numerical id identifying the marker.</p>
5551 <p>This intrinsic does not modify the behavior of the program. Backends that do
5552 not support this intrinisic may ignore it.</p>
5556 <!-- _______________________________________________________________________ -->
5557 <div class="doc_subsubsection">
5558 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
5561 <div class="doc_text">
5565 declare i64 @llvm.readcyclecounter( )
5569 <p>The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
5570 counter register (or similar low latency, high accuracy clocks) on those
5571 targets that support it. On X86, it should map to RDTSC. On Alpha, it
5572 should map to RPCC. As the backing counters overflow quickly (on the order
5573 of 9 seconds on alpha), this should only be used for small timings.</p>
5576 <p>When directly supported, reading the cycle counter should not modify any
5577 memory. Implementations are allowed to either return a application specific
5578 value or a system wide value. On backends without support, this is lowered
5579 to a constant 0.</p>
5583 <!-- ======================================================================= -->
5584 <div class="doc_subsection">
5585 <a name="int_libc">Standard C Library Intrinsics</a>
5588 <div class="doc_text">
5590 <p>LLVM provides intrinsics for a few important standard C library functions.
5591 These intrinsics allow source-language front-ends to pass information about
5592 the alignment of the pointer arguments to the code generator, providing
5593 opportunity for more efficient code generation.</p>
5597 <!-- _______________________________________________________________________ -->
5598 <div class="doc_subsubsection">
5599 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
5602 <div class="doc_text">
5605 <p>This is an overloaded intrinsic. You can use <tt>llvm.memcpy</tt> on any
5606 integer bit width. Not all targets support all bit widths however.</p>
5609 declare void @llvm.memcpy.i8(i8 * <dest>, i8 * <src>,
5610 i8 <len>, i32 <align>)
5611 declare void @llvm.memcpy.i16(i8 * <dest>, i8 * <src>,
5612 i16 <len>, i32 <align>)
5613 declare void @llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
5614 i32 <len>, i32 <align>)
5615 declare void @llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
5616 i64 <len>, i32 <align>)
5620 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
5621 source location to the destination location.</p>
5623 <p>Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
5624 intrinsics do not return a value, and takes an extra alignment argument.</p>
5627 <p>The first argument is a pointer to the destination, the second is a pointer
5628 to the source. The third argument is an integer argument specifying the
5629 number of bytes to copy, and the fourth argument is the alignment of the
5630 source and destination locations.</p>
5632 <p>If the call to this intrinisic has an alignment value that is not 0 or 1,
5633 then the caller guarantees that both the source and destination pointers are
5634 aligned to that boundary.</p>
5637 <p>The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the
5638 source location to the destination location, which are not allowed to
5639 overlap. It copies "len" bytes of memory over. If the argument is known to
5640 be aligned to some boundary, this can be specified as the fourth argument,
5641 otherwise it should be set to 0 or 1.</p>
5645 <!-- _______________________________________________________________________ -->
5646 <div class="doc_subsubsection">
5647 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
5650 <div class="doc_text">
5653 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
5654 width. Not all targets support all bit widths however.</p>
5657 declare void @llvm.memmove.i8(i8 * <dest>, i8 * <src>,
5658 i8 <len>, i32 <align>)
5659 declare void @llvm.memmove.i16(i8 * <dest>, i8 * <src>,
5660 i16 <len>, i32 <align>)
5661 declare void @llvm.memmove.i32(i8 * <dest>, i8 * <src>,
5662 i32 <len>, i32 <align>)
5663 declare void @llvm.memmove.i64(i8 * <dest>, i8 * <src>,
5664 i64 <len>, i32 <align>)
5668 <p>The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the
5669 source location to the destination location. It is similar to the
5670 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to
5673 <p>Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
5674 intrinsics do not return a value, and takes an extra alignment argument.</p>
5677 <p>The first argument is a pointer to the destination, the second is a pointer
5678 to the source. The third argument is an integer argument specifying the
5679 number of bytes to copy, and the fourth argument is the alignment of the
5680 source and destination locations.</p>
5682 <p>If the call to this intrinisic has an alignment value that is not 0 or 1,
5683 then the caller guarantees that the source and destination pointers are
5684 aligned to that boundary.</p>
5687 <p>The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the
5688 source location to the destination location, which may overlap. It copies
5689 "len" bytes of memory over. If the argument is known to be aligned to some
5690 boundary, this can be specified as the fourth argument, otherwise it should
5691 be set to 0 or 1.</p>
5695 <!-- _______________________________________________________________________ -->
5696 <div class="doc_subsubsection">
5697 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
5700 <div class="doc_text">
5703 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
5704 width. Not all targets support all bit widths however.</p>
5707 declare void @llvm.memset.i8(i8 * <dest>, i8 <val>,
5708 i8 <len>, i32 <align>)
5709 declare void @llvm.memset.i16(i8 * <dest>, i8 <val>,
5710 i16 <len>, i32 <align>)
5711 declare void @llvm.memset.i32(i8 * <dest>, i8 <val>,
5712 i32 <len>, i32 <align>)
5713 declare void @llvm.memset.i64(i8 * <dest>, i8 <val>,
5714 i64 <len>, i32 <align>)
5718 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a
5719 particular byte value.</p>
5721 <p>Note that, unlike the standard libc function, the <tt>llvm.memset</tt>
5722 intrinsic does not return a value, and takes an extra alignment argument.</p>
5725 <p>The first argument is a pointer to the destination to fill, the second is the
5726 byte value to fill it with, the third argument is an integer argument
5727 specifying the number of bytes to fill, and the fourth argument is the known
5728 alignment of destination location.</p>
5730 <p>If the call to this intrinisic has an alignment value that is not 0 or 1,
5731 then the caller guarantees that the destination pointer is aligned to that
5735 <p>The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting
5736 at the destination location. If the argument is known to be aligned to some
5737 boundary, this can be specified as the fourth argument, otherwise it should
5738 be set to 0 or 1.</p>
5742 <!-- _______________________________________________________________________ -->
5743 <div class="doc_subsubsection">
5744 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
5747 <div class="doc_text">
5750 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
5751 floating point or vector of floating point type. Not all targets support all
5755 declare float @llvm.sqrt.f32(float %Val)
5756 declare double @llvm.sqrt.f64(double %Val)
5757 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
5758 declare fp128 @llvm.sqrt.f128(fp128 %Val)
5759 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
5763 <p>The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
5764 returning the same value as the libm '<tt>sqrt</tt>' functions would.
5765 Unlike <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined
5766 behavior for negative numbers other than -0.0 (which allows for better
5767 optimization, because there is no need to worry about errno being
5768 set). <tt>llvm.sqrt(-0.0)</tt> is defined to return -0.0 like IEEE sqrt.</p>
5771 <p>The argument and return value are floating point numbers of the same
5775 <p>This function returns the sqrt of the specified operand if it is a
5776 nonnegative floating point number.</p>
5780 <!-- _______________________________________________________________________ -->
5781 <div class="doc_subsubsection">
5782 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
5785 <div class="doc_text">
5788 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
5789 floating point or vector of floating point type. Not all targets support all
5793 declare float @llvm.powi.f32(float %Val, i32 %power)
5794 declare double @llvm.powi.f64(double %Val, i32 %power)
5795 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
5796 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
5797 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
5801 <p>The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
5802 specified (positive or negative) power. The order of evaluation of
5803 multiplications is not defined. When a vector of floating point type is
5804 used, the second argument remains a scalar integer value.</p>
5807 <p>The second argument is an integer power, and the first is a value to raise to
5811 <p>This function returns the first value raised to the second power with an
5812 unspecified sequence of rounding operations.</p>
5816 <!-- _______________________________________________________________________ -->
5817 <div class="doc_subsubsection">
5818 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
5821 <div class="doc_text">
5824 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
5825 floating point or vector of floating point type. Not all targets support all
5829 declare float @llvm.sin.f32(float %Val)
5830 declare double @llvm.sin.f64(double %Val)
5831 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
5832 declare fp128 @llvm.sin.f128(fp128 %Val)
5833 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
5837 <p>The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.</p>
5840 <p>The argument and return value are floating point numbers of the same
5844 <p>This function returns the sine of the specified operand, returning the same
5845 values as the libm <tt>sin</tt> functions would, and handles error conditions
5846 in the same way.</p>
5850 <!-- _______________________________________________________________________ -->
5851 <div class="doc_subsubsection">
5852 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
5855 <div class="doc_text">
5858 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
5859 floating point or vector of floating point type. Not all targets support all
5863 declare float @llvm.cos.f32(float %Val)
5864 declare double @llvm.cos.f64(double %Val)
5865 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
5866 declare fp128 @llvm.cos.f128(fp128 %Val)
5867 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
5871 <p>The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.</p>
5874 <p>The argument and return value are floating point numbers of the same
5878 <p>This function returns the cosine of the specified operand, returning the same
5879 values as the libm <tt>cos</tt> functions would, and handles error conditions
5880 in the same way.</p>
5884 <!-- _______________________________________________________________________ -->
5885 <div class="doc_subsubsection">
5886 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
5889 <div class="doc_text">
5892 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
5893 floating point or vector of floating point type. Not all targets support all
5897 declare float @llvm.pow.f32(float %Val, float %Power)
5898 declare double @llvm.pow.f64(double %Val, double %Power)
5899 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
5900 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
5901 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
5905 <p>The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
5906 specified (positive or negative) power.</p>
5909 <p>The second argument is a floating point power, and the first is a value to
5910 raise to that power.</p>
5913 <p>This function returns the first value raised to the second power, returning
5914 the same values as the libm <tt>pow</tt> functions would, and handles error
5915 conditions in the same way.</p>
5919 <!-- ======================================================================= -->
5920 <div class="doc_subsection">
5921 <a name="int_manip">Bit Manipulation Intrinsics</a>
5924 <div class="doc_text">
5926 <p>LLVM provides intrinsics for a few important bit manipulation operations.
5927 These allow efficient code generation for some algorithms.</p>
5931 <!-- _______________________________________________________________________ -->
5932 <div class="doc_subsubsection">
5933 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
5936 <div class="doc_text">
5939 <p>This is an overloaded intrinsic function. You can use bswap on any integer
5940 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
5943 declare i16 @llvm.bswap.i16(i16 <id>)
5944 declare i32 @llvm.bswap.i32(i32 <id>)
5945 declare i64 @llvm.bswap.i64(i64 <id>)
5949 <p>The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
5950 values with an even number of bytes (positive multiple of 16 bits). These
5951 are useful for performing operations on data that is not in the target's
5952 native byte order.</p>
5955 <p>The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
5956 and low byte of the input i16 swapped. Similarly,
5957 the <tt>llvm.bswap.i32</tt> intrinsic returns an i32 value that has the four
5958 bytes of the input i32 swapped, so that if the input bytes are numbered 0, 1,
5959 2, 3 then the returned i32 will have its bytes in 3, 2, 1, 0 order.
5960 The <tt>llvm.bswap.i48</tt>, <tt>llvm.bswap.i64</tt> and other intrinsics
5961 extend this concept to additional even-byte lengths (6 bytes, 8 bytes and
5962 more, respectively).</p>
5966 <!-- _______________________________________________________________________ -->
5967 <div class="doc_subsubsection">
5968 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
5971 <div class="doc_text">
5974 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
5975 width. Not all targets support all bit widths however.</p>
5978 declare i8 @llvm.ctpop.i8(i8 <src>)
5979 declare i16 @llvm.ctpop.i16(i16 <src>)
5980 declare i32 @llvm.ctpop.i32(i32 <src>)
5981 declare i64 @llvm.ctpop.i64(i64 <src>)
5982 declare i256 @llvm.ctpop.i256(i256 <src>)
5986 <p>The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set
5990 <p>The only argument is the value to be counted. The argument may be of any
5991 integer type. The return type must match the argument type.</p>
5994 <p>The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.</p>
5998 <!-- _______________________________________________________________________ -->
5999 <div class="doc_subsubsection">
6000 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
6003 <div class="doc_text">
6006 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
6007 integer bit width. Not all targets support all bit widths however.</p>
6010 declare i8 @llvm.ctlz.i8 (i8 <src>)
6011 declare i16 @llvm.ctlz.i16(i16 <src>)
6012 declare i32 @llvm.ctlz.i32(i32 <src>)
6013 declare i64 @llvm.ctlz.i64(i64 <src>)
6014 declare i256 @llvm.ctlz.i256(i256 <src>)
6018 <p>The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
6019 leading zeros in a variable.</p>
6022 <p>The only argument is the value to be counted. The argument may be of any
6023 integer type. The return type must match the argument type.</p>
6026 <p>The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant)
6027 zeros in a variable. If the src == 0 then the result is the size in bits of
6028 the type of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.</p>
6032 <!-- _______________________________________________________________________ -->
6033 <div class="doc_subsubsection">
6034 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
6037 <div class="doc_text">
6040 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
6041 integer bit width. Not all targets support all bit widths however.</p>
6044 declare i8 @llvm.cttz.i8 (i8 <src>)
6045 declare i16 @llvm.cttz.i16(i16 <src>)
6046 declare i32 @llvm.cttz.i32(i32 <src>)
6047 declare i64 @llvm.cttz.i64(i64 <src>)
6048 declare i256 @llvm.cttz.i256(i256 <src>)
6052 <p>The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
6056 <p>The only argument is the value to be counted. The argument may be of any
6057 integer type. The return type must match the argument type.</p>
6060 <p>The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant)
6061 zeros in a variable. If the src == 0 then the result is the size in bits of
6062 the type of src. For example, <tt>llvm.cttz(2) = 1</tt>.</p>
6066 <!-- ======================================================================= -->
6067 <div class="doc_subsection">
6068 <a name="int_overflow">Arithmetic with Overflow Intrinsics</a>
6071 <div class="doc_text">
6073 <p>LLVM provides intrinsics for some arithmetic with overflow operations.</p>
6077 <!-- _______________________________________________________________________ -->
6078 <div class="doc_subsubsection">
6079 <a name="int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt>' Intrinsics</a>
6082 <div class="doc_text">
6085 <p>This is an overloaded intrinsic. You can use <tt>llvm.sadd.with.overflow</tt>
6086 on any integer bit width.</p>
6089 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
6090 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6091 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
6095 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
6096 a signed addition of the two arguments, and indicate whether an overflow
6097 occurred during the signed summation.</p>
6100 <p>The arguments (%a and %b) and the first element of the result structure may
6101 be of integer types of any bit width, but they must have the same bit
6102 width. The second element of the result structure must be of
6103 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6104 undergo signed addition.</p>
6107 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
6108 a signed addition of the two variables. They return a structure — the
6109 first element of which is the signed summation, and the second element of
6110 which is a bit specifying if the signed summation resulted in an
6115 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6116 %sum = extractvalue {i32, i1} %res, 0
6117 %obit = extractvalue {i32, i1} %res, 1
6118 br i1 %obit, label %overflow, label %normal
6123 <!-- _______________________________________________________________________ -->
6124 <div class="doc_subsubsection">
6125 <a name="int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt>' Intrinsics</a>
6128 <div class="doc_text">
6131 <p>This is an overloaded intrinsic. You can use <tt>llvm.uadd.with.overflow</tt>
6132 on any integer bit width.</p>
6135 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
6136 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6137 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
6141 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
6142 an unsigned addition of the two arguments, and indicate whether a carry
6143 occurred during the unsigned summation.</p>
6146 <p>The arguments (%a and %b) and the first element of the result structure may
6147 be of integer types of any bit width, but they must have the same bit
6148 width. The second element of the result structure must be of
6149 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6150 undergo unsigned addition.</p>
6153 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
6154 an unsigned addition of the two arguments. They return a structure —
6155 the first element of which is the sum, and the second element of which is a
6156 bit specifying if the unsigned summation resulted in a carry.</p>
6160 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6161 %sum = extractvalue {i32, i1} %res, 0
6162 %obit = extractvalue {i32, i1} %res, 1
6163 br i1 %obit, label %carry, label %normal
6168 <!-- _______________________________________________________________________ -->
6169 <div class="doc_subsubsection">
6170 <a name="int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt>' Intrinsics</a>
6173 <div class="doc_text">
6176 <p>This is an overloaded intrinsic. You can use <tt>llvm.ssub.with.overflow</tt>
6177 on any integer bit width.</p>
6180 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
6181 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6182 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
6186 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
6187 a signed subtraction of the two arguments, and indicate whether an overflow
6188 occurred during the signed subtraction.</p>
6191 <p>The arguments (%a and %b) and the first element of the result structure may
6192 be of integer types of any bit width, but they must have the same bit
6193 width. The second element of the result structure must be of
6194 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6195 undergo signed subtraction.</p>
6198 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
6199 a signed subtraction of the two arguments. They return a structure —
6200 the first element of which is the subtraction, and the second element of
6201 which is a bit specifying if the signed subtraction resulted in an
6206 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6207 %sum = extractvalue {i32, i1} %res, 0
6208 %obit = extractvalue {i32, i1} %res, 1
6209 br i1 %obit, label %overflow, label %normal
6214 <!-- _______________________________________________________________________ -->
6215 <div class="doc_subsubsection">
6216 <a name="int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt>' Intrinsics</a>
6219 <div class="doc_text">
6222 <p>This is an overloaded intrinsic. You can use <tt>llvm.usub.with.overflow</tt>
6223 on any integer bit width.</p>
6226 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
6227 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6228 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
6232 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
6233 an unsigned subtraction of the two arguments, and indicate whether an
6234 overflow occurred during the unsigned subtraction.</p>
6237 <p>The arguments (%a and %b) and the first element of the result structure may
6238 be of integer types of any bit width, but they must have the same bit
6239 width. The second element of the result structure must be of
6240 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6241 undergo unsigned subtraction.</p>
6244 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
6245 an unsigned subtraction of the two arguments. They return a structure —
6246 the first element of which is the subtraction, and the second element of
6247 which is a bit specifying if the unsigned subtraction resulted in an
6252 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6253 %sum = extractvalue {i32, i1} %res, 0
6254 %obit = extractvalue {i32, i1} %res, 1
6255 br i1 %obit, label %overflow, label %normal
6260 <!-- _______________________________________________________________________ -->
6261 <div class="doc_subsubsection">
6262 <a name="int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt>' Intrinsics</a>
6265 <div class="doc_text">
6268 <p>This is an overloaded intrinsic. You can use <tt>llvm.smul.with.overflow</tt>
6269 on any integer bit width.</p>
6272 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
6273 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6274 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
6279 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
6280 a signed multiplication of the two arguments, and indicate whether an
6281 overflow occurred during the signed multiplication.</p>
6284 <p>The arguments (%a and %b) and the first element of the result structure may
6285 be of integer types of any bit width, but they must have the same bit
6286 width. The second element of the result structure must be of
6287 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6288 undergo signed multiplication.</p>
6291 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
6292 a signed multiplication of the two arguments. They return a structure —
6293 the first element of which is the multiplication, and the second element of
6294 which is a bit specifying if the signed multiplication resulted in an
6299 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6300 %sum = extractvalue {i32, i1} %res, 0
6301 %obit = extractvalue {i32, i1} %res, 1
6302 br i1 %obit, label %overflow, label %normal
6307 <!-- _______________________________________________________________________ -->
6308 <div class="doc_subsubsection">
6309 <a name="int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt>' Intrinsics</a>
6312 <div class="doc_text">
6315 <p>This is an overloaded intrinsic. You can use <tt>llvm.umul.with.overflow</tt>
6316 on any integer bit width.</p>
6319 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
6320 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6321 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
6325 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
6326 a unsigned multiplication of the two arguments, and indicate whether an
6327 overflow occurred during the unsigned multiplication.</p>
6330 <p>The arguments (%a and %b) and the first element of the result structure may
6331 be of integer types of any bit width, but they must have the same bit
6332 width. The second element of the result structure must be of
6333 type <tt>i1</tt>. <tt>%a</tt> and <tt>%b</tt> are the two values that will
6334 undergo unsigned multiplication.</p>
6337 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
6338 an unsigned multiplication of the two arguments. They return a structure
6339 — the first element of which is the multiplication, and the second
6340 element of which is a bit specifying if the unsigned multiplication resulted
6345 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6346 %sum = extractvalue {i32, i1} %res, 0
6347 %obit = extractvalue {i32, i1} %res, 1
6348 br i1 %obit, label %overflow, label %normal
6353 <!-- ======================================================================= -->
6354 <div class="doc_subsection">
6355 <a name="int_debugger">Debugger Intrinsics</a>
6358 <div class="doc_text">
6360 <p>The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt>
6361 prefix), are described in
6362 the <a href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source
6363 Level Debugging</a> document.</p>
6367 <!-- ======================================================================= -->
6368 <div class="doc_subsection">
6369 <a name="int_eh">Exception Handling Intrinsics</a>
6372 <div class="doc_text">
6374 <p>The LLVM exception handling intrinsics (which all start with
6375 <tt>llvm.eh.</tt> prefix), are described in
6376 the <a href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
6377 Handling</a> document.</p>
6381 <!-- ======================================================================= -->
6382 <div class="doc_subsection">
6383 <a name="int_trampoline">Trampoline Intrinsic</a>
6386 <div class="doc_text">
6388 <p>This intrinsic makes it possible to excise one parameter, marked with
6389 the <tt>nest</tt> attribute, from a function. The result is a callable
6390 function pointer lacking the nest parameter - the caller does not need to
6391 provide a value for it. Instead, the value to use is stored in advance in a
6392 "trampoline", a block of memory usually allocated on the stack, which also
6393 contains code to splice the nest value into the argument list. This is used
6394 to implement the GCC nested function address extension.</p>
6396 <p>For example, if the function is
6397 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
6398 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as
6401 <div class="doc_code">
6403 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
6404 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
6405 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
6406 %fp = bitcast i8* %p to i32 (i32, i32)*
6410 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
6411 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
6415 <!-- _______________________________________________________________________ -->
6416 <div class="doc_subsubsection">
6417 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
6420 <div class="doc_text">
6424 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
6428 <p>This fills the memory pointed to by <tt>tramp</tt> with code and returns a
6429 function pointer suitable for executing it.</p>
6432 <p>The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
6433 pointers. The <tt>tramp</tt> argument must point to a sufficiently large and
6434 sufficiently aligned block of memory; this memory is written to by the
6435 intrinsic. Note that the size and the alignment are target-specific - LLVM
6436 currently provides no portable way of determining them, so a front-end that
6437 generates this intrinsic needs to have some target-specific knowledge.
6438 The <tt>func</tt> argument must hold a function bitcast to
6439 an <tt>i8*</tt>.</p>
6442 <p>The block of memory pointed to by <tt>tramp</tt> is filled with target
6443 dependent code, turning it into a function. A pointer to this function is
6444 returned, but needs to be bitcast to an <a href="#int_trampoline">appropriate
6445 function pointer type</a> before being called. The new function's signature
6446 is the same as that of <tt>func</tt> with any arguments marked with
6447 the <tt>nest</tt> attribute removed. At most one such <tt>nest</tt> argument
6448 is allowed, and it must be of pointer type. Calling the new function is
6449 equivalent to calling <tt>func</tt> with the same argument list, but
6450 with <tt>nval</tt> used for the missing <tt>nest</tt> argument. If, after
6451 calling <tt>llvm.init.trampoline</tt>, the memory pointed to
6452 by <tt>tramp</tt> is modified, then the effect of any later call to the
6453 returned function pointer is undefined.</p>
6457 <!-- ======================================================================= -->
6458 <div class="doc_subsection">
6459 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
6462 <div class="doc_text">
6464 <p>These intrinsic functions expand the "universal IR" of LLVM to represent
6465 hardware constructs for atomic operations and memory synchronization. This
6466 provides an interface to the hardware, not an interface to the programmer. It
6467 is aimed at a low enough level to allow any programming models or APIs
6468 (Application Programming Interfaces) which need atomic behaviors to map
6469 cleanly onto it. It is also modeled primarily on hardware behavior. Just as
6470 hardware provides a "universal IR" for source languages, it also provides a
6471 starting point for developing a "universal" atomic operation and
6472 synchronization IR.</p>
6474 <p>These do <em>not</em> form an API such as high-level threading libraries,
6475 software transaction memory systems, atomic primitives, and intrinsic
6476 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
6477 application libraries. The hardware interface provided by LLVM should allow
6478 a clean implementation of all of these APIs and parallel programming models.
6479 No one model or paradigm should be selected above others unless the hardware
6480 itself ubiquitously does so.</p>
6484 <!-- _______________________________________________________________________ -->
6485 <div class="doc_subsubsection">
6486 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
6488 <div class="doc_text">
6491 declare void @llvm.memory.barrier( i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>, i1 <device> )
6495 <p>The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
6496 specific pairs of memory access types.</p>
6499 <p>The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
6500 The first four arguments enables a specific barrier as listed below. The
6501 fith argument specifies that the barrier applies to io or device or uncached
6505 <li><tt>ll</tt>: load-load barrier</li>
6506 <li><tt>ls</tt>: load-store barrier</li>
6507 <li><tt>sl</tt>: store-load barrier</li>
6508 <li><tt>ss</tt>: store-store barrier</li>
6509 <li><tt>device</tt>: barrier applies to device and uncached memory also.</li>
6513 <p>This intrinsic causes the system to enforce some ordering constraints upon
6514 the loads and stores of the program. This barrier does not
6515 indicate <em>when</em> any events will occur, it only enforces
6516 an <em>order</em> in which they occur. For any of the specified pairs of load
6517 and store operations (f.ex. load-load, or store-load), all of the first
6518 operations preceding the barrier will complete before any of the second
6519 operations succeeding the barrier begin. Specifically the semantics for each
6520 pairing is as follows:</p>
6523 <li><tt>ll</tt>: All loads before the barrier must complete before any load
6524 after the barrier begins.</li>
6525 <li><tt>ls</tt>: All loads before the barrier must complete before any
6526 store after the barrier begins.</li>
6527 <li><tt>ss</tt>: All stores before the barrier must complete before any
6528 store after the barrier begins.</li>
6529 <li><tt>sl</tt>: All stores before the barrier must complete before any
6530 load after the barrier begins.</li>
6533 <p>These semantics are applied with a logical "and" behavior when more than one
6534 is enabled in a single memory barrier intrinsic.</p>
6536 <p>Backends may implement stronger barriers than those requested when they do
6537 not support as fine grained a barrier as requested. Some architectures do
6538 not need all types of barriers and on such architectures, these become
6543 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
6544 %ptr = bitcast i8* %mallocP to i32*
6547 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
6548 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
6549 <i>; guarantee the above finishes</i>
6550 store i32 8, %ptr <i>; before this begins</i>
6555 <!-- _______________________________________________________________________ -->
6556 <div class="doc_subsubsection">
6557 <a name="int_atomic_cmp_swap">'<tt>llvm.atomic.cmp.swap.*</tt>' Intrinsic</a>
6560 <div class="doc_text">
6563 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.cmp.swap</tt> on
6564 any integer bit width and for different address spaces. Not all targets
6565 support all bit widths however.</p>
6568 declare i8 @llvm.atomic.cmp.swap.i8.p0i8( i8* <ptr>, i8 <cmp>, i8 <val> )
6569 declare i16 @llvm.atomic.cmp.swap.i16.p0i16( i16* <ptr>, i16 <cmp>, i16 <val> )
6570 declare i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* <ptr>, i32 <cmp>, i32 <val> )
6571 declare i64 @llvm.atomic.cmp.swap.i64.p0i64( i64* <ptr>, i64 <cmp>, i64 <val> )
6575 <p>This loads a value in memory and compares it to a given value. If they are
6576 equal, it stores a new value into the memory.</p>
6579 <p>The <tt>llvm.atomic.cmp.swap</tt> intrinsic takes three arguments. The result
6580 as well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
6581 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
6582 this integer type. While any bit width integer may be used, targets may only
6583 lower representations they support in hardware.</p>
6586 <p>This entire intrinsic must be executed atomically. It first loads the value
6587 in memory pointed to by <tt>ptr</tt> and compares it with the
6588 value <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the
6589 memory. The loaded value is yielded in all cases. This provides the
6590 equivalent of an atomic compare-and-swap operation within the SSA
6595 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
6596 %ptr = bitcast i8* %mallocP to i32*
6599 %val1 = add i32 4, 4
6600 %result1 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 4, %val1 )
6601 <i>; yields {i32}:result1 = 4</i>
6602 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
6603 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
6605 %val2 = add i32 1, 1
6606 %result2 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 5, %val2 )
6607 <i>; yields {i32}:result2 = 8</i>
6608 %stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
6610 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
6615 <!-- _______________________________________________________________________ -->
6616 <div class="doc_subsubsection">
6617 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
6619 <div class="doc_text">
6622 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
6623 integer bit width. Not all targets support all bit widths however.</p>
6626 declare i8 @llvm.atomic.swap.i8.p0i8( i8* <ptr>, i8 <val> )
6627 declare i16 @llvm.atomic.swap.i16.p0i16( i16* <ptr>, i16 <val> )
6628 declare i32 @llvm.atomic.swap.i32.p0i32( i32* <ptr>, i32 <val> )
6629 declare i64 @llvm.atomic.swap.i64.p0i64( i64* <ptr>, i64 <val> )
6633 <p>This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
6634 the value from memory. It then stores the value in <tt>val</tt> in the memory
6635 at <tt>ptr</tt>.</p>
6638 <p>The <tt>llvm.atomic.swap</tt> intrinsic takes two arguments. Both
6639 the <tt>val</tt> argument and the result must be integers of the same bit
6640 width. The first argument, <tt>ptr</tt>, must be a pointer to a value of this
6641 integer type. The targets may only lower integer representations they
6645 <p>This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
6646 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
6647 equivalent of an atomic swap operation within the SSA framework.</p>
6651 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
6652 %ptr = bitcast i8* %mallocP to i32*
6655 %val1 = add i32 4, 4
6656 %result1 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val1 )
6657 <i>; yields {i32}:result1 = 4</i>
6658 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
6659 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
6661 %val2 = add i32 1, 1
6662 %result2 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val2 )
6663 <i>; yields {i32}:result2 = 8</i>
6665 %stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
6666 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
6671 <!-- _______________________________________________________________________ -->
6672 <div class="doc_subsubsection">
6673 <a name="int_atomic_load_add">'<tt>llvm.atomic.load.add.*</tt>' Intrinsic</a>
6677 <div class="doc_text">
6680 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.add</tt> on
6681 any integer bit width. Not all targets support all bit widths however.</p>
6684 declare i8 @llvm.atomic.load.add.i8..p0i8( i8* <ptr>, i8 <delta> )
6685 declare i16 @llvm.atomic.load.add.i16..p0i16( i16* <ptr>, i16 <delta> )
6686 declare i32 @llvm.atomic.load.add.i32..p0i32( i32* <ptr>, i32 <delta> )
6687 declare i64 @llvm.atomic.load.add.i64..p0i64( i64* <ptr>, i64 <delta> )
6691 <p>This intrinsic adds <tt>delta</tt> to the value stored in memory
6692 at <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.</p>
6695 <p>The intrinsic takes two arguments, the first a pointer to an integer value
6696 and the second an integer value. The result is also an integer value. These
6697 integer types can have any bit width, but they must all have the same bit
6698 width. The targets may only lower integer representations they support.</p>
6701 <p>This intrinsic does a series of operations atomically. It first loads the
6702 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
6703 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.</p>
6707 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
6708 %ptr = bitcast i8* %mallocP to i32*
6710 %result1 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 4 )
6711 <i>; yields {i32}:result1 = 4</i>
6712 %result2 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 2 )
6713 <i>; yields {i32}:result2 = 8</i>
6714 %result3 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 5 )
6715 <i>; yields {i32}:result3 = 10</i>
6716 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
6721 <!-- _______________________________________________________________________ -->
6722 <div class="doc_subsubsection">
6723 <a name="int_atomic_load_sub">'<tt>llvm.atomic.load.sub.*</tt>' Intrinsic</a>
6727 <div class="doc_text">
6730 <p>This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.sub</tt> on
6731 any integer bit width and for different address spaces. Not all targets
6732 support all bit widths however.</p>
6735 declare i8 @llvm.atomic.load.sub.i8.p0i32( i8* <ptr>, i8 <delta> )
6736 declare i16 @llvm.atomic.load.sub.i16.p0i32( i16* <ptr>, i16 <delta> )
6737 declare i32 @llvm.atomic.load.sub.i32.p0i32( i32* <ptr>, i32 <delta> )
6738 declare i64 @llvm.atomic.load.sub.i64.p0i32( i64* <ptr>, i64 <delta> )
6742 <p>This intrinsic subtracts <tt>delta</tt> to the value stored in memory at
6743 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.</p>
6746 <p>The intrinsic takes two arguments, the first a pointer to an integer value
6747 and the second an integer value. The result is also an integer value. These
6748 integer types can have any bit width, but they must all have the same bit
6749 width. The targets may only lower integer representations they support.</p>
6752 <p>This intrinsic does a series of operations atomically. It first loads the
6753 value stored at <tt>ptr</tt>. It then subtracts <tt>delta</tt>, stores the
6754 result to <tt>ptr</tt>. It yields the original value stored
6755 at <tt>ptr</tt>.</p>
6759 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
6760 %ptr = bitcast i8* %mallocP to i32*
6762 %result1 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 4 )
6763 <i>; yields {i32}:result1 = 8</i>
6764 %result2 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 2 )
6765 <i>; yields {i32}:result2 = 4</i>
6766 %result3 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 5 )
6767 <i>; yields {i32}:result3 = 2</i>
6768 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = -3</i>
6773 <!-- _______________________________________________________________________ -->
6774 <div class="doc_subsubsection">
6775 <a name="int_atomic_load_and">'<tt>llvm.atomic.load.and.*</tt>' Intrinsic</a><br>
6776 <a name="int_atomic_load_nand">'<tt>llvm.atomic.load.nand.*</tt>' Intrinsic</a><br>
6777 <a name="int_atomic_load_or">'<tt>llvm.atomic.load.or.*</tt>' Intrinsic</a><br>
6778 <a name="int_atomic_load_xor">'<tt>llvm.atomic.load.xor.*</tt>' Intrinsic</a><br>
6781 <div class="doc_text">
6784 <p>These are overloaded intrinsics. You can
6785 use <tt>llvm.atomic.load_and</tt>, <tt>llvm.atomic.load_nand</tt>,
6786 <tt>llvm.atomic.load_or</tt>, and <tt>llvm.atomic.load_xor</tt> on any integer
6787 bit width and for different address spaces. Not all targets support all bit
6791 declare i8 @llvm.atomic.load.and.i8.p0i8( i8* <ptr>, i8 <delta> )
6792 declare i16 @llvm.atomic.load.and.i16.p0i16( i16* <ptr>, i16 <delta> )
6793 declare i32 @llvm.atomic.load.and.i32.p0i32( i32* <ptr>, i32 <delta> )
6794 declare i64 @llvm.atomic.load.and.i64.p0i64( i64* <ptr>, i64 <delta> )
6798 declare i8 @llvm.atomic.load.or.i8.p0i8( i8* <ptr>, i8 <delta> )
6799 declare i16 @llvm.atomic.load.or.i16.p0i16( i16* <ptr>, i16 <delta> )
6800 declare i32 @llvm.atomic.load.or.i32.p0i32( i32* <ptr>, i32 <delta> )
6801 declare i64 @llvm.atomic.load.or.i64.p0i64( i64* <ptr>, i64 <delta> )
6805 declare i8 @llvm.atomic.load.nand.i8.p0i32( i8* <ptr>, i8 <delta> )
6806 declare i16 @llvm.atomic.load.nand.i16.p0i32( i16* <ptr>, i16 <delta> )
6807 declare i32 @llvm.atomic.load.nand.i32.p0i32( i32* <ptr>, i32 <delta> )
6808 declare i64 @llvm.atomic.load.nand.i64.p0i32( i64* <ptr>, i64 <delta> )
6812 declare i8 @llvm.atomic.load.xor.i8.p0i32( i8* <ptr>, i8 <delta> )
6813 declare i16 @llvm.atomic.load.xor.i16.p0i32( i16* <ptr>, i16 <delta> )
6814 declare i32 @llvm.atomic.load.xor.i32.p0i32( i32* <ptr>, i32 <delta> )
6815 declare i64 @llvm.atomic.load.xor.i64.p0i32( i64* <ptr>, i64 <delta> )
6819 <p>These intrinsics bitwise the operation (and, nand, or, xor) <tt>delta</tt> to
6820 the value stored in memory at <tt>ptr</tt>. It yields the original value
6821 at <tt>ptr</tt>.</p>
6824 <p>These intrinsics take two arguments, the first a pointer to an integer value
6825 and the second an integer value. The result is also an integer value. These
6826 integer types can have any bit width, but they must all have the same bit
6827 width. The targets may only lower integer representations they support.</p>
6830 <p>These intrinsics does a series of operations atomically. They first load the
6831 value stored at <tt>ptr</tt>. They then do the bitwise
6832 operation <tt>delta</tt>, store the result to <tt>ptr</tt>. They yield the
6833 original value stored at <tt>ptr</tt>.</p>
6837 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
6838 %ptr = bitcast i8* %mallocP to i32*
6839 store i32 0x0F0F, %ptr
6840 %result0 = call i32 @llvm.atomic.load.nand.i32.p0i32( i32* %ptr, i32 0xFF )
6841 <i>; yields {i32}:result0 = 0x0F0F</i>
6842 %result1 = call i32 @llvm.atomic.load.and.i32.p0i32( i32* %ptr, i32 0xFF )
6843 <i>; yields {i32}:result1 = 0xFFFFFFF0</i>
6844 %result2 = call i32 @llvm.atomic.load.or.i32.p0i32( i32* %ptr, i32 0F )
6845 <i>; yields {i32}:result2 = 0xF0</i>
6846 %result3 = call i32 @llvm.atomic.load.xor.i32.p0i32( i32* %ptr, i32 0F )
6847 <i>; yields {i32}:result3 = FF</i>
6848 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = F0</i>
6853 <!-- _______________________________________________________________________ -->
6854 <div class="doc_subsubsection">
6855 <a name="int_atomic_load_max">'<tt>llvm.atomic.load.max.*</tt>' Intrinsic</a><br>
6856 <a name="int_atomic_load_min">'<tt>llvm.atomic.load.min.*</tt>' Intrinsic</a><br>
6857 <a name="int_atomic_load_umax">'<tt>llvm.atomic.load.umax.*</tt>' Intrinsic</a><br>
6858 <a name="int_atomic_load_umin">'<tt>llvm.atomic.load.umin.*</tt>' Intrinsic</a><br>
6861 <div class="doc_text">
6864 <p>These are overloaded intrinsics. You can use <tt>llvm.atomic.load_max</tt>,
6865 <tt>llvm.atomic.load_min</tt>, <tt>llvm.atomic.load_umax</tt>, and
6866 <tt>llvm.atomic.load_umin</tt> on any integer bit width and for different
6867 address spaces. Not all targets support all bit widths however.</p>
6870 declare i8 @llvm.atomic.load.max.i8.p0i8( i8* <ptr>, i8 <delta> )
6871 declare i16 @llvm.atomic.load.max.i16.p0i16( i16* <ptr>, i16 <delta> )
6872 declare i32 @llvm.atomic.load.max.i32.p0i32( i32* <ptr>, i32 <delta> )
6873 declare i64 @llvm.atomic.load.max.i64.p0i64( i64* <ptr>, i64 <delta> )
6877 declare i8 @llvm.atomic.load.min.i8.p0i8( i8* <ptr>, i8 <delta> )
6878 declare i16 @llvm.atomic.load.min.i16.p0i16( i16* <ptr>, i16 <delta> )
6879 declare i32 @llvm.atomic.load.min.i32..p0i32( i32* <ptr>, i32 <delta> )
6880 declare i64 @llvm.atomic.load.min.i64..p0i64( i64* <ptr>, i64 <delta> )
6884 declare i8 @llvm.atomic.load.umax.i8.p0i8( i8* <ptr>, i8 <delta> )
6885 declare i16 @llvm.atomic.load.umax.i16.p0i16( i16* <ptr>, i16 <delta> )
6886 declare i32 @llvm.atomic.load.umax.i32.p0i32( i32* <ptr>, i32 <delta> )
6887 declare i64 @llvm.atomic.load.umax.i64.p0i64( i64* <ptr>, i64 <delta> )
6891 declare i8 @llvm.atomic.load.umin.i8..p0i8( i8* <ptr>, i8 <delta> )
6892 declare i16 @llvm.atomic.load.umin.i16.p0i16( i16* <ptr>, i16 <delta> )
6893 declare i32 @llvm.atomic.load.umin.i32..p0i32( i32* <ptr>, i32 <delta> )
6894 declare i64 @llvm.atomic.load.umin.i64..p0i64( i64* <ptr>, i64 <delta> )
6898 <p>These intrinsics takes the signed or unsigned minimum or maximum of
6899 <tt>delta</tt> and the value stored in memory at <tt>ptr</tt>. It yields the
6900 original value at <tt>ptr</tt>.</p>
6903 <p>These intrinsics take two arguments, the first a pointer to an integer value
6904 and the second an integer value. The result is also an integer value. These
6905 integer types can have any bit width, but they must all have the same bit
6906 width. The targets may only lower integer representations they support.</p>
6909 <p>These intrinsics does a series of operations atomically. They first load the
6910 value stored at <tt>ptr</tt>. They then do the signed or unsigned min or
6911 max <tt>delta</tt> and the value, store the result to <tt>ptr</tt>. They
6912 yield the original value stored at <tt>ptr</tt>.</p>
6916 %mallocP = tail call i8* @malloc(i32 ptrtoint (i32* getelementptr (i32* null, i32 1) to i32))
6917 %ptr = bitcast i8* %mallocP to i32*
6919 %result0 = call i32 @llvm.atomic.load.min.i32.p0i32( i32* %ptr, i32 -2 )
6920 <i>; yields {i32}:result0 = 7</i>
6921 %result1 = call i32 @llvm.atomic.load.max.i32.p0i32( i32* %ptr, i32 8 )
6922 <i>; yields {i32}:result1 = -2</i>
6923 %result2 = call i32 @llvm.atomic.load.umin.i32.p0i32( i32* %ptr, i32 10 )
6924 <i>; yields {i32}:result2 = 8</i>
6925 %result3 = call i32 @llvm.atomic.load.umax.i32.p0i32( i32* %ptr, i32 30 )
6926 <i>; yields {i32}:result3 = 8</i>
6927 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 30</i>
6933 <!-- ======================================================================= -->
6934 <div class="doc_subsection">
6935 <a name="int_memorymarkers">Memory Use Markers</a>
6938 <div class="doc_text">
6940 <p>This class of intrinsics exists to information about the lifetime of memory
6941 objects and ranges where variables are immutable.</p>
6945 <!-- _______________________________________________________________________ -->
6946 <div class="doc_subsubsection">
6947 <a name="int_lifetime_start">'<tt>llvm.lifetime.start</tt>' Intrinsic</a>
6950 <div class="doc_text">
6954 declare void @llvm.lifetime.start(i64 <size>, i8* nocapture <ptr>)
6958 <p>The '<tt>llvm.lifetime.start</tt>' intrinsic specifies the start of a memory
6959 object's lifetime.</p>
6962 <p>The first argument is a constant integer representing the size of the
6963 object, or -1 if it is variable sized. The second argument is a pointer to
6967 <p>This intrinsic indicates that before this point in the code, the value of the
6968 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
6969 never be used and has an undefined value. A load from the pointer that is
6970 preceded by this intrinsic can be replaced with
6971 <tt>'<a href="#undefvalues">undef</a>'</tt>.</p>
6975 <!-- _______________________________________________________________________ -->
6976 <div class="doc_subsubsection">
6977 <a name="int_lifetime_end">'<tt>llvm.lifetime.end</tt>' Intrinsic</a>
6980 <div class="doc_text">
6984 declare void @llvm.lifetime.end(i64 <size>, i8* nocapture <ptr>)
6988 <p>The '<tt>llvm.lifetime.end</tt>' intrinsic specifies the end of a memory
6989 object's lifetime.</p>
6992 <p>The first argument is a constant integer representing the size of the
6993 object, or -1 if it is variable sized. The second argument is a pointer to
6997 <p>This intrinsic indicates that after this point in the code, the value of the
6998 memory pointed to by <tt>ptr</tt> is dead. This means that it is known to
6999 never be used and has an undefined value. Any stores into the memory object
7000 following this intrinsic may be removed as dead.
7004 <!-- _______________________________________________________________________ -->
7005 <div class="doc_subsubsection">
7006 <a name="int_invariant_start">'<tt>llvm.invariant.start</tt>' Intrinsic</a>
7009 <div class="doc_text">
7013 declare {}* @llvm.invariant.start(i64 <size>, i8* nocapture <ptr>) readonly
7017 <p>The '<tt>llvm.invariant.start</tt>' intrinsic specifies that the contents of
7018 a memory object will not change.</p>
7021 <p>The first argument is a constant integer representing the size of the
7022 object, or -1 if it is variable sized. The second argument is a pointer to
7026 <p>This intrinsic indicates that until an <tt>llvm.invariant.end</tt> that uses
7027 the return value, the referenced memory location is constant and
7032 <!-- _______________________________________________________________________ -->
7033 <div class="doc_subsubsection">
7034 <a name="int_invariant_end">'<tt>llvm.invariant.end</tt>' Intrinsic</a>
7037 <div class="doc_text">
7041 declare void @llvm.invariant.end({}* <start>, i64 <size>, i8* nocapture <ptr>)
7045 <p>The '<tt>llvm.invariant.end</tt>' intrinsic specifies that the contents of
7046 a memory object are mutable.</p>
7049 <p>The first argument is the matching <tt>llvm.invariant.start</tt> intrinsic.
7050 The second argument is a constant integer representing the size of the
7051 object, or -1 if it is variable sized and the third argument is a pointer
7055 <p>This intrinsic indicates that the memory is mutable again.</p>
7059 <!-- ======================================================================= -->
7060 <div class="doc_subsection">
7061 <a name="int_general">General Intrinsics</a>
7064 <div class="doc_text">
7066 <p>This class of intrinsics is designed to be generic and has no specific
7071 <!-- _______________________________________________________________________ -->
7072 <div class="doc_subsubsection">
7073 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
7076 <div class="doc_text">
7080 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
7084 <p>The '<tt>llvm.var.annotation</tt>' intrinsic.</p>
7087 <p>The first argument is a pointer to a value, the second is a pointer to a
7088 global string, the third is a pointer to a global string which is the source
7089 file name, and the last argument is the line number.</p>
7092 <p>This intrinsic allows annotation of local variables with arbitrary strings.
7093 This can be useful for special purpose optimizations that want to look for
7094 these annotations. These have no other defined use, they are ignored by code
7095 generation and optimization.</p>
7099 <!-- _______________________________________________________________________ -->
7100 <div class="doc_subsubsection">
7101 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
7104 <div class="doc_text">
7107 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
7108 any integer bit width.</p>
7111 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
7112 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
7113 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
7114 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
7115 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
7119 <p>The '<tt>llvm.annotation</tt>' intrinsic.</p>
7122 <p>The first argument is an integer value (result of some expression), the
7123 second is a pointer to a global string, the third is a pointer to a global
7124 string which is the source file name, and the last argument is the line
7125 number. It returns the value of the first argument.</p>
7128 <p>This intrinsic allows annotations to be put on arbitrary expressions with
7129 arbitrary strings. This can be useful for special purpose optimizations that
7130 want to look for these annotations. These have no other defined use, they
7131 are ignored by code generation and optimization.</p>
7135 <!-- _______________________________________________________________________ -->
7136 <div class="doc_subsubsection">
7137 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
7140 <div class="doc_text">
7144 declare void @llvm.trap()
7148 <p>The '<tt>llvm.trap</tt>' intrinsic.</p>
7154 <p>This intrinsics is lowered to the target dependent trap instruction. If the
7155 target does not have a trap instruction, this intrinsic will be lowered to
7156 the call of the <tt>abort()</tt> function.</p>
7160 <!-- _______________________________________________________________________ -->
7161 <div class="doc_subsubsection">
7162 <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a>
7165 <div class="doc_text">
7169 declare void @llvm.stackprotector( i8* <guard>, i8** <slot> )
7173 <p>The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and
7174 stores it onto the stack at <tt>slot</tt>. The stack slot is adjusted to
7175 ensure that it is placed on the stack before local variables.</p>
7178 <p>The <tt>llvm.stackprotector</tt> intrinsic requires two pointer
7179 arguments. The first argument is the value loaded from the stack
7180 guard <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt>
7181 that has enough space to hold the value of the guard.</p>
7184 <p>This intrinsic causes the prologue/epilogue inserter to force the position of
7185 the <tt>AllocaInst</tt> stack slot to be before local variables on the
7186 stack. This is to ensure that if a local variable on the stack is
7187 overwritten, it will destroy the value of the guard. When the function exits,
7188 the guard on the stack is checked against the original guard. If they're
7189 different, then the program aborts by calling the <tt>__stack_chk_fail()</tt>
7194 <!-- *********************************************************************** -->
7197 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
7198 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
7199 <a href="http://validator.w3.org/check/referer"><img
7200 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
7202 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
7203 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
7204 Last modified: $Date$