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="#globalvars">Global Variables</a></li>
26 <li><a href="#functionstructure">Functions</a></li>
27 <li><a href="#aliasstructure">Aliases</a>
28 <li><a href="#paramattrs">Parameter Attributes</a></li>
29 <li><a href="#gc">Garbage Collector Names</a></li>
30 <li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
31 <li><a href="#datalayout">Data Layout</a></li>
34 <li><a href="#typesystem">Type System</a>
36 <li><a href="#t_classifications">Type Classifications</a></li>
37 <li><a href="#t_primitive">Primitive Types</a>
39 <li><a href="#t_floating">Floating Point Types</a></li>
40 <li><a href="#t_void">Void Type</a></li>
41 <li><a href="#t_label">Label Type</a></li>
44 <li><a href="#t_derived">Derived Types</a>
46 <li><a href="#t_integer">Integer Type</a></li>
47 <li><a href="#t_array">Array Type</a></li>
48 <li><a href="#t_function">Function Type</a></li>
49 <li><a href="#t_pointer">Pointer Type</a></li>
50 <li><a href="#t_struct">Structure Type</a></li>
51 <li><a href="#t_pstruct">Packed Structure Type</a></li>
52 <li><a href="#t_vector">Vector Type</a></li>
53 <li><a href="#t_opaque">Opaque Type</a></li>
58 <li><a href="#constants">Constants</a>
60 <li><a href="#simpleconstants">Simple Constants</a>
61 <li><a href="#aggregateconstants">Aggregate Constants</a>
62 <li><a href="#globalconstants">Global Variable and Function Addresses</a>
63 <li><a href="#undefvalues">Undefined Values</a>
64 <li><a href="#constantexprs">Constant Expressions</a>
67 <li><a href="#othervalues">Other Values</a>
69 <li><a href="#inlineasm">Inline Assembler Expressions</a>
72 <li><a href="#instref">Instruction Reference</a>
74 <li><a href="#terminators">Terminator Instructions</a>
76 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
77 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
78 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
79 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
80 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
81 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
84 <li><a href="#binaryops">Binary Operations</a>
86 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
87 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
88 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
89 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
90 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
91 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
92 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
93 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
94 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
97 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
99 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
100 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
101 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
102 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
103 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
104 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
107 <li><a href="#vectorops">Vector Operations</a>
109 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
110 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
111 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
114 <li><a href="#aggregateops">Aggregate Operations</a>
116 <li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li>
117 <li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li>
120 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
122 <li><a href="#i_malloc">'<tt>malloc</tt>' Instruction</a></li>
123 <li><a href="#i_free">'<tt>free</tt>' Instruction</a></li>
124 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
125 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
126 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
127 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
130 <li><a href="#convertops">Conversion Operations</a>
132 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
133 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
134 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
135 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
136 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
137 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
138 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
139 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
140 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
141 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
142 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
143 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
145 <li><a href="#otherops">Other Operations</a>
147 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
148 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
149 <li><a href="#i_vicmp">'<tt>vicmp</tt>' Instruction</a></li>
150 <li><a href="#i_vfcmp">'<tt>vfcmp</tt>' Instruction</a></li>
151 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
152 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
153 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
154 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
155 <li><a href="#i_getresult">'<tt>getresult</tt>' Instruction</a></li>
160 <li><a href="#intrinsics">Intrinsic Functions</a>
162 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
164 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
165 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
166 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
169 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
171 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
172 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
173 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
176 <li><a href="#int_codegen">Code Generator Intrinsics</a>
178 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
179 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
180 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
181 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
182 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
183 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
184 <li><a href="#int_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
187 <li><a href="#int_libc">Standard C Library Intrinsics</a>
189 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
190 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
191 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
192 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
193 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
194 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
195 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
196 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
199 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
201 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
202 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
203 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
204 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
205 <li><a href="#int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic </a></li>
206 <li><a href="#int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic </a></li>
209 <li><a href="#int_debugger">Debugger intrinsics</a></li>
210 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
211 <li><a href="#int_trampoline">Trampoline Intrinsic</a>
213 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
216 <li><a href="#int_atomics">Atomic intrinsics</a>
218 <li><a href="#int_memory_barrier"><tt>llvm.memory_barrier</tt></a></li>
219 <li><a href="#int_atomic_lcs"><tt>llvm.atomic.lcs</tt></a></li>
220 <li><a href="#int_atomic_las"><tt>llvm.atomic.las</tt></a></li>
221 <li><a href="#int_atomic_swap"><tt>llvm.atomic.swap</tt></a></li>
224 <li><a href="#int_general">General intrinsics</a>
226 <li><a href="#int_var_annotation">
227 <tt>llvm.var.annotation</tt>' Intrinsic</a></li>
228 <li><a href="#int_annotation">
229 <tt>llvm.annotation.*</tt>' Intrinsic</a></li>
230 <li><a href="#int_trap">
231 <tt>llvm.trap</tt>' Intrinsic</a></li>
238 <div class="doc_author">
239 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
240 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
243 <!-- *********************************************************************** -->
244 <div class="doc_section"> <a name="abstract">Abstract </a></div>
245 <!-- *********************************************************************** -->
247 <div class="doc_text">
248 <p>This document is a reference manual for the LLVM assembly language.
249 LLVM is an SSA based representation that provides type safety,
250 low-level operations, flexibility, and the capability of representing
251 'all' high-level languages cleanly. It is the common code
252 representation used throughout all phases of the LLVM compilation
256 <!-- *********************************************************************** -->
257 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
258 <!-- *********************************************************************** -->
260 <div class="doc_text">
262 <p>The LLVM code representation is designed to be used in three
263 different forms: as an in-memory compiler IR, as an on-disk bitcode
264 representation (suitable for fast loading by a Just-In-Time compiler),
265 and as a human readable assembly language representation. This allows
266 LLVM to provide a powerful intermediate representation for efficient
267 compiler transformations and analysis, while providing a natural means
268 to debug and visualize the transformations. The three different forms
269 of LLVM are all equivalent. This document describes the human readable
270 representation and notation.</p>
272 <p>The LLVM representation aims to be light-weight and low-level
273 while being expressive, typed, and extensible at the same time. It
274 aims to be a "universal IR" of sorts, by being at a low enough level
275 that high-level ideas may be cleanly mapped to it (similar to how
276 microprocessors are "universal IR's", allowing many source languages to
277 be mapped to them). By providing type information, LLVM can be used as
278 the target of optimizations: for example, through pointer analysis, it
279 can be proven that a C automatic variable is never accessed outside of
280 the current function... allowing it to be promoted to a simple SSA
281 value instead of a memory location.</p>
285 <!-- _______________________________________________________________________ -->
286 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
288 <div class="doc_text">
290 <p>It is important to note that this document describes 'well formed'
291 LLVM assembly language. There is a difference between what the parser
292 accepts and what is considered 'well formed'. For example, the
293 following instruction is syntactically okay, but not well formed:</p>
295 <div class="doc_code">
297 %x = <a href="#i_add">add</a> i32 1, %x
301 <p>...because the definition of <tt>%x</tt> does not dominate all of
302 its uses. The LLVM infrastructure provides a verification pass that may
303 be used to verify that an LLVM module is well formed. This pass is
304 automatically run by the parser after parsing input assembly and by
305 the optimizer before it outputs bitcode. The violations pointed out
306 by the verifier pass indicate bugs in transformation passes or input to
310 <!-- Describe the typesetting conventions here. -->
312 <!-- *********************************************************************** -->
313 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
314 <!-- *********************************************************************** -->
316 <div class="doc_text">
318 <p>LLVM identifiers come in two basic types: global and local. Global
319 identifiers (functions, global variables) begin with the @ character. Local
320 identifiers (register names, types) begin with the % character. Additionally,
321 there are three different formats for identifiers, for different purposes:
324 <li>Named values are represented as a string of characters with their prefix.
325 For example, %foo, @DivisionByZero, %a.really.long.identifier. The actual
326 regular expression used is '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
327 Identifiers which require other characters in their names can be surrounded
328 with quotes. In this way, anything except a <tt>"</tt> character can
329 be used in a named value.</li>
331 <li>Unnamed values are represented as an unsigned numeric value with their
332 prefix. For example, %12, @2, %44.</li>
334 <li>Constants, which are described in a <a href="#constants">section about
335 constants</a>, below.</li>
338 <p>LLVM requires that values start with a prefix for two reasons: Compilers
339 don't need to worry about name clashes with reserved words, and the set of
340 reserved words may be expanded in the future without penalty. Additionally,
341 unnamed identifiers allow a compiler to quickly come up with a temporary
342 variable without having to avoid symbol table conflicts.</p>
344 <p>Reserved words in LLVM are very similar to reserved words in other
345 languages. There are keywords for different opcodes
346 ('<tt><a href="#i_add">add</a></tt>',
347 '<tt><a href="#i_bitcast">bitcast</a></tt>',
348 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
349 href="#t_void">void</a></tt>', '<tt><a href="#t_primitive">i32</a></tt>', etc...),
350 and others. These reserved words cannot conflict with variable names, because
351 none of them start with a prefix character ('%' or '@').</p>
353 <p>Here is an example of LLVM code to multiply the integer variable
354 '<tt>%X</tt>' by 8:</p>
358 <div class="doc_code">
360 %result = <a href="#i_mul">mul</a> i32 %X, 8
364 <p>After strength reduction:</p>
366 <div class="doc_code">
368 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
372 <p>And the hard way:</p>
374 <div class="doc_code">
376 <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
377 <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
378 %result = <a href="#i_add">add</a> i32 %1, %1
382 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
383 important lexical features of LLVM:</p>
387 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
390 <li>Unnamed temporaries are created when the result of a computation is not
391 assigned to a named value.</li>
393 <li>Unnamed temporaries are numbered sequentially</li>
397 <p>...and it also shows a convention that we follow in this document. When
398 demonstrating instructions, we will follow an instruction with a comment that
399 defines the type and name of value produced. Comments are shown in italic
404 <!-- *********************************************************************** -->
405 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
406 <!-- *********************************************************************** -->
408 <!-- ======================================================================= -->
409 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
412 <div class="doc_text">
414 <p>LLVM programs are composed of "Module"s, each of which is a
415 translation unit of the input programs. Each module consists of
416 functions, global variables, and symbol table entries. Modules may be
417 combined together with the LLVM linker, which merges function (and
418 global variable) definitions, resolves forward declarations, and merges
419 symbol table entries. Here is an example of the "hello world" module:</p>
421 <div class="doc_code">
422 <pre><i>; Declare the string constant as a global constant...</i>
423 <a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a
424 href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
426 <i>; External declaration of the puts function</i>
427 <a href="#functionstructure">declare</a> i32 @puts(i8 *) <i>; i32(i8 *)* </i>
429 <i>; Definition of main function</i>
430 define i32 @main() { <i>; i32()* </i>
431 <i>; Convert [13x i8 ]* to i8 *...</i>
433 href="#i_getelementptr">getelementptr</a> [13 x i8 ]* @.LC0, i64 0, i64 0 <i>; i8 *</i>
435 <i>; Call puts function to write out the string to stdout...</i>
437 href="#i_call">call</a> i32 @puts(i8 * %cast210) <i>; i32</i>
439 href="#i_ret">ret</a> i32 0<br>}<br>
443 <p>This example is made up of a <a href="#globalvars">global variable</a>
444 named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
445 function, and a <a href="#functionstructure">function definition</a>
446 for "<tt>main</tt>".</p>
448 <p>In general, a module is made up of a list of global values,
449 where both functions and global variables are global values. Global values are
450 represented by a pointer to a memory location (in this case, a pointer to an
451 array of char, and a pointer to a function), and have one of the following <a
452 href="#linkage">linkage types</a>.</p>
456 <!-- ======================================================================= -->
457 <div class="doc_subsection">
458 <a name="linkage">Linkage Types</a>
461 <div class="doc_text">
464 All Global Variables and Functions have one of the following types of linkage:
469 <dt><tt><b><a name="linkage_internal">internal</a></b></tt> </dt>
471 <dd>Global values with internal linkage are only directly accessible by
472 objects in the current module. In particular, linking code into a module with
473 an internal global value may cause the internal to be renamed as necessary to
474 avoid collisions. Because the symbol is internal to the module, all
475 references can be updated. This corresponds to the notion of the
476 '<tt>static</tt>' keyword in C.
479 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
481 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
482 the same name when linkage occurs. This is typically used to implement
483 inline functions, templates, or other code which must be generated in each
484 translation unit that uses it. Unreferenced <tt>linkonce</tt> globals are
485 allowed to be discarded.
488 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
490 <dd>"<tt>weak</tt>" linkage is exactly the same as <tt>linkonce</tt> linkage,
491 except that unreferenced <tt>weak</tt> globals may not be discarded. This is
492 used for globals that may be emitted in multiple translation units, but that
493 are not guaranteed to be emitted into every translation unit that uses them.
494 One example of this are common globals in C, such as "<tt>int X;</tt>" at
498 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
500 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
501 pointer to array type. When two global variables with appending linkage are
502 linked together, the two global arrays are appended together. This is the
503 LLVM, typesafe, equivalent of having the system linker append together
504 "sections" with identical names when .o files are linked.
507 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt>: </dt>
508 <dd>The semantics of this linkage follow the ELF model: the symbol is weak
509 until linked, if not linked, the symbol becomes null instead of being an
513 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
515 <dd>If none of the above identifiers are used, the global is externally
516 visible, meaning that it participates in linkage and can be used to resolve
517 external symbol references.
522 The next two types of linkage are targeted for Microsoft Windows platform
523 only. They are designed to support importing (exporting) symbols from (to)
528 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt>: </dt>
530 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
531 or variable via a global pointer to a pointer that is set up by the DLL
532 exporting the symbol. On Microsoft Windows targets, the pointer name is
533 formed by combining <code>_imp__</code> and the function or variable name.
536 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt>: </dt>
538 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
539 pointer to a pointer in a DLL, so that it can be referenced with the
540 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
541 name is formed by combining <code>_imp__</code> and the function or variable
547 <p><a name="linkage_external"></a>For example, since the "<tt>.LC0</tt>"
548 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
549 variable and was linked with this one, one of the two would be renamed,
550 preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
551 external (i.e., lacking any linkage declarations), they are accessible
552 outside of the current module.</p>
553 <p>It is illegal for a function <i>declaration</i>
554 to have any linkage type other than "externally visible", <tt>dllimport</tt>,
555 or <tt>extern_weak</tt>.</p>
556 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt> and <tt>weak</tt>
560 <!-- ======================================================================= -->
561 <div class="doc_subsection">
562 <a name="callingconv">Calling Conventions</a>
565 <div class="doc_text">
567 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
568 and <a href="#i_invoke">invokes</a> can all have an optional calling convention
569 specified for the call. The calling convention of any pair of dynamic
570 caller/callee must match, or the behavior of the program is undefined. The
571 following calling conventions are supported by LLVM, and more may be added in
575 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
577 <dd>This calling convention (the default if no other calling convention is
578 specified) matches the target C calling conventions. This calling convention
579 supports varargs function calls and tolerates some mismatch in the declared
580 prototype and implemented declaration of the function (as does normal C).
583 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
585 <dd>This calling convention attempts to make calls as fast as possible
586 (e.g. by passing things in registers). This calling convention allows the
587 target to use whatever tricks it wants to produce fast code for the target,
588 without having to conform to an externally specified ABI. Implementations of
589 this convention should allow arbitrary
590 <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> to be
591 supported. This calling convention does not support varargs and requires the
592 prototype of all callees to exactly match the prototype of the function
596 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
598 <dd>This calling convention attempts to make code in the caller as efficient
599 as possible under the assumption that the call is not commonly executed. As
600 such, these calls often preserve all registers so that the call does not break
601 any live ranges in the caller side. This calling convention does not support
602 varargs and requires the prototype of all callees to exactly match the
603 prototype of the function definition.
606 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
608 <dd>Any calling convention may be specified by number, allowing
609 target-specific calling conventions to be used. Target specific calling
610 conventions start at 64.
614 <p>More calling conventions can be added/defined on an as-needed basis, to
615 support pascal conventions or any other well-known target-independent
620 <!-- ======================================================================= -->
621 <div class="doc_subsection">
622 <a name="visibility">Visibility Styles</a>
625 <div class="doc_text">
628 All Global Variables and Functions have one of the following visibility styles:
632 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
634 <dd>On ELF, default visibility means that the declaration is visible to other
635 modules and, in shared libraries, means that the declared entity may be
636 overridden. On Darwin, default visibility means that the declaration is
637 visible to other modules. Default visibility corresponds to "external
638 linkage" in the language.
641 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
643 <dd>Two declarations of an object with hidden visibility refer to the same
644 object if they are in the same shared object. Usually, hidden visibility
645 indicates that the symbol will not be placed into the dynamic symbol table,
646 so no other module (executable or shared library) can reference it
650 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
652 <dd>On ELF, protected visibility indicates that the symbol will be placed in
653 the dynamic symbol table, but that references within the defining module will
654 bind to the local symbol. That is, the symbol cannot be overridden by another
661 <!-- ======================================================================= -->
662 <div class="doc_subsection">
663 <a name="globalvars">Global Variables</a>
666 <div class="doc_text">
668 <p>Global variables define regions of memory allocated at compilation time
669 instead of run-time. Global variables may optionally be initialized, may have
670 an explicit section to be placed in, and may have an optional explicit alignment
671 specified. A variable may be defined as "thread_local", which means that it
672 will not be shared by threads (each thread will have a separated copy of the
673 variable). A variable may be defined as a global "constant," which indicates
674 that the contents of the variable will <b>never</b> be modified (enabling better
675 optimization, allowing the global data to be placed in the read-only section of
676 an executable, etc). Note that variables that need runtime initialization
677 cannot be marked "constant" as there is a store to the variable.</p>
680 LLVM explicitly allows <em>declarations</em> of global variables to be marked
681 constant, even if the final definition of the global is not. This capability
682 can be used to enable slightly better optimization of the program, but requires
683 the language definition to guarantee that optimizations based on the
684 'constantness' are valid for the translation units that do not include the
688 <p>As SSA values, global variables define pointer values that are in
689 scope (i.e. they dominate) all basic blocks in the program. Global
690 variables always define a pointer to their "content" type because they
691 describe a region of memory, and all memory objects in LLVM are
692 accessed through pointers.</p>
694 <p>A global variable may be declared to reside in a target-specifc numbered
695 address space. For targets that support them, address spaces may affect how
696 optimizations are performed and/or what target instructions are used to access
697 the variable. The default address space is zero. The address space qualifier
698 must precede any other attributes.</p>
700 <p>LLVM allows an explicit section to be specified for globals. If the target
701 supports it, it will emit globals to the section specified.</p>
703 <p>An explicit alignment may be specified for a global. If not present, or if
704 the alignment is set to zero, the alignment of the global is set by the target
705 to whatever it feels convenient. If an explicit alignment is specified, the
706 global is forced to have at least that much alignment. All alignments must be
709 <p>For example, the following defines a global in a numbered address space with
710 an initializer, section, and alignment:</p>
712 <div class="doc_code">
714 @G = constant float 1.0 addrspace(5), section "foo", align 4
721 <!-- ======================================================================= -->
722 <div class="doc_subsection">
723 <a name="functionstructure">Functions</a>
726 <div class="doc_text">
728 <p>LLVM function definitions consist of the "<tt>define</tt>" keyord,
729 an optional <a href="#linkage">linkage type</a>, an optional
730 <a href="#visibility">visibility style</a>, an optional
731 <a href="#callingconv">calling convention</a>, a return type, an optional
732 <a href="#paramattrs">parameter attribute</a> for the return type, a function
733 name, a (possibly empty) argument list (each with optional
734 <a href="#paramattrs">parameter attributes</a>), an optional section, an
735 optional alignment, an optional <a href="#gc">garbage collector name</a>, an
736 opening curly brace, a list of basic blocks, and a closing curly brace.
738 LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
739 optional <a href="#linkage">linkage type</a>, an optional
740 <a href="#visibility">visibility style</a>, an optional
741 <a href="#callingconv">calling convention</a>, a return type, an optional
742 <a href="#paramattrs">parameter attribute</a> for the return type, a function
743 name, a possibly empty list of arguments, an optional alignment, and an optional
744 <a href="#gc">garbage collector name</a>.</p>
746 <p>A function definition contains a list of basic blocks, forming the CFG for
747 the function. Each basic block may optionally start with a label (giving the
748 basic block a symbol table entry), contains a list of instructions, and ends
749 with a <a href="#terminators">terminator</a> instruction (such as a branch or
750 function return).</p>
752 <p>The first basic block in a function is special in two ways: it is immediately
753 executed on entrance to the function, and it is not allowed to have predecessor
754 basic blocks (i.e. there can not be any branches to the entry block of a
755 function). Because the block can have no predecessors, it also cannot have any
756 <a href="#i_phi">PHI nodes</a>.</p>
758 <p>LLVM allows an explicit section to be specified for functions. If the target
759 supports it, it will emit functions to the section specified.</p>
761 <p>An explicit alignment may be specified for a function. If not present, or if
762 the alignment is set to zero, the alignment of the function is set by the target
763 to whatever it feels convenient. If an explicit alignment is specified, the
764 function is forced to have at least that much alignment. All alignments must be
770 <!-- ======================================================================= -->
771 <div class="doc_subsection">
772 <a name="aliasstructure">Aliases</a>
774 <div class="doc_text">
775 <p>Aliases act as "second name" for the aliasee value (which can be either
776 function, global variable, another alias or bitcast of global value). Aliases
777 may have an optional <a href="#linkage">linkage type</a>, and an
778 optional <a href="#visibility">visibility style</a>.</p>
782 <div class="doc_code">
784 @<Name> = [Linkage] [Visibility] alias <AliaseeTy> @<Aliasee>
792 <!-- ======================================================================= -->
793 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
794 <div class="doc_text">
795 <p>The return type and each parameter of a function type may have a set of
796 <i>parameter attributes</i> associated with them. Parameter attributes are
797 used to communicate additional information about the result or parameters of
798 a function. Parameter attributes are considered to be part of the function,
799 not of the function type, so functions with different parameter attributes
800 can have the same function type.</p>
802 <p>Parameter attributes are simple keywords that follow the type specified. If
803 multiple parameter attributes are needed, they are space separated. For
806 <div class="doc_code">
808 declare i32 @printf(i8* noalias , ...) nounwind
809 declare i32 @atoi(i8*) nounwind readonly
813 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
814 <tt>readonly</tt>) come immediately after the argument list.</p>
816 <p>Currently, only the following parameter attributes are defined:</p>
818 <dt><tt>zeroext</tt></dt>
819 <dd>This indicates that the parameter should be zero extended just before
820 a call to this function.</dd>
822 <dt><tt>signext</tt></dt>
823 <dd>This indicates that the parameter should be sign extended just before
824 a call to this function.</dd>
826 <dt><tt>inreg</tt></dt>
827 <dd>This indicates that the parameter should be placed in register (if
828 possible) during assembling function call. Support for this attribute is
831 <dt><tt>byval</tt></dt>
832 <dd>This indicates that the pointer parameter should really be passed by
833 value to the function. The attribute implies that a hidden copy of the
834 pointee is made between the caller and the callee, so the callee is unable
835 to modify the value in the callee. This attribute is only valid on llvm
836 pointer arguments. It is generally used to pass structs and arrays by
837 value, but is also valid on scalars (even though this is silly).</dd>
839 <dt><tt>sret</tt></dt>
840 <dd>This indicates that the pointer parameter specifies the address of a
841 structure that is the return value of the function in the source program.
842 Loads and stores to the structure are assumed not to trap.
843 May only be applied to the first parameter.</dd>
845 <dt><tt>noalias</tt></dt>
846 <dd>This indicates that the parameter does not alias any global or any other
847 parameter. The caller is responsible for ensuring that this is the case,
848 usually by placing the value in a stack allocation.</dd>
850 <dt><tt>noreturn</tt></dt>
851 <dd>This function attribute indicates that the function never returns. This
852 indicates to LLVM that every call to this function should be treated as if
853 an <tt>unreachable</tt> instruction immediately followed the call.</dd>
855 <dt><tt>nounwind</tt></dt>
856 <dd>This function attribute indicates that no exceptions unwind out of the
857 function. Usually this is because the function makes no use of exceptions,
858 but it may also be that the function catches any exceptions thrown when
861 <dt><tt>nest</tt></dt>
862 <dd>This indicates that the parameter can be excised using the
863 <a href="#int_trampoline">trampoline intrinsics</a>.</dd>
864 <dt><tt>readonly</tt></dt>
865 <dd>This function attribute indicates that the function has no side-effects
866 except for producing a return value or throwing an exception. The value
867 returned must only depend on the function arguments and/or global variables.
868 It may use values obtained by dereferencing pointers.</dd>
869 <dt><tt>readnone</tt></dt>
870 <dd>A <tt>readnone</tt> function has the same restrictions as a <tt>readonly</tt>
871 function, but in addition it is not allowed to dereference any pointer arguments
877 <!-- ======================================================================= -->
878 <div class="doc_subsection">
879 <a name="gc">Garbage Collector Names</a>
882 <div class="doc_text">
883 <p>Each function may specify a garbage collector name, which is simply a
886 <div class="doc_code"><pre
887 >define void @f() gc "name" { ...</pre></div>
889 <p>The compiler declares the supported values of <i>name</i>. Specifying a
890 collector which will cause the compiler to alter its output in order to support
891 the named garbage collection algorithm.</p>
894 <!-- ======================================================================= -->
895 <div class="doc_subsection">
896 <a name="moduleasm">Module-Level Inline Assembly</a>
899 <div class="doc_text">
901 Modules may contain "module-level inline asm" blocks, which corresponds to the
902 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
903 LLVM and treated as a single unit, but may be separated in the .ll file if
904 desired. The syntax is very simple:
907 <div class="doc_code">
909 module asm "inline asm code goes here"
910 module asm "more can go here"
914 <p>The strings can contain any character by escaping non-printable characters.
915 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
920 The inline asm code is simply printed to the machine code .s file when
921 assembly code is generated.
925 <!-- ======================================================================= -->
926 <div class="doc_subsection">
927 <a name="datalayout">Data Layout</a>
930 <div class="doc_text">
931 <p>A module may specify a target specific data layout string that specifies how
932 data is to be laid out in memory. The syntax for the data layout is simply:</p>
933 <pre> target datalayout = "<i>layout specification</i>"</pre>
934 <p>The <i>layout specification</i> consists of a list of specifications
935 separated by the minus sign character ('-'). Each specification starts with a
936 letter and may include other information after the letter to define some
937 aspect of the data layout. The specifications accepted are as follows: </p>
940 <dd>Specifies that the target lays out data in big-endian form. That is, the
941 bits with the most significance have the lowest address location.</dd>
943 <dd>Specifies that hte target lays out data in little-endian form. That is,
944 the bits with the least significance have the lowest address location.</dd>
945 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
946 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
947 <i>preferred</i> alignments. All sizes are in bits. Specifying the <i>pref</i>
948 alignment is optional. If omitted, the preceding <tt>:</tt> should be omitted
950 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
951 <dd>This specifies the alignment for an integer type of a given bit
952 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
953 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
954 <dd>This specifies the alignment for a vector type of a given bit
956 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
957 <dd>This specifies the alignment for a floating point type of a given bit
958 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
960 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
961 <dd>This specifies the alignment for an aggregate type of a given bit
964 <p>When constructing the data layout for a given target, LLVM starts with a
965 default set of specifications which are then (possibly) overriden by the
966 specifications in the <tt>datalayout</tt> keyword. The default specifications
967 are given in this list:</p>
969 <li><tt>E</tt> - big endian</li>
970 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
971 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
972 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
973 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
974 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
975 <li><tt>i64:32:64</tt> - i64 has abi alignment of 32-bits but preferred
976 alignment of 64-bits</li>
977 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
978 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
979 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
980 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
981 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
983 <p>When llvm is determining the alignment for a given type, it uses the
986 <li>If the type sought is an exact match for one of the specifications, that
987 specification is used.</li>
988 <li>If no match is found, and the type sought is an integer type, then the
989 smallest integer type that is larger than the bitwidth of the sought type is
990 used. If none of the specifications are larger than the bitwidth then the the
991 largest integer type is used. For example, given the default specifications
992 above, the i7 type will use the alignment of i8 (next largest) while both
993 i65 and i256 will use the alignment of i64 (largest specified).</li>
994 <li>If no match is found, and the type sought is a vector type, then the
995 largest vector type that is smaller than the sought vector type will be used
996 as a fall back. This happens because <128 x double> can be implemented in
997 terms of 64 <2 x double>, for example.</li>
1001 <!-- *********************************************************************** -->
1002 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
1003 <!-- *********************************************************************** -->
1005 <div class="doc_text">
1007 <p>The LLVM type system is one of the most important features of the
1008 intermediate representation. Being typed enables a number of
1009 optimizations to be performed on the IR directly, without having to do
1010 extra analyses on the side before the transformation. A strong type
1011 system makes it easier to read the generated code and enables novel
1012 analyses and transformations that are not feasible to perform on normal
1013 three address code representations.</p>
1017 <!-- ======================================================================= -->
1018 <div class="doc_subsection"> <a name="t_classifications">Type
1019 Classifications</a> </div>
1020 <div class="doc_text">
1021 <p>The types fall into a few useful
1022 classifications:</p>
1024 <table border="1" cellspacing="0" cellpadding="4">
1026 <tr><th>Classification</th><th>Types</th></tr>
1028 <td><a href="#t_integer">integer</a></td>
1029 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1032 <td><a href="#t_floating">floating point</a></td>
1033 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1036 <td><a name="t_firstclass">first class</a></td>
1037 <td><a href="#t_integer">integer</a>,
1038 <a href="#t_floating">floating point</a>,
1039 <a href="#t_pointer">pointer</a>,
1040 <a href="#t_vector">vector</a>
1041 <a href="#t_struct">structure</a>,
1042 <a href="#t_array">array</a>,
1046 <td><a href="#t_primitive">primitive</a></td>
1047 <td><a href="#t_label">label</a>,
1048 <a href="#t_void">void</a>,
1049 <a href="#t_integer">integer</a>,
1050 <a href="#t_floating">floating point</a>.</td>
1053 <td><a href="#t_derived">derived</a></td>
1054 <td><a href="#t_integer">integer</a>,
1055 <a href="#t_array">array</a>,
1056 <a href="#t_function">function</a>,
1057 <a href="#t_pointer">pointer</a>,
1058 <a href="#t_struct">structure</a>,
1059 <a href="#t_pstruct">packed structure</a>,
1060 <a href="#t_vector">vector</a>,
1061 <a href="#t_opaque">opaque</a>.
1066 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
1067 most important. Values of these types are the only ones which can be
1068 produced by instructions, passed as arguments, or used as operands to
1069 instructions. This means that all structures and arrays must be
1070 manipulated either by pointer or by component.</p>
1073 <!-- ======================================================================= -->
1074 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
1076 <div class="doc_text">
1077 <p>The primitive types are the fundamental building blocks of the LLVM
1082 <!-- _______________________________________________________________________ -->
1083 <div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1085 <div class="doc_text">
1088 <tr><th>Type</th><th>Description</th></tr>
1089 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1090 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1091 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1092 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1093 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1098 <!-- _______________________________________________________________________ -->
1099 <div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1101 <div class="doc_text">
1103 <p>The void type does not represent any value and has no size.</p>
1112 <!-- _______________________________________________________________________ -->
1113 <div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1115 <div class="doc_text">
1117 <p>The label type represents code labels.</p>
1127 <!-- ======================================================================= -->
1128 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1130 <div class="doc_text">
1132 <p>The real power in LLVM comes from the derived types in the system.
1133 This is what allows a programmer to represent arrays, functions,
1134 pointers, and other useful types. Note that these derived types may be
1135 recursive: For example, it is possible to have a two dimensional array.</p>
1139 <!-- _______________________________________________________________________ -->
1140 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1142 <div class="doc_text">
1145 <p>The integer type is a very simple derived type that simply specifies an
1146 arbitrary bit width for the integer type desired. Any bit width from 1 bit to
1147 2^23-1 (about 8 million) can be specified.</p>
1155 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1159 <table class="layout">
1162 <td><tt>i1</tt></td>
1163 <td>a single-bit integer.</td>
1165 <td><tt>i32</tt></td>
1166 <td>a 32-bit integer.</td>
1168 <td><tt>i1942652</tt></td>
1169 <td>a really big integer of over 1 million bits.</td>
1175 <!-- _______________________________________________________________________ -->
1176 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1178 <div class="doc_text">
1182 <p>The array type is a very simple derived type that arranges elements
1183 sequentially in memory. The array type requires a size (number of
1184 elements) and an underlying data type.</p>
1189 [<# elements> x <elementtype>]
1192 <p>The number of elements is a constant integer value; elementtype may
1193 be any type with a size.</p>
1196 <table class="layout">
1198 <td class="left"><tt>[40 x i32]</tt></td>
1199 <td class="left">Array of 40 32-bit integer values.</td>
1202 <td class="left"><tt>[41 x i32]</tt></td>
1203 <td class="left">Array of 41 32-bit integer values.</td>
1206 <td class="left"><tt>[4 x i8]</tt></td>
1207 <td class="left">Array of 4 8-bit integer values.</td>
1210 <p>Here are some examples of multidimensional arrays:</p>
1211 <table class="layout">
1213 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1214 <td class="left">3x4 array of 32-bit integer values.</td>
1217 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1218 <td class="left">12x10 array of single precision floating point values.</td>
1221 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1222 <td class="left">2x3x4 array of 16-bit integer values.</td>
1226 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1227 length array. Normally, accesses past the end of an array are undefined in
1228 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
1229 As a special case, however, zero length arrays are recognized to be variable
1230 length. This allows implementation of 'pascal style arrays' with the LLVM
1231 type "{ i32, [0 x float]}", for example.</p>
1235 <!-- _______________________________________________________________________ -->
1236 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1237 <div class="doc_text">
1241 <p>The function type can be thought of as a function signature. It
1242 consists of a return type and a list of formal parameter types. The
1243 return type of a function type is a scalar type, a void type, or a struct type.
1244 If the return type is a struct type then all struct elements must be of first
1245 class types, and the struct must have at least one element.</p>
1250 <returntype list> (<parameter list>)
1253 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1254 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1255 which indicates that the function takes a variable number of arguments.
1256 Variable argument functions can access their arguments with the <a
1257 href="#int_varargs">variable argument handling intrinsic</a> functions.
1258 '<tt><returntype list></tt>' is a comma-separated list of
1259 <a href="#t_firstclass">first class</a> type specifiers.</p>
1262 <table class="layout">
1264 <td class="left"><tt>i32 (i32)</tt></td>
1265 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1267 </tr><tr class="layout">
1268 <td class="left"><tt>float (i16 signext, i32 *) *
1270 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1271 an <tt>i16</tt> that should be sign extended and a
1272 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1275 </tr><tr class="layout">
1276 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1277 <td class="left">A vararg function that takes at least one
1278 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1279 which returns an integer. This is the signature for <tt>printf</tt> in
1282 </tr><tr class="layout">
1283 <td class="left"><tt>{i32, i32} (i32)</tt></td>
1284 <td class="left">A function taking an <tt>i32></tt>, returning two
1285 <tt> i32 </tt> values as an aggregate of type <tt>{ i32, i32 }</tt>
1291 <!-- _______________________________________________________________________ -->
1292 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1293 <div class="doc_text">
1295 <p>The structure type is used to represent a collection of data members
1296 together in memory. The packing of the field types is defined to match
1297 the ABI of the underlying processor. The elements of a structure may
1298 be any type that has a size.</p>
1299 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1300 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1301 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1304 <pre> { <type list> }<br></pre>
1306 <table class="layout">
1308 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1309 <td class="left">A triple of three <tt>i32</tt> values</td>
1310 </tr><tr class="layout">
1311 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1312 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1313 second element is a <a href="#t_pointer">pointer</a> to a
1314 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1315 an <tt>i32</tt>.</td>
1320 <!-- _______________________________________________________________________ -->
1321 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1323 <div class="doc_text">
1325 <p>The packed structure type is used to represent a collection of data members
1326 together in memory. There is no padding between fields. Further, the alignment
1327 of a packed structure is 1 byte. The elements of a packed structure may
1328 be any type that has a size.</p>
1329 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1330 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1331 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1334 <pre> < { <type list> } > <br></pre>
1336 <table class="layout">
1338 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1339 <td class="left">A triple of three <tt>i32</tt> values</td>
1340 </tr><tr class="layout">
1341 <td class="left"><tt>< { float, i32 (i32)* } ></tt></td>
1342 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1343 second element is a <a href="#t_pointer">pointer</a> to a
1344 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1345 an <tt>i32</tt>.</td>
1350 <!-- _______________________________________________________________________ -->
1351 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1352 <div class="doc_text">
1354 <p>As in many languages, the pointer type represents a pointer or
1355 reference to another object, which must live in memory. Pointer types may have
1356 an optional address space attribute defining the target-specific numbered
1357 address space where the pointed-to object resides. The default address space is
1360 <pre> <type> *<br></pre>
1362 <table class="layout">
1364 <td class="left"><tt>[4x i32]*</tt></td>
1365 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1366 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1369 <td class="left"><tt>i32 (i32 *) *</tt></td>
1370 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
1371 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1375 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1376 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1377 that resides in address space #5.</td>
1382 <!-- _______________________________________________________________________ -->
1383 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1384 <div class="doc_text">
1388 <p>A vector type is a simple derived type that represents a vector
1389 of elements. Vector types are used when multiple primitive data
1390 are operated in parallel using a single instruction (SIMD).
1391 A vector type requires a size (number of
1392 elements) and an underlying primitive data type. Vectors must have a power
1393 of two length (1, 2, 4, 8, 16 ...). Vector types are
1394 considered <a href="#t_firstclass">first class</a>.</p>
1399 < <# elements> x <elementtype> >
1402 <p>The number of elements is a constant integer value; elementtype may
1403 be any integer or floating point type.</p>
1407 <table class="layout">
1409 <td class="left"><tt><4 x i32></tt></td>
1410 <td class="left">Vector of 4 32-bit integer values.</td>
1413 <td class="left"><tt><8 x float></tt></td>
1414 <td class="left">Vector of 8 32-bit floating-point values.</td>
1417 <td class="left"><tt><2 x i64></tt></td>
1418 <td class="left">Vector of 2 64-bit integer values.</td>
1423 <!-- _______________________________________________________________________ -->
1424 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1425 <div class="doc_text">
1429 <p>Opaque types are used to represent unknown types in the system. This
1430 corresponds (for example) to the C notion of a forward declared structure type.
1431 In LLVM, opaque types can eventually be resolved to any type (not just a
1432 structure type).</p>
1442 <table class="layout">
1444 <td class="left"><tt>opaque</tt></td>
1445 <td class="left">An opaque type.</td>
1451 <!-- *********************************************************************** -->
1452 <div class="doc_section"> <a name="constants">Constants</a> </div>
1453 <!-- *********************************************************************** -->
1455 <div class="doc_text">
1457 <p>LLVM has several different basic types of constants. This section describes
1458 them all and their syntax.</p>
1462 <!-- ======================================================================= -->
1463 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1465 <div class="doc_text">
1468 <dt><b>Boolean constants</b></dt>
1470 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1471 constants of the <tt><a href="#t_primitive">i1</a></tt> type.
1474 <dt><b>Integer constants</b></dt>
1476 <dd>Standard integers (such as '4') are constants of the <a
1477 href="#t_integer">integer</a> type. Negative numbers may be used with
1481 <dt><b>Floating point constants</b></dt>
1483 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1484 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1485 notation (see below). The assembler requires the exact decimal value of
1486 a floating-point constant. For example, the assembler accepts 1.25 but
1487 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
1488 constants must have a <a href="#t_floating">floating point</a> type. </dd>
1490 <dt><b>Null pointer constants</b></dt>
1492 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1493 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1497 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1498 of floating point constants. For example, the form '<tt>double
1499 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1500 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1501 (and the only time that they are generated by the disassembler) is when a
1502 floating point constant must be emitted but it cannot be represented as a
1503 decimal floating point number. For example, NaN's, infinities, and other
1504 special values are represented in their IEEE hexadecimal format so that
1505 assembly and disassembly do not cause any bits to change in the constants.</p>
1509 <!-- ======================================================================= -->
1510 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1513 <div class="doc_text">
1514 <p>Aggregate constants arise from aggregation of simple constants
1515 and smaller aggregate constants.</p>
1518 <dt><b>Structure constants</b></dt>
1520 <dd>Structure constants are represented with notation similar to structure
1521 type definitions (a comma separated list of elements, surrounded by braces
1522 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
1523 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>". Structure constants
1524 must have <a href="#t_struct">structure type</a>, and the number and
1525 types of elements must match those specified by the type.
1528 <dt><b>Array constants</b></dt>
1530 <dd>Array constants are represented with notation similar to array type
1531 definitions (a comma separated list of elements, surrounded by square brackets
1532 (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74 ]</tt>". Array
1533 constants must have <a href="#t_array">array type</a>, and the number and
1534 types of elements must match those specified by the type.
1537 <dt><b>Vector constants</b></dt>
1539 <dd>Vector constants are represented with notation similar to vector type
1540 definitions (a comma separated list of elements, surrounded by
1541 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32 42,
1542 i32 11, i32 74, i32 100 ></tt>". Vector constants must have <a
1543 href="#t_vector">vector type</a>, and the number and types of elements must
1544 match those specified by the type.
1547 <dt><b>Zero initialization</b></dt>
1549 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1550 value to zero of <em>any</em> type, including scalar and aggregate types.
1551 This is often used to avoid having to print large zero initializers (e.g. for
1552 large arrays) and is always exactly equivalent to using explicit zero
1559 <!-- ======================================================================= -->
1560 <div class="doc_subsection">
1561 <a name="globalconstants">Global Variable and Function Addresses</a>
1564 <div class="doc_text">
1566 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1567 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1568 constants. These constants are explicitly referenced when the <a
1569 href="#identifiers">identifier for the global</a> is used and always have <a
1570 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1573 <div class="doc_code">
1577 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1583 <!-- ======================================================================= -->
1584 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1585 <div class="doc_text">
1586 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1587 no specific value. Undefined values may be of any type and be used anywhere
1588 a constant is permitted.</p>
1590 <p>Undefined values indicate to the compiler that the program is well defined
1591 no matter what value is used, giving the compiler more freedom to optimize.
1595 <!-- ======================================================================= -->
1596 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1599 <div class="doc_text">
1601 <p>Constant expressions are used to allow expressions involving other constants
1602 to be used as constants. Constant expressions may be of any <a
1603 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1604 that does not have side effects (e.g. load and call are not supported). The
1605 following is the syntax for constant expressions:</p>
1608 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1609 <dd>Truncate a constant to another type. The bit size of CST must be larger
1610 than the bit size of TYPE. Both types must be integers.</dd>
1612 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1613 <dd>Zero extend a constant to another type. The bit size of CST must be
1614 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1616 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1617 <dd>Sign extend a constant to another type. The bit size of CST must be
1618 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1620 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1621 <dd>Truncate a floating point constant to another floating point type. The
1622 size of CST must be larger than the size of TYPE. Both types must be
1623 floating point.</dd>
1625 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1626 <dd>Floating point extend a constant to another type. The size of CST must be
1627 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1629 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
1630 <dd>Convert a floating point constant to the corresponding unsigned integer
1631 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1632 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1633 of the same number of elements. If the value won't fit in the integer type,
1634 the results are undefined.</dd>
1636 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
1637 <dd>Convert a floating point constant to the corresponding signed integer
1638 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1639 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1640 of the same number of elements. If the value won't fit in the integer type,
1641 the results are undefined.</dd>
1643 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
1644 <dd>Convert an unsigned integer constant to the corresponding floating point
1645 constant. TYPE must be a scalar or vector floating point type. CST must be of
1646 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1647 of the same number of elements. If the value won't fit in the floating point
1648 type, the results are undefined.</dd>
1650 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
1651 <dd>Convert a signed integer constant to the corresponding floating point
1652 constant. TYPE must be a scalar or vector floating point type. CST must be of
1653 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1654 of the same number of elements. If the value won't fit in the floating point
1655 type, the results are undefined.</dd>
1657 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
1658 <dd>Convert a pointer typed constant to the corresponding integer constant
1659 TYPE must be an integer type. CST must be of pointer type. The CST value is
1660 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1662 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
1663 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1664 pointer type. CST must be of integer type. The CST value is zero extended,
1665 truncated, or unchanged to make it fit in a pointer size. This one is
1666 <i>really</i> dangerous!</dd>
1668 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
1669 <dd>Convert a constant, CST, to another TYPE. The size of CST and TYPE must be
1670 identical (same number of bits). The conversion is done as if the CST value
1671 was stored to memory and read back as TYPE. In other words, no bits change
1672 with this operator, just the type. This can be used for conversion of
1673 vector types to any other type, as long as they have the same bit width. For
1674 pointers it is only valid to cast to another pointer type.
1677 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1679 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1680 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1681 instruction, the index list may have zero or more indexes, which are required
1682 to make sense for the type of "CSTPTR".</dd>
1684 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1686 <dd>Perform the <a href="#i_select">select operation</a> on
1689 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
1690 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
1692 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
1693 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
1695 <dt><b><tt>vicmp COND ( VAL1, VAL2 )</tt></b></dt>
1696 <dd>Performs the <a href="#i_vicmp">vicmp operation</a> on constants.</dd>
1698 <dt><b><tt>vfcmp COND ( VAL1, VAL2 )</tt></b></dt>
1699 <dd>Performs the <a href="#i_vfcmp">vfcmp operation</a> on constants.</dd>
1701 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1703 <dd>Perform the <a href="#i_extractelement">extractelement
1704 operation</a> on constants.
1706 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1708 <dd>Perform the <a href="#i_insertelement">insertelement
1709 operation</a> on constants.</dd>
1712 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
1714 <dd>Perform the <a href="#i_shufflevector">shufflevector
1715 operation</a> on constants.</dd>
1717 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1719 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1720 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1721 binary</a> operations. The constraints on operands are the same as those for
1722 the corresponding instruction (e.g. no bitwise operations on floating point
1723 values are allowed).</dd>
1727 <!-- *********************************************************************** -->
1728 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1729 <!-- *********************************************************************** -->
1731 <!-- ======================================================================= -->
1732 <div class="doc_subsection">
1733 <a name="inlineasm">Inline Assembler Expressions</a>
1736 <div class="doc_text">
1739 LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
1740 Module-Level Inline Assembly</a>) through the use of a special value. This
1741 value represents the inline assembler as a string (containing the instructions
1742 to emit), a list of operand constraints (stored as a string), and a flag that
1743 indicates whether or not the inline asm expression has side effects. An example
1744 inline assembler expression is:
1747 <div class="doc_code">
1749 i32 (i32) asm "bswap $0", "=r,r"
1754 Inline assembler expressions may <b>only</b> be used as the callee operand of
1755 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
1758 <div class="doc_code">
1760 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
1765 Inline asms with side effects not visible in the constraint list must be marked
1766 as having side effects. This is done through the use of the
1767 '<tt>sideeffect</tt>' keyword, like so:
1770 <div class="doc_code">
1772 call void asm sideeffect "eieio", ""()
1776 <p>TODO: The format of the asm and constraints string still need to be
1777 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1778 need to be documented).
1783 <!-- *********************************************************************** -->
1784 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1785 <!-- *********************************************************************** -->
1787 <div class="doc_text">
1789 <p>The LLVM instruction set consists of several different
1790 classifications of instructions: <a href="#terminators">terminator
1791 instructions</a>, <a href="#binaryops">binary instructions</a>,
1792 <a href="#bitwiseops">bitwise binary instructions</a>, <a
1793 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1794 instructions</a>.</p>
1798 <!-- ======================================================================= -->
1799 <div class="doc_subsection"> <a name="terminators">Terminator
1800 Instructions</a> </div>
1802 <div class="doc_text">
1804 <p>As mentioned <a href="#functionstructure">previously</a>, every
1805 basic block in a program ends with a "Terminator" instruction, which
1806 indicates which block should be executed after the current block is
1807 finished. These terminator instructions typically yield a '<tt>void</tt>'
1808 value: they produce control flow, not values (the one exception being
1809 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1810 <p>There are six different terminator instructions: the '<a
1811 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1812 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1813 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1814 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1815 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1819 <!-- _______________________________________________________________________ -->
1820 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1821 Instruction</a> </div>
1822 <div class="doc_text">
1824 <pre> ret <type> <value> <i>; Return a value from a non-void function</i>
1825 ret void <i>; Return from void function</i>
1826 ret <type> <value>, <type> <value> <i>; Return two values from a non-void function </i>
1831 <p>The '<tt>ret</tt>' instruction is used to return control flow (and a
1832 value) from a function back to the caller.</p>
1833 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
1834 returns value(s) and then causes control flow, and one that just causes
1835 control flow to occur.</p>
1839 <p>The '<tt>ret</tt>' instruction may return zero, one or multiple values.
1840 The type of each return value must be a '<a href="#t_firstclass">first
1841 class</a>' type. Note that a function is not <a href="#wellformed">well
1842 formed</a> if there exists a '<tt>ret</tt>' instruction inside of the
1843 function that returns values that do not match the return type of the
1848 <p>When the '<tt>ret</tt>' instruction is executed, control flow
1849 returns back to the calling function's context. If the caller is a "<a
1850 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1851 the instruction after the call. If the caller was an "<a
1852 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1853 at the beginning of the "normal" destination block. If the instruction
1854 returns a value, that value shall set the call or invoke instruction's
1855 return value. If the instruction returns multiple values then these
1856 values can only be accessed through a '<a href="#i_getresult"><tt>getresult</tt>
1857 </a>' instruction.</p>
1862 ret i32 5 <i>; Return an integer value of 5</i>
1863 ret void <i>; Return from a void function</i>
1864 ret i32 4, i8 2 <i>; Return two values 4 and 2 </i>
1867 <!-- _______________________________________________________________________ -->
1868 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1869 <div class="doc_text">
1871 <pre> br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1874 <p>The '<tt>br</tt>' instruction is used to cause control flow to
1875 transfer to a different basic block in the current function. There are
1876 two forms of this instruction, corresponding to a conditional branch
1877 and an unconditional branch.</p>
1879 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1880 single '<tt>i1</tt>' value and two '<tt>label</tt>' values. The
1881 unconditional form of the '<tt>br</tt>' instruction takes a single
1882 '<tt>label</tt>' value as a target.</p>
1884 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
1885 argument is evaluated. If the value is <tt>true</tt>, control flows
1886 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1887 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1889 <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
1890 href="#i_ret">ret</a> i32 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> i32 0<br></pre>
1892 <!-- _______________________________________________________________________ -->
1893 <div class="doc_subsubsection">
1894 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
1897 <div class="doc_text">
1901 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
1906 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
1907 several different places. It is a generalization of the '<tt>br</tt>'
1908 instruction, allowing a branch to occur to one of many possible
1914 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
1915 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
1916 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
1917 table is not allowed to contain duplicate constant entries.</p>
1921 <p>The <tt>switch</tt> instruction specifies a table of values and
1922 destinations. When the '<tt>switch</tt>' instruction is executed, this
1923 table is searched for the given value. If the value is found, control flow is
1924 transfered to the corresponding destination; otherwise, control flow is
1925 transfered to the default destination.</p>
1927 <h5>Implementation:</h5>
1929 <p>Depending on properties of the target machine and the particular
1930 <tt>switch</tt> instruction, this instruction may be code generated in different
1931 ways. For example, it could be generated as a series of chained conditional
1932 branches or with a lookup table.</p>
1937 <i>; Emulate a conditional br instruction</i>
1938 %Val = <a href="#i_zext">zext</a> i1 %value to i32
1939 switch i32 %Val, label %truedest [i32 0, label %falsedest ]
1941 <i>; Emulate an unconditional br instruction</i>
1942 switch i32 0, label %dest [ ]
1944 <i>; Implement a jump table:</i>
1945 switch i32 %val, label %otherwise [ i32 0, label %onzero
1947 i32 2, label %ontwo ]
1951 <!-- _______________________________________________________________________ -->
1952 <div class="doc_subsubsection">
1953 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
1956 <div class="doc_text">
1961 <result> = invoke [<a href="#callingconv">cconv</a>] <ptr to function ty> <function ptr val>(<function args>)
1962 to label <normal label> unwind label <exception label>
1967 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
1968 function, with the possibility of control flow transfer to either the
1969 '<tt>normal</tt>' label or the
1970 '<tt>exception</tt>' label. If the callee function returns with the
1971 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
1972 "normal" label. If the callee (or any indirect callees) returns with the "<a
1973 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
1974 continued at the dynamically nearest "exception" label. If the callee function
1975 returns multiple values then individual return values are only accessible through
1976 a '<tt><a href="#i_getresult">getresult</a></tt>' instruction.</p>
1980 <p>This instruction requires several arguments:</p>
1984 The optional "cconv" marker indicates which <a href="#callingconv">calling
1985 convention</a> the call should use. If none is specified, the call defaults
1986 to using C calling conventions.
1988 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
1989 function value being invoked. In most cases, this is a direct function
1990 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
1991 an arbitrary pointer to function value.
1994 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
1995 function to be invoked. </li>
1997 <li>'<tt>function args</tt>': argument list whose types match the function
1998 signature argument types. If the function signature indicates the function
1999 accepts a variable number of arguments, the extra arguments can be
2002 <li>'<tt>normal label</tt>': the label reached when the called function
2003 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
2005 <li>'<tt>exception label</tt>': the label reached when a callee returns with
2006 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
2012 <p>This instruction is designed to operate as a standard '<tt><a
2013 href="#i_call">call</a></tt>' instruction in most regards. The primary
2014 difference is that it establishes an association with a label, which is used by
2015 the runtime library to unwind the stack.</p>
2017 <p>This instruction is used in languages with destructors to ensure that proper
2018 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
2019 exception. Additionally, this is important for implementation of
2020 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
2024 %retval = invoke i32 @Test(i32 15) to label %Continue
2025 unwind label %TestCleanup <i>; {i32}:retval set</i>
2026 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
2027 unwind label %TestCleanup <i>; {i32}:retval set</i>
2032 <!-- _______________________________________________________________________ -->
2034 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
2035 Instruction</a> </div>
2037 <div class="doc_text">
2046 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
2047 at the first callee in the dynamic call stack which used an <a
2048 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
2049 primarily used to implement exception handling.</p>
2053 <p>The '<tt>unwind</tt>' instruction causes execution of the current function to
2054 immediately halt. The dynamic call stack is then searched for the first <a
2055 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
2056 execution continues at the "exceptional" destination block specified by the
2057 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
2058 dynamic call chain, undefined behavior results.</p>
2061 <!-- _______________________________________________________________________ -->
2063 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
2064 Instruction</a> </div>
2066 <div class="doc_text">
2075 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
2076 instruction is used to inform the optimizer that a particular portion of the
2077 code is not reachable. This can be used to indicate that the code after a
2078 no-return function cannot be reached, and other facts.</p>
2082 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
2087 <!-- ======================================================================= -->
2088 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
2089 <div class="doc_text">
2090 <p>Binary operators are used to do most of the computation in a
2091 program. They require two operands of the same type, execute an operation on them, and
2092 produce a single value. The operands might represent
2093 multiple data, as is the case with the <a href="#t_vector">vector</a> data type.
2094 The result value has the same type as its operands.</p>
2095 <p>There are several different binary operators:</p>
2097 <!-- _______________________________________________________________________ -->
2098 <div class="doc_subsubsection"> <a name="i_add">'<tt>add</tt>'
2099 Instruction</a> </div>
2100 <div class="doc_text">
2102 <pre> <result> = add <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2105 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
2107 <p>The two arguments to the '<tt>add</tt>' instruction must be either <a
2108 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a> values.
2109 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2110 Both arguments must have identical types.</p>
2112 <p>The value produced is the integer or floating point sum of the two
2114 <p>If an integer sum has unsigned overflow, the result returned is the
2115 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2117 <p>Because LLVM integers use a two's complement representation, this
2118 instruction is appropriate for both signed and unsigned integers.</p>
2120 <pre> <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
2123 <!-- _______________________________________________________________________ -->
2124 <div class="doc_subsubsection"> <a name="i_sub">'<tt>sub</tt>'
2125 Instruction</a> </div>
2126 <div class="doc_text">
2128 <pre> <result> = sub <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2131 <p>The '<tt>sub</tt>' instruction returns the difference of its two
2133 <p>Note that the '<tt>sub</tt>' instruction is used to represent the '<tt>neg</tt>'
2134 instruction present in most other intermediate representations.</p>
2136 <p>The two arguments to the '<tt>sub</tt>' instruction must be either <a
2137 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
2139 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2140 Both arguments must have identical types.</p>
2142 <p>The value produced is the integer or floating point difference of
2143 the two operands.</p>
2144 <p>If an integer difference has unsigned overflow, the result returned is the
2145 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2147 <p>Because LLVM integers use a two's complement representation, this
2148 instruction is appropriate for both signed and unsigned integers.</p>
2151 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
2152 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
2155 <!-- _______________________________________________________________________ -->
2156 <div class="doc_subsubsection"> <a name="i_mul">'<tt>mul</tt>'
2157 Instruction</a> </div>
2158 <div class="doc_text">
2160 <pre> <result> = mul <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2163 <p>The '<tt>mul</tt>' instruction returns the product of its two
2166 <p>The two arguments to the '<tt>mul</tt>' instruction must be either <a
2167 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
2169 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2170 Both arguments must have identical types.</p>
2172 <p>The value produced is the integer or floating point product of the
2174 <p>If the result of an integer multiplication has unsigned overflow,
2175 the result returned is the mathematical result modulo
2176 2<sup>n</sup>, where n is the bit width of the result.</p>
2177 <p>Because LLVM integers use a two's complement representation, and the
2178 result is the same width as the operands, this instruction returns the
2179 correct result for both signed and unsigned integers. If a full product
2180 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands
2181 should be sign-extended or zero-extended as appropriate to the
2182 width of the full product.</p>
2184 <pre> <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
2187 <!-- _______________________________________________________________________ -->
2188 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
2190 <div class="doc_text">
2192 <pre> <result> = udiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2195 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two
2198 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
2199 <a href="#t_integer">integer</a> values. Both arguments must have identical
2200 types. This instruction can also take <a href="#t_vector">vector</a> versions
2201 of the values in which case the elements must be integers.</p>
2203 <p>The value produced is the unsigned integer quotient of the two operands.</p>
2204 <p>Note that unsigned integer division and signed integer division are distinct
2205 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
2206 <p>Division by zero leads to undefined behavior.</p>
2208 <pre> <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2211 <!-- _______________________________________________________________________ -->
2212 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
2214 <div class="doc_text">
2216 <pre> <result> = sdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2219 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
2222 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
2223 <a href="#t_integer">integer</a> values. Both arguments must have identical
2224 types. This instruction can also take <a href="#t_vector">vector</a> versions
2225 of the values in which case the elements must be integers.</p>
2227 <p>The value produced is the signed integer quotient of the two operands rounded towards zero.</p>
2228 <p>Note that signed integer division and unsigned integer division are distinct
2229 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
2230 <p>Division by zero leads to undefined behavior. Overflow also leads to
2231 undefined behavior; this is a rare case, but can occur, for example,
2232 by doing a 32-bit division of -2147483648 by -1.</p>
2234 <pre> <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2237 <!-- _______________________________________________________________________ -->
2238 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
2239 Instruction</a> </div>
2240 <div class="doc_text">
2242 <pre> <result> = fdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2245 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
2248 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
2249 <a href="#t_floating">floating point</a> values. Both arguments must have
2250 identical types. This instruction can also take <a href="#t_vector">vector</a>
2251 versions of floating point values.</p>
2253 <p>The value produced is the floating point quotient of the two operands.</p>
2255 <pre> <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
2258 <!-- _______________________________________________________________________ -->
2259 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
2261 <div class="doc_text">
2263 <pre> <result> = urem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2266 <p>The '<tt>urem</tt>' instruction returns the remainder from the
2267 unsigned division of its two arguments.</p>
2269 <p>The two arguments to the '<tt>urem</tt>' instruction must be
2270 <a href="#t_integer">integer</a> values. Both arguments must have identical
2271 types. This instruction can also take <a href="#t_vector">vector</a> versions
2272 of the values in which case the elements must be integers.</p>
2274 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
2275 This instruction always performs an unsigned division to get the remainder.</p>
2276 <p>Note that unsigned integer remainder and signed integer remainder are
2277 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
2278 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
2280 <pre> <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2284 <!-- _______________________________________________________________________ -->
2285 <div class="doc_subsubsection"> <a name="i_srem">'<tt>srem</tt>'
2286 Instruction</a> </div>
2287 <div class="doc_text">
2289 <pre> <result> = srem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2292 <p>The '<tt>srem</tt>' instruction returns the remainder from the
2293 signed division of its two operands. This instruction can also take
2294 <a href="#t_vector">vector</a> versions of the values in which case
2295 the elements must be integers.</p>
2298 <p>The two arguments to the '<tt>srem</tt>' instruction must be
2299 <a href="#t_integer">integer</a> values. Both arguments must have identical
2302 <p>This instruction returns the <i>remainder</i> of a division (where the result
2303 has the same sign as the dividend, <tt>var1</tt>), not the <i>modulo</i>
2304 operator (where the result has the same sign as the divisor, <tt>var2</tt>) of
2305 a value. For more information about the difference, see <a
2306 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
2307 Math Forum</a>. For a table of how this is implemented in various languages,
2308 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
2309 Wikipedia: modulo operation</a>.</p>
2310 <p>Note that signed integer remainder and unsigned integer remainder are
2311 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
2312 <p>Taking the remainder of a division by zero leads to undefined behavior.
2313 Overflow also leads to undefined behavior; this is a rare case, but can occur,
2314 for example, by taking the remainder of a 32-bit division of -2147483648 by -1.
2315 (The remainder doesn't actually overflow, but this rule lets srem be
2316 implemented using instructions that return both the result of the division
2317 and the remainder.)</p>
2319 <pre> <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2323 <!-- _______________________________________________________________________ -->
2324 <div class="doc_subsubsection"> <a name="i_frem">'<tt>frem</tt>'
2325 Instruction</a> </div>
2326 <div class="doc_text">
2328 <pre> <result> = frem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2331 <p>The '<tt>frem</tt>' instruction returns the remainder from the
2332 division of its two operands.</p>
2334 <p>The two arguments to the '<tt>frem</tt>' instruction must be
2335 <a href="#t_floating">floating point</a> values. Both arguments must have
2336 identical types. This instruction can also take <a href="#t_vector">vector</a>
2337 versions of floating point values.</p>
2339 <p>This instruction returns the <i>remainder</i> of a division.
2340 The remainder has the same sign as the dividend.</p>
2342 <pre> <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
2346 <!-- ======================================================================= -->
2347 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
2348 Operations</a> </div>
2349 <div class="doc_text">
2350 <p>Bitwise binary operators are used to do various forms of
2351 bit-twiddling in a program. They are generally very efficient
2352 instructions and can commonly be strength reduced from other
2353 instructions. They require two operands of the same type, execute an operation on them,
2354 and produce a single value. The resulting value is the same type as its operands.</p>
2357 <!-- _______________________________________________________________________ -->
2358 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
2359 Instruction</a> </div>
2360 <div class="doc_text">
2362 <pre> <result> = shl <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2367 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2368 the left a specified number of bits.</p>
2372 <p>Both arguments to the '<tt>shl</tt>' instruction must be the same <a
2373 href="#t_integer">integer</a> type. '<tt>var2</tt>' is treated as an
2378 <p>The value produced is <tt>var1</tt> * 2<sup><tt>var2</tt></sup> mod 2<sup>n</sup>,
2379 where n is the width of the result. If <tt>var2</tt> is (statically or dynamically) negative or
2380 equal to or larger than the number of bits in <tt>var1</tt>, the result is undefined.</p>
2382 <h5>Example:</h5><pre>
2383 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
2384 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
2385 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
2386 <result> = shl i32 1, 32 <i>; undefined</i>
2389 <!-- _______________________________________________________________________ -->
2390 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
2391 Instruction</a> </div>
2392 <div class="doc_text">
2394 <pre> <result> = lshr <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2398 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
2399 operand shifted to the right a specified number of bits with zero fill.</p>
2402 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
2403 <a href="#t_integer">integer</a> type. '<tt>var2</tt>' is treated as an
2408 <p>This instruction always performs a logical shift right operation. The most
2409 significant bits of the result will be filled with zero bits after the
2410 shift. If <tt>var2</tt> is (statically or dynamically) equal to or larger than
2411 the number of bits in <tt>var1</tt>, the result is undefined.</p>
2415 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
2416 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
2417 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
2418 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
2419 <result> = lshr i32 1, 32 <i>; undefined</i>
2423 <!-- _______________________________________________________________________ -->
2424 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
2425 Instruction</a> </div>
2426 <div class="doc_text">
2429 <pre> <result> = ashr <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2433 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
2434 operand shifted to the right a specified number of bits with sign extension.</p>
2437 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
2438 <a href="#t_integer">integer</a> type. '<tt>var2</tt>' is treated as an
2442 <p>This instruction always performs an arithmetic shift right operation,
2443 The most significant bits of the result will be filled with the sign bit
2444 of <tt>var1</tt>. If <tt>var2</tt> is (statically or dynamically) equal to or
2445 larger than the number of bits in <tt>var1</tt>, the result is undefined.
2450 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
2451 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
2452 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
2453 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
2454 <result> = ashr i32 1, 32 <i>; undefined</i>
2458 <!-- _______________________________________________________________________ -->
2459 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
2460 Instruction</a> </div>
2461 <div class="doc_text">
2463 <pre> <result> = and <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2466 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
2467 its two operands.</p>
2469 <p>The two arguments to the '<tt>and</tt>' instruction must be <a
2470 href="#t_integer">integer</a> values. Both arguments must have
2471 identical types.</p>
2473 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
2475 <div style="align: center">
2476 <table border="1" cellspacing="0" cellpadding="4">
2507 <pre> <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
2508 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
2509 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
2512 <!-- _______________________________________________________________________ -->
2513 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
2514 <div class="doc_text">
2516 <pre> <result> = or <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2519 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
2520 or of its two operands.</p>
2522 <p>The two arguments to the '<tt>or</tt>' instruction must be <a
2523 href="#t_integer">integer</a> values. Both arguments must have
2524 identical types.</p>
2526 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
2528 <div style="align: center">
2529 <table border="1" cellspacing="0" cellpadding="4">
2560 <pre> <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
2561 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
2562 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
2565 <!-- _______________________________________________________________________ -->
2566 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
2567 Instruction</a> </div>
2568 <div class="doc_text">
2570 <pre> <result> = xor <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2573 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
2574 or of its two operands. The <tt>xor</tt> is used to implement the
2575 "one's complement" operation, which is the "~" operator in C.</p>
2577 <p>The two arguments to the '<tt>xor</tt>' instruction must be <a
2578 href="#t_integer">integer</a> values. Both arguments must have
2579 identical types.</p>
2581 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
2583 <div style="align: center">
2584 <table border="1" cellspacing="0" cellpadding="4">
2616 <pre> <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
2617 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
2618 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
2619 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
2623 <!-- ======================================================================= -->
2624 <div class="doc_subsection">
2625 <a name="vectorops">Vector Operations</a>
2628 <div class="doc_text">
2630 <p>LLVM supports several instructions to represent vector operations in a
2631 target-independent manner. These instructions cover the element-access and
2632 vector-specific operations needed to process vectors effectively. While LLVM
2633 does directly support these vector operations, many sophisticated algorithms
2634 will want to use target-specific intrinsics to take full advantage of a specific
2639 <!-- _______________________________________________________________________ -->
2640 <div class="doc_subsubsection">
2641 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2644 <div class="doc_text">
2649 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
2655 The '<tt>extractelement</tt>' instruction extracts a single scalar
2656 element from a vector at a specified index.
2663 The first operand of an '<tt>extractelement</tt>' instruction is a
2664 value of <a href="#t_vector">vector</a> type. The second operand is
2665 an index indicating the position from which to extract the element.
2666 The index may be a variable.</p>
2671 The result is a scalar of the same type as the element type of
2672 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2673 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2674 results are undefined.
2680 %result = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
2685 <!-- _______________________________________________________________________ -->
2686 <div class="doc_subsubsection">
2687 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
2690 <div class="doc_text">
2695 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
2701 The '<tt>insertelement</tt>' instruction inserts a scalar
2702 element into a vector at a specified index.
2709 The first operand of an '<tt>insertelement</tt>' instruction is a
2710 value of <a href="#t_vector">vector</a> type. The second operand is a
2711 scalar value whose type must equal the element type of the first
2712 operand. The third operand is an index indicating the position at
2713 which to insert the value. The index may be a variable.</p>
2718 The result is a vector of the same type as <tt>val</tt>. Its
2719 element values are those of <tt>val</tt> except at position
2720 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
2721 exceeds the length of <tt>val</tt>, the results are undefined.
2727 %result = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
2731 <!-- _______________________________________________________________________ -->
2732 <div class="doc_subsubsection">
2733 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
2736 <div class="doc_text">
2741 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <n x i32> <mask> <i>; yields <n x <ty>></i>
2747 The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
2748 from two input vectors, returning a vector of the same type.
2754 The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
2755 with types that match each other and types that match the result of the
2756 instruction. The third argument is a shuffle mask, which has the same number
2757 of elements as the other vector type, but whose element type is always 'i32'.
2761 The shuffle mask operand is required to be a constant vector with either
2762 constant integer or undef values.
2768 The elements of the two input vectors are numbered from left to right across
2769 both of the vectors. The shuffle mask operand specifies, for each element of
2770 the result vector, which element of the two input registers the result element
2771 gets. The element selector may be undef (meaning "don't care") and the second
2772 operand may be undef if performing a shuffle from only one vector.
2778 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
2779 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
2780 %result = shufflevector <4 x i32> %v1, <4 x i32> undef,
2781 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
2786 <!-- ======================================================================= -->
2787 <div class="doc_subsection">
2788 <a name="aggregateops">Aggregate Operations</a>
2791 <div class="doc_text">
2793 <p>LLVM supports several instructions for working with aggregate values.
2798 <!-- _______________________________________________________________________ -->
2799 <div class="doc_subsubsection">
2800 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
2803 <div class="doc_text">
2808 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
2814 The '<tt>extractvalue</tt>' instruction extracts the value of a struct field
2815 or array element from an aggregate value.
2822 The first operand of an '<tt>extractvalue</tt>' instruction is a
2823 value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a>
2824 type. The operands are constant indices to specify which value to extract
2825 in the same manner as indices in a
2826 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
2832 The result is the value at the position in the aggregate specified by
2839 %result = extractvalue {i32, float} %agg, i32 0 <i>; yields i32</i>
2844 <!-- _______________________________________________________________________ -->
2845 <div class="doc_subsubsection">
2846 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
2849 <div class="doc_text">
2854 <result> = insertvalue <aggregate type> <val>, <ty> <val>, i32 <idx> <i>; yields <n x <ty>></i>
2860 The '<tt>insertvalue</tt>' instruction inserts a value
2861 into a struct field or array element in an aggregate.
2868 The first operand of an '<tt>insertvalue</tt>' instruction is a
2869 value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a> type.
2870 The second operand is a first-class value to insert.
2871 type of the first operand. The following operands are constant indices
2872 indicating the position at which to insert the value in the same manner as
2874 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
2875 The value to insert must have the same type as the value identified
2881 The result is an aggregate of the same type as <tt>val</tt>. Its
2882 value is that of <tt>val</tt> except that the value at the position
2883 specified by the indices is that of <tt>elt</tt>.
2889 %result = insertvalue {i32, float} %agg, i32 1, i32 0 <i>; yields {i32, float}</i>
2894 <!-- ======================================================================= -->
2895 <div class="doc_subsection">
2896 <a name="memoryops">Memory Access and Addressing Operations</a>
2899 <div class="doc_text">
2901 <p>A key design point of an SSA-based representation is how it
2902 represents memory. In LLVM, no memory locations are in SSA form, which
2903 makes things very simple. This section describes how to read, write,
2904 allocate, and free memory in LLVM.</p>
2908 <!-- _______________________________________________________________________ -->
2909 <div class="doc_subsubsection">
2910 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
2913 <div class="doc_text">
2918 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2923 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
2924 heap and returns a pointer to it. The object is always allocated in the generic
2925 address space (address space zero).</p>
2929 <p>The '<tt>malloc</tt>' instruction allocates
2930 <tt>sizeof(<type>)*NumElements</tt>
2931 bytes of memory from the operating system and returns a pointer of the
2932 appropriate type to the program. If "NumElements" is specified, it is the
2933 number of elements allocated, otherwise "NumElements" is defaulted to be one.
2934 If a constant alignment is specified, the value result of the allocation is guaranteed to
2935 be aligned to at least that boundary. If not specified, or if zero, the target can
2936 choose to align the allocation on any convenient boundary.</p>
2938 <p>'<tt>type</tt>' must be a sized type.</p>
2942 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
2943 a pointer is returned. The result of a zero byte allocattion is undefined. The
2944 result is null if there is insufficient memory available.</p>
2949 %array = malloc [4 x i8 ] <i>; yields {[%4 x i8]*}:array</i>
2951 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
2952 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
2953 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
2954 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
2955 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
2959 <!-- _______________________________________________________________________ -->
2960 <div class="doc_subsubsection">
2961 <a name="i_free">'<tt>free</tt>' Instruction</a>
2964 <div class="doc_text">
2969 free <type> <value> <i>; yields {void}</i>
2974 <p>The '<tt>free</tt>' instruction returns memory back to the unused
2975 memory heap to be reallocated in the future.</p>
2979 <p>'<tt>value</tt>' shall be a pointer value that points to a value
2980 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
2985 <p>Access to the memory pointed to by the pointer is no longer defined
2986 after this instruction executes. If the pointer is null, the operation
2992 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
2993 free [4 x i8]* %array
2997 <!-- _______________________________________________________________________ -->
2998 <div class="doc_subsubsection">
2999 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
3002 <div class="doc_text">
3007 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3012 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
3013 currently executing function, to be automatically released when this function
3014 returns to its caller. The object is always allocated in the generic address
3015 space (address space zero).</p>
3019 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
3020 bytes of memory on the runtime stack, returning a pointer of the
3021 appropriate type to the program. If "NumElements" is specified, it is the
3022 number of elements allocated, otherwise "NumElements" is defaulted to be one.
3023 If a constant alignment is specified, the value result of the allocation is guaranteed
3024 to be aligned to at least that boundary. If not specified, or if zero, the target
3025 can choose to align the allocation on any convenient boundary.</p>
3027 <p>'<tt>type</tt>' may be any sized type.</p>
3031 <p>Memory is allocated; a pointer is returned. The operation is undefiend if
3032 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
3033 memory is automatically released when the function returns. The '<tt>alloca</tt>'
3034 instruction is commonly used to represent automatic variables that must
3035 have an address available. When the function returns (either with the <tt><a
3036 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
3037 instructions), the memory is reclaimed. Allocating zero bytes
3038 is legal, but the result is undefined.</p>
3043 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
3044 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
3045 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
3046 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
3050 <!-- _______________________________________________________________________ -->
3051 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
3052 Instruction</a> </div>
3053 <div class="doc_text">
3055 <pre> <result> = load <ty>* <pointer>[, align <alignment>]<br> <result> = volatile load <ty>* <pointer>[, align <alignment>]<br></pre>
3057 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
3059 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
3060 address from which to load. The pointer must point to a <a
3061 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
3062 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
3063 the number or order of execution of this <tt>load</tt> with other
3064 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
3067 The optional constant "align" argument specifies the alignment of the operation
3068 (that is, the alignment of the memory address). A value of 0 or an
3069 omitted "align" argument means that the operation has the preferential
3070 alignment for the target. It is the responsibility of the code emitter
3071 to ensure that the alignment information is correct. Overestimating
3072 the alignment results in an undefined behavior. Underestimating the
3073 alignment may produce less efficient code. An alignment of 1 is always
3077 <p>The location of memory pointed to is loaded.</p>
3079 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3081 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
3082 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
3085 <!-- _______________________________________________________________________ -->
3086 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
3087 Instruction</a> </div>
3088 <div class="doc_text">
3090 <pre> store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3091 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3094 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
3096 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
3097 to store and an address at which to store it. The type of the '<tt><pointer></tt>'
3098 operand must be a pointer to the <a href="#t_firstclass">first class</a> type
3099 of the '<tt><value></tt>'
3100 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
3101 optimizer is not allowed to modify the number or order of execution of
3102 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
3103 href="#i_store">store</a></tt> instructions.</p>
3105 The optional constant "align" argument specifies the alignment of the operation
3106 (that is, the alignment of the memory address). A value of 0 or an
3107 omitted "align" argument means that the operation has the preferential
3108 alignment for the target. It is the responsibility of the code emitter
3109 to ensure that the alignment information is correct. Overestimating
3110 the alignment results in an undefined behavior. Underestimating the
3111 alignment may produce less efficient code. An alignment of 1 is always
3115 <p>The contents of memory are updated to contain '<tt><value></tt>'
3116 at the location specified by the '<tt><pointer></tt>' operand.</p>
3118 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3119 store i32 3, i32* %ptr <i>; yields {void}</i>
3120 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
3124 <!-- _______________________________________________________________________ -->
3125 <div class="doc_subsubsection">
3126 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
3129 <div class="doc_text">
3132 <result> = getelementptr <ty>* <ptrval>{, <ty> <idx>}*
3138 The '<tt>getelementptr</tt>' instruction is used to get the address of a
3139 subelement of an aggregate data structure.</p>
3143 <p>This instruction takes a list of integer operands that indicate what
3144 elements of the aggregate object to index to. The actual types of the arguments
3145 provided depend on the type of the first pointer argument. The
3146 '<tt>getelementptr</tt>' instruction is used to index down through the type
3147 levels of a structure or to a specific index in an array. When indexing into a
3148 structure, only <tt>i32</tt> integer constants are allowed. When indexing
3149 into an array or pointer, only integers of 32 or 64 bits are allowed; 32-bit
3150 values will be sign extended to 64-bits if required.</p>
3152 <p>For example, let's consider a C code fragment and how it gets
3153 compiled to LLVM:</p>
3155 <div class="doc_code">
3168 int *foo(struct ST *s) {
3169 return &s[1].Z.B[5][13];
3174 <p>The LLVM code generated by the GCC frontend is:</p>
3176 <div class="doc_code">
3178 %RT = type { i8 , [10 x [20 x i32]], i8 }
3179 %ST = type { i32, double, %RT }
3181 define i32* %foo(%ST* %s) {
3183 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
3191 <p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
3192 on the pointer type that is being indexed into. <a href="#t_pointer">Pointer</a>
3193 and <a href="#t_array">array</a> types can use a 32-bit or 64-bit
3194 <a href="#t_integer">integer</a> type but the value will always be sign extended
3195 to 64-bits. <a href="#t_struct">Structure</a> and <a href="#t_pstruct">packed
3196 structure</a> types require <tt>i32</tt> <b>constants</b>.</p>
3198 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
3199 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
3200 }</tt>' type, a structure. The second index indexes into the third element of
3201 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
3202 i8 }</tt>' type, another structure. The third index indexes into the second
3203 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
3204 array. The two dimensions of the array are subscripted into, yielding an
3205 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
3206 to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
3208 <p>Note that it is perfectly legal to index partially through a
3209 structure, returning a pointer to an inner element. Because of this,
3210 the LLVM code for the given testcase is equivalent to:</p>
3213 define i32* %foo(%ST* %s) {
3214 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
3215 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
3216 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
3217 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
3218 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
3223 <p>Note that it is undefined to access an array out of bounds: array and
3224 pointer indexes must always be within the defined bounds of the array type.
3225 The one exception for this rule is zero length arrays. These arrays are
3226 defined to be accessible as variable length arrays, which requires access
3227 beyond the zero'th element.</p>
3229 <p>The getelementptr instruction is often confusing. For some more insight
3230 into how it works, see <a href="GetElementPtr.html">the getelementptr
3236 <i>; yields [12 x i8]*:aptr</i>
3237 %aptr = getelementptr {i32, [12 x i8]}* %sptr, i64 0, i32 1
3241 <!-- ======================================================================= -->
3242 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
3244 <div class="doc_text">
3245 <p>The instructions in this category are the conversion instructions (casting)
3246 which all take a single operand and a type. They perform various bit conversions
3250 <!-- _______________________________________________________________________ -->
3251 <div class="doc_subsubsection">
3252 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
3254 <div class="doc_text">
3258 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
3263 The '<tt>trunc</tt>' instruction truncates its operand to the type <tt>ty2</tt>.
3268 The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
3269 be an <a href="#t_integer">integer</a> type, and a type that specifies the size
3270 and type of the result, which must be an <a href="#t_integer">integer</a>
3271 type. The bit size of <tt>value</tt> must be larger than the bit size of
3272 <tt>ty2</tt>. Equal sized types are not allowed.</p>
3276 The '<tt>trunc</tt>' instruction truncates the high order bits in <tt>value</tt>
3277 and converts the remaining bits to <tt>ty2</tt>. Since the source size must be
3278 larger than the destination size, <tt>trunc</tt> cannot be a <i>no-op cast</i>.
3279 It will always truncate bits.</p>
3283 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
3284 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
3285 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
3289 <!-- _______________________________________________________________________ -->
3290 <div class="doc_subsubsection">
3291 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
3293 <div class="doc_text">
3297 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
3301 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
3306 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
3307 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3308 also be of <a href="#t_integer">integer</a> type. The bit size of the
3309 <tt>value</tt> must be smaller than the bit size of the destination type,
3313 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
3314 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
3316 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
3320 %X = zext i32 257 to i64 <i>; yields i64:257</i>
3321 %Y = zext i1 true to i32 <i>; yields i32:1</i>
3325 <!-- _______________________________________________________________________ -->
3326 <div class="doc_subsubsection">
3327 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
3329 <div class="doc_text">
3333 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
3337 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
3341 The '<tt>sext</tt>' instruction takes a value to cast, which must be of
3342 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3343 also be of <a href="#t_integer">integer</a> type. The bit size of the
3344 <tt>value</tt> must be smaller than the bit size of the destination type,
3349 The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
3350 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size of
3351 the type <tt>ty2</tt>.</p>
3353 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
3357 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
3358 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
3362 <!-- _______________________________________________________________________ -->
3363 <div class="doc_subsubsection">
3364 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
3367 <div class="doc_text">
3372 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
3376 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
3381 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
3382 point</a> value to cast and a <a href="#t_floating">floating point</a> type to
3383 cast it to. The size of <tt>value</tt> must be larger than the size of
3384 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
3385 <i>no-op cast</i>.</p>
3388 <p> The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
3389 <a href="#t_floating">floating point</a> type to a smaller
3390 <a href="#t_floating">floating point</a> type. If the value cannot fit within
3391 the destination type, <tt>ty2</tt>, then the results are undefined.</p>
3395 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
3396 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
3400 <!-- _______________________________________________________________________ -->
3401 <div class="doc_subsubsection">
3402 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
3404 <div class="doc_text">
3408 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
3412 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
3413 floating point value.</p>
3416 <p>The '<tt>fpext</tt>' instruction takes a
3417 <a href="#t_floating">floating point</a> <tt>value</tt> to cast,
3418 and a <a href="#t_floating">floating point</a> type to cast it to. The source
3419 type must be smaller than the destination type.</p>
3422 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
3423 <a href="#t_floating">floating point</a> type to a larger
3424 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
3425 used to make a <i>no-op cast</i> because it always changes bits. Use
3426 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
3430 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
3431 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
3435 <!-- _______________________________________________________________________ -->
3436 <div class="doc_subsubsection">
3437 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
3439 <div class="doc_text">
3443 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
3447 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
3448 unsigned integer equivalent of type <tt>ty2</tt>.
3452 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
3453 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3454 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3455 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3456 vector integer type with the same number of elements as <tt>ty</tt></p>
3459 <p> The '<tt>fptoui</tt>' instruction converts its
3460 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3461 towards zero) unsigned integer value. If the value cannot fit in <tt>ty2</tt>,
3462 the results are undefined.</p>
3466 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
3467 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
3468 %X = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
3472 <!-- _______________________________________________________________________ -->
3473 <div class="doc_subsubsection">
3474 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
3476 <div class="doc_text">
3480 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
3484 <p>The '<tt>fptosi</tt>' instruction converts
3485 <a href="#t_floating">floating point</a> <tt>value</tt> to type <tt>ty2</tt>.
3489 <p> The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
3490 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3491 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3492 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3493 vector integer type with the same number of elements as <tt>ty</tt></p>
3496 <p>The '<tt>fptosi</tt>' instruction converts its
3497 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3498 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
3499 the results are undefined.</p>
3503 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
3504 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
3505 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
3509 <!-- _______________________________________________________________________ -->
3510 <div class="doc_subsubsection">
3511 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
3513 <div class="doc_text">
3517 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3521 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
3522 integer and converts that value to the <tt>ty2</tt> type.</p>
3525 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
3526 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3527 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3528 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3529 floating point type with the same number of elements as <tt>ty</tt></p>
3532 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
3533 integer quantity and converts it to the corresponding floating point value. If
3534 the value cannot fit in the floating point value, the results are undefined.</p>
3538 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
3539 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
3543 <!-- _______________________________________________________________________ -->
3544 <div class="doc_subsubsection">
3545 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
3547 <div class="doc_text">
3551 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3555 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed
3556 integer and converts that value to the <tt>ty2</tt> type.</p>
3559 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
3560 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3561 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3562 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3563 floating point type with the same number of elements as <tt>ty</tt></p>
3566 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed
3567 integer quantity and converts it to the corresponding floating point value. If
3568 the value cannot fit in the floating point value, the results are undefined.</p>
3572 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
3573 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
3577 <!-- _______________________________________________________________________ -->
3578 <div class="doc_subsubsection">
3579 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
3581 <div class="doc_text">
3585 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
3589 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
3590 the integer type <tt>ty2</tt>.</p>
3593 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
3594 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
3595 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.
3598 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
3599 <tt>ty2</tt> by interpreting the pointer value as an integer and either
3600 truncating or zero extending that value to the size of the integer type. If
3601 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
3602 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
3603 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
3608 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
3609 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
3613 <!-- _______________________________________________________________________ -->
3614 <div class="doc_subsubsection">
3615 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
3617 <div class="doc_text">
3621 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
3625 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to
3626 a pointer type, <tt>ty2</tt>.</p>
3629 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
3630 value to cast, and a type to cast it to, which must be a
3631 <a href="#t_pointer">pointer</a> type.
3634 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
3635 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
3636 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
3637 size of a pointer then a truncation is done. If <tt>value</tt> is smaller than
3638 the size of a pointer then a zero extension is done. If they are the same size,
3639 nothing is done (<i>no-op cast</i>).</p>
3643 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
3644 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
3645 %Y = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
3649 <!-- _______________________________________________________________________ -->
3650 <div class="doc_subsubsection">
3651 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
3653 <div class="doc_text">
3657 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
3661 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3662 <tt>ty2</tt> without changing any bits.</p>
3665 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be
3666 a first class value, and a type to cast it to, which must also be a <a
3667 href="#t_firstclass">first class</a> type. The bit sizes of <tt>value</tt>
3668 and the destination type, <tt>ty2</tt>, must be identical. If the source
3669 type is a pointer, the destination type must also be a pointer.</p>
3672 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3673 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
3674 this conversion. The conversion is done as if the <tt>value</tt> had been
3675 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only be
3676 converted to other pointer types with this instruction. To convert pointers to
3677 other types, use the <a href="#i_inttoptr">inttoptr</a> or
3678 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
3682 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
3683 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
3684 %Z = bitcast <2xint> %V to i64; <i>; yields i64: %V</i>
3688 <!-- ======================================================================= -->
3689 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
3690 <div class="doc_text">
3691 <p>The instructions in this category are the "miscellaneous"
3692 instructions, which defy better classification.</p>
3695 <!-- _______________________________________________________________________ -->
3696 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
3698 <div class="doc_text">
3700 <pre> <result> = icmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3703 <p>The '<tt>icmp</tt>' instruction returns a boolean value based on comparison
3704 of its two integer or pointer operands.</p>
3706 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
3707 the condition code indicating the kind of comparison to perform. It is not
3708 a value, just a keyword. The possible condition code are:
3710 <li><tt>eq</tt>: equal</li>
3711 <li><tt>ne</tt>: not equal </li>
3712 <li><tt>ugt</tt>: unsigned greater than</li>
3713 <li><tt>uge</tt>: unsigned greater or equal</li>
3714 <li><tt>ult</tt>: unsigned less than</li>
3715 <li><tt>ule</tt>: unsigned less or equal</li>
3716 <li><tt>sgt</tt>: signed greater than</li>
3717 <li><tt>sge</tt>: signed greater or equal</li>
3718 <li><tt>slt</tt>: signed less than</li>
3719 <li><tt>sle</tt>: signed less or equal</li>
3721 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
3722 <a href="#t_pointer">pointer</a> typed. They must also be identical types.</p>
3724 <p>The '<tt>icmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3725 the condition code given as <tt>cond</tt>. The comparison performed always
3726 yields a <a href="#t_primitive">i1</a> result, as follows:
3728 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
3729 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3731 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
3732 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3733 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
3734 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3735 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
3736 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3737 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
3738 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3739 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
3740 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3741 <li><tt>sgt</tt>: interprets the operands as signed values and yields
3742 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3743 <li><tt>sge</tt>: interprets the operands as signed values and yields
3744 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3745 <li><tt>slt</tt>: interprets the operands as signed values and yields
3746 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3747 <li><tt>sle</tt>: interprets the operands as signed values and yields
3748 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3750 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
3751 values are compared as if they were integers.</p>
3754 <pre> <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
3755 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
3756 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
3757 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
3758 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
3759 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
3763 <!-- _______________________________________________________________________ -->
3764 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
3766 <div class="doc_text">
3768 <pre> <result> = fcmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3771 <p>The '<tt>fcmp</tt>' instruction returns a boolean value based on comparison
3772 of its floating point operands.</p>
3774 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
3775 the condition code indicating the kind of comparison to perform. It is not
3776 a value, just a keyword. The possible condition code are:
3778 <li><tt>false</tt>: no comparison, always returns false</li>
3779 <li><tt>oeq</tt>: ordered and equal</li>
3780 <li><tt>ogt</tt>: ordered and greater than </li>
3781 <li><tt>oge</tt>: ordered and greater than or equal</li>
3782 <li><tt>olt</tt>: ordered and less than </li>
3783 <li><tt>ole</tt>: ordered and less than or equal</li>
3784 <li><tt>one</tt>: ordered and not equal</li>
3785 <li><tt>ord</tt>: ordered (no nans)</li>
3786 <li><tt>ueq</tt>: unordered or equal</li>
3787 <li><tt>ugt</tt>: unordered or greater than </li>
3788 <li><tt>uge</tt>: unordered or greater than or equal</li>
3789 <li><tt>ult</tt>: unordered or less than </li>
3790 <li><tt>ule</tt>: unordered or less than or equal</li>
3791 <li><tt>une</tt>: unordered or not equal</li>
3792 <li><tt>uno</tt>: unordered (either nans)</li>
3793 <li><tt>true</tt>: no comparison, always returns true</li>
3795 <p><i>Ordered</i> means that neither operand is a QNAN while
3796 <i>unordered</i> means that either operand may be a QNAN.</p>
3797 <p>The <tt>val1</tt> and <tt>val2</tt> arguments must be
3798 <a href="#t_floating">floating point</a> typed. They must have identical
3801 <p>The '<tt>fcmp</tt>' instruction compares <tt>var1</tt> and <tt>var2</tt>
3802 according to the condition code given as <tt>cond</tt>. The comparison performed
3803 always yields a <a href="#t_primitive">i1</a> result, as follows:
3805 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
3806 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3807 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3808 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3809 <tt>var1</tt> is greather than <tt>var2</tt>.</li>
3810 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3811 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3812 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3813 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3814 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3815 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3816 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3817 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3818 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
3819 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
3820 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3821 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
3822 <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3823 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
3824 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3825 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
3826 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3827 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
3828 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3829 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
3830 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3831 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
3832 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
3836 <pre> <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
3837 <result> = icmp one float 4.0, 5.0 <i>; yields: result=true</i>
3838 <result> = icmp olt float 4.0, 5.0 <i>; yields: result=true</i>
3839 <result> = icmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
3843 <!-- _______________________________________________________________________ -->
3844 <div class="doc_subsubsection">
3845 <a name="i_vicmp">'<tt>vicmp</tt>' Instruction</a>
3847 <div class="doc_text">
3849 <pre> <result> = vicmp <cond> <ty> <var1>, <var2> <i>; yields {ty}:result</i>
3852 <p>The '<tt>vicmp</tt>' instruction returns an integer vector value based on
3853 element-wise comparison of its two integer vector operands.</p>
3855 <p>The '<tt>vicmp</tt>' instruction takes three operands. The first operand is
3856 the condition code indicating the kind of comparison to perform. It is not
3857 a value, just a keyword. The possible condition code are:
3859 <li><tt>eq</tt>: equal</li>
3860 <li><tt>ne</tt>: not equal </li>
3861 <li><tt>ugt</tt>: unsigned greater than</li>
3862 <li><tt>uge</tt>: unsigned greater or equal</li>
3863 <li><tt>ult</tt>: unsigned less than</li>
3864 <li><tt>ule</tt>: unsigned less or equal</li>
3865 <li><tt>sgt</tt>: signed greater than</li>
3866 <li><tt>sge</tt>: signed greater or equal</li>
3867 <li><tt>slt</tt>: signed less than</li>
3868 <li><tt>sle</tt>: signed less or equal</li>
3870 <p>The remaining two arguments must be <a href="#t_vector">vector</a> of
3871 <a href="#t_integer">integer</a> typed. They must also be identical types.</p>
3873 <p>The '<tt>vicmp</tt>' instruction compares <tt>var1</tt> and <tt>var2</tt>
3874 according to the condition code given as <tt>cond</tt>. The comparison yields a
3875 <a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, of
3876 identical type as the values being compared. The most significant bit in each
3877 element is 1 if the element-wise comparison evaluates to true, and is 0
3878 otherwise. All other bits of the result are undefined. The condition codes
3879 are evaluated identically to the <a href="#i_icmp">'<tt>icmp</tt>'
3884 <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>
3885 <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>
3889 <!-- _______________________________________________________________________ -->
3890 <div class="doc_subsubsection">
3891 <a name="i_vfcmp">'<tt>vfcmp</tt>' Instruction</a>
3893 <div class="doc_text">
3895 <pre> <result> = vfcmp <cond> <ty> <var1>, <var2></pre>
3897 <p>The '<tt>vfcmp</tt>' instruction returns an integer vector value based on
3898 element-wise comparison of its two floating point vector operands. The output
3899 elements have the same width as the input elements.</p>
3901 <p>The '<tt>vfcmp</tt>' instruction takes three operands. The first operand is
3902 the condition code indicating the kind of comparison to perform. It is not
3903 a value, just a keyword. The possible condition code are:
3905 <li><tt>false</tt>: no comparison, always returns false</li>
3906 <li><tt>oeq</tt>: ordered and equal</li>
3907 <li><tt>ogt</tt>: ordered and greater than </li>
3908 <li><tt>oge</tt>: ordered and greater than or equal</li>
3909 <li><tt>olt</tt>: ordered and less than </li>
3910 <li><tt>ole</tt>: ordered and less than or equal</li>
3911 <li><tt>one</tt>: ordered and not equal</li>
3912 <li><tt>ord</tt>: ordered (no nans)</li>
3913 <li><tt>ueq</tt>: unordered or equal</li>
3914 <li><tt>ugt</tt>: unordered or greater than </li>
3915 <li><tt>uge</tt>: unordered or greater than or equal</li>
3916 <li><tt>ult</tt>: unordered or less than </li>
3917 <li><tt>ule</tt>: unordered or less than or equal</li>
3918 <li><tt>une</tt>: unordered or not equal</li>
3919 <li><tt>uno</tt>: unordered (either nans)</li>
3920 <li><tt>true</tt>: no comparison, always returns true</li>
3922 <p>The remaining two arguments must be <a href="#t_vector">vector</a> of
3923 <a href="#t_floating">floating point</a> typed. They must also be identical
3926 <p>The '<tt>vfcmp</tt>' instruction compares <tt>var1</tt> and <tt>var2</tt>
3927 according to the condition code given as <tt>cond</tt>. The comparison yields a
3928 <a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, with
3929 an identical number of elements as the values being compared, and each element
3930 having identical with to the width of the floating point elements. The most
3931 significant bit in each element is 1 if the element-wise comparison evaluates to
3932 true, and is 0 otherwise. All other bits of the result are undefined. The
3933 condition codes are evaluated identically to the
3934 <a href="#i_fcmp">'<tt>fcmp</tt>' instruction</a>.
3938 <result> = vfcmp oeq <2 x float> < float 4, float 0 >, < float 5, float 0 > <i>; yields: result=<2 x i32> < i32 0, i32 -1 ></i>
3939 <result> = vfcmp ult <2 x double> < double 1, double 2 >, < double 2, double 2> <i>; yields: result=<2 x i64> < i64 -1, i64 0 ></i>
3943 <!-- _______________________________________________________________________ -->
3944 <div class="doc_subsubsection"> <a name="i_phi">'<tt>phi</tt>'
3945 Instruction</a> </div>
3946 <div class="doc_text">
3948 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
3950 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
3951 the SSA graph representing the function.</p>
3953 <p>The type of the incoming values is specified with the first type
3954 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
3955 as arguments, with one pair for each predecessor basic block of the
3956 current block. Only values of <a href="#t_firstclass">first class</a>
3957 type may be used as the value arguments to the PHI node. Only labels
3958 may be used as the label arguments.</p>
3959 <p>There must be no non-phi instructions between the start of a basic
3960 block and the PHI instructions: i.e. PHI instructions must be first in
3963 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
3964 specified by the pair corresponding to the predecessor basic block that executed
3965 just prior to the current block.</p>
3967 <pre>Loop: ; Infinite loop that counts from 0 on up...<br> %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]<br> %nextindvar = add i32 %indvar, 1<br> br label %Loop<br></pre>
3970 <!-- _______________________________________________________________________ -->
3971 <div class="doc_subsubsection">
3972 <a name="i_select">'<tt>select</tt>' Instruction</a>
3975 <div class="doc_text">
3980 <result> = select i1 <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
3986 The '<tt>select</tt>' instruction is used to choose one value based on a
3987 condition, without branching.
3994 The '<tt>select</tt>' instruction requires a boolean value indicating the condition, and two values of the same <a href="#t_firstclass">first class</a> type.
4000 If the boolean condition evaluates to true, the instruction returns the first
4001 value argument; otherwise, it returns the second value argument.
4007 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
4012 <!-- _______________________________________________________________________ -->
4013 <div class="doc_subsubsection">
4014 <a name="i_call">'<tt>call</tt>' Instruction</a>
4017 <div class="doc_text">
4021 <result> = [tail] call [<a href="#callingconv">cconv</a>] <ty> [<fnty>*] <fnptrval>(<param list>)
4026 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
4030 <p>This instruction requires several arguments:</p>
4034 <p>The optional "tail" marker indicates whether the callee function accesses
4035 any allocas or varargs in the caller. If the "tail" marker is present, the
4036 function call is eligible for tail call optimization. Note that calls may
4037 be marked "tail" even if they do not occur before a <a
4038 href="#i_ret"><tt>ret</tt></a> instruction.
4041 <p>The optional "cconv" marker indicates which <a href="#callingconv">calling
4042 convention</a> the call should use. If none is specified, the call defaults
4043 to using C calling conventions.
4046 <p>'<tt>ty</tt>': the type of the call instruction itself which is also
4047 the type of the return value. Functions that return no value are marked
4048 <tt><a href="#t_void">void</a></tt>.</p>
4051 <p>'<tt>fnty</tt>': shall be the signature of the pointer to function
4052 value being invoked. The argument types must match the types implied by
4053 this signature. This type can be omitted if the function is not varargs
4054 and if the function type does not return a pointer to a function.</p>
4057 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
4058 be invoked. In most cases, this is a direct function invocation, but
4059 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
4060 to function value.</p>
4063 <p>'<tt>function args</tt>': argument list whose types match the
4064 function signature argument types. All arguments must be of
4065 <a href="#t_firstclass">first class</a> type. If the function signature
4066 indicates the function accepts a variable number of arguments, the extra
4067 arguments can be specified.</p>
4073 <p>The '<tt>call</tt>' instruction is used to cause control flow to
4074 transfer to a specified function, with its incoming arguments bound to
4075 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
4076 instruction in the called function, control flow continues with the
4077 instruction after the function call, and the return value of the
4078 function is bound to the result argument. If the callee returns multiple
4079 values then the return values of the function are only accessible through
4080 the '<tt><a href="#i_getresult">getresult</a></tt>' instruction.</p>
4085 %retval = call i32 @test(i32 %argc)
4086 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42) <i>; yields i32</i>
4087 %X = tail call i32 @foo() <i>; yields i32</i>
4088 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
4089 call void %foo(i8 97 signext)
4091 %struct.A = type { i32, i8 }
4092 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
4093 %gr = getresult %struct.A %r, 0 <i>; yields i32</i>
4094 %gr1 = getresult %struct.A %r, 1 <i>; yields i8</i>
4099 <!-- _______________________________________________________________________ -->
4100 <div class="doc_subsubsection">
4101 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
4104 <div class="doc_text">
4109 <resultval> = va_arg <va_list*> <arglist>, <argty>
4114 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
4115 the "variable argument" area of a function call. It is used to implement the
4116 <tt>va_arg</tt> macro in C.</p>
4120 <p>This instruction takes a <tt>va_list*</tt> value and the type of
4121 the argument. It returns a value of the specified argument type and
4122 increments the <tt>va_list</tt> to point to the next argument. The
4123 actual type of <tt>va_list</tt> is target specific.</p>
4127 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
4128 type from the specified <tt>va_list</tt> and causes the
4129 <tt>va_list</tt> to point to the next argument. For more information,
4130 see the variable argument handling <a href="#int_varargs">Intrinsic
4133 <p>It is legal for this instruction to be called in a function which does not
4134 take a variable number of arguments, for example, the <tt>vfprintf</tt>
4137 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
4138 href="#intrinsics">intrinsic function</a> because it takes a type as an
4143 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
4147 <!-- _______________________________________________________________________ -->
4148 <div class="doc_subsubsection">
4149 <a name="i_getresult">'<tt>getresult</tt>' Instruction</a>
4152 <div class="doc_text">
4156 <resultval> = getresult <type> <retval>, <index>
4161 <p> The '<tt>getresult</tt>' instruction is used to extract individual values
4162 from a '<tt><a href="#i_call">call</a></tt>'
4163 or '<tt><a href="#i_invoke">invoke</a></tt>' instruction that returns multiple
4168 <p>The '<tt>getresult</tt>' instruction takes a call or invoke value as its
4169 first argument, or an undef value. The value must have <a
4170 href="#t_struct">structure type</a>. The second argument is a constant
4171 unsigned index value which must be in range for the number of values returned
4176 <p>The '<tt>getresult</tt>' instruction extracts the element identified by
4177 '<tt>index</tt>' from the aggregate value.</p>
4182 %struct.A = type { i32, i8 }
4184 %r = call %struct.A @foo()
4185 %gr = getresult %struct.A %r, 0 <i>; yields i32:%gr</i>
4186 %gr1 = getresult %struct.A %r, 1 <i>; yields i8:%gr1</i>
4193 <!-- *********************************************************************** -->
4194 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
4195 <!-- *********************************************************************** -->
4197 <div class="doc_text">
4199 <p>LLVM supports the notion of an "intrinsic function". These functions have
4200 well known names and semantics and are required to follow certain restrictions.
4201 Overall, these intrinsics represent an extension mechanism for the LLVM
4202 language that does not require changing all of the transformations in LLVM when
4203 adding to the language (or the bitcode reader/writer, the parser, etc...).</p>
4205 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
4206 prefix is reserved in LLVM for intrinsic names; thus, function names may not
4207 begin with this prefix. Intrinsic functions must always be external functions:
4208 you cannot define the body of intrinsic functions. Intrinsic functions may
4209 only be used in call or invoke instructions: it is illegal to take the address
4210 of an intrinsic function. Additionally, because intrinsic functions are part
4211 of the LLVM language, it is required if any are added that they be documented
4214 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
4215 a family of functions that perform the same operation but on different data
4216 types. Because LLVM can represent over 8 million different integer types,
4217 overloading is used commonly to allow an intrinsic function to operate on any
4218 integer type. One or more of the argument types or the result type can be
4219 overloaded to accept any integer type. Argument types may also be defined as
4220 exactly matching a previous argument's type or the result type. This allows an
4221 intrinsic function which accepts multiple arguments, but needs all of them to
4222 be of the same type, to only be overloaded with respect to a single argument or
4225 <p>Overloaded intrinsics will have the names of its overloaded argument types
4226 encoded into its function name, each preceded by a period. Only those types
4227 which are overloaded result in a name suffix. Arguments whose type is matched
4228 against another type do not. For example, the <tt>llvm.ctpop</tt> function can
4229 take an integer of any width and returns an integer of exactly the same integer
4230 width. This leads to a family of functions such as
4231 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29 %val)</tt>.
4232 Only one type, the return type, is overloaded, and only one type suffix is
4233 required. Because the argument's type is matched against the return type, it
4234 does not require its own name suffix.</p>
4236 <p>To learn how to add an intrinsic function, please see the
4237 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.
4242 <!-- ======================================================================= -->
4243 <div class="doc_subsection">
4244 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
4247 <div class="doc_text">
4249 <p>Variable argument support is defined in LLVM with the <a
4250 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
4251 intrinsic functions. These functions are related to the similarly
4252 named macros defined in the <tt><stdarg.h></tt> header file.</p>
4254 <p>All of these functions operate on arguments that use a
4255 target-specific value type "<tt>va_list</tt>". The LLVM assembly
4256 language reference manual does not define what this type is, so all
4257 transformations should be prepared to handle these functions regardless of
4260 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
4261 instruction and the variable argument handling intrinsic functions are
4264 <div class="doc_code">
4266 define i32 @test(i32 %X, ...) {
4267 ; Initialize variable argument processing
4269 %ap2 = bitcast i8** %ap to i8*
4270 call void @llvm.va_start(i8* %ap2)
4272 ; Read a single integer argument
4273 %tmp = va_arg i8** %ap, i32
4275 ; Demonstrate usage of llvm.va_copy and llvm.va_end
4277 %aq2 = bitcast i8** %aq to i8*
4278 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
4279 call void @llvm.va_end(i8* %aq2)
4281 ; Stop processing of arguments.
4282 call void @llvm.va_end(i8* %ap2)
4286 declare void @llvm.va_start(i8*)
4287 declare void @llvm.va_copy(i8*, i8*)
4288 declare void @llvm.va_end(i8*)
4294 <!-- _______________________________________________________________________ -->
4295 <div class="doc_subsubsection">
4296 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
4300 <div class="doc_text">
4302 <pre> declare void %llvm.va_start(i8* <arglist>)<br></pre>
4304 <P>The '<tt>llvm.va_start</tt>' intrinsic initializes
4305 <tt>*<arglist></tt> for subsequent use by <tt><a
4306 href="#i_va_arg">va_arg</a></tt>.</p>
4310 <P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
4314 <P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
4315 macro available in C. In a target-dependent way, it initializes the
4316 <tt>va_list</tt> element to which the argument points, so that the next call to
4317 <tt>va_arg</tt> will produce the first variable argument passed to the function.
4318 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
4319 last argument of the function as the compiler can figure that out.</p>
4323 <!-- _______________________________________________________________________ -->
4324 <div class="doc_subsubsection">
4325 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
4328 <div class="doc_text">
4330 <pre> declare void @llvm.va_end(i8* <arglist>)<br></pre>
4333 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
4334 which has been initialized previously with <tt><a href="#int_va_start">llvm.va_start</a></tt>
4335 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
4339 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
4343 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
4344 macro available in C. In a target-dependent way, it destroys the
4345 <tt>va_list</tt> element to which the argument points. Calls to <a
4346 href="#int_va_start"><tt>llvm.va_start</tt></a> and <a href="#int_va_copy">
4347 <tt>llvm.va_copy</tt></a> must be matched exactly with calls to
4348 <tt>llvm.va_end</tt>.</p>
4352 <!-- _______________________________________________________________________ -->
4353 <div class="doc_subsubsection">
4354 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
4357 <div class="doc_text">
4362 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
4367 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
4368 from the source argument list to the destination argument list.</p>
4372 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
4373 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
4378 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
4379 macro available in C. In a target-dependent way, it copies the source
4380 <tt>va_list</tt> element into the destination <tt>va_list</tt> element. This
4381 intrinsic is necessary because the <tt><a href="#int_va_start">
4382 llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
4383 example, memory allocation.</p>
4387 <!-- ======================================================================= -->
4388 <div class="doc_subsection">
4389 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
4392 <div class="doc_text">
4395 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
4396 Collection</a> requires the implementation and generation of these intrinsics.
4397 These intrinsics allow identification of <a href="#int_gcroot">GC roots on the
4398 stack</a>, as well as garbage collector implementations that require <a
4399 href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a> barriers.
4400 Front-ends for type-safe garbage collected languages should generate these
4401 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
4402 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
4405 <p>The garbage collection intrinsics only operate on objects in the generic
4406 address space (address space zero).</p>
4410 <!-- _______________________________________________________________________ -->
4411 <div class="doc_subsubsection">
4412 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
4415 <div class="doc_text">
4420 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
4425 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
4426 the code generator, and allows some metadata to be associated with it.</p>
4430 <p>The first argument specifies the address of a stack object that contains the
4431 root pointer. The second pointer (which must be either a constant or a global
4432 value address) contains the meta-data to be associated with the root.</p>
4436 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
4437 location. At compile-time, the code generator generates information to allow
4438 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
4439 intrinsic may only be used in a function which <a href="#gc">specifies a GC
4445 <!-- _______________________________________________________________________ -->
4446 <div class="doc_subsubsection">
4447 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
4450 <div class="doc_text">
4455 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
4460 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
4461 locations, allowing garbage collector implementations that require read
4466 <p>The second argument is the address to read from, which should be an address
4467 allocated from the garbage collector. The first object is a pointer to the
4468 start of the referenced object, if needed by the language runtime (otherwise
4473 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
4474 instruction, but may be replaced with substantially more complex code by the
4475 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
4476 may only be used in a function which <a href="#gc">specifies a GC
4482 <!-- _______________________________________________________________________ -->
4483 <div class="doc_subsubsection">
4484 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
4487 <div class="doc_text">
4492 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
4497 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
4498 locations, allowing garbage collector implementations that require write
4499 barriers (such as generational or reference counting collectors).</p>
4503 <p>The first argument is the reference to store, the second is the start of the
4504 object to store it to, and the third is the address of the field of Obj to
4505 store to. If the runtime does not require a pointer to the object, Obj may be
4510 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
4511 instruction, but may be replaced with substantially more complex code by the
4512 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
4513 may only be used in a function which <a href="#gc">specifies a GC
4520 <!-- ======================================================================= -->
4521 <div class="doc_subsection">
4522 <a name="int_codegen">Code Generator Intrinsics</a>
4525 <div class="doc_text">
4527 These intrinsics are provided by LLVM to expose special features that may only
4528 be implemented with code generator support.
4533 <!-- _______________________________________________________________________ -->
4534 <div class="doc_subsubsection">
4535 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
4538 <div class="doc_text">
4542 declare i8 *@llvm.returnaddress(i32 <level>)
4548 The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
4549 target-specific value indicating the return address of the current function
4550 or one of its callers.
4556 The argument to this intrinsic indicates which function to return the address
4557 for. Zero indicates the calling function, one indicates its caller, etc. The
4558 argument is <b>required</b> to be a constant integer value.
4564 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
4565 the return address of the specified call frame, or zero if it cannot be
4566 identified. The value returned by this intrinsic is likely to be incorrect or 0
4567 for arguments other than zero, so it should only be used for debugging purposes.
4571 Note that calling this intrinsic does not prevent function inlining or other
4572 aggressive transformations, so the value returned may not be that of the obvious
4573 source-language caller.
4578 <!-- _______________________________________________________________________ -->
4579 <div class="doc_subsubsection">
4580 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
4583 <div class="doc_text">
4587 declare i8 *@llvm.frameaddress(i32 <level>)
4593 The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
4594 target-specific frame pointer value for the specified stack frame.
4600 The argument to this intrinsic indicates which function to return the frame
4601 pointer for. Zero indicates the calling function, one indicates its caller,
4602 etc. The argument is <b>required</b> to be a constant integer value.
4608 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
4609 the frame address of the specified call frame, or zero if it cannot be
4610 identified. The value returned by this intrinsic is likely to be incorrect or 0
4611 for arguments other than zero, so it should only be used for debugging purposes.
4615 Note that calling this intrinsic does not prevent function inlining or other
4616 aggressive transformations, so the value returned may not be that of the obvious
4617 source-language caller.
4621 <!-- _______________________________________________________________________ -->
4622 <div class="doc_subsubsection">
4623 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
4626 <div class="doc_text">
4630 declare i8 *@llvm.stacksave()
4636 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
4637 the function stack, for use with <a href="#int_stackrestore">
4638 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
4639 features like scoped automatic variable sized arrays in C99.
4645 This intrinsic returns a opaque pointer value that can be passed to <a
4646 href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
4647 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
4648 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
4649 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
4650 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
4651 that were allocated after the <tt>llvm.stacksave</tt> was executed.
4656 <!-- _______________________________________________________________________ -->
4657 <div class="doc_subsubsection">
4658 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
4661 <div class="doc_text">
4665 declare void @llvm.stackrestore(i8 * %ptr)
4671 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
4672 the function stack to the state it was in when the corresponding <a
4673 href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
4674 useful for implementing language features like scoped automatic variable sized
4681 See the description for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.
4687 <!-- _______________________________________________________________________ -->
4688 <div class="doc_subsubsection">
4689 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
4692 <div class="doc_text">
4696 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>)
4703 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
4704 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
4706 effect on the behavior of the program but can change its performance
4713 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
4714 determining if the fetch should be for a read (0) or write (1), and
4715 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
4716 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
4717 <tt>locality</tt> arguments must be constant integers.
4723 This intrinsic does not modify the behavior of the program. In particular,
4724 prefetches cannot trap and do not produce a value. On targets that support this
4725 intrinsic, the prefetch can provide hints to the processor cache for better
4731 <!-- _______________________________________________________________________ -->
4732 <div class="doc_subsubsection">
4733 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
4736 <div class="doc_text">
4740 declare void @llvm.pcmarker(i32 <id>)
4747 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
4749 code to simulators and other tools. The method is target specific, but it is
4750 expected that the marker will use exported symbols to transmit the PC of the marker.
4751 The marker makes no guarantees that it will remain with any specific instruction
4752 after optimizations. It is possible that the presence of a marker will inhibit
4753 optimizations. The intended use is to be inserted after optimizations to allow
4754 correlations of simulation runs.
4760 <tt>id</tt> is a numerical id identifying the marker.
4766 This intrinsic does not modify the behavior of the program. Backends that do not
4767 support this intrinisic may ignore it.
4772 <!-- _______________________________________________________________________ -->
4773 <div class="doc_subsubsection">
4774 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
4777 <div class="doc_text">
4781 declare i64 @llvm.readcyclecounter( )
4788 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
4789 counter register (or similar low latency, high accuracy clocks) on those targets
4790 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
4791 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
4792 should only be used for small timings.
4798 When directly supported, reading the cycle counter should not modify any memory.
4799 Implementations are allowed to either return a application specific value or a
4800 system wide value. On backends without support, this is lowered to a constant 0.
4805 <!-- ======================================================================= -->
4806 <div class="doc_subsection">
4807 <a name="int_libc">Standard C Library Intrinsics</a>
4810 <div class="doc_text">
4812 LLVM provides intrinsics for a few important standard C library functions.
4813 These intrinsics allow source-language front-ends to pass information about the
4814 alignment of the pointer arguments to the code generator, providing opportunity
4815 for more efficient code generation.
4820 <!-- _______________________________________________________________________ -->
4821 <div class="doc_subsubsection">
4822 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
4825 <div class="doc_text">
4829 declare void @llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
4830 i32 <len>, i32 <align>)
4831 declare void @llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
4832 i64 <len>, i32 <align>)
4838 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4839 location to the destination location.
4843 Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
4844 intrinsics do not return a value, and takes an extra alignment argument.
4850 The first argument is a pointer to the destination, the second is a pointer to
4851 the source. The third argument is an integer argument
4852 specifying the number of bytes to copy, and the fourth argument is the alignment
4853 of the source and destination locations.
4857 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4858 the caller guarantees that both the source and destination pointers are aligned
4865 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4866 location to the destination location, which are not allowed to overlap. It
4867 copies "len" bytes of memory over. If the argument is known to be aligned to
4868 some boundary, this can be specified as the fourth argument, otherwise it should
4874 <!-- _______________________________________________________________________ -->
4875 <div class="doc_subsubsection">
4876 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
4879 <div class="doc_text">
4883 declare void @llvm.memmove.i32(i8 * <dest>, i8 * <src>,
4884 i32 <len>, i32 <align>)
4885 declare void @llvm.memmove.i64(i8 * <dest>, i8 * <src>,
4886 i64 <len>, i32 <align>)
4892 The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
4893 location to the destination location. It is similar to the
4894 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to overlap.
4898 Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
4899 intrinsics do not return a value, and takes an extra alignment argument.
4905 The first argument is a pointer to the destination, the second is a pointer to
4906 the source. The third argument is an integer argument
4907 specifying the number of bytes to copy, and the fourth argument is the alignment
4908 of the source and destination locations.
4912 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4913 the caller guarantees that the source and destination pointers are aligned to
4920 The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
4921 location to the destination location, which may overlap. It
4922 copies "len" bytes of memory over. If the argument is known to be aligned to
4923 some boundary, this can be specified as the fourth argument, otherwise it should
4929 <!-- _______________________________________________________________________ -->
4930 <div class="doc_subsubsection">
4931 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
4934 <div class="doc_text">
4938 declare void @llvm.memset.i32(i8 * <dest>, i8 <val>,
4939 i32 <len>, i32 <align>)
4940 declare void @llvm.memset.i64(i8 * <dest>, i8 <val>,
4941 i64 <len>, i32 <align>)
4947 The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
4952 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
4953 does not return a value, and takes an extra alignment argument.
4959 The first argument is a pointer to the destination to fill, the second is the
4960 byte value to fill it with, the third argument is an integer
4961 argument specifying the number of bytes to fill, and the fourth argument is the
4962 known alignment of destination location.
4966 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4967 the caller guarantees that the destination pointer is aligned to that boundary.
4973 The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
4975 destination location. If the argument is known to be aligned to some boundary,
4976 this can be specified as the fourth argument, otherwise it should be set to 0 or
4982 <!-- _______________________________________________________________________ -->
4983 <div class="doc_subsubsection">
4984 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
4987 <div class="doc_text">
4990 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
4991 floating point or vector of floating point type. Not all targets support all
4994 declare float @llvm.sqrt.f32(float %Val)
4995 declare double @llvm.sqrt.f64(double %Val)
4996 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
4997 declare fp128 @llvm.sqrt.f128(fp128 %Val)
4998 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
5004 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
5005 returning the same value as the libm '<tt>sqrt</tt>' functions would. Unlike
5006 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
5007 negative numbers other than -0.0 (which allows for better optimization, because
5008 there is no need to worry about errno being set). <tt>llvm.sqrt(-0.0)</tt> is
5009 defined to return -0.0 like IEEE sqrt.
5015 The argument and return value are floating point numbers of the same type.
5021 This function returns the sqrt of the specified operand if it is a nonnegative
5022 floating point number.
5026 <!-- _______________________________________________________________________ -->
5027 <div class="doc_subsubsection">
5028 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
5031 <div class="doc_text">
5034 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
5035 floating point or vector of floating point type. Not all targets support all
5038 declare float @llvm.powi.f32(float %Val, i32 %power)
5039 declare double @llvm.powi.f64(double %Val, i32 %power)
5040 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
5041 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
5042 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
5048 The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
5049 specified (positive or negative) power. The order of evaluation of
5050 multiplications is not defined. When a vector of floating point type is
5051 used, the second argument remains a scalar integer value.
5057 The second argument is an integer power, and the first is a value to raise to
5064 This function returns the first value raised to the second power with an
5065 unspecified sequence of rounding operations.</p>
5068 <!-- _______________________________________________________________________ -->
5069 <div class="doc_subsubsection">
5070 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
5073 <div class="doc_text">
5076 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
5077 floating point or vector of floating point type. Not all targets support all
5080 declare float @llvm.sin.f32(float %Val)
5081 declare double @llvm.sin.f64(double %Val)
5082 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
5083 declare fp128 @llvm.sin.f128(fp128 %Val)
5084 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
5090 The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.
5096 The argument and return value are floating point numbers of the same type.
5102 This function returns the sine of the specified operand, returning the
5103 same values as the libm <tt>sin</tt> functions would, and handles error
5104 conditions in the same way.</p>
5107 <!-- _______________________________________________________________________ -->
5108 <div class="doc_subsubsection">
5109 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
5112 <div class="doc_text">
5115 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
5116 floating point or vector of floating point type. Not all targets support all
5119 declare float @llvm.cos.f32(float %Val)
5120 declare double @llvm.cos.f64(double %Val)
5121 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
5122 declare fp128 @llvm.cos.f128(fp128 %Val)
5123 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
5129 The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.
5135 The argument and return value are floating point numbers of the same type.
5141 This function returns the cosine of the specified operand, returning the
5142 same values as the libm <tt>cos</tt> functions would, and handles error
5143 conditions in the same way.</p>
5146 <!-- _______________________________________________________________________ -->
5147 <div class="doc_subsubsection">
5148 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
5151 <div class="doc_text">
5154 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
5155 floating point or vector of floating point type. Not all targets support all
5158 declare float @llvm.pow.f32(float %Val, float %Power)
5159 declare double @llvm.pow.f64(double %Val, double %Power)
5160 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
5161 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
5162 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
5168 The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
5169 specified (positive or negative) power.
5175 The second argument is a floating point power, and the first is a value to
5176 raise to that power.
5182 This function returns the first value raised to the second power,
5184 same values as the libm <tt>pow</tt> functions would, and handles error
5185 conditions in the same way.</p>
5189 <!-- ======================================================================= -->
5190 <div class="doc_subsection">
5191 <a name="int_manip">Bit Manipulation Intrinsics</a>
5194 <div class="doc_text">
5196 LLVM provides intrinsics for a few important bit manipulation operations.
5197 These allow efficient code generation for some algorithms.
5202 <!-- _______________________________________________________________________ -->
5203 <div class="doc_subsubsection">
5204 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
5207 <div class="doc_text">
5210 <p>This is an overloaded intrinsic function. You can use bswap on any integer
5211 type that is an even number of bytes (i.e. BitWidth % 16 == 0).
5213 declare i16 @llvm.bswap.i16(i16 <id>)
5214 declare i32 @llvm.bswap.i32(i32 <id>)
5215 declare i64 @llvm.bswap.i64(i64 <id>)
5221 The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
5222 values with an even number of bytes (positive multiple of 16 bits). These are
5223 useful for performing operations on data that is not in the target's native
5230 The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
5231 and low byte of the input i16 swapped. Similarly, the <tt>llvm.bswap.i32</tt>
5232 intrinsic returns an i32 value that has the four bytes of the input i32
5233 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the returned
5234 i32 will have its bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i48</tt>,
5235 <tt>llvm.bswap.i64</tt> and other intrinsics extend this concept to
5236 additional even-byte lengths (6 bytes, 8 bytes and more, respectively).
5241 <!-- _______________________________________________________________________ -->
5242 <div class="doc_subsubsection">
5243 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
5246 <div class="doc_text">
5249 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
5250 width. Not all targets support all bit widths however.
5252 declare i8 @llvm.ctpop.i8 (i8 <src>)
5253 declare i16 @llvm.ctpop.i16(i16 <src>)
5254 declare i32 @llvm.ctpop.i32(i32 <src>)
5255 declare i64 @llvm.ctpop.i64(i64 <src>)
5256 declare i256 @llvm.ctpop.i256(i256 <src>)
5262 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
5269 The only argument is the value to be counted. The argument may be of any
5270 integer type. The return type must match the argument type.
5276 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
5280 <!-- _______________________________________________________________________ -->
5281 <div class="doc_subsubsection">
5282 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
5285 <div class="doc_text">
5288 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
5289 integer bit width. Not all targets support all bit widths however.
5291 declare i8 @llvm.ctlz.i8 (i8 <src>)
5292 declare i16 @llvm.ctlz.i16(i16 <src>)
5293 declare i32 @llvm.ctlz.i32(i32 <src>)
5294 declare i64 @llvm.ctlz.i64(i64 <src>)
5295 declare i256 @llvm.ctlz.i256(i256 <src>)
5301 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
5302 leading zeros in a variable.
5308 The only argument is the value to be counted. The argument may be of any
5309 integer type. The return type must match the argument type.
5315 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
5316 in a variable. If the src == 0 then the result is the size in bits of the type
5317 of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.
5323 <!-- _______________________________________________________________________ -->
5324 <div class="doc_subsubsection">
5325 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
5328 <div class="doc_text">
5331 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
5332 integer bit width. Not all targets support all bit widths however.
5334 declare i8 @llvm.cttz.i8 (i8 <src>)
5335 declare i16 @llvm.cttz.i16(i16 <src>)
5336 declare i32 @llvm.cttz.i32(i32 <src>)
5337 declare i64 @llvm.cttz.i64(i64 <src>)
5338 declare i256 @llvm.cttz.i256(i256 <src>)
5344 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
5351 The only argument is the value to be counted. The argument may be of any
5352 integer type. The return type must match the argument type.
5358 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
5359 in a variable. If the src == 0 then the result is the size in bits of the type
5360 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
5364 <!-- _______________________________________________________________________ -->
5365 <div class="doc_subsubsection">
5366 <a name="int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic</a>
5369 <div class="doc_text">
5372 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.select</tt>
5373 on any integer bit width.
5375 declare i17 @llvm.part.select.i17 (i17 %val, i32 %loBit, i32 %hiBit)
5376 declare i29 @llvm.part.select.i29 (i29 %val, i32 %loBit, i32 %hiBit)
5380 <p>The '<tt>llvm.part.select</tt>' family of intrinsic functions selects a
5381 range of bits from an integer value and returns them in the same bit width as
5382 the original value.</p>
5385 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5386 any bit width but they must have the same bit width. The second and third
5387 arguments must be <tt>i32</tt> type since they specify only a bit index.</p>
5390 <p>The operation of the '<tt>llvm.part.select</tt>' intrinsic has two modes
5391 of operation: forwards and reverse. If <tt>%loBit</tt> is greater than
5392 <tt>%hiBits</tt> then the intrinsic operates in reverse mode. Otherwise it
5393 operates in forward mode.</p>
5394 <p>In forward mode, this intrinsic is the equivalent of shifting <tt>%val</tt>
5395 right by <tt>%loBit</tt> bits and then ANDing it with a mask with
5396 only the <tt>%hiBit - %loBit</tt> bits set, as follows:</p>
5398 <li>The <tt>%val</tt> is shifted right (LSHR) by the number of bits specified
5399 by <tt>%loBits</tt>. This normalizes the value to the low order bits.</li>
5400 <li>The <tt>%loBits</tt> value is subtracted from the <tt>%hiBits</tt> value
5401 to determine the number of bits to retain.</li>
5402 <li>A mask of the retained bits is created by shifting a -1 value.</li>
5403 <li>The mask is ANDed with <tt>%val</tt> to produce the result.
5405 <p>In reverse mode, a similar computation is made except that the bits are
5406 returned in the reverse order. So, for example, if <tt>X</tt> has the value
5407 <tt>i16 0x0ACF (101011001111)</tt> and we apply
5408 <tt>part.select(i16 X, 8, 3)</tt> to it, we get back the value
5409 <tt>i16 0x0026 (000000100110)</tt>.</p>
5412 <div class="doc_subsubsection">
5413 <a name="int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic</a>
5416 <div class="doc_text">
5419 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.set</tt>
5420 on any integer bit width.
5422 declare i17 @llvm.part.set.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
5423 declare i29 @llvm.part.set.i29.i9 (i29 %val, i9 %repl, i32 %lo, i32 %hi)
5427 <p>The '<tt>llvm.part.set</tt>' family of intrinsic functions replaces a range
5428 of bits in an integer value with another integer value. It returns the integer
5429 with the replaced bits.</p>
5432 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5433 any bit width but they must have the same bit width. <tt>%val</tt> is the value
5434 whose bits will be replaced. The second argument, <tt>%repl</tt> may be an
5435 integer of any bit width. The third and fourth arguments must be <tt>i32</tt>
5436 type since they specify only a bit index.</p>
5439 <p>The operation of the '<tt>llvm.part.set</tt>' intrinsic has two modes
5440 of operation: forwards and reverse. If <tt>%lo</tt> is greater than
5441 <tt>%hi</tt> then the intrinsic operates in reverse mode. Otherwise it
5442 operates in forward mode.</p>
5443 <p>For both modes, the <tt>%repl</tt> value is prepared for use by either
5444 truncating it down to the size of the replacement area or zero extending it
5445 up to that size.</p>
5446 <p>In forward mode, the bits between <tt>%lo</tt> and <tt>%hi</tt> (inclusive)
5447 are replaced with corresponding bits from <tt>%repl</tt>. That is the 0th bit
5448 in <tt>%repl</tt> replaces the <tt>%lo</tt>th bit in <tt>%val</tt> and etc. up
5449 to the <tt>%hi</tt>th bit.
5450 <p>In reverse mode, a similar computation is made except that the bits are
5451 reversed. That is, the <tt>0</tt>th bit in <tt>%repl</tt> replaces the
5452 <tt>%hi</tt> bit in <tt>%val</tt> and etc. down to the <tt>%lo</tt>th bit.
5455 llvm.part.set(0xFFFF, 0, 4, 7) -> 0xFF0F
5456 llvm.part.set(0xFFFF, 0, 7, 4) -> 0xFF0F
5457 llvm.part.set(0xFFFF, 1, 7, 4) -> 0xFF8F
5458 llvm.part.set(0xFFFF, F, 8, 3) -> 0xFFE7
5459 llvm.part.set(0xFFFF, 0, 3, 8) -> 0xFE07
5463 <!-- ======================================================================= -->
5464 <div class="doc_subsection">
5465 <a name="int_debugger">Debugger Intrinsics</a>
5468 <div class="doc_text">
5470 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
5471 are described in the <a
5472 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
5473 Debugging</a> document.
5478 <!-- ======================================================================= -->
5479 <div class="doc_subsection">
5480 <a name="int_eh">Exception Handling Intrinsics</a>
5483 <div class="doc_text">
5484 <p> The LLVM exception handling intrinsics (which all start with
5485 <tt>llvm.eh.</tt> prefix), are described in the <a
5486 href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
5487 Handling</a> document. </p>
5490 <!-- ======================================================================= -->
5491 <div class="doc_subsection">
5492 <a name="int_trampoline">Trampoline Intrinsic</a>
5495 <div class="doc_text">
5497 This intrinsic makes it possible to excise one parameter, marked with
5498 the <tt>nest</tt> attribute, from a function. The result is a callable
5499 function pointer lacking the nest parameter - the caller does not need
5500 to provide a value for it. Instead, the value to use is stored in
5501 advance in a "trampoline", a block of memory usually allocated
5502 on the stack, which also contains code to splice the nest value into the
5503 argument list. This is used to implement the GCC nested function address
5507 For example, if the function is
5508 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
5509 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as follows:</p>
5511 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
5512 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
5513 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
5514 %fp = bitcast i8* %p to i32 (i32, i32)*
5516 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
5517 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
5520 <!-- _______________________________________________________________________ -->
5521 <div class="doc_subsubsection">
5522 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
5524 <div class="doc_text">
5527 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
5531 This fills the memory pointed to by <tt>tramp</tt> with code
5532 and returns a function pointer suitable for executing it.
5536 The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
5537 pointers. The <tt>tramp</tt> argument must point to a sufficiently large
5538 and sufficiently aligned block of memory; this memory is written to by the
5539 intrinsic. Note that the size and the alignment are target-specific - LLVM
5540 currently provides no portable way of determining them, so a front-end that
5541 generates this intrinsic needs to have some target-specific knowledge.
5542 The <tt>func</tt> argument must hold a function bitcast to an <tt>i8*</tt>.
5546 The block of memory pointed to by <tt>tramp</tt> is filled with target
5547 dependent code, turning it into a function. A pointer to this function is
5548 returned, but needs to be bitcast to an
5549 <a href="#int_trampoline">appropriate function pointer type</a>
5550 before being called. The new function's signature is the same as that of
5551 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
5552 removed. At most one such <tt>nest</tt> argument is allowed, and it must be
5553 of pointer type. Calling the new function is equivalent to calling
5554 <tt>func</tt> with the same argument list, but with <tt>nval</tt> used for the
5555 missing <tt>nest</tt> argument. If, after calling
5556 <tt>llvm.init.trampoline</tt>, the memory pointed to by <tt>tramp</tt> is
5557 modified, then the effect of any later call to the returned function pointer is
5562 <!-- ======================================================================= -->
5563 <div class="doc_subsection">
5564 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
5567 <div class="doc_text">
5569 These intrinsic functions expand the "universal IR" of LLVM to represent
5570 hardware constructs for atomic operations and memory synchronization. This
5571 provides an interface to the hardware, not an interface to the programmer. It
5572 is aimed at a low enough level to allow any programming models or APIs which
5573 need atomic behaviors to map cleanly onto it. It is also modeled primarily on
5574 hardware behavior. Just as hardware provides a "universal IR" for source
5575 languages, it also provides a starting point for developing a "universal"
5576 atomic operation and synchronization IR.
5579 These do <em>not</em> form an API such as high-level threading libraries,
5580 software transaction memory systems, atomic primitives, and intrinsic
5581 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
5582 application libraries. The hardware interface provided by LLVM should allow
5583 a clean implementation of all of these APIs and parallel programming models.
5584 No one model or paradigm should be selected above others unless the hardware
5585 itself ubiquitously does so.
5590 <!-- _______________________________________________________________________ -->
5591 <div class="doc_subsubsection">
5592 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
5594 <div class="doc_text">
5597 declare void @llvm.memory.barrier( i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>,
5603 The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
5604 specific pairs of memory access types.
5608 The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
5609 The first four arguments enables a specific barrier as listed below. The fith
5610 argument specifies that the barrier applies to io or device or uncached memory.
5614 <li><tt>ll</tt>: load-load barrier</li>
5615 <li><tt>ls</tt>: load-store barrier</li>
5616 <li><tt>sl</tt>: store-load barrier</li>
5617 <li><tt>ss</tt>: store-store barrier</li>
5618 <li><tt>device</tt>: barrier applies to device and uncached memory also.
5622 This intrinsic causes the system to enforce some ordering constraints upon
5623 the loads and stores of the program. This barrier does not indicate
5624 <em>when</em> any events will occur, it only enforces an <em>order</em> in
5625 which they occur. For any of the specified pairs of load and store operations
5626 (f.ex. load-load, or store-load), all of the first operations preceding the
5627 barrier will complete before any of the second operations succeeding the
5628 barrier begin. Specifically the semantics for each pairing is as follows:
5631 <li><tt>ll</tt>: All loads before the barrier must complete before any load
5632 after the barrier begins.</li>
5634 <li><tt>ls</tt>: All loads before the barrier must complete before any
5635 store after the barrier begins.</li>
5636 <li><tt>ss</tt>: All stores before the barrier must complete before any
5637 store after the barrier begins.</li>
5638 <li><tt>sl</tt>: All stores before the barrier must complete before any
5639 load after the barrier begins.</li>
5642 These semantics are applied with a logical "and" behavior when more than one
5643 is enabled in a single memory barrier intrinsic.
5646 Backends may implement stronger barriers than those requested when they do not
5647 support as fine grained a barrier as requested. Some architectures do not
5648 need all types of barriers and on such architectures, these become noops.
5655 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
5656 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
5657 <i>; guarantee the above finishes</i>
5658 store i32 8, %ptr <i>; before this begins</i>
5662 <!-- _______________________________________________________________________ -->
5663 <div class="doc_subsubsection">
5664 <a name="int_atomic_lcs">'<tt>llvm.atomic.lcs.*</tt>' Intrinsic</a>
5666 <div class="doc_text">
5669 This is an overloaded intrinsic. You can use <tt>llvm.atomic.lcs</tt> on any
5670 integer bit width. Not all targets support all bit widths however.</p>
5673 declare i8 @llvm.atomic.lcs.i8( i8* <ptr>, i8 <cmp>, i8 <val> )
5674 declare i16 @llvm.atomic.lcs.i16( i16* <ptr>, i16 <cmp>, i16 <val> )
5675 declare i32 @llvm.atomic.lcs.i32( i32* <ptr>, i32 <cmp>, i32 <val> )
5676 declare i64 @llvm.atomic.lcs.i64( i64* <ptr>, i64 <cmp>, i64 <val> )
5681 This loads a value in memory and compares it to a given value. If they are
5682 equal, it stores a new value into the memory.
5686 The <tt>llvm.atomic.lcs</tt> intrinsic takes three arguments. The result as
5687 well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
5688 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
5689 this integer type. While any bit width integer may be used, targets may only
5690 lower representations they support in hardware.
5695 This entire intrinsic must be executed atomically. It first loads the value
5696 in memory pointed to by <tt>ptr</tt> and compares it with the value
5697 <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the memory. The
5698 loaded value is yielded in all cases. This provides the equivalent of an
5699 atomic compare-and-swap operation within the SSA framework.
5707 %val1 = add i32 4, 4
5708 %result1 = call i32 @llvm.atomic.lcs.i32( i32* %ptr, i32 4, %val1 )
5709 <i>; yields {i32}:result1 = 4</i>
5710 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
5711 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
5713 %val2 = add i32 1, 1
5714 %result2 = call i32 @llvm.atomic.lcs.i32( i32* %ptr, i32 5, %val2 )
5715 <i>; yields {i32}:result2 = 8</i>
5716 %stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
5718 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
5722 <!-- _______________________________________________________________________ -->
5723 <div class="doc_subsubsection">
5724 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
5726 <div class="doc_text">
5730 This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
5731 integer bit width. Not all targets support all bit widths however.</p>
5733 declare i8 @llvm.atomic.swap.i8( i8* <ptr>, i8 <val> )
5734 declare i16 @llvm.atomic.swap.i16( i16* <ptr>, i16 <val> )
5735 declare i32 @llvm.atomic.swap.i32( i32* <ptr>, i32 <val> )
5736 declare i64 @llvm.atomic.swap.i64( i64* <ptr>, i64 <val> )
5741 This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
5742 the value from memory. It then stores the value in <tt>val</tt> in the memory
5748 The <tt>llvm.atomic.ls</tt> intrinsic takes two arguments. Both the
5749 <tt>val</tt> argument and the result must be integers of the same bit width.
5750 The first argument, <tt>ptr</tt>, must be a pointer to a value of this
5751 integer type. The targets may only lower integer representations they
5756 This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
5757 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
5758 equivalent of an atomic swap operation within the SSA framework.
5766 %val1 = add i32 4, 4
5767 %result1 = call i32 @llvm.atomic.swap.i32( i32* %ptr, i32 %val1 )
5768 <i>; yields {i32}:result1 = 4</i>
5769 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
5770 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
5772 %val2 = add i32 1, 1
5773 %result2 = call i32 @llvm.atomic.swap.i32( i32* %ptr, i32 %val2 )
5774 <i>; yields {i32}:result2 = 8</i>
5776 %stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
5777 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
5781 <!-- _______________________________________________________________________ -->
5782 <div class="doc_subsubsection">
5783 <a name="int_atomic_las">'<tt>llvm.atomic.las.*</tt>' Intrinsic</a>
5786 <div class="doc_text">
5789 This is an overloaded intrinsic. You can use <tt>llvm.atomic.las</tt> on any
5790 integer bit width. Not all targets support all bit widths however.</p>
5792 declare i8 @llvm.atomic.las.i8.( i8* <ptr>, i8 <delta> )
5793 declare i16 @llvm.atomic.las.i16.( i16* <ptr>, i16 <delta> )
5794 declare i32 @llvm.atomic.las.i32.( i32* <ptr>, i32 <delta> )
5795 declare i64 @llvm.atomic.las.i64.( i64* <ptr>, i64 <delta> )
5800 This intrinsic adds <tt>delta</tt> to the value stored in memory at
5801 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
5806 The intrinsic takes two arguments, the first a pointer to an integer value
5807 and the second an integer value. The result is also an integer value. These
5808 integer types can have any bit width, but they must all have the same bit
5809 width. The targets may only lower integer representations they support.
5813 This intrinsic does a series of operations atomically. It first loads the
5814 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
5815 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
5822 %result1 = call i32 @llvm.atomic.las.i32( i32* %ptr, i32 4 )
5823 <i>; yields {i32}:result1 = 4</i>
5824 %result2 = call i32 @llvm.atomic.las.i32( i32* %ptr, i32 2 )
5825 <i>; yields {i32}:result2 = 8</i>
5826 %result3 = call i32 @llvm.atomic.las.i32( i32* %ptr, i32 5 )
5827 <i>; yields {i32}:result3 = 10</i>
5828 %memval = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
5833 <!-- ======================================================================= -->
5834 <div class="doc_subsection">
5835 <a name="int_general">General Intrinsics</a>
5838 <div class="doc_text">
5839 <p> This class of intrinsics is designed to be generic and has
5840 no specific purpose. </p>
5843 <!-- _______________________________________________________________________ -->
5844 <div class="doc_subsubsection">
5845 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
5848 <div class="doc_text">
5852 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
5858 The '<tt>llvm.var.annotation</tt>' intrinsic
5864 The first argument is a pointer to a value, the second is a pointer to a
5865 global string, the third is a pointer to a global string which is the source
5866 file name, and the last argument is the line number.
5872 This intrinsic allows annotation of local variables with arbitrary strings.
5873 This can be useful for special purpose optimizations that want to look for these
5874 annotations. These have no other defined use, they are ignored by code
5875 generation and optimization.
5879 <!-- _______________________________________________________________________ -->
5880 <div class="doc_subsubsection">
5881 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
5884 <div class="doc_text">
5887 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
5888 any integer bit width.
5891 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
5892 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
5893 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
5894 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
5895 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
5901 The '<tt>llvm.annotation</tt>' intrinsic.
5907 The first argument is an integer value (result of some expression),
5908 the second is a pointer to a global string, the third is a pointer to a global
5909 string which is the source file name, and the last argument is the line number.
5910 It returns the value of the first argument.
5916 This intrinsic allows annotations to be put on arbitrary expressions
5917 with arbitrary strings. This can be useful for special purpose optimizations
5918 that want to look for these annotations. These have no other defined use, they
5919 are ignored by code generation and optimization.
5922 <!-- _______________________________________________________________________ -->
5923 <div class="doc_subsubsection">
5924 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
5927 <div class="doc_text">
5931 declare void @llvm.trap()
5937 The '<tt>llvm.trap</tt>' intrinsic
5949 This intrinsics is lowered to the target dependent trap instruction. If the
5950 target does not have a trap instruction, this intrinsic will be lowered to the
5951 call of the abort() function.
5955 <!-- *********************************************************************** -->
5958 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
5959 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
5960 <a href="http://validator.w3.org/check/referer"><img
5961 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
5963 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
5964 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
5965 Last modified: $Date$