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></li>
24 <li><a href="#callingconv">Calling Conventions</a></li>
25 <li><a href="#namedtypes">Named Types</a></li>
26 <li><a href="#globalvars">Global Variables</a></li>
27 <li><a href="#functionstructure">Functions</a></li>
28 <li><a href="#aliasstructure">Aliases</a></li>
29 <li><a href="#paramattrs">Parameter Attributes</a></li>
30 <li><a href="#fnattrs">Function Attributes</a></li>
31 <li><a href="#gc">Garbage Collector Names</a></li>
32 <li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
33 <li><a href="#datalayout">Data Layout</a></li>
36 <li><a href="#typesystem">Type System</a>
38 <li><a href="#t_classifications">Type Classifications</a></li>
39 <li><a href="#t_primitive">Primitive Types</a>
41 <li><a href="#t_floating">Floating Point Types</a></li>
42 <li><a href="#t_void">Void Type</a></li>
43 <li><a href="#t_label">Label Type</a></li>
46 <li><a href="#t_derived">Derived Types</a>
48 <li><a href="#t_integer">Integer Type</a></li>
49 <li><a href="#t_array">Array Type</a></li>
50 <li><a href="#t_function">Function Type</a></li>
51 <li><a href="#t_pointer">Pointer Type</a></li>
52 <li><a href="#t_struct">Structure Type</a></li>
53 <li><a href="#t_pstruct">Packed Structure Type</a></li>
54 <li><a href="#t_vector">Vector Type</a></li>
55 <li><a href="#t_opaque">Opaque Type</a></li>
60 <li><a href="#constants">Constants</a>
62 <li><a href="#simpleconstants">Simple Constants</a></li>
63 <li><a href="#aggregateconstants">Aggregate Constants</a></li>
64 <li><a href="#globalconstants">Global Variable and Function Addresses</a></li>
65 <li><a href="#undefvalues">Undefined Values</a></li>
66 <li><a href="#constantexprs">Constant Expressions</a></li>
69 <li><a href="#othervalues">Other Values</a>
71 <li><a href="#inlineasm">Inline Assembler Expressions</a></li>
74 <li><a href="#instref">Instruction Reference</a>
76 <li><a href="#terminators">Terminator Instructions</a>
78 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
79 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
80 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
81 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
82 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
83 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
86 <li><a href="#binaryops">Binary Operations</a>
88 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
89 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
90 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
91 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
92 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
93 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
94 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
95 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
96 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
99 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
101 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
102 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
103 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
104 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
105 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
106 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
109 <li><a href="#vectorops">Vector Operations</a>
111 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
112 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
113 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
116 <li><a href="#aggregateops">Aggregate Operations</a>
118 <li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li>
119 <li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li>
122 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
124 <li><a href="#i_malloc">'<tt>malloc</tt>' Instruction</a></li>
125 <li><a href="#i_free">'<tt>free</tt>' Instruction</a></li>
126 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
127 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
128 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
129 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
132 <li><a href="#convertops">Conversion Operations</a>
134 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
135 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
136 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
137 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
138 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
139 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
140 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
141 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
142 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
143 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
144 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
145 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
148 <li><a href="#otherops">Other Operations</a>
150 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
151 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
152 <li><a href="#i_vicmp">'<tt>vicmp</tt>' Instruction</a></li>
153 <li><a href="#i_vfcmp">'<tt>vfcmp</tt>' Instruction</a></li>
154 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
155 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
156 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
157 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
162 <li><a href="#intrinsics">Intrinsic Functions</a>
164 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
166 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
167 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
168 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
171 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
173 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
174 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
175 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
178 <li><a href="#int_codegen">Code Generator Intrinsics</a>
180 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
181 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
182 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
183 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
184 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
185 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
186 <li><a href="#int_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
189 <li><a href="#int_libc">Standard C Library Intrinsics</a>
191 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
192 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
193 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
194 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
195 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
196 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
197 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
198 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
201 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
203 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
204 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
205 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
206 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
207 <li><a href="#int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic </a></li>
208 <li><a href="#int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic </a></li>
211 <li><a href="#int_debugger">Debugger intrinsics</a></li>
212 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
213 <li><a href="#int_trampoline">Trampoline Intrinsic</a>
215 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
218 <li><a href="#int_atomics">Atomic intrinsics</a>
220 <li><a href="#int_memory_barrier"><tt>llvm.memory_barrier</tt></a></li>
221 <li><a href="#int_atomic_cmp_swap"><tt>llvm.atomic.cmp.swap</tt></a></li>
222 <li><a href="#int_atomic_swap"><tt>llvm.atomic.swap</tt></a></li>
223 <li><a href="#int_atomic_load_add"><tt>llvm.atomic.load.add</tt></a></li>
224 <li><a href="#int_atomic_load_sub"><tt>llvm.atomic.load.sub</tt></a></li>
225 <li><a href="#int_atomic_load_and"><tt>llvm.atomic.load.and</tt></a></li>
226 <li><a href="#int_atomic_load_nand"><tt>llvm.atomic.load.nand</tt></a></li>
227 <li><a href="#int_atomic_load_or"><tt>llvm.atomic.load.or</tt></a></li>
228 <li><a href="#int_atomic_load_xor"><tt>llvm.atomic.load.xor</tt></a></li>
229 <li><a href="#int_atomic_load_max"><tt>llvm.atomic.load.max</tt></a></li>
230 <li><a href="#int_atomic_load_min"><tt>llvm.atomic.load.min</tt></a></li>
231 <li><a href="#int_atomic_load_umax"><tt>llvm.atomic.load.umax</tt></a></li>
232 <li><a href="#int_atomic_load_umin"><tt>llvm.atomic.load.umin</tt></a></li>
235 <li><a href="#int_general">General intrinsics</a>
237 <li><a href="#int_var_annotation">
238 '<tt>llvm.var.annotation</tt>' Intrinsic</a></li>
239 <li><a href="#int_annotation">
240 '<tt>llvm.annotation.*</tt>' Intrinsic</a></li>
241 <li><a href="#int_trap">
242 '<tt>llvm.trap</tt>' Intrinsic</a></li>
243 <li><a href="#int_stackprotector">
244 '<tt>llvm.stackprotector</tt>' Intrinsic</a></li>
251 <div class="doc_author">
252 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
253 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
256 <!-- *********************************************************************** -->
257 <div class="doc_section"> <a name="abstract">Abstract </a></div>
258 <!-- *********************************************************************** -->
260 <div class="doc_text">
261 <p>This document is a reference manual for the LLVM assembly language.
262 LLVM is a Static Single Assignment (SSA) based representation that provides
263 type safety, low-level operations, flexibility, and the capability of
264 representing 'all' high-level languages cleanly. It is the common code
265 representation used throughout all phases of the LLVM compilation
269 <!-- *********************************************************************** -->
270 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
271 <!-- *********************************************************************** -->
273 <div class="doc_text">
275 <p>The LLVM code representation is designed to be used in three
276 different forms: as an in-memory compiler IR, as an on-disk bitcode
277 representation (suitable for fast loading by a Just-In-Time compiler),
278 and as a human readable assembly language representation. This allows
279 LLVM to provide a powerful intermediate representation for efficient
280 compiler transformations and analysis, while providing a natural means
281 to debug and visualize the transformations. The three different forms
282 of LLVM are all equivalent. This document describes the human readable
283 representation and notation.</p>
285 <p>The LLVM representation aims to be light-weight and low-level
286 while being expressive, typed, and extensible at the same time. It
287 aims to be a "universal IR" of sorts, by being at a low enough level
288 that high-level ideas may be cleanly mapped to it (similar to how
289 microprocessors are "universal IR's", allowing many source languages to
290 be mapped to them). By providing type information, LLVM can be used as
291 the target of optimizations: for example, through pointer analysis, it
292 can be proven that a C automatic variable is never accessed outside of
293 the current function... allowing it to be promoted to a simple SSA
294 value instead of a memory location.</p>
298 <!-- _______________________________________________________________________ -->
299 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
301 <div class="doc_text">
303 <p>It is important to note that this document describes 'well formed'
304 LLVM assembly language. There is a difference between what the parser
305 accepts and what is considered 'well formed'. For example, the
306 following instruction is syntactically okay, but not well formed:</p>
308 <div class="doc_code">
310 %x = <a href="#i_add">add</a> i32 1, %x
314 <p>...because the definition of <tt>%x</tt> does not dominate all of
315 its uses. The LLVM infrastructure provides a verification pass that may
316 be used to verify that an LLVM module is well formed. This pass is
317 automatically run by the parser after parsing input assembly and by
318 the optimizer before it outputs bitcode. The violations pointed out
319 by the verifier pass indicate bugs in transformation passes or input to
323 <!-- Describe the typesetting conventions here. -->
325 <!-- *********************************************************************** -->
326 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
327 <!-- *********************************************************************** -->
329 <div class="doc_text">
331 <p>LLVM identifiers come in two basic types: global and local. Global
332 identifiers (functions, global variables) begin with the @ character. Local
333 identifiers (register names, types) begin with the % character. Additionally,
334 there are three different formats for identifiers, for different purposes:</p>
337 <li>Named values are represented as a string of characters with their prefix.
338 For example, %foo, @DivisionByZero, %a.really.long.identifier. The actual
339 regular expression used is '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
340 Identifiers which require other characters in their names can be surrounded
341 with quotes. Special characters may be escaped using "\xx" where xx is the
342 ASCII code for the character in hexadecimal. In this way, any character can
343 be used in a name value, even quotes themselves.
345 <li>Unnamed values are represented as an unsigned numeric value with their
346 prefix. For example, %12, @2, %44.</li>
348 <li>Constants, which are described in a <a href="#constants">section about
349 constants</a>, below.</li>
352 <p>LLVM requires that values start with a prefix for two reasons: Compilers
353 don't need to worry about name clashes with reserved words, and the set of
354 reserved words may be expanded in the future without penalty. Additionally,
355 unnamed identifiers allow a compiler to quickly come up with a temporary
356 variable without having to avoid symbol table conflicts.</p>
358 <p>Reserved words in LLVM are very similar to reserved words in other
359 languages. There are keywords for different opcodes
360 ('<tt><a href="#i_add">add</a></tt>',
361 '<tt><a href="#i_bitcast">bitcast</a></tt>',
362 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
363 href="#t_void">void</a></tt>', '<tt><a href="#t_primitive">i32</a></tt>', etc...),
364 and others. These reserved words cannot conflict with variable names, because
365 none of them start with a prefix character ('%' or '@').</p>
367 <p>Here is an example of LLVM code to multiply the integer variable
368 '<tt>%X</tt>' by 8:</p>
372 <div class="doc_code">
374 %result = <a href="#i_mul">mul</a> i32 %X, 8
378 <p>After strength reduction:</p>
380 <div class="doc_code">
382 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
386 <p>And the hard way:</p>
388 <div class="doc_code">
390 <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
391 <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
392 %result = <a href="#i_add">add</a> i32 %1, %1
396 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
397 important lexical features of LLVM:</p>
401 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
404 <li>Unnamed temporaries are created when the result of a computation is not
405 assigned to a named value.</li>
407 <li>Unnamed temporaries are numbered sequentially</li>
411 <p>...and it also shows a convention that we follow in this document. When
412 demonstrating instructions, we will follow an instruction with a comment that
413 defines the type and name of value produced. Comments are shown in italic
418 <!-- *********************************************************************** -->
419 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
420 <!-- *********************************************************************** -->
422 <!-- ======================================================================= -->
423 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
426 <div class="doc_text">
428 <p>LLVM programs are composed of "Module"s, each of which is a
429 translation unit of the input programs. Each module consists of
430 functions, global variables, and symbol table entries. Modules may be
431 combined together with the LLVM linker, which merges function (and
432 global variable) definitions, resolves forward declarations, and merges
433 symbol table entries. Here is an example of the "hello world" module:</p>
435 <div class="doc_code">
436 <pre><i>; Declare the string constant as a global constant...</i>
437 <a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a
438 href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
440 <i>; External declaration of the puts function</i>
441 <a href="#functionstructure">declare</a> i32 @puts(i8 *) <i>; i32(i8 *)* </i>
443 <i>; Definition of main function</i>
444 define i32 @main() { <i>; i32()* </i>
445 <i>; Convert [13 x i8]* to i8 *...</i>
447 href="#i_getelementptr">getelementptr</a> [13 x i8]* @.LC0, i64 0, i64 0 <i>; i8 *</i>
449 <i>; Call puts function to write out the string to stdout...</i>
451 href="#i_call">call</a> i32 @puts(i8 * %cast210) <i>; i32</i>
453 href="#i_ret">ret</a> i32 0<br>}<br>
457 <p>This example is made up of a <a href="#globalvars">global variable</a>
458 named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
459 function, and a <a href="#functionstructure">function definition</a>
460 for "<tt>main</tt>".</p>
462 <p>In general, a module is made up of a list of global values,
463 where both functions and global variables are global values. Global values are
464 represented by a pointer to a memory location (in this case, a pointer to an
465 array of char, and a pointer to a function), and have one of the following <a
466 href="#linkage">linkage types</a>.</p>
470 <!-- ======================================================================= -->
471 <div class="doc_subsection">
472 <a name="linkage">Linkage Types</a>
475 <div class="doc_text">
478 All Global Variables and Functions have one of the following types of linkage:
483 <dt><tt><b><a name="linkage_private">private</a></b></tt>: </dt>
485 <dd>Global values with private linkage are only directly accessible by
486 objects in the current module. In particular, linking code into a module with
487 an private global value may cause the private to be renamed as necessary to
488 avoid collisions. Because the symbol is private to the module, all
489 references can be updated. This doesn't show up in any symbol table in the
493 <dt><tt><b><a name="linkage_internal">internal</a></b></tt>: </dt>
495 <dd> Similar to private, but the value shows as a local symbol (STB_LOCAL in
496 the case of ELF) in the object file. This corresponds to the notion of the
497 '<tt>static</tt>' keyword in C.
500 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
502 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
503 the same name when linkage occurs. This is typically used to implement
504 inline functions, templates, or other code which must be generated in each
505 translation unit that uses it. Unreferenced <tt>linkonce</tt> globals are
506 allowed to be discarded.
509 <dt><tt><b><a name="linkage_common">common</a></b></tt>: </dt>
511 <dd>"<tt>common</tt>" linkage is exactly the same as <tt>linkonce</tt>
512 linkage, except that unreferenced <tt>common</tt> globals may not be
513 discarded. This is used for globals that may be emitted in multiple
514 translation units, but that are not guaranteed to be emitted into every
515 translation unit that uses them. One example of this is tentative
516 definitions in C, such as "<tt>int X;</tt>" at global scope.
519 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
521 <dd>"<tt>weak</tt>" linkage is the same as <tt>common</tt> linkage, except
522 that some targets may choose to emit different assembly sequences for them
523 for target-dependent reasons. This is used for globals that are declared
524 "weak" in C source code.
527 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
529 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
530 pointer to array type. When two global variables with appending linkage are
531 linked together, the two global arrays are appended together. This is the
532 LLVM, typesafe, equivalent of having the system linker append together
533 "sections" with identical names when .o files are linked.
536 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt>: </dt>
537 <dd>The semantics of this linkage follow the ELF object file model: the
538 symbol is weak until linked, if not linked, the symbol becomes null instead
539 of being an undefined reference.
542 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
544 <dd>If none of the above identifiers are used, the global is externally
545 visible, meaning that it participates in linkage and can be used to resolve
546 external symbol references.
551 The next two types of linkage are targeted for Microsoft Windows platform
552 only. They are designed to support importing (exporting) symbols from (to)
553 DLLs (Dynamic Link Libraries).
557 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt>: </dt>
559 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
560 or variable via a global pointer to a pointer that is set up by the DLL
561 exporting the symbol. On Microsoft Windows targets, the pointer name is
562 formed by combining <code>__imp_</code> and the function or variable name.
565 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt>: </dt>
567 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
568 pointer to a pointer in a DLL, so that it can be referenced with the
569 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
570 name is formed by combining <code>__imp_</code> and the function or variable
576 <p>For example, since the "<tt>.LC0</tt>"
577 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
578 variable and was linked with this one, one of the two would be renamed,
579 preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
580 external (i.e., lacking any linkage declarations), they are accessible
581 outside of the current module.</p>
582 <p>It is illegal for a function <i>declaration</i>
583 to have any linkage type other than "externally visible", <tt>dllimport</tt>,
584 or <tt>extern_weak</tt>.</p>
585 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt> and <tt>weak</tt>
589 <!-- ======================================================================= -->
590 <div class="doc_subsection">
591 <a name="callingconv">Calling Conventions</a>
594 <div class="doc_text">
596 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
597 and <a href="#i_invoke">invokes</a> can all have an optional calling convention
598 specified for the call. The calling convention of any pair of dynamic
599 caller/callee must match, or the behavior of the program is undefined. The
600 following calling conventions are supported by LLVM, and more may be added in
604 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
606 <dd>This calling convention (the default if no other calling convention is
607 specified) matches the target C calling conventions. This calling convention
608 supports varargs function calls and tolerates some mismatch in the declared
609 prototype and implemented declaration of the function (as does normal C).
612 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
614 <dd>This calling convention attempts to make calls as fast as possible
615 (e.g. by passing things in registers). This calling convention allows the
616 target to use whatever tricks it wants to produce fast code for the target,
617 without having to conform to an externally specified ABI (Application Binary
618 Interface). Implementations of this convention should allow arbitrary
619 <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> to be
620 supported. This calling convention does not support varargs and requires the
621 prototype of all callees to exactly match the prototype of the function
625 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
627 <dd>This calling convention attempts to make code in the caller as efficient
628 as possible under the assumption that the call is not commonly executed. As
629 such, these calls often preserve all registers so that the call does not break
630 any live ranges in the caller side. This calling convention does not support
631 varargs and requires the prototype of all callees to exactly match the
632 prototype of the function definition.
635 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
637 <dd>Any calling convention may be specified by number, allowing
638 target-specific calling conventions to be used. Target specific calling
639 conventions start at 64.
643 <p>More calling conventions can be added/defined on an as-needed basis, to
644 support pascal conventions or any other well-known target-independent
649 <!-- ======================================================================= -->
650 <div class="doc_subsection">
651 <a name="visibility">Visibility Styles</a>
654 <div class="doc_text">
657 All Global Variables and Functions have one of the following visibility styles:
661 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
663 <dd>On targets that use the ELF object file format, default visibility means
664 that the declaration is visible to other
665 modules and, in shared libraries, means that the declared entity may be
666 overridden. On Darwin, default visibility means that the declaration is
667 visible to other modules. Default visibility corresponds to "external
668 linkage" in the language.
671 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
673 <dd>Two declarations of an object with hidden visibility refer to the same
674 object if they are in the same shared object. Usually, hidden visibility
675 indicates that the symbol will not be placed into the dynamic symbol table,
676 so no other module (executable or shared library) can reference it
680 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
682 <dd>On ELF, protected visibility indicates that the symbol will be placed in
683 the dynamic symbol table, but that references within the defining module will
684 bind to the local symbol. That is, the symbol cannot be overridden by another
691 <!-- ======================================================================= -->
692 <div class="doc_subsection">
693 <a name="namedtypes">Named Types</a>
696 <div class="doc_text">
698 <p>LLVM IR allows you to specify name aliases for certain types. This can make
699 it easier to read the IR and make the IR more condensed (particularly when
700 recursive types are involved). An example of a name specification is:
703 <div class="doc_code">
705 %mytype = type { %mytype*, i32 }
709 <p>You may give a name to any <a href="#typesystem">type</a> except "<a
710 href="t_void">void</a>". Type name aliases may be used anywhere a type is
711 expected with the syntax "%mytype".</p>
713 <p>Note that type names are aliases for the structural type that they indicate,
714 and that you can therefore specify multiple names for the same type. This often
715 leads to confusing behavior when dumping out a .ll file. Since LLVM IR uses
716 structural typing, the name is not part of the type. When printing out LLVM IR,
717 the printer will pick <em>one name</em> to render all types of a particular
718 shape. This means that if you have code where two different source types end up
719 having the same LLVM type, that the dumper will sometimes print the "wrong" or
720 unexpected type. This is an important design point and isn't going to
726 <!-- ======================================================================= -->
727 <div class="doc_subsection">
728 <a name="globalvars">Global Variables</a>
731 <div class="doc_text">
733 <p>Global variables define regions of memory allocated at compilation time
734 instead of run-time. Global variables may optionally be initialized, may have
735 an explicit section to be placed in, and may have an optional explicit alignment
736 specified. A variable may be defined as "thread_local", which means that it
737 will not be shared by threads (each thread will have a separated copy of the
738 variable). A variable may be defined as a global "constant," which indicates
739 that the contents of the variable will <b>never</b> be modified (enabling better
740 optimization, allowing the global data to be placed in the read-only section of
741 an executable, etc). Note that variables that need runtime initialization
742 cannot be marked "constant" as there is a store to the variable.</p>
745 LLVM explicitly allows <em>declarations</em> of global variables to be marked
746 constant, even if the final definition of the global is not. This capability
747 can be used to enable slightly better optimization of the program, but requires
748 the language definition to guarantee that optimizations based on the
749 'constantness' are valid for the translation units that do not include the
753 <p>As SSA values, global variables define pointer values that are in
754 scope (i.e. they dominate) all basic blocks in the program. Global
755 variables always define a pointer to their "content" type because they
756 describe a region of memory, and all memory objects in LLVM are
757 accessed through pointers.</p>
759 <p>A global variable may be declared to reside in a target-specifc numbered
760 address space. For targets that support them, address spaces may affect how
761 optimizations are performed and/or what target instructions are used to access
762 the variable. The default address space is zero. The address space qualifier
763 must precede any other attributes.</p>
765 <p>LLVM allows an explicit section to be specified for globals. If the target
766 supports it, it will emit globals to the section specified.</p>
768 <p>An explicit alignment may be specified for a global. If not present, or if
769 the alignment is set to zero, the alignment of the global is set by the target
770 to whatever it feels convenient. If an explicit alignment is specified, the
771 global is forced to have at least that much alignment. All alignments must be
774 <p>For example, the following defines a global in a numbered address space with
775 an initializer, section, and alignment:</p>
777 <div class="doc_code">
779 @G = addrspace(5) constant float 1.0, section "foo", align 4
786 <!-- ======================================================================= -->
787 <div class="doc_subsection">
788 <a name="functionstructure">Functions</a>
791 <div class="doc_text">
793 <p>LLVM function definitions consist of the "<tt>define</tt>" keyord,
794 an optional <a href="#linkage">linkage type</a>, an optional
795 <a href="#visibility">visibility style</a>, an optional
796 <a href="#callingconv">calling convention</a>, a return type, an optional
797 <a href="#paramattrs">parameter attribute</a> for the return type, a function
798 name, a (possibly empty) argument list (each with optional
799 <a href="#paramattrs">parameter attributes</a>), optional
800 <a href="#fnattrs">function attributes</a>, an optional section,
801 an optional alignment, an optional <a href="#gc">garbage collector name</a>,
802 an opening curly brace, a list of basic blocks, and a closing curly brace.
804 LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
805 optional <a href="#linkage">linkage type</a>, an optional
806 <a href="#visibility">visibility style</a>, an optional
807 <a href="#callingconv">calling convention</a>, a return type, an optional
808 <a href="#paramattrs">parameter attribute</a> for the return type, a function
809 name, a possibly empty list of arguments, an optional alignment, and an optional
810 <a href="#gc">garbage collector name</a>.</p>
812 <p>A function definition contains a list of basic blocks, forming the CFG
813 (Control Flow Graph) for
814 the function. Each basic block may optionally start with a label (giving the
815 basic block a symbol table entry), contains a list of instructions, and ends
816 with a <a href="#terminators">terminator</a> instruction (such as a branch or
817 function return).</p>
819 <p>The first basic block in a function is special in two ways: it is immediately
820 executed on entrance to the function, and it is not allowed to have predecessor
821 basic blocks (i.e. there can not be any branches to the entry block of a
822 function). Because the block can have no predecessors, it also cannot have any
823 <a href="#i_phi">PHI nodes</a>.</p>
825 <p>LLVM allows an explicit section to be specified for functions. If the target
826 supports it, it will emit functions to the section specified.</p>
828 <p>An explicit alignment may be specified for a function. If not present, or if
829 the alignment is set to zero, the alignment of the function is set by the target
830 to whatever it feels convenient. If an explicit alignment is specified, the
831 function is forced to have at least that much alignment. All alignments must be
836 <div class="doc_code">
838 define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
839 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
840 <ResultType> @<FunctionName> ([argument list])
841 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
842 [<a href="#gc">gc</a>] { ... }
849 <!-- ======================================================================= -->
850 <div class="doc_subsection">
851 <a name="aliasstructure">Aliases</a>
853 <div class="doc_text">
854 <p>Aliases act as "second name" for the aliasee value (which can be either
855 function, global variable, another alias or bitcast of global value). Aliases
856 may have an optional <a href="#linkage">linkage type</a>, and an
857 optional <a href="#visibility">visibility style</a>.</p>
861 <div class="doc_code">
863 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
871 <!-- ======================================================================= -->
872 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
873 <div class="doc_text">
874 <p>The return type and each parameter of a function type may have a set of
875 <i>parameter attributes</i> associated with them. Parameter attributes are
876 used to communicate additional information about the result or parameters of
877 a function. Parameter attributes are considered to be part of the function,
878 not of the function type, so functions with different parameter attributes
879 can have the same function type.</p>
881 <p>Parameter attributes are simple keywords that follow the type specified. If
882 multiple parameter attributes are needed, they are space separated. For
885 <div class="doc_code">
887 declare i32 @printf(i8* noalias , ...)
888 declare i32 @atoi(i8 zeroext)
889 declare signext i8 @returns_signed_char()
893 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
894 <tt>readonly</tt>) come immediately after the argument list.</p>
896 <p>Currently, only the following parameter attributes are defined:</p>
898 <dt><tt>zeroext</tt></dt>
899 <dd>This indicates to the code generator that the parameter or return value
900 should be zero-extended to a 32-bit value by the caller (for a parameter)
901 or the callee (for a return value).</dd>
903 <dt><tt>signext</tt></dt>
904 <dd>This indicates to the code generator that the parameter or return value
905 should be sign-extended to a 32-bit value by the caller (for a parameter)
906 or the callee (for a return value).</dd>
908 <dt><tt>inreg</tt></dt>
909 <dd>This indicates that this parameter or return value should be treated
910 in a special target-dependent fashion during while emitting code for a
911 function call or return (usually, by putting it in a register as opposed
912 to memory, though some targets use it to distinguish between two different
913 kinds of registers). Use of this attribute is target-specific.</dd>
915 <dt><tt><a name="byval">byval</a></tt></dt>
916 <dd>This indicates that the pointer parameter should really be passed by
917 value to the function. The attribute implies that a hidden copy of the
918 pointee is made between the caller and the callee, so the callee is unable
919 to modify the value in the callee. This attribute is only valid on LLVM
920 pointer arguments. It is generally used to pass structs and arrays by
921 value, but is also valid on pointers to scalars. The copy is considered to
922 belong to the caller not the callee (for example,
923 <tt><a href="#readonly">readonly</a></tt> functions should not write to
924 <tt>byval</tt> parameters). This is not a valid attribute for return
927 <dt><tt>sret</tt></dt>
928 <dd>This indicates that the pointer parameter specifies the address of a
929 structure that is the return value of the function in the source program.
930 This pointer must be guaranteed by the caller to be valid: loads and stores
931 to the structure may be assumed by the callee to not to trap. This may only
932 be applied to the first parameter. This is not a valid attribute for
935 <dt><tt>noalias</tt></dt>
936 <dd>This indicates that the pointer does not alias any global or any other
937 parameter. The caller is responsible for ensuring that this is the
938 case. On a function return value, <tt>noalias</tt> additionally indicates
939 that the pointer does not alias any other pointers visible to the
940 caller. For further details, please see the discussion of the NoAlias
942 <a href="http://llvm.org/docs/AliasAnalysis.html#MustMayNo">alias
945 <dt><tt>nocapture</tt></dt>
946 <dd>This indicates that the callee does not make any copies of the pointer
947 that outlive the callee itself. This is not a valid attribute for return
950 <dt><tt>nest</tt></dt>
951 <dd>This indicates that the pointer parameter can be excised using the
952 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
953 attribute for return values.</dd>
958 <!-- ======================================================================= -->
959 <div class="doc_subsection">
960 <a name="gc">Garbage Collector Names</a>
963 <div class="doc_text">
964 <p>Each function may specify a garbage collector name, which is simply a
967 <div class="doc_code"><pre
968 >define void @f() gc "name" { ...</pre></div>
970 <p>The compiler declares the supported values of <i>name</i>. Specifying a
971 collector which will cause the compiler to alter its output in order to support
972 the named garbage collection algorithm.</p>
975 <!-- ======================================================================= -->
976 <div class="doc_subsection">
977 <a name="fnattrs">Function Attributes</a>
980 <div class="doc_text">
982 <p>Function attributes are set to communicate additional information about
983 a function. Function attributes are considered to be part of the function,
984 not of the function type, so functions with different parameter attributes
985 can have the same function type.</p>
987 <p>Function attributes are simple keywords that follow the type specified. If
988 multiple attributes are needed, they are space separated. For
991 <div class="doc_code">
993 define void @f() noinline { ... }
994 define void @f() alwaysinline { ... }
995 define void @f() alwaysinline optsize { ... }
996 define void @f() optsize
1001 <dt><tt>alwaysinline</tt></dt>
1002 <dd>This attribute indicates that the inliner should attempt to inline this
1003 function into callers whenever possible, ignoring any active inlining size
1004 threshold for this caller.</dd>
1006 <dt><tt>noinline</tt></dt>
1007 <dd>This attribute indicates that the inliner should never inline this function
1008 in any situation. This attribute may not be used together with the
1009 <tt>alwaysinline</tt> attribute.</dd>
1011 <dt><tt>optsize</tt></dt>
1012 <dd>This attribute suggests that optimization passes and code generator passes
1013 make choices that keep the code size of this function low, and otherwise do
1014 optimizations specifically to reduce code size.</dd>
1016 <dt><tt>noreturn</tt></dt>
1017 <dd>This function attribute indicates that the function never returns normally.
1018 This produces undefined behavior at runtime if the function ever does
1019 dynamically return.</dd>
1021 <dt><tt>nounwind</tt></dt>
1022 <dd>This function attribute indicates that the function never returns with an
1023 unwind or exceptional control flow. If the function does unwind, its runtime
1024 behavior is undefined.</dd>
1026 <dt><tt>readnone</tt></dt>
1027 <dd>This attribute indicates that the function computes its result (or the
1028 exception it throws) based strictly on its arguments, without dereferencing any
1029 pointer arguments or otherwise accessing any mutable state (e.g. memory, control
1030 registers, etc) visible to caller functions. It does not write through any
1031 pointer arguments (including <tt><a href="#byval">byval</a></tt> arguments) and
1032 never changes any state visible to callers.</dd>
1034 <dt><tt><a name="readonly">readonly</a></tt></dt>
1035 <dd>This attribute indicates that the function does not write through any
1036 pointer arguments (including <tt><a href="#byval">byval</a></tt> arguments)
1037 or otherwise modify any state (e.g. memory, control registers, etc) visible to
1038 caller functions. It may dereference pointer arguments and read state that may
1039 be set in the caller. A readonly function always returns the same value (or
1040 throws the same exception) when called with the same set of arguments and global
1043 <dt><tt><a name="ssp">ssp</a></tt></dt>
1044 <dd>This attribute indicates that the function should emit a stack smashing
1045 protector. It is in the form of a "canary"—a random value placed on the
1046 stack before the local variables that's checked upon return from the function to
1047 see if it has been overwritten. A heuristic is used to determine if a function
1048 needs stack protectors or not.
1050 <p>If a function that has an <tt>ssp</tt> attribute is inlined into a function
1051 that doesn't have an <tt>ssp</tt> attribute, then the resulting function will
1052 have an <tt>ssp</tt> attribute.</p></dd>
1054 <dt><tt>sspreq</tt></dt>
1055 <dd>This attribute indicates that the function should <em>always</em> emit a
1056 stack smashing protector. This overrides the <tt><a href="#ssp">ssp</a></tt>
1059 <p>If a function that has an <tt>sspreq</tt> attribute is inlined into a
1060 function that doesn't have an <tt>sspreq</tt> attribute or which has
1061 an <tt>ssp</tt> attribute, then the resulting function will have
1062 an <tt>sspreq</tt> attribute.</p></dd>
1067 <!-- ======================================================================= -->
1068 <div class="doc_subsection">
1069 <a name="moduleasm">Module-Level Inline Assembly</a>
1072 <div class="doc_text">
1074 Modules may contain "module-level inline asm" blocks, which corresponds to the
1075 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
1076 LLVM and treated as a single unit, but may be separated in the .ll file if
1077 desired. The syntax is very simple:
1080 <div class="doc_code">
1082 module asm "inline asm code goes here"
1083 module asm "more can go here"
1087 <p>The strings can contain any character by escaping non-printable characters.
1088 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1093 The inline asm code is simply printed to the machine code .s file when
1094 assembly code is generated.
1098 <!-- ======================================================================= -->
1099 <div class="doc_subsection">
1100 <a name="datalayout">Data Layout</a>
1103 <div class="doc_text">
1104 <p>A module may specify a target specific data layout string that specifies how
1105 data is to be laid out in memory. The syntax for the data layout is simply:</p>
1106 <pre> target datalayout = "<i>layout specification</i>"</pre>
1107 <p>The <i>layout specification</i> consists of a list of specifications
1108 separated by the minus sign character ('-'). Each specification starts with a
1109 letter and may include other information after the letter to define some
1110 aspect of the data layout. The specifications accepted are as follows: </p>
1113 <dd>Specifies that the target lays out data in big-endian form. That is, the
1114 bits with the most significance have the lowest address location.</dd>
1116 <dd>Specifies that the target lays out data in little-endian form. That is,
1117 the bits with the least significance have the lowest address location.</dd>
1118 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1119 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1120 <i>preferred</i> alignments. All sizes are in bits. Specifying the <i>pref</i>
1121 alignment is optional. If omitted, the preceding <tt>:</tt> should be omitted
1123 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1124 <dd>This specifies the alignment for an integer type of a given bit
1125 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1126 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1127 <dd>This specifies the alignment for a vector type of a given bit
1129 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1130 <dd>This specifies the alignment for a floating point type of a given bit
1131 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
1133 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1134 <dd>This specifies the alignment for an aggregate type of a given bit
1137 <p>When constructing the data layout for a given target, LLVM starts with a
1138 default set of specifications which are then (possibly) overriden by the
1139 specifications in the <tt>datalayout</tt> keyword. The default specifications
1140 are given in this list:</p>
1142 <li><tt>E</tt> - big endian</li>
1143 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
1144 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1145 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1146 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1147 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1148 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1149 alignment of 64-bits</li>
1150 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1151 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1152 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1153 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1154 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1156 <p>When LLVM is determining the alignment for a given type, it uses the
1157 following rules:</p>
1159 <li>If the type sought is an exact match for one of the specifications, that
1160 specification is used.</li>
1161 <li>If no match is found, and the type sought is an integer type, then the
1162 smallest integer type that is larger than the bitwidth of the sought type is
1163 used. If none of the specifications are larger than the bitwidth then the the
1164 largest integer type is used. For example, given the default specifications
1165 above, the i7 type will use the alignment of i8 (next largest) while both
1166 i65 and i256 will use the alignment of i64 (largest specified).</li>
1167 <li>If no match is found, and the type sought is a vector type, then the
1168 largest vector type that is smaller than the sought vector type will be used
1169 as a fall back. This happens because <128 x double> can be implemented
1170 in terms of 64 <2 x double>, for example.</li>
1174 <!-- *********************************************************************** -->
1175 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
1176 <!-- *********************************************************************** -->
1178 <div class="doc_text">
1180 <p>The LLVM type system is one of the most important features of the
1181 intermediate representation. Being typed enables a number of
1182 optimizations to be performed on the intermediate representation directly,
1183 without having to do
1184 extra analyses on the side before the transformation. A strong type
1185 system makes it easier to read the generated code and enables novel
1186 analyses and transformations that are not feasible to perform on normal
1187 three address code representations.</p>
1191 <!-- ======================================================================= -->
1192 <div class="doc_subsection"> <a name="t_classifications">Type
1193 Classifications</a> </div>
1194 <div class="doc_text">
1195 <p>The types fall into a few useful
1196 classifications:</p>
1198 <table border="1" cellspacing="0" cellpadding="4">
1200 <tr><th>Classification</th><th>Types</th></tr>
1202 <td><a href="#t_integer">integer</a></td>
1203 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1206 <td><a href="#t_floating">floating point</a></td>
1207 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1210 <td><a name="t_firstclass">first class</a></td>
1211 <td><a href="#t_integer">integer</a>,
1212 <a href="#t_floating">floating point</a>,
1213 <a href="#t_pointer">pointer</a>,
1214 <a href="#t_vector">vector</a>,
1215 <a href="#t_struct">structure</a>,
1216 <a href="#t_array">array</a>,
1217 <a href="#t_label">label</a>.
1221 <td><a href="#t_primitive">primitive</a></td>
1222 <td><a href="#t_label">label</a>,
1223 <a href="#t_void">void</a>,
1224 <a href="#t_floating">floating point</a>.</td>
1227 <td><a href="#t_derived">derived</a></td>
1228 <td><a href="#t_integer">integer</a>,
1229 <a href="#t_array">array</a>,
1230 <a href="#t_function">function</a>,
1231 <a href="#t_pointer">pointer</a>,
1232 <a href="#t_struct">structure</a>,
1233 <a href="#t_pstruct">packed structure</a>,
1234 <a href="#t_vector">vector</a>,
1235 <a href="#t_opaque">opaque</a>.
1241 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
1242 most important. Values of these types are the only ones which can be
1243 produced by instructions, passed as arguments, or used as operands to
1247 <!-- ======================================================================= -->
1248 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
1250 <div class="doc_text">
1251 <p>The primitive types are the fundamental building blocks of the LLVM
1256 <!-- _______________________________________________________________________ -->
1257 <div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1259 <div class="doc_text">
1262 <tr><th>Type</th><th>Description</th></tr>
1263 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1264 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1265 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1266 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1267 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1272 <!-- _______________________________________________________________________ -->
1273 <div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1275 <div class="doc_text">
1277 <p>The void type does not represent any value and has no size.</p>
1286 <!-- _______________________________________________________________________ -->
1287 <div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1289 <div class="doc_text">
1291 <p>The label type represents code labels.</p>
1301 <!-- ======================================================================= -->
1302 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1304 <div class="doc_text">
1306 <p>The real power in LLVM comes from the derived types in the system.
1307 This is what allows a programmer to represent arrays, functions,
1308 pointers, and other useful types. Note that these derived types may be
1309 recursive: For example, it is possible to have a two dimensional array.</p>
1313 <!-- _______________________________________________________________________ -->
1314 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1316 <div class="doc_text">
1319 <p>The integer type is a very simple derived type that simply specifies an
1320 arbitrary bit width for the integer type desired. Any bit width from 1 bit to
1321 2^23-1 (about 8 million) can be specified.</p>
1329 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1333 <table class="layout">
1336 <td><tt>i1</tt></td>
1337 <td>a single-bit integer.</td>
1339 <td><tt>i32</tt></td>
1340 <td>a 32-bit integer.</td>
1342 <td><tt>i1942652</tt></td>
1343 <td>a really big integer of over 1 million bits.</td>
1348 <p>Note that the code generator does not yet support large integer types
1349 to be used as function return types. The specific limit on how large a
1350 return type the code generator can currently handle is target-dependent;
1351 currently it's often 64 bits for 32-bit targets and 128 bits for 64-bit
1356 <!-- _______________________________________________________________________ -->
1357 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1359 <div class="doc_text">
1363 <p>The array type is a very simple derived type that arranges elements
1364 sequentially in memory. The array type requires a size (number of
1365 elements) and an underlying data type.</p>
1370 [<# elements> x <elementtype>]
1373 <p>The number of elements is a constant integer value; elementtype may
1374 be any type with a size.</p>
1377 <table class="layout">
1379 <td class="left"><tt>[40 x i32]</tt></td>
1380 <td class="left">Array of 40 32-bit integer values.</td>
1383 <td class="left"><tt>[41 x i32]</tt></td>
1384 <td class="left">Array of 41 32-bit integer values.</td>
1387 <td class="left"><tt>[4 x i8]</tt></td>
1388 <td class="left">Array of 4 8-bit integer values.</td>
1391 <p>Here are some examples of multidimensional arrays:</p>
1392 <table class="layout">
1394 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1395 <td class="left">3x4 array of 32-bit integer values.</td>
1398 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1399 <td class="left">12x10 array of single precision floating point values.</td>
1402 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1403 <td class="left">2x3x4 array of 16-bit integer values.</td>
1407 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1408 length array. Normally, accesses past the end of an array are undefined in
1409 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
1410 As a special case, however, zero length arrays are recognized to be variable
1411 length. This allows implementation of 'pascal style arrays' with the LLVM
1412 type "{ i32, [0 x float]}", for example.</p>
1414 <p>Note that the code generator does not yet support large aggregate types
1415 to be used as function return types. The specific limit on how large an
1416 aggregate return type the code generator can currently handle is
1417 target-dependent, and also dependent on the aggregate element types.</p>
1421 <!-- _______________________________________________________________________ -->
1422 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1423 <div class="doc_text">
1427 <p>The function type can be thought of as a function signature. It
1428 consists of a return type and a list of formal parameter types. The
1429 return type of a function type is a scalar type, a void type, or a struct type.
1430 If the return type is a struct type then all struct elements must be of first
1431 class types, and the struct must have at least one element.</p>
1436 <returntype list> (<parameter list>)
1439 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1440 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1441 which indicates that the function takes a variable number of arguments.
1442 Variable argument functions can access their arguments with the <a
1443 href="#int_varargs">variable argument handling intrinsic</a> functions.
1444 '<tt><returntype list></tt>' is a comma-separated list of
1445 <a href="#t_firstclass">first class</a> type specifiers.</p>
1448 <table class="layout">
1450 <td class="left"><tt>i32 (i32)</tt></td>
1451 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1453 </tr><tr class="layout">
1454 <td class="left"><tt>float (i16 signext, i32 *) *
1456 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1457 an <tt>i16</tt> that should be sign extended and a
1458 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1461 </tr><tr class="layout">
1462 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1463 <td class="left">A vararg function that takes at least one
1464 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1465 which returns an integer. This is the signature for <tt>printf</tt> in
1468 </tr><tr class="layout">
1469 <td class="left"><tt>{i32, i32} (i32)</tt></td>
1470 <td class="left">A function taking an <tt>i32</tt>, returning two
1471 <tt>i32</tt> values as an aggregate of type <tt>{ i32, i32 }</tt>
1477 <!-- _______________________________________________________________________ -->
1478 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1479 <div class="doc_text">
1481 <p>The structure type is used to represent a collection of data members
1482 together in memory. The packing of the field types is defined to match
1483 the ABI of the underlying processor. The elements of a structure may
1484 be any type that has a size.</p>
1485 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1486 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1487 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1490 <pre> { <type list> }<br></pre>
1492 <table class="layout">
1494 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1495 <td class="left">A triple of three <tt>i32</tt> values</td>
1496 </tr><tr class="layout">
1497 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1498 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1499 second element is a <a href="#t_pointer">pointer</a> to a
1500 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1501 an <tt>i32</tt>.</td>
1505 <p>Note that the code generator does not yet support large aggregate types
1506 to be used as function return types. The specific limit on how large an
1507 aggregate return type the code generator can currently handle is
1508 target-dependent, and also dependent on the aggregate element types.</p>
1512 <!-- _______________________________________________________________________ -->
1513 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1515 <div class="doc_text">
1517 <p>The packed structure type is used to represent a collection of data members
1518 together in memory. There is no padding between fields. Further, the alignment
1519 of a packed structure is 1 byte. The elements of a packed structure may
1520 be any type that has a size.</p>
1521 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1522 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1523 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1526 <pre> < { <type list> } > <br></pre>
1528 <table class="layout">
1530 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1531 <td class="left">A triple of three <tt>i32</tt> values</td>
1532 </tr><tr class="layout">
1534 <tt>< { float, i32 (i32)* } ></tt></td>
1535 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1536 second element is a <a href="#t_pointer">pointer</a> to a
1537 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1538 an <tt>i32</tt>.</td>
1543 <!-- _______________________________________________________________________ -->
1544 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1545 <div class="doc_text">
1547 <p>As in many languages, the pointer type represents a pointer or
1548 reference to another object, which must live in memory. Pointer types may have
1549 an optional address space attribute defining the target-specific numbered
1550 address space where the pointed-to object resides. The default address space is
1553 <pre> <type> *<br></pre>
1555 <table class="layout">
1557 <td class="left"><tt>[4 x i32]*</tt></td>
1558 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1559 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1562 <td class="left"><tt>i32 (i32 *) *</tt></td>
1563 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
1564 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1568 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1569 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1570 that resides in address space #5.</td>
1575 <!-- _______________________________________________________________________ -->
1576 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1577 <div class="doc_text">
1581 <p>A vector type is a simple derived type that represents a vector
1582 of elements. Vector types are used when multiple primitive data
1583 are operated in parallel using a single instruction (SIMD).
1584 A vector type requires a size (number of
1585 elements) and an underlying primitive data type. Vectors must have a power
1586 of two length (1, 2, 4, 8, 16 ...). Vector types are
1587 considered <a href="#t_firstclass">first class</a>.</p>
1592 < <# elements> x <elementtype> >
1595 <p>The number of elements is a constant integer value; elementtype may
1596 be any integer or floating point type.</p>
1600 <table class="layout">
1602 <td class="left"><tt><4 x i32></tt></td>
1603 <td class="left">Vector of 4 32-bit integer values.</td>
1606 <td class="left"><tt><8 x float></tt></td>
1607 <td class="left">Vector of 8 32-bit floating-point values.</td>
1610 <td class="left"><tt><2 x i64></tt></td>
1611 <td class="left">Vector of 2 64-bit integer values.</td>
1615 <p>Note that the code generator does not yet support large vector types
1616 to be used as function return types. The specific limit on how large a
1617 vector return type codegen can currently handle is target-dependent;
1618 currently it's often a few times longer than a hardware vector register.</p>
1622 <!-- _______________________________________________________________________ -->
1623 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1624 <div class="doc_text">
1628 <p>Opaque types are used to represent unknown types in the system. This
1629 corresponds (for example) to the C notion of a forward declared structure type.
1630 In LLVM, opaque types can eventually be resolved to any type (not just a
1631 structure type).</p>
1641 <table class="layout">
1643 <td class="left"><tt>opaque</tt></td>
1644 <td class="left">An opaque type.</td>
1650 <!-- *********************************************************************** -->
1651 <div class="doc_section"> <a name="constants">Constants</a> </div>
1652 <!-- *********************************************************************** -->
1654 <div class="doc_text">
1656 <p>LLVM has several different basic types of constants. This section describes
1657 them all and their syntax.</p>
1661 <!-- ======================================================================= -->
1662 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1664 <div class="doc_text">
1667 <dt><b>Boolean constants</b></dt>
1669 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1670 constants of the <tt><a href="#t_primitive">i1</a></tt> type.
1673 <dt><b>Integer constants</b></dt>
1675 <dd>Standard integers (such as '4') are constants of the <a
1676 href="#t_integer">integer</a> type. Negative numbers may be used with
1680 <dt><b>Floating point constants</b></dt>
1682 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1683 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1684 notation (see below). The assembler requires the exact decimal value of
1685 a floating-point constant. For example, the assembler accepts 1.25 but
1686 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
1687 constants must have a <a href="#t_floating">floating point</a> type. </dd>
1689 <dt><b>Null pointer constants</b></dt>
1691 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1692 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1696 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1697 of floating point constants. For example, the form '<tt>double
1698 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1699 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1700 (and the only time that they are generated by the disassembler) is when a
1701 floating point constant must be emitted but it cannot be represented as a
1702 decimal floating point number. For example, NaN's, infinities, and other
1703 special values are represented in their IEEE hexadecimal format so that
1704 assembly and disassembly do not cause any bits to change in the constants.</p>
1708 <!-- ======================================================================= -->
1709 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1712 <div class="doc_text">
1713 <p>Aggregate constants arise from aggregation of simple constants
1714 and smaller aggregate constants.</p>
1717 <dt><b>Structure constants</b></dt>
1719 <dd>Structure constants are represented with notation similar to structure
1720 type definitions (a comma separated list of elements, surrounded by braces
1721 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
1722 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>". Structure constants
1723 must have <a href="#t_struct">structure type</a>, and the number and
1724 types of elements must match those specified by the type.
1727 <dt><b>Array constants</b></dt>
1729 <dd>Array constants are represented with notation similar to array type
1730 definitions (a comma separated list of elements, surrounded by square brackets
1731 (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74 ]</tt>". Array
1732 constants must have <a href="#t_array">array type</a>, and the number and
1733 types of elements must match those specified by the type.
1736 <dt><b>Vector constants</b></dt>
1738 <dd>Vector constants are represented with notation similar to vector type
1739 definitions (a comma separated list of elements, surrounded by
1740 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32 42,
1741 i32 11, i32 74, i32 100 ></tt>". Vector constants must have <a
1742 href="#t_vector">vector type</a>, and the number and types of elements must
1743 match those specified by the type.
1746 <dt><b>Zero initialization</b></dt>
1748 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1749 value to zero of <em>any</em> type, including scalar and aggregate types.
1750 This is often used to avoid having to print large zero initializers (e.g. for
1751 large arrays) and is always exactly equivalent to using explicit zero
1758 <!-- ======================================================================= -->
1759 <div class="doc_subsection">
1760 <a name="globalconstants">Global Variable and Function Addresses</a>
1763 <div class="doc_text">
1765 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1766 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1767 constants. These constants are explicitly referenced when the <a
1768 href="#identifiers">identifier for the global</a> is used and always have <a
1769 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1772 <div class="doc_code">
1776 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1782 <!-- ======================================================================= -->
1783 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1784 <div class="doc_text">
1785 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1786 no specific value. Undefined values may be of any type and be used anywhere
1787 a constant is permitted.</p>
1789 <p>Undefined values indicate to the compiler that the program is well defined
1790 no matter what value is used, giving the compiler more freedom to optimize.
1794 <!-- ======================================================================= -->
1795 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1798 <div class="doc_text">
1800 <p>Constant expressions are used to allow expressions involving other constants
1801 to be used as constants. Constant expressions may be of any <a
1802 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1803 that does not have side effects (e.g. load and call are not supported). The
1804 following is the syntax for constant expressions:</p>
1807 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1808 <dd>Truncate a constant to another type. The bit size of CST must be larger
1809 than the bit size of TYPE. Both types must be integers.</dd>
1811 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1812 <dd>Zero extend a constant to another type. The bit size of CST must be
1813 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1815 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1816 <dd>Sign extend a constant to another type. The bit size of CST must be
1817 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1819 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1820 <dd>Truncate a floating point constant to another floating point type. The
1821 size of CST must be larger than the size of TYPE. Both types must be
1822 floating point.</dd>
1824 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1825 <dd>Floating point extend a constant to another type. The size of CST must be
1826 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1828 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
1829 <dd>Convert a floating point constant to the corresponding unsigned integer
1830 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1831 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1832 of the same number of elements. If the value won't fit in the integer type,
1833 the results are undefined.</dd>
1835 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
1836 <dd>Convert a floating point constant to the corresponding signed integer
1837 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1838 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1839 of the same number of elements. If the value won't fit in the integer type,
1840 the results are undefined.</dd>
1842 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
1843 <dd>Convert an unsigned integer constant to the corresponding floating point
1844 constant. TYPE must be a scalar or vector floating point type. CST must be of
1845 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1846 of the same number of elements. If the value won't fit in the floating point
1847 type, the results are undefined.</dd>
1849 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
1850 <dd>Convert a signed integer constant to the corresponding floating point
1851 constant. TYPE must be a scalar or vector floating point type. CST must be of
1852 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1853 of the same number of elements. If the value won't fit in the floating point
1854 type, the results are undefined.</dd>
1856 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
1857 <dd>Convert a pointer typed constant to the corresponding integer constant
1858 TYPE must be an integer type. CST must be of pointer type. The CST value is
1859 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1861 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
1862 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1863 pointer type. CST must be of integer type. The CST value is zero extended,
1864 truncated, or unchanged to make it fit in a pointer size. This one is
1865 <i>really</i> dangerous!</dd>
1867 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
1868 <dd>Convert a constant, CST, to another TYPE. The size of CST and TYPE must be
1869 identical (same number of bits). The conversion is done as if the CST value
1870 was stored to memory and read back as TYPE. In other words, no bits change
1871 with this operator, just the type. This can be used for conversion of
1872 vector types to any other type, as long as they have the same bit width. For
1873 pointers it is only valid to cast to another pointer type. It is not valid
1874 to bitcast to or from an aggregate type.
1877 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1879 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1880 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1881 instruction, the index list may have zero or more indexes, which are required
1882 to make sense for the type of "CSTPTR".</dd>
1884 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1886 <dd>Perform the <a href="#i_select">select operation</a> on
1889 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
1890 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
1892 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
1893 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
1895 <dt><b><tt>vicmp COND ( VAL1, VAL2 )</tt></b></dt>
1896 <dd>Performs the <a href="#i_vicmp">vicmp operation</a> on constants.</dd>
1898 <dt><b><tt>vfcmp COND ( VAL1, VAL2 )</tt></b></dt>
1899 <dd>Performs the <a href="#i_vfcmp">vfcmp operation</a> on constants.</dd>
1901 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1903 <dd>Perform the <a href="#i_extractelement">extractelement
1904 operation</a> on constants.</dd>
1906 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1908 <dd>Perform the <a href="#i_insertelement">insertelement
1909 operation</a> on constants.</dd>
1912 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
1914 <dd>Perform the <a href="#i_shufflevector">shufflevector
1915 operation</a> on constants.</dd>
1917 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1919 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1920 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1921 binary</a> operations. The constraints on operands are the same as those for
1922 the corresponding instruction (e.g. no bitwise operations on floating point
1923 values are allowed).</dd>
1927 <!-- *********************************************************************** -->
1928 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1929 <!-- *********************************************************************** -->
1931 <!-- ======================================================================= -->
1932 <div class="doc_subsection">
1933 <a name="inlineasm">Inline Assembler Expressions</a>
1936 <div class="doc_text">
1939 LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
1940 Module-Level Inline Assembly</a>) through the use of a special value. This
1941 value represents the inline assembler as a string (containing the instructions
1942 to emit), a list of operand constraints (stored as a string), and a flag that
1943 indicates whether or not the inline asm expression has side effects. An example
1944 inline assembler expression is:
1947 <div class="doc_code">
1949 i32 (i32) asm "bswap $0", "=r,r"
1954 Inline assembler expressions may <b>only</b> be used as the callee operand of
1955 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
1958 <div class="doc_code">
1960 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
1965 Inline asms with side effects not visible in the constraint list must be marked
1966 as having side effects. This is done through the use of the
1967 '<tt>sideeffect</tt>' keyword, like so:
1970 <div class="doc_code">
1972 call void asm sideeffect "eieio", ""()
1976 <p>TODO: The format of the asm and constraints string still need to be
1977 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1978 need to be documented). This is probably best done by reference to another
1979 document that covers inline asm from a holistic perspective.
1984 <!-- *********************************************************************** -->
1985 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1986 <!-- *********************************************************************** -->
1988 <div class="doc_text">
1990 <p>The LLVM instruction set consists of several different
1991 classifications of instructions: <a href="#terminators">terminator
1992 instructions</a>, <a href="#binaryops">binary instructions</a>,
1993 <a href="#bitwiseops">bitwise binary instructions</a>, <a
1994 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1995 instructions</a>.</p>
1999 <!-- ======================================================================= -->
2000 <div class="doc_subsection"> <a name="terminators">Terminator
2001 Instructions</a> </div>
2003 <div class="doc_text">
2005 <p>As mentioned <a href="#functionstructure">previously</a>, every
2006 basic block in a program ends with a "Terminator" instruction, which
2007 indicates which block should be executed after the current block is
2008 finished. These terminator instructions typically yield a '<tt>void</tt>'
2009 value: they produce control flow, not values (the one exception being
2010 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
2011 <p>There are six different terminator instructions: the '<a
2012 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
2013 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
2014 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
2015 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
2016 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
2020 <!-- _______________________________________________________________________ -->
2021 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
2022 Instruction</a> </div>
2023 <div class="doc_text">
2026 ret <type> <value> <i>; Return a value from a non-void function</i>
2027 ret void <i>; Return from void function</i>
2032 <p>The '<tt>ret</tt>' instruction is used to return control flow (and
2033 optionally a value) from a function back to the caller.</p>
2034 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
2035 returns a value and then causes control flow, and one that just causes
2036 control flow to occur.</p>
2040 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument,
2041 the return value. The type of the return value must be a
2042 '<a href="#t_firstclass">first class</a>' type.</p>
2044 <p>A function is not <a href="#wellformed">well formed</a> if
2045 it it has a non-void return type and contains a '<tt>ret</tt>'
2046 instruction with no return value or a return value with a type that
2047 does not match its type, or if it has a void return type and contains
2048 a '<tt>ret</tt>' instruction with a return value.</p>
2052 <p>When the '<tt>ret</tt>' instruction is executed, control flow
2053 returns back to the calling function's context. If the caller is a "<a
2054 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
2055 the instruction after the call. If the caller was an "<a
2056 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
2057 at the beginning of the "normal" destination block. If the instruction
2058 returns a value, that value shall set the call or invoke instruction's
2064 ret i32 5 <i>; Return an integer value of 5</i>
2065 ret void <i>; Return from a void function</i>
2066 ret { i32, i8 } { i32 4, i8 2 } <i>; Return an aggregate of values 4 and 2</i>
2069 <p>Note that the code generator does not yet fully support large
2070 return values. The specific sizes that are currently supported are
2071 dependent on the target. For integers, on 32-bit targets the limit
2072 is often 64 bits, and on 64-bit targets the limit is often 128 bits.
2073 For aggregate types, the current limits are dependent on the element
2074 types; for example targets are often limited to 2 total integer
2075 elements and 2 total floating-point elements.</p>
2078 <!-- _______________________________________________________________________ -->
2079 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
2080 <div class="doc_text">
2082 <pre> br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
2085 <p>The '<tt>br</tt>' instruction is used to cause control flow to
2086 transfer to a different basic block in the current function. There are
2087 two forms of this instruction, corresponding to a conditional branch
2088 and an unconditional branch.</p>
2090 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
2091 single '<tt>i1</tt>' value and two '<tt>label</tt>' values. The
2092 unconditional form of the '<tt>br</tt>' instruction takes a single
2093 '<tt>label</tt>' value as a target.</p>
2095 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
2096 argument is evaluated. If the value is <tt>true</tt>, control flows
2097 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
2098 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
2100 <pre>Test:<br> %cond = <a href="#i_icmp">icmp</a> eq, i32 %a, %b<br> br i1 %cond, label %IfEqual, label %IfUnequal<br>IfEqual:<br> <a
2101 href="#i_ret">ret</a> i32 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> i32 0<br></pre>
2103 <!-- _______________________________________________________________________ -->
2104 <div class="doc_subsubsection">
2105 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
2108 <div class="doc_text">
2112 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
2117 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
2118 several different places. It is a generalization of the '<tt>br</tt>'
2119 instruction, allowing a branch to occur to one of many possible
2125 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
2126 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
2127 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
2128 table is not allowed to contain duplicate constant entries.</p>
2132 <p>The <tt>switch</tt> instruction specifies a table of values and
2133 destinations. When the '<tt>switch</tt>' instruction is executed, this
2134 table is searched for the given value. If the value is found, control flow is
2135 transfered to the corresponding destination; otherwise, control flow is
2136 transfered to the default destination.</p>
2138 <h5>Implementation:</h5>
2140 <p>Depending on properties of the target machine and the particular
2141 <tt>switch</tt> instruction, this instruction may be code generated in different
2142 ways. For example, it could be generated as a series of chained conditional
2143 branches or with a lookup table.</p>
2148 <i>; Emulate a conditional br instruction</i>
2149 %Val = <a href="#i_zext">zext</a> i1 %value to i32
2150 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
2152 <i>; Emulate an unconditional br instruction</i>
2153 switch i32 0, label %dest [ ]
2155 <i>; Implement a jump table:</i>
2156 switch i32 %val, label %otherwise [ i32 0, label %onzero
2158 i32 2, label %ontwo ]
2162 <!-- _______________________________________________________________________ -->
2163 <div class="doc_subsubsection">
2164 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
2167 <div class="doc_text">
2172 <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>]
2173 to label <normal label> unwind label <exception label>
2178 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
2179 function, with the possibility of control flow transfer to either the
2180 '<tt>normal</tt>' label or the
2181 '<tt>exception</tt>' label. If the callee function returns with the
2182 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
2183 "normal" label. If the callee (or any indirect callees) returns with the "<a
2184 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
2185 continued at the dynamically nearest "exception" label.</p>
2189 <p>This instruction requires several arguments:</p>
2193 The optional "cconv" marker indicates which <a href="#callingconv">calling
2194 convention</a> the call should use. If none is specified, the call defaults
2195 to using C calling conventions.
2198 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
2199 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>',
2200 and '<tt>inreg</tt>' attributes are valid here.</li>
2202 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
2203 function value being invoked. In most cases, this is a direct function
2204 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
2205 an arbitrary pointer to function value.
2208 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
2209 function to be invoked. </li>
2211 <li>'<tt>function args</tt>': argument list whose types match the function
2212 signature argument types. If the function signature indicates the function
2213 accepts a variable number of arguments, the extra arguments can be
2216 <li>'<tt>normal label</tt>': the label reached when the called function
2217 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
2219 <li>'<tt>exception label</tt>': the label reached when a callee returns with
2220 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
2222 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
2223 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
2224 '<tt>readnone</tt>' attributes are valid here.</li>
2229 <p>This instruction is designed to operate as a standard '<tt><a
2230 href="#i_call">call</a></tt>' instruction in most regards. The primary
2231 difference is that it establishes an association with a label, which is used by
2232 the runtime library to unwind the stack.</p>
2234 <p>This instruction is used in languages with destructors to ensure that proper
2235 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
2236 exception. Additionally, this is important for implementation of
2237 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
2241 %retval = invoke i32 @Test(i32 15) to label %Continue
2242 unwind label %TestCleanup <i>; {i32}:retval set</i>
2243 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
2244 unwind label %TestCleanup <i>; {i32}:retval set</i>
2249 <!-- _______________________________________________________________________ -->
2251 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
2252 Instruction</a> </div>
2254 <div class="doc_text">
2263 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
2264 at the first callee in the dynamic call stack which used an <a
2265 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
2266 primarily used to implement exception handling.</p>
2270 <p>The '<tt>unwind</tt>' instruction causes execution of the current function to
2271 immediately halt. The dynamic call stack is then searched for the first <a
2272 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
2273 execution continues at the "exceptional" destination block specified by the
2274 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
2275 dynamic call chain, undefined behavior results.</p>
2278 <!-- _______________________________________________________________________ -->
2280 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
2281 Instruction</a> </div>
2283 <div class="doc_text">
2292 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
2293 instruction is used to inform the optimizer that a particular portion of the
2294 code is not reachable. This can be used to indicate that the code after a
2295 no-return function cannot be reached, and other facts.</p>
2299 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
2304 <!-- ======================================================================= -->
2305 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
2306 <div class="doc_text">
2307 <p>Binary operators are used to do most of the computation in a
2308 program. They require two operands of the same type, execute an operation on them, and
2309 produce a single value. The operands might represent
2310 multiple data, as is the case with the <a href="#t_vector">vector</a> data type.
2311 The result value has the same type as its operands.</p>
2312 <p>There are several different binary operators:</p>
2314 <!-- _______________________________________________________________________ -->
2315 <div class="doc_subsubsection">
2316 <a name="i_add">'<tt>add</tt>' Instruction</a>
2319 <div class="doc_text">
2324 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2329 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
2333 <p>The two arguments to the '<tt>add</tt>' instruction must be <a
2334 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>, or
2335 <a href="#t_vector">vector</a> values. Both arguments must have identical
2340 <p>The value produced is the integer or floating point sum of the two
2343 <p>If an integer sum has unsigned overflow, the result returned is the
2344 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2347 <p>Because LLVM integers use a two's complement representation, this
2348 instruction is appropriate for both signed and unsigned integers.</p>
2353 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
2356 <!-- _______________________________________________________________________ -->
2357 <div class="doc_subsubsection">
2358 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
2361 <div class="doc_text">
2366 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2371 <p>The '<tt>sub</tt>' instruction returns the difference of its two
2374 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
2375 '<tt>neg</tt>' instruction present in most other intermediate
2376 representations.</p>
2380 <p>The two arguments to the '<tt>sub</tt>' instruction must be <a
2381 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>,
2382 or <a href="#t_vector">vector</a> values. Both arguments must have identical
2387 <p>The value produced is the integer or floating point difference of
2388 the two operands.</p>
2390 <p>If an integer difference has unsigned overflow, the result returned is the
2391 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2394 <p>Because LLVM integers use a two's complement representation, this
2395 instruction is appropriate for both signed and unsigned integers.</p>
2399 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
2400 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
2404 <!-- _______________________________________________________________________ -->
2405 <div class="doc_subsubsection">
2406 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
2409 <div class="doc_text">
2412 <pre> <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2415 <p>The '<tt>mul</tt>' instruction returns the product of its two
2420 <p>The two arguments to the '<tt>mul</tt>' instruction must be <a
2421 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>,
2422 or <a href="#t_vector">vector</a> values. Both arguments must have identical
2427 <p>The value produced is the integer or floating point product of the
2430 <p>If the result of an integer multiplication has unsigned overflow,
2431 the result returned is the mathematical result modulo
2432 2<sup>n</sup>, where n is the bit width of the result.</p>
2433 <p>Because LLVM integers use a two's complement representation, and the
2434 result is the same width as the operands, this instruction returns the
2435 correct result for both signed and unsigned integers. If a full product
2436 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands
2437 should be sign-extended or zero-extended as appropriate to the
2438 width of the full product.</p>
2440 <pre> <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
2444 <!-- _______________________________________________________________________ -->
2445 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
2447 <div class="doc_text">
2449 <pre> <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2452 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two
2457 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
2458 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2459 values. Both arguments must have identical types.</p>
2463 <p>The value produced is the unsigned integer quotient of the two operands.</p>
2464 <p>Note that unsigned integer division and signed integer division are distinct
2465 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
2466 <p>Division by zero leads to undefined behavior.</p>
2468 <pre> <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2471 <!-- _______________________________________________________________________ -->
2472 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
2474 <div class="doc_text">
2477 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2482 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
2487 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
2488 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2489 values. Both arguments must have identical types.</p>
2492 <p>The value produced is the signed integer quotient of the two operands rounded towards zero.</p>
2493 <p>Note that signed integer division and unsigned integer division are distinct
2494 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
2495 <p>Division by zero leads to undefined behavior. Overflow also leads to
2496 undefined behavior; this is a rare case, but can occur, for example,
2497 by doing a 32-bit division of -2147483648 by -1.</p>
2499 <pre> <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2502 <!-- _______________________________________________________________________ -->
2503 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
2504 Instruction</a> </div>
2505 <div class="doc_text">
2508 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2512 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
2517 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
2518 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a>
2519 of floating point values. Both arguments must have identical types.</p>
2523 <p>The value produced is the floating point quotient of the two operands.</p>
2528 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
2532 <!-- _______________________________________________________________________ -->
2533 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
2535 <div class="doc_text">
2537 <pre> <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2540 <p>The '<tt>urem</tt>' instruction returns the remainder from the
2541 unsigned division of its two arguments.</p>
2543 <p>The two arguments to the '<tt>urem</tt>' instruction must be
2544 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2545 values. Both arguments must have identical types.</p>
2547 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
2548 This instruction always performs an unsigned division to get the remainder.</p>
2549 <p>Note that unsigned integer remainder and signed integer remainder are
2550 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
2551 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
2553 <pre> <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2557 <!-- _______________________________________________________________________ -->
2558 <div class="doc_subsubsection">
2559 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
2562 <div class="doc_text">
2567 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2572 <p>The '<tt>srem</tt>' instruction returns the remainder from the
2573 signed division of its two operands. This instruction can also take
2574 <a href="#t_vector">vector</a> versions of the values in which case
2575 the elements must be integers.</p>
2579 <p>The two arguments to the '<tt>srem</tt>' instruction must be
2580 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2581 values. Both arguments must have identical types.</p>
2585 <p>This instruction returns the <i>remainder</i> of a division (where the result
2586 has the same sign as the dividend, <tt>op1</tt>), not the <i>modulo</i>
2587 operator (where the result has the same sign as the divisor, <tt>op2</tt>) of
2588 a value. For more information about the difference, see <a
2589 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
2590 Math Forum</a>. For a table of how this is implemented in various languages,
2591 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
2592 Wikipedia: modulo operation</a>.</p>
2593 <p>Note that signed integer remainder and unsigned integer remainder are
2594 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
2595 <p>Taking the remainder of a division by zero leads to undefined behavior.
2596 Overflow also leads to undefined behavior; this is a rare case, but can occur,
2597 for example, by taking the remainder of a 32-bit division of -2147483648 by -1.
2598 (The remainder doesn't actually overflow, but this rule lets srem be
2599 implemented using instructions that return both the result of the division
2600 and the remainder.)</p>
2602 <pre> <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2606 <!-- _______________________________________________________________________ -->
2607 <div class="doc_subsubsection">
2608 <a name="i_frem">'<tt>frem</tt>' Instruction</a> </div>
2610 <div class="doc_text">
2613 <pre> <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2616 <p>The '<tt>frem</tt>' instruction returns the remainder from the
2617 division of its two operands.</p>
2619 <p>The two arguments to the '<tt>frem</tt>' instruction must be
2620 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a>
2621 of floating point values. Both arguments must have identical types.</p>
2625 <p>This instruction returns the <i>remainder</i> of a division.
2626 The remainder has the same sign as the dividend.</p>
2631 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
2635 <!-- ======================================================================= -->
2636 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
2637 Operations</a> </div>
2638 <div class="doc_text">
2639 <p>Bitwise binary operators are used to do various forms of
2640 bit-twiddling in a program. They are generally very efficient
2641 instructions and can commonly be strength reduced from other
2642 instructions. They require two operands of the same type, execute an operation on them,
2643 and produce a single value. The resulting value is the same type as its operands.</p>
2646 <!-- _______________________________________________________________________ -->
2647 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
2648 Instruction</a> </div>
2649 <div class="doc_text">
2651 <pre> <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2656 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2657 the left a specified number of bits.</p>
2661 <p>Both arguments to the '<tt>shl</tt>' instruction must be the same <a
2662 href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2663 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
2667 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod 2<sup>n</sup>,
2668 where n is the width of the result. If <tt>op2</tt> is (statically or dynamically) negative or
2669 equal to or larger than the number of bits in <tt>op1</tt>, the result is undefined.
2670 If the arguments are vectors, each vector element of <tt>op1</tt> is shifted by the
2671 corresponding shift amount in <tt>op2</tt>.</p>
2673 <h5>Example:</h5><pre>
2674 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
2675 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
2676 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
2677 <result> = shl i32 1, 32 <i>; undefined</i>
2678 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
2681 <!-- _______________________________________________________________________ -->
2682 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
2683 Instruction</a> </div>
2684 <div class="doc_text">
2686 <pre> <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2690 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
2691 operand shifted to the right a specified number of bits with zero fill.</p>
2694 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
2695 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2696 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
2700 <p>This instruction always performs a logical shift right operation. The most
2701 significant bits of the result will be filled with zero bits after the
2702 shift. If <tt>op2</tt> is (statically or dynamically) equal to or larger than
2703 the number of bits in <tt>op1</tt>, the result is undefined. If the arguments are
2704 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding shift
2705 amount in <tt>op2</tt>.</p>
2709 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
2710 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
2711 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
2712 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
2713 <result> = lshr i32 1, 32 <i>; undefined</i>
2714 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
2718 <!-- _______________________________________________________________________ -->
2719 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
2720 Instruction</a> </div>
2721 <div class="doc_text">
2724 <pre> <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2728 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
2729 operand shifted to the right a specified number of bits with sign extension.</p>
2732 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
2733 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2734 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
2737 <p>This instruction always performs an arithmetic shift right operation,
2738 The most significant bits of the result will be filled with the sign bit
2739 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
2740 larger than the number of bits in <tt>op1</tt>, the result is undefined. If the
2741 arguments are vectors, each vector element of <tt>op1</tt> is shifted by the
2742 corresponding shift amount in <tt>op2</tt>.</p>
2746 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
2747 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
2748 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
2749 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
2750 <result> = ashr i32 1, 32 <i>; undefined</i>
2751 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
2755 <!-- _______________________________________________________________________ -->
2756 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
2757 Instruction</a> </div>
2759 <div class="doc_text">
2764 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2769 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
2770 its two operands.</p>
2774 <p>The two arguments to the '<tt>and</tt>' instruction must be
2775 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2776 values. Both arguments must have identical types.</p>
2779 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
2782 <table border="1" cellspacing="0" cellpadding="4">
2814 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
2815 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
2816 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
2819 <!-- _______________________________________________________________________ -->
2820 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
2821 <div class="doc_text">
2823 <pre> <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2826 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
2827 or of its two operands.</p>
2830 <p>The two arguments to the '<tt>or</tt>' instruction must be
2831 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2832 values. Both arguments must have identical types.</p>
2834 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
2837 <table border="1" cellspacing="0" cellpadding="4">
2868 <pre> <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
2869 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
2870 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
2873 <!-- _______________________________________________________________________ -->
2874 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
2875 Instruction</a> </div>
2876 <div class="doc_text">
2878 <pre> <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2881 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
2882 or of its two operands. The <tt>xor</tt> is used to implement the
2883 "one's complement" operation, which is the "~" operator in C.</p>
2885 <p>The two arguments to the '<tt>xor</tt>' instruction must be
2886 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2887 values. Both arguments must have identical types.</p>
2891 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
2894 <table border="1" cellspacing="0" cellpadding="4">
2926 <pre> <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
2927 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
2928 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
2929 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
2933 <!-- ======================================================================= -->
2934 <div class="doc_subsection">
2935 <a name="vectorops">Vector Operations</a>
2938 <div class="doc_text">
2940 <p>LLVM supports several instructions to represent vector operations in a
2941 target-independent manner. These instructions cover the element-access and
2942 vector-specific operations needed to process vectors effectively. While LLVM
2943 does directly support these vector operations, many sophisticated algorithms
2944 will want to use target-specific intrinsics to take full advantage of a specific
2949 <!-- _______________________________________________________________________ -->
2950 <div class="doc_subsubsection">
2951 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2954 <div class="doc_text">
2959 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
2965 The '<tt>extractelement</tt>' instruction extracts a single scalar
2966 element from a vector at a specified index.
2973 The first operand of an '<tt>extractelement</tt>' instruction is a
2974 value of <a href="#t_vector">vector</a> type. The second operand is
2975 an index indicating the position from which to extract the element.
2976 The index may be a variable.</p>
2981 The result is a scalar of the same type as the element type of
2982 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2983 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2984 results are undefined.
2990 %result = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
2995 <!-- _______________________________________________________________________ -->
2996 <div class="doc_subsubsection">
2997 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
3000 <div class="doc_text">
3005 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
3011 The '<tt>insertelement</tt>' instruction inserts a scalar
3012 element into a vector at a specified index.
3019 The first operand of an '<tt>insertelement</tt>' instruction is a
3020 value of <a href="#t_vector">vector</a> type. The second operand is a
3021 scalar value whose type must equal the element type of the first
3022 operand. The third operand is an index indicating the position at
3023 which to insert the value. The index may be a variable.</p>
3028 The result is a vector of the same type as <tt>val</tt>. Its
3029 element values are those of <tt>val</tt> except at position
3030 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
3031 exceeds the length of <tt>val</tt>, the results are undefined.
3037 %result = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
3041 <!-- _______________________________________________________________________ -->
3042 <div class="doc_subsubsection">
3043 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
3046 <div class="doc_text">
3051 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
3057 The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
3058 from two input vectors, returning a vector with the same element type as
3059 the input and length that is the same as the shuffle mask.
3065 The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
3066 with types that match each other. The third argument is a shuffle mask whose
3067 element type is always 'i32'. The result of the instruction is a vector whose
3068 length is the same as the shuffle mask and whose element type is the same as
3069 the element type of the first two operands.
3073 The shuffle mask operand is required to be a constant vector with either
3074 constant integer or undef values.
3080 The elements of the two input vectors are numbered from left to right across
3081 both of the vectors. The shuffle mask operand specifies, for each element of
3082 the result vector, which element of the two input vectors the result element
3083 gets. The element selector may be undef (meaning "don't care") and the second
3084 operand may be undef if performing a shuffle from only one vector.
3090 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
3091 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
3092 %result = shufflevector <4 x i32> %v1, <4 x i32> undef,
3093 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
3094 %result = shufflevector <8 x i32> %v1, <8 x i32> undef,
3095 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i>
3096 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
3097 <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>
3102 <!-- ======================================================================= -->
3103 <div class="doc_subsection">
3104 <a name="aggregateops">Aggregate Operations</a>
3107 <div class="doc_text">
3109 <p>LLVM supports several instructions for working with aggregate values.
3114 <!-- _______________________________________________________________________ -->
3115 <div class="doc_subsubsection">
3116 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
3119 <div class="doc_text">
3124 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
3130 The '<tt>extractvalue</tt>' instruction extracts the value of a struct field
3131 or array element from an aggregate value.
3138 The first operand of an '<tt>extractvalue</tt>' instruction is a
3139 value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a>
3140 type. The operands are constant indices to specify which value to extract
3141 in a similar manner as indices in a
3142 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
3148 The result is the value at the position in the aggregate specified by
3155 %result = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
3160 <!-- _______________________________________________________________________ -->
3161 <div class="doc_subsubsection">
3162 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
3165 <div class="doc_text">
3170 <result> = insertvalue <aggregate type> <val>, <ty> <val>, <idx> <i>; yields <n x <ty>></i>
3176 The '<tt>insertvalue</tt>' instruction inserts a value
3177 into a struct field or array element in an aggregate.
3184 The first operand of an '<tt>insertvalue</tt>' instruction is a
3185 value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a> type.
3186 The second operand is a first-class value to insert.
3187 The following operands are constant indices
3188 indicating the position at which to insert the value in a similar manner as
3190 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
3191 The value to insert must have the same type as the value identified
3198 The result is an aggregate of the same type as <tt>val</tt>. Its
3199 value is that of <tt>val</tt> except that the value at the position
3200 specified by the indices is that of <tt>elt</tt>.
3206 %result = insertvalue {i32, float} %agg, i32 1, 0 <i>; yields {i32, float}</i>
3211 <!-- ======================================================================= -->
3212 <div class="doc_subsection">
3213 <a name="memoryops">Memory Access and Addressing Operations</a>
3216 <div class="doc_text">
3218 <p>A key design point of an SSA-based representation is how it
3219 represents memory. In LLVM, no memory locations are in SSA form, which
3220 makes things very simple. This section describes how to read, write,
3221 allocate, and free memory in LLVM.</p>
3225 <!-- _______________________________________________________________________ -->
3226 <div class="doc_subsubsection">
3227 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
3230 <div class="doc_text">
3235 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3240 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
3241 heap and returns a pointer to it. The object is always allocated in the generic
3242 address space (address space zero).</p>
3246 <p>The '<tt>malloc</tt>' instruction allocates
3247 <tt>sizeof(<type>)*NumElements</tt>
3248 bytes of memory from the operating system and returns a pointer of the
3249 appropriate type to the program. If "NumElements" is specified, it is the
3250 number of elements allocated, otherwise "NumElements" is defaulted to be one.
3251 If a constant alignment is specified, the value result of the allocation is guaranteed to
3252 be aligned to at least that boundary. If not specified, or if zero, the target can
3253 choose to align the allocation on any convenient boundary.</p>
3255 <p>'<tt>type</tt>' must be a sized type.</p>
3259 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
3260 a pointer is returned. The result of a zero byte allocation is undefined. The
3261 result is null if there is insufficient memory available.</p>
3266 %array = malloc [4 x i8] <i>; yields {[%4 x i8]*}:array</i>
3268 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
3269 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
3270 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
3271 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
3272 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
3275 <p>Note that the code generator does not yet respect the
3276 alignment value.</p>
3280 <!-- _______________________________________________________________________ -->
3281 <div class="doc_subsubsection">
3282 <a name="i_free">'<tt>free</tt>' Instruction</a>
3285 <div class="doc_text">
3290 free <type> <value> <i>; yields {void}</i>
3295 <p>The '<tt>free</tt>' instruction returns memory back to the unused
3296 memory heap to be reallocated in the future.</p>
3300 <p>'<tt>value</tt>' shall be a pointer value that points to a value
3301 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
3306 <p>Access to the memory pointed to by the pointer is no longer defined
3307 after this instruction executes. If the pointer is null, the operation
3313 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
3314 free [4 x i8]* %array
3318 <!-- _______________________________________________________________________ -->
3319 <div class="doc_subsubsection">
3320 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
3323 <div class="doc_text">
3328 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3333 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
3334 currently executing function, to be automatically released when this function
3335 returns to its caller. The object is always allocated in the generic address
3336 space (address space zero).</p>
3340 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
3341 bytes of memory on the runtime stack, returning a pointer of the
3342 appropriate type to the program. If "NumElements" is specified, it is the
3343 number of elements allocated, otherwise "NumElements" is defaulted to be one.
3344 If a constant alignment is specified, the value result of the allocation is guaranteed
3345 to be aligned to at least that boundary. If not specified, or if zero, the target
3346 can choose to align the allocation on any convenient boundary.</p>
3348 <p>'<tt>type</tt>' may be any sized type.</p>
3352 <p>Memory is allocated; a pointer is returned. The operation is undefiend if
3353 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
3354 memory is automatically released when the function returns. The '<tt>alloca</tt>'
3355 instruction is commonly used to represent automatic variables that must
3356 have an address available. When the function returns (either with the <tt><a
3357 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
3358 instructions), the memory is reclaimed. Allocating zero bytes
3359 is legal, but the result is undefined.</p>
3364 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
3365 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
3366 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
3367 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
3371 <!-- _______________________________________________________________________ -->
3372 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
3373 Instruction</a> </div>
3374 <div class="doc_text">
3376 <pre> <result> = load <ty>* <pointer>[, align <alignment>]<br> <result> = volatile load <ty>* <pointer>[, align <alignment>]<br></pre>
3378 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
3380 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
3381 address from which to load. The pointer must point to a <a
3382 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
3383 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
3384 the number or order of execution of this <tt>load</tt> with other
3385 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
3388 The optional constant "align" argument specifies the alignment of the operation
3389 (that is, the alignment of the memory address). A value of 0 or an
3390 omitted "align" argument means that the operation has the preferential
3391 alignment for the target. It is the responsibility of the code emitter
3392 to ensure that the alignment information is correct. Overestimating
3393 the alignment results in an undefined behavior. Underestimating the
3394 alignment may produce less efficient code. An alignment of 1 is always
3398 <p>The location of memory pointed to is loaded.</p>
3400 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3402 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
3403 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
3406 <!-- _______________________________________________________________________ -->
3407 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
3408 Instruction</a> </div>
3409 <div class="doc_text">
3411 <pre> store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3412 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3415 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
3417 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
3418 to store and an address at which to store it. The type of the '<tt><pointer></tt>'
3419 operand must be a pointer to the <a href="#t_firstclass">first class</a> type
3420 of the '<tt><value></tt>'
3421 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
3422 optimizer is not allowed to modify the number or order of execution of
3423 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
3424 href="#i_store">store</a></tt> instructions.</p>
3426 The optional constant "align" argument specifies the alignment of the operation
3427 (that is, the alignment of the memory address). A value of 0 or an
3428 omitted "align" argument means that the operation has the preferential
3429 alignment for the target. It is the responsibility of the code emitter
3430 to ensure that the alignment information is correct. Overestimating
3431 the alignment results in an undefined behavior. Underestimating the
3432 alignment may produce less efficient code. An alignment of 1 is always
3436 <p>The contents of memory are updated to contain '<tt><value></tt>'
3437 at the location specified by the '<tt><pointer></tt>' operand.</p>
3439 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3440 store i32 3, i32* %ptr <i>; yields {void}</i>
3441 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
3445 <!-- _______________________________________________________________________ -->
3446 <div class="doc_subsubsection">
3447 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
3450 <div class="doc_text">
3453 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
3459 The '<tt>getelementptr</tt>' instruction is used to get the address of a
3460 subelement of an aggregate data structure. It performs address calculation only
3461 and does not access memory.</p>
3465 <p>The first argument is always a pointer, and forms the basis of the
3466 calculation. The remaining arguments are indices, that indicate which of the
3467 elements of the aggregate object are indexed. The interpretation of each index
3468 is dependent on the type being indexed into. The first index always indexes the
3469 pointer value given as the first argument, the second index indexes a value of
3470 the type pointed to (not necessarily the value directly pointed to, since the
3471 first index can be non-zero), etc. The first type indexed into must be a pointer
3472 value, subsequent types can be arrays, vectors and structs. Note that subsequent
3473 types being indexed into can never be pointers, since that would require loading
3474 the pointer before continuing calculation.</p>
3476 <p>The type of each index argument depends on the type it is indexing into.
3477 When indexing into a (packed) structure, only <tt>i32</tt> integer
3478 <b>constants</b> are allowed. When indexing into an array, pointer or vector,
3479 only integers of 32 or 64 bits are allowed (also non-constants). 32-bit values
3480 will be sign extended to 64-bits if required.</p>
3482 <p>For example, let's consider a C code fragment and how it gets
3483 compiled to LLVM:</p>
3485 <div class="doc_code">
3498 int *foo(struct ST *s) {
3499 return &s[1].Z.B[5][13];
3504 <p>The LLVM code generated by the GCC frontend is:</p>
3506 <div class="doc_code">
3508 %RT = <a href="#namedtypes">type</a> { i8 , [10 x [20 x i32]], i8 }
3509 %ST = <a href="#namedtypes">type</a> { i32, double, %RT }
3511 define i32* %foo(%ST* %s) {
3513 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
3521 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
3522 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
3523 }</tt>' type, a structure. The second index indexes into the third element of
3524 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
3525 i8 }</tt>' type, another structure. The third index indexes into the second
3526 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
3527 array. The two dimensions of the array are subscripted into, yielding an
3528 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
3529 to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
3531 <p>Note that it is perfectly legal to index partially through a
3532 structure, returning a pointer to an inner element. Because of this,
3533 the LLVM code for the given testcase is equivalent to:</p>
3536 define i32* %foo(%ST* %s) {
3537 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
3538 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
3539 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
3540 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
3541 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
3546 <p>Note that it is undefined to access an array out of bounds: array and
3547 pointer indexes must always be within the defined bounds of the array type.
3548 The one exception for this rule is zero length arrays. These arrays are
3549 defined to be accessible as variable length arrays, which requires access
3550 beyond the zero'th element.</p>
3552 <p>The getelementptr instruction is often confusing. For some more insight
3553 into how it works, see <a href="GetElementPtr.html">the getelementptr
3559 <i>; yields [12 x i8]*:aptr</i>
3560 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
3561 <i>; yields i8*:vptr</i>
3562 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
3563 <i>; yields i8*:eptr</i>
3564 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
3568 <!-- ======================================================================= -->
3569 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
3571 <div class="doc_text">
3572 <p>The instructions in this category are the conversion instructions (casting)
3573 which all take a single operand and a type. They perform various bit conversions
3577 <!-- _______________________________________________________________________ -->
3578 <div class="doc_subsubsection">
3579 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
3581 <div class="doc_text">
3585 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
3590 The '<tt>trunc</tt>' instruction truncates its operand to the type <tt>ty2</tt>.
3595 The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
3596 be an <a href="#t_integer">integer</a> type, and a type that specifies the size
3597 and type of the result, which must be an <a href="#t_integer">integer</a>
3598 type. The bit size of <tt>value</tt> must be larger than the bit size of
3599 <tt>ty2</tt>. Equal sized types are not allowed.</p>
3603 The '<tt>trunc</tt>' instruction truncates the high order bits in <tt>value</tt>
3604 and converts the remaining bits to <tt>ty2</tt>. Since the source size must be
3605 larger than the destination size, <tt>trunc</tt> cannot be a <i>no-op cast</i>.
3606 It will always truncate bits.</p>
3610 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
3611 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
3612 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
3616 <!-- _______________________________________________________________________ -->
3617 <div class="doc_subsubsection">
3618 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
3620 <div class="doc_text">
3624 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
3628 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
3633 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
3634 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3635 also be of <a href="#t_integer">integer</a> type. The bit size of the
3636 <tt>value</tt> must be smaller than the bit size of the destination type,
3640 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
3641 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
3643 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
3647 %X = zext i32 257 to i64 <i>; yields i64:257</i>
3648 %Y = zext i1 true to i32 <i>; yields i32:1</i>
3652 <!-- _______________________________________________________________________ -->
3653 <div class="doc_subsubsection">
3654 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
3656 <div class="doc_text">
3660 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
3664 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
3668 The '<tt>sext</tt>' instruction takes a value to cast, which must be of
3669 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3670 also be of <a href="#t_integer">integer</a> type. The bit size of the
3671 <tt>value</tt> must be smaller than the bit size of the destination type,
3676 The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
3677 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size of
3678 the type <tt>ty2</tt>.</p>
3680 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
3684 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
3685 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
3689 <!-- _______________________________________________________________________ -->
3690 <div class="doc_subsubsection">
3691 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
3694 <div class="doc_text">
3699 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
3703 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
3708 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
3709 point</a> value to cast and a <a href="#t_floating">floating point</a> type to
3710 cast it to. The size of <tt>value</tt> must be larger than the size of
3711 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
3712 <i>no-op cast</i>.</p>
3715 <p> The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
3716 <a href="#t_floating">floating point</a> type to a smaller
3717 <a href="#t_floating">floating point</a> type. If the value cannot fit within
3718 the destination type, <tt>ty2</tt>, then the results are undefined.</p>
3722 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
3723 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
3727 <!-- _______________________________________________________________________ -->
3728 <div class="doc_subsubsection">
3729 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
3731 <div class="doc_text">
3735 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
3739 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
3740 floating point value.</p>
3743 <p>The '<tt>fpext</tt>' instruction takes a
3744 <a href="#t_floating">floating point</a> <tt>value</tt> to cast,
3745 and a <a href="#t_floating">floating point</a> type to cast it to. The source
3746 type must be smaller than the destination type.</p>
3749 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
3750 <a href="#t_floating">floating point</a> type to a larger
3751 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
3752 used to make a <i>no-op cast</i> because it always changes bits. Use
3753 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
3757 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
3758 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
3762 <!-- _______________________________________________________________________ -->
3763 <div class="doc_subsubsection">
3764 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
3766 <div class="doc_text">
3770 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
3774 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
3775 unsigned integer equivalent of type <tt>ty2</tt>.
3779 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
3780 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3781 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3782 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3783 vector integer type with the same number of elements as <tt>ty</tt></p>
3786 <p> The '<tt>fptoui</tt>' instruction converts its
3787 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3788 towards zero) unsigned integer value. If the value cannot fit in <tt>ty2</tt>,
3789 the results are undefined.</p>
3793 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
3794 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
3795 %X = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
3799 <!-- _______________________________________________________________________ -->
3800 <div class="doc_subsubsection">
3801 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
3803 <div class="doc_text">
3807 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
3811 <p>The '<tt>fptosi</tt>' instruction converts
3812 <a href="#t_floating">floating point</a> <tt>value</tt> to type <tt>ty2</tt>.
3816 <p> The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
3817 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3818 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3819 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3820 vector integer type with the same number of elements as <tt>ty</tt></p>
3823 <p>The '<tt>fptosi</tt>' instruction converts its
3824 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3825 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
3826 the results are undefined.</p>
3830 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
3831 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
3832 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
3836 <!-- _______________________________________________________________________ -->
3837 <div class="doc_subsubsection">
3838 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
3840 <div class="doc_text">
3844 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3848 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
3849 integer and converts that value to the <tt>ty2</tt> type.</p>
3852 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
3853 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3854 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3855 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3856 floating point type with the same number of elements as <tt>ty</tt></p>
3859 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
3860 integer quantity and converts it to the corresponding floating point value. If
3861 the value cannot fit in the floating point value, the results are undefined.</p>
3865 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
3866 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
3870 <!-- _______________________________________________________________________ -->
3871 <div class="doc_subsubsection">
3872 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
3874 <div class="doc_text">
3878 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3882 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed
3883 integer and converts that value to the <tt>ty2</tt> type.</p>
3886 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
3887 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3888 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3889 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3890 floating point type with the same number of elements as <tt>ty</tt></p>
3893 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed
3894 integer quantity and converts it to the corresponding floating point value. If
3895 the value cannot fit in the floating point value, the results are undefined.</p>
3899 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
3900 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
3904 <!-- _______________________________________________________________________ -->
3905 <div class="doc_subsubsection">
3906 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
3908 <div class="doc_text">
3912 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
3916 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
3917 the integer type <tt>ty2</tt>.</p>
3920 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
3921 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
3922 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.</p>
3925 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
3926 <tt>ty2</tt> by interpreting the pointer value as an integer and either
3927 truncating or zero extending that value to the size of the integer type. If
3928 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
3929 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
3930 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
3935 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
3936 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
3940 <!-- _______________________________________________________________________ -->
3941 <div class="doc_subsubsection">
3942 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
3944 <div class="doc_text">
3948 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
3952 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to
3953 a pointer type, <tt>ty2</tt>.</p>
3956 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
3957 value to cast, and a type to cast it to, which must be a
3958 <a href="#t_pointer">pointer</a> type.</p>
3961 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
3962 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
3963 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
3964 size of a pointer then a truncation is done. If <tt>value</tt> is smaller than
3965 the size of a pointer then a zero extension is done. If they are the same size,
3966 nothing is done (<i>no-op cast</i>).</p>
3970 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
3971 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
3972 %Y = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
3976 <!-- _______________________________________________________________________ -->
3977 <div class="doc_subsubsection">
3978 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
3980 <div class="doc_text">
3984 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
3989 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3990 <tt>ty2</tt> without changing any bits.</p>
3994 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be
3995 a non-aggregate first class value, and a type to cast it to, which must also be
3996 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes of
3998 and the destination type, <tt>ty2</tt>, must be identical. If the source
3999 type is a pointer, the destination type must also be a pointer. This
4000 instruction supports bitwise conversion of vectors to integers and to vectors
4001 of other types (as long as they have the same size).</p>
4004 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
4005 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
4006 this conversion. The conversion is done as if the <tt>value</tt> had been
4007 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only be
4008 converted to other pointer types with this instruction. To convert pointers to
4009 other types, use the <a href="#i_inttoptr">inttoptr</a> or
4010 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
4014 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
4015 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
4016 %Z = bitcast <2 x int> %V to i64; <i>; yields i64: %V</i>
4020 <!-- ======================================================================= -->
4021 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
4022 <div class="doc_text">
4023 <p>The instructions in this category are the "miscellaneous"
4024 instructions, which defy better classification.</p>
4027 <!-- _______________________________________________________________________ -->
4028 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
4030 <div class="doc_text">
4032 <pre> <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4035 <p>The '<tt>icmp</tt>' instruction returns a boolean value or
4036 a vector of boolean values based on comparison
4037 of its two integer, integer vector, or pointer operands.</p>
4039 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
4040 the condition code indicating the kind of comparison to perform. It is not
4041 a value, just a keyword. The possible condition code are:
4044 <li><tt>eq</tt>: equal</li>
4045 <li><tt>ne</tt>: not equal </li>
4046 <li><tt>ugt</tt>: unsigned greater than</li>
4047 <li><tt>uge</tt>: unsigned greater or equal</li>
4048 <li><tt>ult</tt>: unsigned less than</li>
4049 <li><tt>ule</tt>: unsigned less or equal</li>
4050 <li><tt>sgt</tt>: signed greater than</li>
4051 <li><tt>sge</tt>: signed greater or equal</li>
4052 <li><tt>slt</tt>: signed less than</li>
4053 <li><tt>sle</tt>: signed less or equal</li>
4055 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
4056 <a href="#t_pointer">pointer</a>
4057 or integer <a href="#t_vector">vector</a> typed.
4058 They must also be identical types.</p>
4060 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to
4061 the condition code given as <tt>cond</tt>. The comparison performed always
4062 yields either an <a href="#t_primitive"><tt>i1</tt></a> or vector of <tt>i1</tt> result, as follows:
4065 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
4066 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
4068 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
4069 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.</li>
4070 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
4071 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4072 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
4073 <tt>true</tt> if <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4074 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
4075 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
4076 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
4077 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4078 <li><tt>sgt</tt>: interprets the operands as signed values and yields
4079 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4080 <li><tt>sge</tt>: interprets the operands as signed values and yields
4081 <tt>true</tt> if <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4082 <li><tt>slt</tt>: interprets the operands as signed values and yields
4083 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
4084 <li><tt>sle</tt>: interprets the operands as signed values and yields
4085 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4087 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
4088 values are compared as if they were integers.</p>
4089 <p>If the operands are integer vectors, then they are compared
4090 element by element. The result is an <tt>i1</tt> vector with
4091 the same number of elements as the values being compared.
4092 Otherwise, the result is an <tt>i1</tt>.
4096 <pre> <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
4097 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
4098 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
4099 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
4100 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
4101 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
4104 <p>Note that the code generator does not yet support vector types with
4105 the <tt>icmp</tt> instruction.</p>
4109 <!-- _______________________________________________________________________ -->
4110 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
4112 <div class="doc_text">
4114 <pre> <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4117 <p>The '<tt>fcmp</tt>' instruction returns a boolean value
4118 or vector of boolean values based on comparison
4119 of its operands.</p>
4121 If the operands are floating point scalars, then the result
4122 type is a boolean (<a href="#t_primitive"><tt>i1</tt></a>).
4124 <p>If the operands are floating point vectors, then the result type
4125 is a vector of boolean with the same number of elements as the
4126 operands being compared.</p>
4128 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
4129 the condition code indicating the kind of comparison to perform. It is not
4130 a value, just a keyword. The possible condition code are:</p>
4132 <li><tt>false</tt>: no comparison, always returns false</li>
4133 <li><tt>oeq</tt>: ordered and equal</li>
4134 <li><tt>ogt</tt>: ordered and greater than </li>
4135 <li><tt>oge</tt>: ordered and greater than or equal</li>
4136 <li><tt>olt</tt>: ordered and less than </li>
4137 <li><tt>ole</tt>: ordered and less than or equal</li>
4138 <li><tt>one</tt>: ordered and not equal</li>
4139 <li><tt>ord</tt>: ordered (no nans)</li>
4140 <li><tt>ueq</tt>: unordered or equal</li>
4141 <li><tt>ugt</tt>: unordered or greater than </li>
4142 <li><tt>uge</tt>: unordered or greater than or equal</li>
4143 <li><tt>ult</tt>: unordered or less than </li>
4144 <li><tt>ule</tt>: unordered or less than or equal</li>
4145 <li><tt>une</tt>: unordered or not equal</li>
4146 <li><tt>uno</tt>: unordered (either nans)</li>
4147 <li><tt>true</tt>: no comparison, always returns true</li>
4149 <p><i>Ordered</i> means that neither operand is a QNAN while
4150 <i>unordered</i> means that either operand may be a QNAN.</p>
4151 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be
4152 either a <a href="#t_floating">floating point</a> type
4153 or a <a href="#t_vector">vector</a> of floating point type.
4154 They must have identical types.</p>
4156 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4157 according to the condition code given as <tt>cond</tt>.
4158 If the operands are vectors, then the vectors are compared
4160 Each comparison performed
4161 always yields an <a href="#t_primitive">i1</a> result, as follows:</p>
4163 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
4164 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4165 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
4166 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4167 <tt>op1</tt> is greather than <tt>op2</tt>.</li>
4168 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4169 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4170 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4171 <tt>op1</tt> is less than <tt>op2</tt>.</li>
4172 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4173 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4174 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4175 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
4176 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
4177 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
4178 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
4179 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
4180 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4181 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
4182 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4183 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
4184 <tt>op1</tt> is less than <tt>op2</tt>.</li>
4185 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
4186 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4187 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
4188 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
4189 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
4190 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
4194 <pre> <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
4195 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
4196 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
4197 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
4200 <p>Note that the code generator does not yet support vector types with
4201 the <tt>fcmp</tt> instruction.</p>
4205 <!-- _______________________________________________________________________ -->
4206 <div class="doc_subsubsection">
4207 <a name="i_vicmp">'<tt>vicmp</tt>' Instruction</a>
4209 <div class="doc_text">
4211 <pre> <result> = vicmp <cond> <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4214 <p>The '<tt>vicmp</tt>' instruction returns an integer vector value based on
4215 element-wise comparison of its two integer vector operands.</p>
4217 <p>The '<tt>vicmp</tt>' instruction takes three operands. The first operand is
4218 the condition code indicating the kind of comparison to perform. It is not
4219 a value, just a keyword. The possible condition code are:</p>
4221 <li><tt>eq</tt>: equal</li>
4222 <li><tt>ne</tt>: not equal </li>
4223 <li><tt>ugt</tt>: unsigned greater than</li>
4224 <li><tt>uge</tt>: unsigned greater or equal</li>
4225 <li><tt>ult</tt>: unsigned less than</li>
4226 <li><tt>ule</tt>: unsigned less or equal</li>
4227 <li><tt>sgt</tt>: signed greater than</li>
4228 <li><tt>sge</tt>: signed greater or equal</li>
4229 <li><tt>slt</tt>: signed less than</li>
4230 <li><tt>sle</tt>: signed less or equal</li>
4232 <p>The remaining two arguments must be <a href="#t_vector">vector</a> or
4233 <a href="#t_integer">integer</a> typed. They must also be identical types.</p>
4235 <p>The '<tt>vicmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4236 according to the condition code given as <tt>cond</tt>. The comparison yields a
4237 <a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, of
4238 identical type as the values being compared. The most significant bit in each
4239 element is 1 if the element-wise comparison evaluates to true, and is 0
4240 otherwise. All other bits of the result are undefined. The condition codes
4241 are evaluated identically to the <a href="#i_icmp">'<tt>icmp</tt>'
4242 instruction</a>.</p>
4246 <result> = vicmp eq <2 x i32> < i32 4, i32 0>, < i32 5, i32 0> <i>; yields: result=<2 x i32> < i32 0, i32 -1 ></i>
4247 <result> = vicmp ult <2 x i8 > < i8 1, i8 2>, < i8 2, i8 2 > <i>; yields: result=<2 x i8> < i8 -1, i8 0 ></i>
4251 <!-- _______________________________________________________________________ -->
4252 <div class="doc_subsubsection">
4253 <a name="i_vfcmp">'<tt>vfcmp</tt>' Instruction</a>
4255 <div class="doc_text">
4257 <pre> <result> = vfcmp <cond> <ty> <op1>, <op2></pre>
4259 <p>The '<tt>vfcmp</tt>' instruction returns an integer vector value based on
4260 element-wise comparison of its two floating point vector operands. The output
4261 elements have the same width as the input elements.</p>
4263 <p>The '<tt>vfcmp</tt>' instruction takes three operands. The first operand is
4264 the condition code indicating the kind of comparison to perform. It is not
4265 a value, just a keyword. The possible condition code are:</p>
4267 <li><tt>false</tt>: no comparison, always returns false</li>
4268 <li><tt>oeq</tt>: ordered and equal</li>
4269 <li><tt>ogt</tt>: ordered and greater than </li>
4270 <li><tt>oge</tt>: ordered and greater than or equal</li>
4271 <li><tt>olt</tt>: ordered and less than </li>
4272 <li><tt>ole</tt>: ordered and less than or equal</li>
4273 <li><tt>one</tt>: ordered and not equal</li>
4274 <li><tt>ord</tt>: ordered (no nans)</li>
4275 <li><tt>ueq</tt>: unordered or equal</li>
4276 <li><tt>ugt</tt>: unordered or greater than </li>
4277 <li><tt>uge</tt>: unordered or greater than or equal</li>
4278 <li><tt>ult</tt>: unordered or less than </li>
4279 <li><tt>ule</tt>: unordered or less than or equal</li>
4280 <li><tt>une</tt>: unordered or not equal</li>
4281 <li><tt>uno</tt>: unordered (either nans)</li>
4282 <li><tt>true</tt>: no comparison, always returns true</li>
4284 <p>The remaining two arguments must be <a href="#t_vector">vector</a> of
4285 <a href="#t_floating">floating point</a> typed. They must also be identical
4288 <p>The '<tt>vfcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4289 according to the condition code given as <tt>cond</tt>. The comparison yields a
4290 <a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, with
4291 an identical number of elements as the values being compared, and each element
4292 having identical with to the width of the floating point elements. The most
4293 significant bit in each element is 1 if the element-wise comparison evaluates to
4294 true, and is 0 otherwise. All other bits of the result are undefined. The
4295 condition codes are evaluated identically to the
4296 <a href="#i_fcmp">'<tt>fcmp</tt>' instruction</a>.</p>
4300 <i>; yields: result=<2 x i32> < i32 0, i32 -1 ></i>
4301 <result> = vfcmp oeq <2 x float> < float 4, float 0 >, < float 5, float 0 >
4303 <i>; yields: result=<2 x i64> < i64 -1, i64 0 ></i>
4304 <result> = vfcmp ult <2 x double> < double 1, double 2 >, < double 2, double 2>
4308 <!-- _______________________________________________________________________ -->
4309 <div class="doc_subsubsection">
4310 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
4313 <div class="doc_text">
4317 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
4319 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
4320 the SSA graph representing the function.</p>
4323 <p>The type of the incoming values is specified with the first type
4324 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
4325 as arguments, with one pair for each predecessor basic block of the
4326 current block. Only values of <a href="#t_firstclass">first class</a>
4327 type may be used as the value arguments to the PHI node. Only labels
4328 may be used as the label arguments.</p>
4330 <p>There must be no non-phi instructions between the start of a basic
4331 block and the PHI instructions: i.e. PHI instructions must be first in
4336 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
4337 specified by the pair corresponding to the predecessor basic block that executed
4338 just prior to the current block.</p>
4342 Loop: ; Infinite loop that counts from 0 on up...
4343 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
4344 %nextindvar = add i32 %indvar, 1
4349 <!-- _______________________________________________________________________ -->
4350 <div class="doc_subsubsection">
4351 <a name="i_select">'<tt>select</tt>' Instruction</a>
4354 <div class="doc_text">
4359 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
4361 <i>selty</i> is either i1 or {<N x i1>}
4367 The '<tt>select</tt>' instruction is used to choose one value based on a
4368 condition, without branching.
4375 The '<tt>select</tt>' instruction requires an 'i1' value or
4376 a vector of 'i1' values indicating the
4377 condition, and two values of the same <a href="#t_firstclass">first class</a>
4378 type. If the val1/val2 are vectors and
4379 the condition is a scalar, then entire vectors are selected, not
4380 individual elements.
4386 If the condition is an i1 and it evaluates to 1, the instruction returns the first
4387 value argument; otherwise, it returns the second value argument.
4390 If the condition is a vector of i1, then the value arguments must
4391 be vectors of the same size, and the selection is done element
4398 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
4401 <p>Note that the code generator does not yet support conditions
4402 with vector type.</p>
4407 <!-- _______________________________________________________________________ -->
4408 <div class="doc_subsubsection">
4409 <a name="i_call">'<tt>call</tt>' Instruction</a>
4412 <div class="doc_text">
4416 <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>]
4421 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
4425 <p>This instruction requires several arguments:</p>
4429 <p>The optional "tail" marker indicates whether the callee function accesses
4430 any allocas or varargs in the caller. If the "tail" marker is present, the
4431 function call is eligible for tail call optimization. Note that calls may
4432 be marked "tail" even if they do not occur before a <a
4433 href="#i_ret"><tt>ret</tt></a> instruction.</p>
4436 <p>The optional "cconv" marker indicates which <a href="#callingconv">calling
4437 convention</a> the call should use. If none is specified, the call defaults
4438 to using C calling conventions.</p>
4442 <p>The optional <a href="#paramattrs">Parameter Attributes</a> list for
4443 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>',
4444 and '<tt>inreg</tt>' attributes are valid here.</p>
4448 <p>'<tt>ty</tt>': the type of the call instruction itself which is also
4449 the type of the return value. Functions that return no value are marked
4450 <tt><a href="#t_void">void</a></tt>.</p>
4453 <p>'<tt>fnty</tt>': shall be the signature of the pointer to function
4454 value being invoked. The argument types must match the types implied by
4455 this signature. This type can be omitted if the function is not varargs
4456 and if the function type does not return a pointer to a function.</p>
4459 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
4460 be invoked. In most cases, this is a direct function invocation, but
4461 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
4462 to function value.</p>
4465 <p>'<tt>function args</tt>': argument list whose types match the
4466 function signature argument types. All arguments must be of
4467 <a href="#t_firstclass">first class</a> type. If the function signature
4468 indicates the function accepts a variable number of arguments, the extra
4469 arguments can be specified.</p>
4472 <p>The optional <a href="#fnattrs">function attributes</a> list. Only
4473 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
4474 '<tt>readnone</tt>' attributes are valid here.</p>
4480 <p>The '<tt>call</tt>' instruction is used to cause control flow to
4481 transfer to a specified function, with its incoming arguments bound to
4482 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
4483 instruction in the called function, control flow continues with the
4484 instruction after the function call, and the return value of the
4485 function is bound to the result argument.</p>
4490 %retval = call i32 @test(i32 %argc)
4491 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42) <i>; yields i32</i>
4492 %X = tail call i32 @foo() <i>; yields i32</i>
4493 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
4494 call void %foo(i8 97 signext)
4496 %struct.A = type { i32, i8 }
4497 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
4498 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
4499 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
4500 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
4501 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
4506 <!-- _______________________________________________________________________ -->
4507 <div class="doc_subsubsection">
4508 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
4511 <div class="doc_text">
4516 <resultval> = va_arg <va_list*> <arglist>, <argty>
4521 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
4522 the "variable argument" area of a function call. It is used to implement the
4523 <tt>va_arg</tt> macro in C.</p>
4527 <p>This instruction takes a <tt>va_list*</tt> value and the type of
4528 the argument. It returns a value of the specified argument type and
4529 increments the <tt>va_list</tt> to point to the next argument. The
4530 actual type of <tt>va_list</tt> is target specific.</p>
4534 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
4535 type from the specified <tt>va_list</tt> and causes the
4536 <tt>va_list</tt> to point to the next argument. For more information,
4537 see the variable argument handling <a href="#int_varargs">Intrinsic
4540 <p>It is legal for this instruction to be called in a function which does not
4541 take a variable number of arguments, for example, the <tt>vfprintf</tt>
4544 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
4545 href="#intrinsics">intrinsic function</a> because it takes a type as an
4550 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
4552 <p>Note that the code generator does not yet fully support va_arg
4553 on many targets. Also, it does not currently support va_arg with
4554 aggregate types on any target.</p>
4558 <!-- *********************************************************************** -->
4559 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
4560 <!-- *********************************************************************** -->
4562 <div class="doc_text">
4564 <p>LLVM supports the notion of an "intrinsic function". These functions have
4565 well known names and semantics and are required to follow certain restrictions.
4566 Overall, these intrinsics represent an extension mechanism for the LLVM
4567 language that does not require changing all of the transformations in LLVM when
4568 adding to the language (or the bitcode reader/writer, the parser, etc...).</p>
4570 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
4571 prefix is reserved in LLVM for intrinsic names; thus, function names may not
4572 begin with this prefix. Intrinsic functions must always be external functions:
4573 you cannot define the body of intrinsic functions. Intrinsic functions may
4574 only be used in call or invoke instructions: it is illegal to take the address
4575 of an intrinsic function. Additionally, because intrinsic functions are part
4576 of the LLVM language, it is required if any are added that they be documented
4579 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
4580 a family of functions that perform the same operation but on different data
4581 types. Because LLVM can represent over 8 million different integer types,
4582 overloading is used commonly to allow an intrinsic function to operate on any
4583 integer type. One or more of the argument types or the result type can be
4584 overloaded to accept any integer type. Argument types may also be defined as
4585 exactly matching a previous argument's type or the result type. This allows an
4586 intrinsic function which accepts multiple arguments, but needs all of them to
4587 be of the same type, to only be overloaded with respect to a single argument or
4590 <p>Overloaded intrinsics will have the names of its overloaded argument types
4591 encoded into its function name, each preceded by a period. Only those types
4592 which are overloaded result in a name suffix. Arguments whose type is matched
4593 against another type do not. For example, the <tt>llvm.ctpop</tt> function can
4594 take an integer of any width and returns an integer of exactly the same integer
4595 width. This leads to a family of functions such as
4596 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29 %val)</tt>.
4597 Only one type, the return type, is overloaded, and only one type suffix is
4598 required. Because the argument's type is matched against the return type, it
4599 does not require its own name suffix.</p>
4601 <p>To learn how to add an intrinsic function, please see the
4602 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.
4607 <!-- ======================================================================= -->
4608 <div class="doc_subsection">
4609 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
4612 <div class="doc_text">
4614 <p>Variable argument support is defined in LLVM with the <a
4615 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
4616 intrinsic functions. These functions are related to the similarly
4617 named macros defined in the <tt><stdarg.h></tt> header file.</p>
4619 <p>All of these functions operate on arguments that use a
4620 target-specific value type "<tt>va_list</tt>". The LLVM assembly
4621 language reference manual does not define what this type is, so all
4622 transformations should be prepared to handle these functions regardless of
4625 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
4626 instruction and the variable argument handling intrinsic functions are
4629 <div class="doc_code">
4631 define i32 @test(i32 %X, ...) {
4632 ; Initialize variable argument processing
4634 %ap2 = bitcast i8** %ap to i8*
4635 call void @llvm.va_start(i8* %ap2)
4637 ; Read a single integer argument
4638 %tmp = va_arg i8** %ap, i32
4640 ; Demonstrate usage of llvm.va_copy and llvm.va_end
4642 %aq2 = bitcast i8** %aq to i8*
4643 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
4644 call void @llvm.va_end(i8* %aq2)
4646 ; Stop processing of arguments.
4647 call void @llvm.va_end(i8* %ap2)
4651 declare void @llvm.va_start(i8*)
4652 declare void @llvm.va_copy(i8*, i8*)
4653 declare void @llvm.va_end(i8*)
4659 <!-- _______________________________________________________________________ -->
4660 <div class="doc_subsubsection">
4661 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
4665 <div class="doc_text">
4667 <pre> declare void %llvm.va_start(i8* <arglist>)<br></pre>
4669 <p>The '<tt>llvm.va_start</tt>' intrinsic initializes
4670 <tt>*<arglist></tt> for subsequent use by <tt><a
4671 href="#i_va_arg">va_arg</a></tt>.</p>
4675 <p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
4679 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
4680 macro available in C. In a target-dependent way, it initializes the
4681 <tt>va_list</tt> element to which the argument points, so that the next call to
4682 <tt>va_arg</tt> will produce the first variable argument passed to the function.
4683 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
4684 last argument of the function as the compiler can figure that out.</p>
4688 <!-- _______________________________________________________________________ -->
4689 <div class="doc_subsubsection">
4690 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
4693 <div class="doc_text">
4695 <pre> declare void @llvm.va_end(i8* <arglist>)<br></pre>
4698 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
4699 which has been initialized previously with <tt><a href="#int_va_start">llvm.va_start</a></tt>
4700 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
4704 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
4708 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
4709 macro available in C. In a target-dependent way, it destroys the
4710 <tt>va_list</tt> element to which the argument points. Calls to <a
4711 href="#int_va_start"><tt>llvm.va_start</tt></a> and <a href="#int_va_copy">
4712 <tt>llvm.va_copy</tt></a> must be matched exactly with calls to
4713 <tt>llvm.va_end</tt>.</p>
4717 <!-- _______________________________________________________________________ -->
4718 <div class="doc_subsubsection">
4719 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
4722 <div class="doc_text">
4727 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
4732 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
4733 from the source argument list to the destination argument list.</p>
4737 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
4738 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
4743 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
4744 macro available in C. In a target-dependent way, it copies the source
4745 <tt>va_list</tt> element into the destination <tt>va_list</tt> element. This
4746 intrinsic is necessary because the <tt><a href="#int_va_start">
4747 llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
4748 example, memory allocation.</p>
4752 <!-- ======================================================================= -->
4753 <div class="doc_subsection">
4754 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
4757 <div class="doc_text">
4760 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
4761 Collection</a> (GC) requires the implementation and generation of these
4763 These intrinsics allow identification of <a href="#int_gcroot">GC roots on the
4764 stack</a>, as well as garbage collector implementations that require <a
4765 href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a> barriers.
4766 Front-ends for type-safe garbage collected languages should generate these
4767 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
4768 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
4771 <p>The garbage collection intrinsics only operate on objects in the generic
4772 address space (address space zero).</p>
4776 <!-- _______________________________________________________________________ -->
4777 <div class="doc_subsubsection">
4778 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
4781 <div class="doc_text">
4786 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
4791 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
4792 the code generator, and allows some metadata to be associated with it.</p>
4796 <p>The first argument specifies the address of a stack object that contains the
4797 root pointer. The second pointer (which must be either a constant or a global
4798 value address) contains the meta-data to be associated with the root.</p>
4802 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
4803 location. At compile-time, the code generator generates information to allow
4804 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
4805 intrinsic may only be used in a function which <a href="#gc">specifies a GC
4811 <!-- _______________________________________________________________________ -->
4812 <div class="doc_subsubsection">
4813 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
4816 <div class="doc_text">
4821 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
4826 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
4827 locations, allowing garbage collector implementations that require read
4832 <p>The second argument is the address to read from, which should be an address
4833 allocated from the garbage collector. The first object is a pointer to the
4834 start of the referenced object, if needed by the language runtime (otherwise
4839 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
4840 instruction, but may be replaced with substantially more complex code by the
4841 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
4842 may only be used in a function which <a href="#gc">specifies a GC
4848 <!-- _______________________________________________________________________ -->
4849 <div class="doc_subsubsection">
4850 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
4853 <div class="doc_text">
4858 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
4863 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
4864 locations, allowing garbage collector implementations that require write
4865 barriers (such as generational or reference counting collectors).</p>
4869 <p>The first argument is the reference to store, the second is the start of the
4870 object to store it to, and the third is the address of the field of Obj to
4871 store to. If the runtime does not require a pointer to the object, Obj may be
4876 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
4877 instruction, but may be replaced with substantially more complex code by the
4878 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
4879 may only be used in a function which <a href="#gc">specifies a GC
4886 <!-- ======================================================================= -->
4887 <div class="doc_subsection">
4888 <a name="int_codegen">Code Generator Intrinsics</a>
4891 <div class="doc_text">
4893 These intrinsics are provided by LLVM to expose special features that may only
4894 be implemented with code generator support.
4899 <!-- _______________________________________________________________________ -->
4900 <div class="doc_subsubsection">
4901 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
4904 <div class="doc_text">
4908 declare i8 *@llvm.returnaddress(i32 <level>)
4914 The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
4915 target-specific value indicating the return address of the current function
4916 or one of its callers.
4922 The argument to this intrinsic indicates which function to return the address
4923 for. Zero indicates the calling function, one indicates its caller, etc. The
4924 argument is <b>required</b> to be a constant integer value.
4930 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
4931 the return address of the specified call frame, or zero if it cannot be
4932 identified. The value returned by this intrinsic is likely to be incorrect or 0
4933 for arguments other than zero, so it should only be used for debugging purposes.
4937 Note that calling this intrinsic does not prevent function inlining or other
4938 aggressive transformations, so the value returned may not be that of the obvious
4939 source-language caller.
4944 <!-- _______________________________________________________________________ -->
4945 <div class="doc_subsubsection">
4946 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
4949 <div class="doc_text">
4953 declare i8 *@llvm.frameaddress(i32 <level>)
4959 The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
4960 target-specific frame pointer value for the specified stack frame.
4966 The argument to this intrinsic indicates which function to return the frame
4967 pointer for. Zero indicates the calling function, one indicates its caller,
4968 etc. The argument is <b>required</b> to be a constant integer value.
4974 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
4975 the frame address of the specified call frame, or zero if it cannot be
4976 identified. The value returned by this intrinsic is likely to be incorrect or 0
4977 for arguments other than zero, so it should only be used for debugging purposes.
4981 Note that calling this intrinsic does not prevent function inlining or other
4982 aggressive transformations, so the value returned may not be that of the obvious
4983 source-language caller.
4987 <!-- _______________________________________________________________________ -->
4988 <div class="doc_subsubsection">
4989 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
4992 <div class="doc_text">
4996 declare i8 *@llvm.stacksave()
5002 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
5003 the function stack, for use with <a href="#int_stackrestore">
5004 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
5005 features like scoped automatic variable sized arrays in C99.
5011 This intrinsic returns a opaque pointer value that can be passed to <a
5012 href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
5013 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
5014 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
5015 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
5016 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
5017 that were allocated after the <tt>llvm.stacksave</tt> was executed.
5022 <!-- _______________________________________________________________________ -->
5023 <div class="doc_subsubsection">
5024 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
5027 <div class="doc_text">
5031 declare void @llvm.stackrestore(i8 * %ptr)
5037 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
5038 the function stack to the state it was in when the corresponding <a
5039 href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
5040 useful for implementing language features like scoped automatic variable sized
5047 See the description for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.
5053 <!-- _______________________________________________________________________ -->
5054 <div class="doc_subsubsection">
5055 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
5058 <div class="doc_text">
5062 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>)
5069 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
5070 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
5072 effect on the behavior of the program but can change its performance
5079 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
5080 determining if the fetch should be for a read (0) or write (1), and
5081 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
5082 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
5083 <tt>locality</tt> arguments must be constant integers.
5089 This intrinsic does not modify the behavior of the program. In particular,
5090 prefetches cannot trap and do not produce a value. On targets that support this
5091 intrinsic, the prefetch can provide hints to the processor cache for better
5097 <!-- _______________________________________________________________________ -->
5098 <div class="doc_subsubsection">
5099 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
5102 <div class="doc_text">
5106 declare void @llvm.pcmarker(i32 <id>)
5113 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
5115 code to simulators and other tools. The method is target specific, but it is
5116 expected that the marker will use exported symbols to transmit the PC of the
5118 The marker makes no guarantees that it will remain with any specific instruction
5119 after optimizations. It is possible that the presence of a marker will inhibit
5120 optimizations. The intended use is to be inserted after optimizations to allow
5121 correlations of simulation runs.
5127 <tt>id</tt> is a numerical id identifying the marker.
5133 This intrinsic does not modify the behavior of the program. Backends that do not
5134 support this intrinisic may ignore it.
5139 <!-- _______________________________________________________________________ -->
5140 <div class="doc_subsubsection">
5141 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
5144 <div class="doc_text">
5148 declare i64 @llvm.readcyclecounter( )
5155 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
5156 counter register (or similar low latency, high accuracy clocks) on those targets
5157 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
5158 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
5159 should only be used for small timings.
5165 When directly supported, reading the cycle counter should not modify any memory.
5166 Implementations are allowed to either return a application specific value or a
5167 system wide value. On backends without support, this is lowered to a constant 0.
5172 <!-- ======================================================================= -->
5173 <div class="doc_subsection">
5174 <a name="int_libc">Standard C Library Intrinsics</a>
5177 <div class="doc_text">
5179 LLVM provides intrinsics for a few important standard C library functions.
5180 These intrinsics allow source-language front-ends to pass information about the
5181 alignment of the pointer arguments to the code generator, providing opportunity
5182 for more efficient code generation.
5187 <!-- _______________________________________________________________________ -->
5188 <div class="doc_subsubsection">
5189 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
5192 <div class="doc_text">
5195 <p>This is an overloaded intrinsic. You can use llvm.memcpy on any integer bit
5196 width. Not all targets support all bit widths however.</p>
5198 declare void @llvm.memcpy.i8(i8 * <dest>, i8 * <src>,
5199 i8 <len>, i32 <align>)
5200 declare void @llvm.memcpy.i16(i8 * <dest>, i8 * <src>,
5201 i16 <len>, i32 <align>)
5202 declare void @llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
5203 i32 <len>, i32 <align>)
5204 declare void @llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
5205 i64 <len>, i32 <align>)
5211 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
5212 location to the destination location.
5216 Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
5217 intrinsics do not return a value, and takes an extra alignment argument.
5223 The first argument is a pointer to the destination, the second is a pointer to
5224 the source. The third argument is an integer argument
5225 specifying the number of bytes to copy, and the fourth argument is the alignment
5226 of the source and destination locations.
5230 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5231 the caller guarantees that both the source and destination pointers are aligned
5238 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
5239 location to the destination location, which are not allowed to overlap. It
5240 copies "len" bytes of memory over. If the argument is known to be aligned to
5241 some boundary, this can be specified as the fourth argument, otherwise it should
5247 <!-- _______________________________________________________________________ -->
5248 <div class="doc_subsubsection">
5249 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
5252 <div class="doc_text">
5255 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
5256 width. Not all targets support all bit widths however.</p>
5258 declare void @llvm.memmove.i8(i8 * <dest>, i8 * <src>,
5259 i8 <len>, i32 <align>)
5260 declare void @llvm.memmove.i16(i8 * <dest>, i8 * <src>,
5261 i16 <len>, i32 <align>)
5262 declare void @llvm.memmove.i32(i8 * <dest>, i8 * <src>,
5263 i32 <len>, i32 <align>)
5264 declare void @llvm.memmove.i64(i8 * <dest>, i8 * <src>,
5265 i64 <len>, i32 <align>)
5271 The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
5272 location to the destination location. It is similar to the
5273 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to overlap.
5277 Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
5278 intrinsics do not return a value, and takes an extra alignment argument.
5284 The first argument is a pointer to the destination, the second is a pointer to
5285 the source. The third argument is an integer argument
5286 specifying the number of bytes to copy, and the fourth argument is the alignment
5287 of the source and destination locations.
5291 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5292 the caller guarantees that the source and destination pointers are aligned to
5299 The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
5300 location to the destination location, which may overlap. It
5301 copies "len" bytes of memory over. If the argument is known to be aligned to
5302 some boundary, this can be specified as the fourth argument, otherwise it should
5308 <!-- _______________________________________________________________________ -->
5309 <div class="doc_subsubsection">
5310 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
5313 <div class="doc_text">
5316 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
5317 width. Not all targets support all bit widths however.</p>
5319 declare void @llvm.memset.i8(i8 * <dest>, i8 <val>,
5320 i8 <len>, i32 <align>)
5321 declare void @llvm.memset.i16(i8 * <dest>, i8 <val>,
5322 i16 <len>, i32 <align>)
5323 declare void @llvm.memset.i32(i8 * <dest>, i8 <val>,
5324 i32 <len>, i32 <align>)
5325 declare void @llvm.memset.i64(i8 * <dest>, i8 <val>,
5326 i64 <len>, i32 <align>)
5332 The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
5337 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
5338 does not return a value, and takes an extra alignment argument.
5344 The first argument is a pointer to the destination to fill, the second is the
5345 byte value to fill it with, the third argument is an integer
5346 argument specifying the number of bytes to fill, and the fourth argument is the
5347 known alignment of destination location.
5351 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5352 the caller guarantees that the destination pointer is aligned to that boundary.
5358 The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
5360 destination location. If the argument is known to be aligned to some boundary,
5361 this can be specified as the fourth argument, otherwise it should be set to 0 or
5367 <!-- _______________________________________________________________________ -->
5368 <div class="doc_subsubsection">
5369 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
5372 <div class="doc_text">
5375 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
5376 floating point or vector of floating point type. Not all targets support all
5379 declare float @llvm.sqrt.f32(float %Val)
5380 declare double @llvm.sqrt.f64(double %Val)
5381 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
5382 declare fp128 @llvm.sqrt.f128(fp128 %Val)
5383 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
5389 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
5390 returning the same value as the libm '<tt>sqrt</tt>' functions would. Unlike
5391 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
5392 negative numbers other than -0.0 (which allows for better optimization, because
5393 there is no need to worry about errno being set). <tt>llvm.sqrt(-0.0)</tt> is
5394 defined to return -0.0 like IEEE sqrt.
5400 The argument and return value are floating point numbers of the same type.
5406 This function returns the sqrt of the specified operand if it is a nonnegative
5407 floating point number.
5411 <!-- _______________________________________________________________________ -->
5412 <div class="doc_subsubsection">
5413 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
5416 <div class="doc_text">
5419 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
5420 floating point or vector of floating point type. Not all targets support all
5423 declare float @llvm.powi.f32(float %Val, i32 %power)
5424 declare double @llvm.powi.f64(double %Val, i32 %power)
5425 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
5426 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
5427 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
5433 The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
5434 specified (positive or negative) power. The order of evaluation of
5435 multiplications is not defined. When a vector of floating point type is
5436 used, the second argument remains a scalar integer value.
5442 The second argument is an integer power, and the first is a value to raise to
5449 This function returns the first value raised to the second power with an
5450 unspecified sequence of rounding operations.</p>
5453 <!-- _______________________________________________________________________ -->
5454 <div class="doc_subsubsection">
5455 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
5458 <div class="doc_text">
5461 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
5462 floating point or vector of floating point type. Not all targets support all
5465 declare float @llvm.sin.f32(float %Val)
5466 declare double @llvm.sin.f64(double %Val)
5467 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
5468 declare fp128 @llvm.sin.f128(fp128 %Val)
5469 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
5475 The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.
5481 The argument and return value are floating point numbers of the same type.
5487 This function returns the sine of the specified operand, returning the
5488 same values as the libm <tt>sin</tt> functions would, and handles error
5489 conditions in the same way.</p>
5492 <!-- _______________________________________________________________________ -->
5493 <div class="doc_subsubsection">
5494 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
5497 <div class="doc_text">
5500 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
5501 floating point or vector of floating point type. Not all targets support all
5504 declare float @llvm.cos.f32(float %Val)
5505 declare double @llvm.cos.f64(double %Val)
5506 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
5507 declare fp128 @llvm.cos.f128(fp128 %Val)
5508 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
5514 The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.
5520 The argument and return value are floating point numbers of the same type.
5526 This function returns the cosine of the specified operand, returning the
5527 same values as the libm <tt>cos</tt> functions would, and handles error
5528 conditions in the same way.</p>
5531 <!-- _______________________________________________________________________ -->
5532 <div class="doc_subsubsection">
5533 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
5536 <div class="doc_text">
5539 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
5540 floating point or vector of floating point type. Not all targets support all
5543 declare float @llvm.pow.f32(float %Val, float %Power)
5544 declare double @llvm.pow.f64(double %Val, double %Power)
5545 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
5546 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
5547 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
5553 The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
5554 specified (positive or negative) power.
5560 The second argument is a floating point power, and the first is a value to
5561 raise to that power.
5567 This function returns the first value raised to the second power,
5569 same values as the libm <tt>pow</tt> functions would, and handles error
5570 conditions in the same way.</p>
5574 <!-- ======================================================================= -->
5575 <div class="doc_subsection">
5576 <a name="int_manip">Bit Manipulation Intrinsics</a>
5579 <div class="doc_text">
5581 LLVM provides intrinsics for a few important bit manipulation operations.
5582 These allow efficient code generation for some algorithms.
5587 <!-- _______________________________________________________________________ -->
5588 <div class="doc_subsubsection">
5589 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
5592 <div class="doc_text">
5595 <p>This is an overloaded intrinsic function. You can use bswap on any integer
5596 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
5598 declare i16 @llvm.bswap.i16(i16 <id>)
5599 declare i32 @llvm.bswap.i32(i32 <id>)
5600 declare i64 @llvm.bswap.i64(i64 <id>)
5606 The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
5607 values with an even number of bytes (positive multiple of 16 bits). These are
5608 useful for performing operations on data that is not in the target's native
5615 The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
5616 and low byte of the input i16 swapped. Similarly, the <tt>llvm.bswap.i32</tt>
5617 intrinsic returns an i32 value that has the four bytes of the input i32
5618 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the returned
5619 i32 will have its bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i48</tt>,
5620 <tt>llvm.bswap.i64</tt> and other intrinsics extend this concept to
5621 additional even-byte lengths (6 bytes, 8 bytes and more, respectively).
5626 <!-- _______________________________________________________________________ -->
5627 <div class="doc_subsubsection">
5628 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
5631 <div class="doc_text">
5634 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
5635 width. Not all targets support all bit widths however.</p>
5637 declare i8 @llvm.ctpop.i8 (i8 <src>)
5638 declare i16 @llvm.ctpop.i16(i16 <src>)
5639 declare i32 @llvm.ctpop.i32(i32 <src>)
5640 declare i64 @llvm.ctpop.i64(i64 <src>)
5641 declare i256 @llvm.ctpop.i256(i256 <src>)
5647 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
5654 The only argument is the value to be counted. The argument may be of any
5655 integer type. The return type must match the argument type.
5661 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
5665 <!-- _______________________________________________________________________ -->
5666 <div class="doc_subsubsection">
5667 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
5670 <div class="doc_text">
5673 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
5674 integer bit width. Not all targets support all bit widths however.</p>
5676 declare i8 @llvm.ctlz.i8 (i8 <src>)
5677 declare i16 @llvm.ctlz.i16(i16 <src>)
5678 declare i32 @llvm.ctlz.i32(i32 <src>)
5679 declare i64 @llvm.ctlz.i64(i64 <src>)
5680 declare i256 @llvm.ctlz.i256(i256 <src>)
5686 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
5687 leading zeros in a variable.
5693 The only argument is the value to be counted. The argument may be of any
5694 integer type. The return type must match the argument type.
5700 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
5701 in a variable. If the src == 0 then the result is the size in bits of the type
5702 of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.
5708 <!-- _______________________________________________________________________ -->
5709 <div class="doc_subsubsection">
5710 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
5713 <div class="doc_text">
5716 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
5717 integer bit width. Not all targets support all bit widths however.</p>
5719 declare i8 @llvm.cttz.i8 (i8 <src>)
5720 declare i16 @llvm.cttz.i16(i16 <src>)
5721 declare i32 @llvm.cttz.i32(i32 <src>)
5722 declare i64 @llvm.cttz.i64(i64 <src>)
5723 declare i256 @llvm.cttz.i256(i256 <src>)
5729 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
5736 The only argument is the value to be counted. The argument may be of any
5737 integer type. The return type must match the argument type.
5743 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
5744 in a variable. If the src == 0 then the result is the size in bits of the type
5745 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
5749 <!-- _______________________________________________________________________ -->
5750 <div class="doc_subsubsection">
5751 <a name="int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic</a>
5754 <div class="doc_text">
5757 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.select</tt>
5758 on any integer bit width.</p>
5760 declare i17 @llvm.part.select.i17 (i17 %val, i32 %loBit, i32 %hiBit)
5761 declare i29 @llvm.part.select.i29 (i29 %val, i32 %loBit, i32 %hiBit)
5765 <p>The '<tt>llvm.part.select</tt>' family of intrinsic functions selects a
5766 range of bits from an integer value and returns them in the same bit width as
5767 the original value.</p>
5770 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5771 any bit width but they must have the same bit width. The second and third
5772 arguments must be <tt>i32</tt> type since they specify only a bit index.</p>
5775 <p>The operation of the '<tt>llvm.part.select</tt>' intrinsic has two modes
5776 of operation: forwards and reverse. If <tt>%loBit</tt> is greater than
5777 <tt>%hiBits</tt> then the intrinsic operates in reverse mode. Otherwise it
5778 operates in forward mode.</p>
5779 <p>In forward mode, this intrinsic is the equivalent of shifting <tt>%val</tt>
5780 right by <tt>%loBit</tt> bits and then ANDing it with a mask with
5781 only the <tt>%hiBit - %loBit</tt> bits set, as follows:</p>
5783 <li>The <tt>%val</tt> is shifted right (LSHR) by the number of bits specified
5784 by <tt>%loBits</tt>. This normalizes the value to the low order bits.</li>
5785 <li>The <tt>%loBits</tt> value is subtracted from the <tt>%hiBits</tt> value
5786 to determine the number of bits to retain.</li>
5787 <li>A mask of the retained bits is created by shifting a -1 value.</li>
5788 <li>The mask is ANDed with <tt>%val</tt> to produce the result.</li>
5790 <p>In reverse mode, a similar computation is made except that the bits are
5791 returned in the reverse order. So, for example, if <tt>X</tt> has the value
5792 <tt>i16 0x0ACF (101011001111)</tt> and we apply
5793 <tt>part.select(i16 X, 8, 3)</tt> to it, we get back the value
5794 <tt>i16 0x0026 (000000100110)</tt>.</p>
5797 <div class="doc_subsubsection">
5798 <a name="int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic</a>
5801 <div class="doc_text">
5804 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.set</tt>
5805 on any integer bit width.</p>
5807 declare i17 @llvm.part.set.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
5808 declare i29 @llvm.part.set.i29.i9 (i29 %val, i9 %repl, i32 %lo, i32 %hi)
5812 <p>The '<tt>llvm.part.set</tt>' family of intrinsic functions replaces a range
5813 of bits in an integer value with another integer value. It returns the integer
5814 with the replaced bits.</p>
5817 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5818 any bit width but they must have the same bit width. <tt>%val</tt> is the value
5819 whose bits will be replaced. The second argument, <tt>%repl</tt> may be an
5820 integer of any bit width. The third and fourth arguments must be <tt>i32</tt>
5821 type since they specify only a bit index.</p>
5824 <p>The operation of the '<tt>llvm.part.set</tt>' intrinsic has two modes
5825 of operation: forwards and reverse. If <tt>%lo</tt> is greater than
5826 <tt>%hi</tt> then the intrinsic operates in reverse mode. Otherwise it
5827 operates in forward mode.</p>
5828 <p>For both modes, the <tt>%repl</tt> value is prepared for use by either
5829 truncating it down to the size of the replacement area or zero extending it
5830 up to that size.</p>
5831 <p>In forward mode, the bits between <tt>%lo</tt> and <tt>%hi</tt> (inclusive)
5832 are replaced with corresponding bits from <tt>%repl</tt>. That is the 0th bit
5833 in <tt>%repl</tt> replaces the <tt>%lo</tt>th bit in <tt>%val</tt> and etc. up
5834 to the <tt>%hi</tt>th bit.</p>
5835 <p>In reverse mode, a similar computation is made except that the bits are
5836 reversed. That is, the <tt>0</tt>th bit in <tt>%repl</tt> replaces the
5837 <tt>%hi</tt> bit in <tt>%val</tt> and etc. down to the <tt>%lo</tt>th bit.</p>
5840 llvm.part.set(0xFFFF, 0, 4, 7) -> 0xFF0F
5841 llvm.part.set(0xFFFF, 0, 7, 4) -> 0xFF0F
5842 llvm.part.set(0xFFFF, 1, 7, 4) -> 0xFF8F
5843 llvm.part.set(0xFFFF, F, 8, 3) -> 0xFFE7
5844 llvm.part.set(0xFFFF, 0, 3, 8) -> 0xFE07
5848 <!-- ======================================================================= -->
5849 <div class="doc_subsection">
5850 <a name="int_debugger">Debugger Intrinsics</a>
5853 <div class="doc_text">
5855 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
5856 are described in the <a
5857 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
5858 Debugging</a> document.
5863 <!-- ======================================================================= -->
5864 <div class="doc_subsection">
5865 <a name="int_eh">Exception Handling Intrinsics</a>
5868 <div class="doc_text">
5869 <p> The LLVM exception handling intrinsics (which all start with
5870 <tt>llvm.eh.</tt> prefix), are described in the <a
5871 href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
5872 Handling</a> document. </p>
5875 <!-- ======================================================================= -->
5876 <div class="doc_subsection">
5877 <a name="int_trampoline">Trampoline Intrinsic</a>
5880 <div class="doc_text">
5882 This intrinsic makes it possible to excise one parameter, marked with
5883 the <tt>nest</tt> attribute, from a function. The result is a callable
5884 function pointer lacking the nest parameter - the caller does not need
5885 to provide a value for it. Instead, the value to use is stored in
5886 advance in a "trampoline", a block of memory usually allocated
5887 on the stack, which also contains code to splice the nest value into the
5888 argument list. This is used to implement the GCC nested function address
5892 For example, if the function is
5893 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
5894 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as follows:</p>
5896 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
5897 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
5898 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
5899 %fp = bitcast i8* %p to i32 (i32, i32)*
5901 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
5902 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
5905 <!-- _______________________________________________________________________ -->
5906 <div class="doc_subsubsection">
5907 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
5909 <div class="doc_text">
5912 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
5916 This fills the memory pointed to by <tt>tramp</tt> with code
5917 and returns a function pointer suitable for executing it.
5921 The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
5922 pointers. The <tt>tramp</tt> argument must point to a sufficiently large
5923 and sufficiently aligned block of memory; this memory is written to by the
5924 intrinsic. Note that the size and the alignment are target-specific - LLVM
5925 currently provides no portable way of determining them, so a front-end that
5926 generates this intrinsic needs to have some target-specific knowledge.
5927 The <tt>func</tt> argument must hold a function bitcast to an <tt>i8*</tt>.
5931 The block of memory pointed to by <tt>tramp</tt> is filled with target
5932 dependent code, turning it into a function. A pointer to this function is
5933 returned, but needs to be bitcast to an
5934 <a href="#int_trampoline">appropriate function pointer type</a>
5935 before being called. The new function's signature is the same as that of
5936 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
5937 removed. At most one such <tt>nest</tt> argument is allowed, and it must be
5938 of pointer type. Calling the new function is equivalent to calling
5939 <tt>func</tt> with the same argument list, but with <tt>nval</tt> used for the
5940 missing <tt>nest</tt> argument. If, after calling
5941 <tt>llvm.init.trampoline</tt>, the memory pointed to by <tt>tramp</tt> is
5942 modified, then the effect of any later call to the returned function pointer is
5947 <!-- ======================================================================= -->
5948 <div class="doc_subsection">
5949 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
5952 <div class="doc_text">
5954 These intrinsic functions expand the "universal IR" of LLVM to represent
5955 hardware constructs for atomic operations and memory synchronization. This
5956 provides an interface to the hardware, not an interface to the programmer. It
5957 is aimed at a low enough level to allow any programming models or APIs
5958 (Application Programming Interfaces) which
5959 need atomic behaviors to map cleanly onto it. It is also modeled primarily on
5960 hardware behavior. Just as hardware provides a "universal IR" for source
5961 languages, it also provides a starting point for developing a "universal"
5962 atomic operation and synchronization IR.
5965 These do <em>not</em> form an API such as high-level threading libraries,
5966 software transaction memory systems, atomic primitives, and intrinsic
5967 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
5968 application libraries. The hardware interface provided by LLVM should allow
5969 a clean implementation of all of these APIs and parallel programming models.
5970 No one model or paradigm should be selected above others unless the hardware
5971 itself ubiquitously does so.
5976 <!-- _______________________________________________________________________ -->
5977 <div class="doc_subsubsection">
5978 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
5980 <div class="doc_text">
5983 declare void @llvm.memory.barrier( i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>,
5989 The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
5990 specific pairs of memory access types.
5994 The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
5995 The first four arguments enables a specific barrier as listed below. The fith
5996 argument specifies that the barrier applies to io or device or uncached memory.
6000 <li><tt>ll</tt>: load-load barrier</li>
6001 <li><tt>ls</tt>: load-store barrier</li>
6002 <li><tt>sl</tt>: store-load barrier</li>
6003 <li><tt>ss</tt>: store-store barrier</li>
6004 <li><tt>device</tt>: barrier applies to device and uncached memory also.</li>
6008 This intrinsic causes the system to enforce some ordering constraints upon
6009 the loads and stores of the program. This barrier does not indicate
6010 <em>when</em> any events will occur, it only enforces an <em>order</em> in
6011 which they occur. For any of the specified pairs of load and store operations
6012 (f.ex. load-load, or store-load), all of the first operations preceding the
6013 barrier will complete before any of the second operations succeeding the
6014 barrier begin. Specifically the semantics for each pairing is as follows:
6017 <li><tt>ll</tt>: All loads before the barrier must complete before any load
6018 after the barrier begins.</li>
6020 <li><tt>ls</tt>: All loads before the barrier must complete before any
6021 store after the barrier begins.</li>
6022 <li><tt>ss</tt>: All stores before the barrier must complete before any
6023 store after the barrier begins.</li>
6024 <li><tt>sl</tt>: All stores before the barrier must complete before any
6025 load after the barrier begins.</li>
6028 These semantics are applied with a logical "and" behavior when more than one
6029 is enabled in a single memory barrier intrinsic.
6032 Backends may implement stronger barriers than those requested when they do not
6033 support as fine grained a barrier as requested. Some architectures do not
6034 need all types of barriers and on such architectures, these become noops.
6041 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
6042 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
6043 <i>; guarantee the above finishes</i>
6044 store i32 8, %ptr <i>; before this begins</i>
6048 <!-- _______________________________________________________________________ -->
6049 <div class="doc_subsubsection">
6050 <a name="int_atomic_cmp_swap">'<tt>llvm.atomic.cmp.swap.*</tt>' Intrinsic</a>
6052 <div class="doc_text">
6055 This is an overloaded intrinsic. You can use <tt>llvm.atomic.cmp.swap</tt> on
6056 any integer bit width and for different address spaces. Not all targets
6057 support all bit widths however.</p>
6060 declare i8 @llvm.atomic.cmp.swap.i8.p0i8( i8* <ptr>, i8 <cmp>, i8 <val> )
6061 declare i16 @llvm.atomic.cmp.swap.i16.p0i16( i16* <ptr>, i16 <cmp>, i16 <val> )
6062 declare i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* <ptr>, i32 <cmp>, i32 <val> )
6063 declare i64 @llvm.atomic.cmp.swap.i64.p0i64( i64* <ptr>, i64 <cmp>, i64 <val> )
6068 This loads a value in memory and compares it to a given value. If they are
6069 equal, it stores a new value into the memory.
6073 The <tt>llvm.atomic.cmp.swap</tt> intrinsic takes three arguments. The result as
6074 well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
6075 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
6076 this integer type. While any bit width integer may be used, targets may only
6077 lower representations they support in hardware.
6082 This entire intrinsic must be executed atomically. It first loads the value
6083 in memory pointed to by <tt>ptr</tt> and compares it with the value
6084 <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the memory. The
6085 loaded value is yielded in all cases. This provides the equivalent of an
6086 atomic compare-and-swap operation within the SSA framework.
6094 %val1 = add i32 4, 4
6095 %result1 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 4, %val1 )
6096 <i>; yields {i32}:result1 = 4</i>
6097 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
6098 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
6100 %val2 = add i32 1, 1
6101 %result2 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 5, %val2 )
6102 <i>; yields {i32}:result2 = 8</i>
6103 %stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
6105 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
6109 <!-- _______________________________________________________________________ -->
6110 <div class="doc_subsubsection">
6111 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
6113 <div class="doc_text">
6117 This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
6118 integer bit width. Not all targets support all bit widths however.</p>
6120 declare i8 @llvm.atomic.swap.i8.p0i8( i8* <ptr>, i8 <val> )
6121 declare i16 @llvm.atomic.swap.i16.p0i16( i16* <ptr>, i16 <val> )
6122 declare i32 @llvm.atomic.swap.i32.p0i32( i32* <ptr>, i32 <val> )
6123 declare i64 @llvm.atomic.swap.i64.p0i64( i64* <ptr>, i64 <val> )
6128 This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
6129 the value from memory. It then stores the value in <tt>val</tt> in the memory
6135 The <tt>llvm.atomic.swap</tt> intrinsic takes two arguments. Both the
6136 <tt>val</tt> argument and the result must be integers of the same bit width.
6137 The first argument, <tt>ptr</tt>, must be a pointer to a value of this
6138 integer type. The targets may only lower integer representations they
6143 This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
6144 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
6145 equivalent of an atomic swap operation within the SSA framework.
6153 %val1 = add i32 4, 4
6154 %result1 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val1 )
6155 <i>; yields {i32}:result1 = 4</i>
6156 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
6157 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
6159 %val2 = add i32 1, 1
6160 %result2 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val2 )
6161 <i>; yields {i32}:result2 = 8</i>
6163 %stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
6164 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
6168 <!-- _______________________________________________________________________ -->
6169 <div class="doc_subsubsection">
6170 <a name="int_atomic_load_add">'<tt>llvm.atomic.load.add.*</tt>' Intrinsic</a>
6173 <div class="doc_text">
6176 This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.add</tt> on any
6177 integer bit width. Not all targets support all bit widths however.</p>
6179 declare i8 @llvm.atomic.load.add.i8..p0i8( i8* <ptr>, i8 <delta> )
6180 declare i16 @llvm.atomic.load.add.i16..p0i16( i16* <ptr>, i16 <delta> )
6181 declare i32 @llvm.atomic.load.add.i32..p0i32( i32* <ptr>, i32 <delta> )
6182 declare i64 @llvm.atomic.load.add.i64..p0i64( i64* <ptr>, i64 <delta> )
6187 This intrinsic adds <tt>delta</tt> to the value stored in memory at
6188 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
6193 The intrinsic takes two arguments, the first a pointer to an integer value
6194 and the second an integer value. The result is also an integer value. These
6195 integer types can have any bit width, but they must all have the same bit
6196 width. The targets may only lower integer representations they support.
6200 This intrinsic does a series of operations atomically. It first loads the
6201 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
6202 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
6209 %result1 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 4 )
6210 <i>; yields {i32}:result1 = 4</i>
6211 %result2 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 2 )
6212 <i>; yields {i32}:result2 = 8</i>
6213 %result3 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 5 )
6214 <i>; yields {i32}:result3 = 10</i>
6215 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
6219 <!-- _______________________________________________________________________ -->
6220 <div class="doc_subsubsection">
6221 <a name="int_atomic_load_sub">'<tt>llvm.atomic.load.sub.*</tt>' Intrinsic</a>
6224 <div class="doc_text">
6227 This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.sub</tt> on
6228 any integer bit width and for different address spaces. Not all targets
6229 support all bit widths however.</p>
6231 declare i8 @llvm.atomic.load.sub.i8.p0i32( i8* <ptr>, i8 <delta> )
6232 declare i16 @llvm.atomic.load.sub.i16.p0i32( i16* <ptr>, i16 <delta> )
6233 declare i32 @llvm.atomic.load.sub.i32.p0i32( i32* <ptr>, i32 <delta> )
6234 declare i64 @llvm.atomic.load.sub.i64.p0i32( i64* <ptr>, i64 <delta> )
6239 This intrinsic subtracts <tt>delta</tt> to the value stored in memory at
6240 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
6245 The intrinsic takes two arguments, the first a pointer to an integer value
6246 and the second an integer value. The result is also an integer value. These
6247 integer types can have any bit width, but they must all have the same bit
6248 width. The targets may only lower integer representations they support.
6252 This intrinsic does a series of operations atomically. It first loads the
6253 value stored at <tt>ptr</tt>. It then subtracts <tt>delta</tt>, stores the
6254 result to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
6261 %result1 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 4 )
6262 <i>; yields {i32}:result1 = 8</i>
6263 %result2 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 2 )
6264 <i>; yields {i32}:result2 = 4</i>
6265 %result3 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 5 )
6266 <i>; yields {i32}:result3 = 2</i>
6267 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = -3</i>
6271 <!-- _______________________________________________________________________ -->
6272 <div class="doc_subsubsection">
6273 <a name="int_atomic_load_and">'<tt>llvm.atomic.load.and.*</tt>' Intrinsic</a><br>
6274 <a name="int_atomic_load_nand">'<tt>llvm.atomic.load.nand.*</tt>' Intrinsic</a><br>
6275 <a name="int_atomic_load_or">'<tt>llvm.atomic.load.or.*</tt>' Intrinsic</a><br>
6276 <a name="int_atomic_load_xor">'<tt>llvm.atomic.load.xor.*</tt>' Intrinsic</a><br>
6279 <div class="doc_text">
6282 These are overloaded intrinsics. You can use <tt>llvm.atomic.load_and</tt>,
6283 <tt>llvm.atomic.load_nand</tt>, <tt>llvm.atomic.load_or</tt>, and
6284 <tt>llvm.atomic.load_xor</tt> on any integer bit width and for different
6285 address spaces. Not all targets support all bit widths however.</p>
6287 declare i8 @llvm.atomic.load.and.i8.p0i8( i8* <ptr>, i8 <delta> )
6288 declare i16 @llvm.atomic.load.and.i16.p0i16( i16* <ptr>, i16 <delta> )
6289 declare i32 @llvm.atomic.load.and.i32.p0i32( i32* <ptr>, i32 <delta> )
6290 declare i64 @llvm.atomic.load.and.i64.p0i64( i64* <ptr>, i64 <delta> )
6295 declare i8 @llvm.atomic.load.or.i8.p0i8( i8* <ptr>, i8 <delta> )
6296 declare i16 @llvm.atomic.load.or.i16.p0i16( i16* <ptr>, i16 <delta> )
6297 declare i32 @llvm.atomic.load.or.i32.p0i32( i32* <ptr>, i32 <delta> )
6298 declare i64 @llvm.atomic.load.or.i64.p0i64( i64* <ptr>, i64 <delta> )
6303 declare i8 @llvm.atomic.load.nand.i8.p0i32( i8* <ptr>, i8 <delta> )
6304 declare i16 @llvm.atomic.load.nand.i16.p0i32( i16* <ptr>, i16 <delta> )
6305 declare i32 @llvm.atomic.load.nand.i32.p0i32( i32* <ptr>, i32 <delta> )
6306 declare i64 @llvm.atomic.load.nand.i64.p0i32( i64* <ptr>, i64 <delta> )
6311 declare i8 @llvm.atomic.load.xor.i8.p0i32( i8* <ptr>, i8 <delta> )
6312 declare i16 @llvm.atomic.load.xor.i16.p0i32( i16* <ptr>, i16 <delta> )
6313 declare i32 @llvm.atomic.load.xor.i32.p0i32( i32* <ptr>, i32 <delta> )
6314 declare i64 @llvm.atomic.load.xor.i64.p0i32( i64* <ptr>, i64 <delta> )
6319 These intrinsics bitwise the operation (and, nand, or, xor) <tt>delta</tt> to
6320 the value stored in memory at <tt>ptr</tt>. It yields the original value
6326 These intrinsics take two arguments, the first a pointer to an integer value
6327 and the second an integer value. The result is also an integer value. These
6328 integer types can have any bit width, but they must all have the same bit
6329 width. The targets may only lower integer representations they support.
6333 These intrinsics does a series of operations atomically. They first load the
6334 value stored at <tt>ptr</tt>. They then do the bitwise operation
6335 <tt>delta</tt>, store the result to <tt>ptr</tt>. They yield the original
6336 value stored at <tt>ptr</tt>.
6342 store i32 0x0F0F, %ptr
6343 %result0 = call i32 @llvm.atomic.load.nand.i32.p0i32( i32* %ptr, i32 0xFF )
6344 <i>; yields {i32}:result0 = 0x0F0F</i>
6345 %result1 = call i32 @llvm.atomic.load.and.i32.p0i32( i32* %ptr, i32 0xFF )
6346 <i>; yields {i32}:result1 = 0xFFFFFFF0</i>
6347 %result2 = call i32 @llvm.atomic.load.or.i32.p0i32( i32* %ptr, i32 0F )
6348 <i>; yields {i32}:result2 = 0xF0</i>
6349 %result3 = call i32 @llvm.atomic.load.xor.i32.p0i32( i32* %ptr, i32 0F )
6350 <i>; yields {i32}:result3 = FF</i>
6351 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = F0</i>
6356 <!-- _______________________________________________________________________ -->
6357 <div class="doc_subsubsection">
6358 <a name="int_atomic_load_max">'<tt>llvm.atomic.load.max.*</tt>' Intrinsic</a><br>
6359 <a name="int_atomic_load_min">'<tt>llvm.atomic.load.min.*</tt>' Intrinsic</a><br>
6360 <a name="int_atomic_load_umax">'<tt>llvm.atomic.load.umax.*</tt>' Intrinsic</a><br>
6361 <a name="int_atomic_load_umin">'<tt>llvm.atomic.load.umin.*</tt>' Intrinsic</a><br>
6364 <div class="doc_text">
6367 These are overloaded intrinsics. You can use <tt>llvm.atomic.load_max</tt>,
6368 <tt>llvm.atomic.load_min</tt>, <tt>llvm.atomic.load_umax</tt>, and
6369 <tt>llvm.atomic.load_umin</tt> on any integer bit width and for different
6370 address spaces. Not all targets
6371 support all bit widths however.</p>
6373 declare i8 @llvm.atomic.load.max.i8.p0i8( i8* <ptr>, i8 <delta> )
6374 declare i16 @llvm.atomic.load.max.i16.p0i16( i16* <ptr>, i16 <delta> )
6375 declare i32 @llvm.atomic.load.max.i32.p0i32( i32* <ptr>, i32 <delta> )
6376 declare i64 @llvm.atomic.load.max.i64.p0i64( i64* <ptr>, i64 <delta> )
6381 declare i8 @llvm.atomic.load.min.i8.p0i8( i8* <ptr>, i8 <delta> )
6382 declare i16 @llvm.atomic.load.min.i16.p0i16( i16* <ptr>, i16 <delta> )
6383 declare i32 @llvm.atomic.load.min.i32..p0i32( i32* <ptr>, i32 <delta> )
6384 declare i64 @llvm.atomic.load.min.i64..p0i64( i64* <ptr>, i64 <delta> )
6389 declare i8 @llvm.atomic.load.umax.i8.p0i8( i8* <ptr>, i8 <delta> )
6390 declare i16 @llvm.atomic.load.umax.i16.p0i16( i16* <ptr>, i16 <delta> )
6391 declare i32 @llvm.atomic.load.umax.i32.p0i32( i32* <ptr>, i32 <delta> )
6392 declare i64 @llvm.atomic.load.umax.i64.p0i64( i64* <ptr>, i64 <delta> )
6397 declare i8 @llvm.atomic.load.umin.i8..p0i8( i8* <ptr>, i8 <delta> )
6398 declare i16 @llvm.atomic.load.umin.i16.p0i16( i16* <ptr>, i16 <delta> )
6399 declare i32 @llvm.atomic.load.umin.i32..p0i32( i32* <ptr>, i32 <delta> )
6400 declare i64 @llvm.atomic.load.umin.i64..p0i64( i64* <ptr>, i64 <delta> )
6405 These intrinsics takes the signed or unsigned minimum or maximum of
6406 <tt>delta</tt> and the value stored in memory at <tt>ptr</tt>. It yields the
6407 original value at <tt>ptr</tt>.
6412 These intrinsics take two arguments, the first a pointer to an integer value
6413 and the second an integer value. The result is also an integer value. These
6414 integer types can have any bit width, but they must all have the same bit
6415 width. The targets may only lower integer representations they support.
6419 These intrinsics does a series of operations atomically. They first load the
6420 value stored at <tt>ptr</tt>. They then do the signed or unsigned min or max
6421 <tt>delta</tt> and the value, store the result to <tt>ptr</tt>. They yield
6422 the original value stored at <tt>ptr</tt>.
6429 %result0 = call i32 @llvm.atomic.load.min.i32.p0i32( i32* %ptr, i32 -2 )
6430 <i>; yields {i32}:result0 = 7</i>
6431 %result1 = call i32 @llvm.atomic.load.max.i32.p0i32( i32* %ptr, i32 8 )
6432 <i>; yields {i32}:result1 = -2</i>
6433 %result2 = call i32 @llvm.atomic.load.umin.i32.p0i32( i32* %ptr, i32 10 )
6434 <i>; yields {i32}:result2 = 8</i>
6435 %result3 = call i32 @llvm.atomic.load.umax.i32.p0i32( i32* %ptr, i32 30 )
6436 <i>; yields {i32}:result3 = 8</i>
6437 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 30</i>
6441 <!-- ======================================================================= -->
6442 <div class="doc_subsection">
6443 <a name="int_general">General Intrinsics</a>
6446 <div class="doc_text">
6447 <p> This class of intrinsics is designed to be generic and has
6448 no specific purpose. </p>
6451 <!-- _______________________________________________________________________ -->
6452 <div class="doc_subsubsection">
6453 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
6456 <div class="doc_text">
6460 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
6466 The '<tt>llvm.var.annotation</tt>' intrinsic
6472 The first argument is a pointer to a value, the second is a pointer to a
6473 global string, the third is a pointer to a global string which is the source
6474 file name, and the last argument is the line number.
6480 This intrinsic allows annotation of local variables with arbitrary strings.
6481 This can be useful for special purpose optimizations that want to look for these
6482 annotations. These have no other defined use, they are ignored by code
6483 generation and optimization.
6487 <!-- _______________________________________________________________________ -->
6488 <div class="doc_subsubsection">
6489 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
6492 <div class="doc_text">
6495 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
6496 any integer bit width.
6499 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
6500 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
6501 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
6502 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
6503 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
6509 The '<tt>llvm.annotation</tt>' intrinsic.
6515 The first argument is an integer value (result of some expression),
6516 the second is a pointer to a global string, the third is a pointer to a global
6517 string which is the source file name, and the last argument is the line number.
6518 It returns the value of the first argument.
6524 This intrinsic allows annotations to be put on arbitrary expressions
6525 with arbitrary strings. This can be useful for special purpose optimizations
6526 that want to look for these annotations. These have no other defined use, they
6527 are ignored by code generation and optimization.
6531 <!-- _______________________________________________________________________ -->
6532 <div class="doc_subsubsection">
6533 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
6536 <div class="doc_text">
6540 declare void @llvm.trap()
6546 The '<tt>llvm.trap</tt>' intrinsic
6558 This intrinsics is lowered to the target dependent trap instruction. If the
6559 target does not have a trap instruction, this intrinsic will be lowered to the
6560 call of the abort() function.
6564 <!-- _______________________________________________________________________ -->
6565 <div class="doc_subsubsection">
6566 <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a>
6568 <div class="doc_text">
6571 declare void @llvm.stackprotector( i8* <guard>, i8** <slot> )
6576 The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and stores
6577 it onto the stack at <tt>slot</tt>. The stack slot is adjusted to ensure that
6578 it is placed on the stack before local variables.
6582 The <tt>llvm.stackprotector</tt> intrinsic requires two pointer arguments. The
6583 first argument is the value loaded from the stack guard
6584 <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt> that
6585 has enough space to hold the value of the guard.
6589 This intrinsic causes the prologue/epilogue inserter to force the position of
6590 the <tt>AllocaInst</tt> stack slot to be before local variables on the
6591 stack. This is to ensure that if a local variable on the stack is overwritten,
6592 it will destroy the value of the guard. When the function exits, the guard on
6593 the stack is checked against the original guard. If they're different, then
6594 the program aborts by calling the <tt>__stack_chk_fail()</tt> function.
6598 <!-- *********************************************************************** -->
6601 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
6602 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
6603 <a href="http://validator.w3.org/check/referer"><img
6604 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
6606 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
6607 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
6608 Last modified: $Date$