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 tail call optimization to be supported.
590 This calling convention does not support varargs and requires the prototype of
591 all callees to exactly match the prototype of the function definition.
594 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
596 <dd>This calling convention attempts to make code in the caller as efficient
597 as possible under the assumption that the call is not commonly executed. As
598 such, these calls often preserve all registers so that the call does not break
599 any live ranges in the caller side. This calling convention does not support
600 varargs and requires the prototype of all callees to exactly match the
601 prototype of the function definition.
604 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
606 <dd>Any calling convention may be specified by number, allowing
607 target-specific calling conventions to be used. Target specific calling
608 conventions start at 64.
612 <p>More calling conventions can be added/defined on an as-needed basis, to
613 support pascal conventions or any other well-known target-independent
618 <!-- ======================================================================= -->
619 <div class="doc_subsection">
620 <a name="visibility">Visibility Styles</a>
623 <div class="doc_text">
626 All Global Variables and Functions have one of the following visibility styles:
630 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
632 <dd>On ELF, default visibility means that the declaration is visible to other
633 modules and, in shared libraries, means that the declared entity may be
634 overridden. On Darwin, default visibility means that the declaration is
635 visible to other modules. Default visibility corresponds to "external
636 linkage" in the language.
639 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
641 <dd>Two declarations of an object with hidden visibility refer to the same
642 object if they are in the same shared object. Usually, hidden visibility
643 indicates that the symbol will not be placed into the dynamic symbol table,
644 so no other module (executable or shared library) can reference it
648 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
650 <dd>On ELF, protected visibility indicates that the symbol will be placed in
651 the dynamic symbol table, but that references within the defining module will
652 bind to the local symbol. That is, the symbol cannot be overridden by another
659 <!-- ======================================================================= -->
660 <div class="doc_subsection">
661 <a name="globalvars">Global Variables</a>
664 <div class="doc_text">
666 <p>Global variables define regions of memory allocated at compilation time
667 instead of run-time. Global variables may optionally be initialized, may have
668 an explicit section to be placed in, and may have an optional explicit alignment
669 specified. A variable may be defined as "thread_local", which means that it
670 will not be shared by threads (each thread will have a separated copy of the
671 variable). A variable may be defined as a global "constant," which indicates
672 that the contents of the variable will <b>never</b> be modified (enabling better
673 optimization, allowing the global data to be placed in the read-only section of
674 an executable, etc). Note that variables that need runtime initialization
675 cannot be marked "constant" as there is a store to the variable.</p>
678 LLVM explicitly allows <em>declarations</em> of global variables to be marked
679 constant, even if the final definition of the global is not. This capability
680 can be used to enable slightly better optimization of the program, but requires
681 the language definition to guarantee that optimizations based on the
682 'constantness' are valid for the translation units that do not include the
686 <p>As SSA values, global variables define pointer values that are in
687 scope (i.e. they dominate) all basic blocks in the program. Global
688 variables always define a pointer to their "content" type because they
689 describe a region of memory, and all memory objects in LLVM are
690 accessed through pointers.</p>
692 <p>A global variable may be declared to reside in a target-specifc numbered
693 address space. For targets that support them, address spaces may affect how
694 optimizations are performed and/or what target instructions are used to access
695 the variable. The default address space is zero. The address space qualifier
696 must precede any other attributes.</p>
698 <p>LLVM allows an explicit section to be specified for globals. If the target
699 supports it, it will emit globals to the section specified.</p>
701 <p>An explicit alignment may be specified for a global. If not present, or if
702 the alignment is set to zero, the alignment of the global is set by the target
703 to whatever it feels convenient. If an explicit alignment is specified, the
704 global is forced to have at least that much alignment. All alignments must be
707 <p>For example, the following defines a global in a numbered address space with
708 an initializer, section, and alignment:</p>
710 <div class="doc_code">
712 @G = constant float 1.0 addrspace(5), section "foo", align 4
719 <!-- ======================================================================= -->
720 <div class="doc_subsection">
721 <a name="functionstructure">Functions</a>
724 <div class="doc_text">
726 <p>LLVM function definitions consist of the "<tt>define</tt>" keyord,
727 an optional <a href="#linkage">linkage type</a>, an optional
728 <a href="#visibility">visibility style</a>, an optional
729 <a href="#callingconv">calling convention</a>, a return type, an optional
730 <a href="#paramattrs">parameter attribute</a> for the return type, a function
731 name, a (possibly empty) argument list (each with optional
732 <a href="#paramattrs">parameter attributes</a>), an optional section, an
733 optional alignment, an optional <a href="#gc">garbage collector name</a>, an
734 opening curly brace, a list of basic blocks, and a closing curly brace.
736 LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
737 optional <a href="#linkage">linkage type</a>, an optional
738 <a href="#visibility">visibility style</a>, an optional
739 <a href="#callingconv">calling convention</a>, a return type, an optional
740 <a href="#paramattrs">parameter attribute</a> for the return type, a function
741 name, a possibly empty list of arguments, an optional alignment, and an optional
742 <a href="#gc">garbage collector name</a>.</p>
744 <p>A function definition contains a list of basic blocks, forming the CFG for
745 the function. Each basic block may optionally start with a label (giving the
746 basic block a symbol table entry), contains a list of instructions, and ends
747 with a <a href="#terminators">terminator</a> instruction (such as a branch or
748 function return).</p>
750 <p>The first basic block in a function is special in two ways: it is immediately
751 executed on entrance to the function, and it is not allowed to have predecessor
752 basic blocks (i.e. there can not be any branches to the entry block of a
753 function). Because the block can have no predecessors, it also cannot have any
754 <a href="#i_phi">PHI nodes</a>.</p>
756 <p>LLVM allows an explicit section to be specified for functions. If the target
757 supports it, it will emit functions to the section specified.</p>
759 <p>An explicit alignment may be specified for a function. If not present, or if
760 the alignment is set to zero, the alignment of the function is set by the target
761 to whatever it feels convenient. If an explicit alignment is specified, the
762 function is forced to have at least that much alignment. All alignments must be
768 <!-- ======================================================================= -->
769 <div class="doc_subsection">
770 <a name="aliasstructure">Aliases</a>
772 <div class="doc_text">
773 <p>Aliases act as "second name" for the aliasee value (which can be either
774 function, global variable, another alias or bitcast of global value). Aliases
775 may have an optional <a href="#linkage">linkage type</a>, and an
776 optional <a href="#visibility">visibility style</a>.</p>
780 <div class="doc_code">
782 @<Name> = [Linkage] [Visibility] alias <AliaseeTy> @<Aliasee>
790 <!-- ======================================================================= -->
791 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
792 <div class="doc_text">
793 <p>The return type and each parameter of a function type may have a set of
794 <i>parameter attributes</i> associated with them. Parameter attributes are
795 used to communicate additional information about the result or parameters of
796 a function. Parameter attributes are considered to be part of the function,
797 not of the function type, so functions with different parameter attributes
798 can have the same function type.</p>
800 <p>Parameter attributes are simple keywords that follow the type specified. If
801 multiple parameter attributes are needed, they are space separated. For
804 <div class="doc_code">
806 declare i32 @printf(i8* noalias , ...) nounwind
807 declare i32 @atoi(i8*) nounwind readonly
811 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
812 <tt>readonly</tt>) come immediately after the argument list.</p>
814 <p>Currently, only the following parameter attributes are defined:</p>
816 <dt><tt>zeroext</tt></dt>
817 <dd>This indicates that the parameter should be zero extended just before
818 a call to this function.</dd>
820 <dt><tt>signext</tt></dt>
821 <dd>This indicates that the parameter should be sign extended just before
822 a call to this function.</dd>
824 <dt><tt>inreg</tt></dt>
825 <dd>This indicates that the parameter should be placed in register (if
826 possible) during assembling function call. Support for this attribute is
829 <dt><tt>byval</tt></dt>
830 <dd>This indicates that the pointer parameter should really be passed by
831 value to the function. The attribute implies that a hidden copy of the
832 pointee is made between the caller and the callee, so the callee is unable
833 to modify the value in the callee. This attribute is only valid on llvm
834 pointer arguments. It is generally used to pass structs and arrays by
835 value, but is also valid on scalars (even though this is silly).</dd>
837 <dt><tt>sret</tt></dt>
838 <dd>This indicates that the pointer parameter specifies the address of a
839 structure that is the return value of the function in the source program.
840 Loads and stores to the structure are assumed not to trap.
841 May only be applied to the first parameter.</dd>
843 <dt><tt>noalias</tt></dt>
844 <dd>This indicates that the parameter does not alias any global or any other
845 parameter. The caller is responsible for ensuring that this is the case,
846 usually by placing the value in a stack allocation.</dd>
848 <dt><tt>noreturn</tt></dt>
849 <dd>This function attribute indicates that the function never returns. This
850 indicates to LLVM that every call to this function should be treated as if
851 an <tt>unreachable</tt> instruction immediately followed the call.</dd>
853 <dt><tt>nounwind</tt></dt>
854 <dd>This function attribute indicates that no exceptions unwind out of the
855 function. Usually this is because the function makes no use of exceptions,
856 but it may also be that the function catches any exceptions thrown when
859 <dt><tt>nest</tt></dt>
860 <dd>This indicates that the parameter can be excised using the
861 <a href="#int_trampoline">trampoline intrinsics</a>.</dd>
862 <dt><tt>readonly</tt></dt>
863 <dd>This function attribute indicates that the function has no side-effects
864 except for producing a return value or throwing an exception. The value
865 returned must only depend on the function arguments and/or global variables.
866 It may use values obtained by dereferencing pointers.</dd>
867 <dt><tt>readnone</tt></dt>
868 <dd>A <tt>readnone</tt> function has the same restrictions as a <tt>readonly</tt>
869 function, but in addition it is not allowed to dereference any pointer arguments
875 <!-- ======================================================================= -->
876 <div class="doc_subsection">
877 <a name="gc">Garbage Collector Names</a>
880 <div class="doc_text">
881 <p>Each function may specify a garbage collector name, which is simply a
884 <div class="doc_code"><pre
885 >define void @f() gc "name" { ...</pre></div>
887 <p>The compiler declares the supported values of <i>name</i>. Specifying a
888 collector which will cause the compiler to alter its output in order to support
889 the named garbage collection algorithm.</p>
892 <!-- ======================================================================= -->
893 <div class="doc_subsection">
894 <a name="moduleasm">Module-Level Inline Assembly</a>
897 <div class="doc_text">
899 Modules may contain "module-level inline asm" blocks, which corresponds to the
900 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
901 LLVM and treated as a single unit, but may be separated in the .ll file if
902 desired. The syntax is very simple:
905 <div class="doc_code">
907 module asm "inline asm code goes here"
908 module asm "more can go here"
912 <p>The strings can contain any character by escaping non-printable characters.
913 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
918 The inline asm code is simply printed to the machine code .s file when
919 assembly code is generated.
923 <!-- ======================================================================= -->
924 <div class="doc_subsection">
925 <a name="datalayout">Data Layout</a>
928 <div class="doc_text">
929 <p>A module may specify a target specific data layout string that specifies how
930 data is to be laid out in memory. The syntax for the data layout is simply:</p>
931 <pre> target datalayout = "<i>layout specification</i>"</pre>
932 <p>The <i>layout specification</i> consists of a list of specifications
933 separated by the minus sign character ('-'). Each specification starts with a
934 letter and may include other information after the letter to define some
935 aspect of the data layout. The specifications accepted are as follows: </p>
938 <dd>Specifies that the target lays out data in big-endian form. That is, the
939 bits with the most significance have the lowest address location.</dd>
941 <dd>Specifies that hte target lays out data in little-endian form. That is,
942 the bits with the least significance have the lowest address location.</dd>
943 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
944 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
945 <i>preferred</i> alignments. All sizes are in bits. Specifying the <i>pref</i>
946 alignment is optional. If omitted, the preceding <tt>:</tt> should be omitted
948 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
949 <dd>This specifies the alignment for an integer type of a given bit
950 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
951 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
952 <dd>This specifies the alignment for a vector type of a given bit
954 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
955 <dd>This specifies the alignment for a floating point type of a given bit
956 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
958 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
959 <dd>This specifies the alignment for an aggregate type of a given bit
962 <p>When constructing the data layout for a given target, LLVM starts with a
963 default set of specifications which are then (possibly) overriden by the
964 specifications in the <tt>datalayout</tt> keyword. The default specifications
965 are given in this list:</p>
967 <li><tt>E</tt> - big endian</li>
968 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
969 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
970 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
971 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
972 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
973 <li><tt>i64:32:64</tt> - i64 has abi alignment of 32-bits but preferred
974 alignment of 64-bits</li>
975 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
976 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
977 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
978 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
979 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
981 <p>When llvm is determining the alignment for a given type, it uses the
984 <li>If the type sought is an exact match for one of the specifications, that
985 specification is used.</li>
986 <li>If no match is found, and the type sought is an integer type, then the
987 smallest integer type that is larger than the bitwidth of the sought type is
988 used. If none of the specifications are larger than the bitwidth then the the
989 largest integer type is used. For example, given the default specifications
990 above, the i7 type will use the alignment of i8 (next largest) while both
991 i65 and i256 will use the alignment of i64 (largest specified).</li>
992 <li>If no match is found, and the type sought is a vector type, then the
993 largest vector type that is smaller than the sought vector type will be used
994 as a fall back. This happens because <128 x double> can be implemented in
995 terms of 64 <2 x double>, for example.</li>
999 <!-- *********************************************************************** -->
1000 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
1001 <!-- *********************************************************************** -->
1003 <div class="doc_text">
1005 <p>The LLVM type system is one of the most important features of the
1006 intermediate representation. Being typed enables a number of
1007 optimizations to be performed on the IR directly, without having to do
1008 extra analyses on the side before the transformation. A strong type
1009 system makes it easier to read the generated code and enables novel
1010 analyses and transformations that are not feasible to perform on normal
1011 three address code representations.</p>
1015 <!-- ======================================================================= -->
1016 <div class="doc_subsection"> <a name="t_classifications">Type
1017 Classifications</a> </div>
1018 <div class="doc_text">
1019 <p>The types fall into a few useful
1020 classifications:</p>
1022 <table border="1" cellspacing="0" cellpadding="4">
1024 <tr><th>Classification</th><th>Types</th></tr>
1026 <td><a href="#t_integer">integer</a></td>
1027 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1030 <td><a href="#t_floating">floating point</a></td>
1031 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1034 <td><a name="t_firstclass">first class</a></td>
1035 <td><a href="#t_integer">integer</a>,
1036 <a href="#t_floating">floating point</a>,
1037 <a href="#t_pointer">pointer</a>,
1038 <a href="#t_vector">vector</a>
1039 <a href="#t_struct">structure</a>,
1040 <a href="#t_array">array</a>,
1044 <td><a href="#t_primitive">primitive</a></td>
1045 <td><a href="#t_label">label</a>,
1046 <a href="#t_void">void</a>,
1047 <a href="#t_integer">integer</a>,
1048 <a href="#t_floating">floating point</a>.</td>
1051 <td><a href="#t_derived">derived</a></td>
1052 <td><a href="#t_integer">integer</a>,
1053 <a href="#t_array">array</a>,
1054 <a href="#t_function">function</a>,
1055 <a href="#t_pointer">pointer</a>,
1056 <a href="#t_struct">structure</a>,
1057 <a href="#t_pstruct">packed structure</a>,
1058 <a href="#t_vector">vector</a>,
1059 <a href="#t_opaque">opaque</a>.
1064 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
1065 most important. Values of these types are the only ones which can be
1066 produced by instructions, passed as arguments, or used as operands to
1067 instructions. This means that all structures and arrays must be
1068 manipulated either by pointer or by component.</p>
1071 <!-- ======================================================================= -->
1072 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
1074 <div class="doc_text">
1075 <p>The primitive types are the fundamental building blocks of the LLVM
1080 <!-- _______________________________________________________________________ -->
1081 <div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1083 <div class="doc_text">
1086 <tr><th>Type</th><th>Description</th></tr>
1087 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1088 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1089 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1090 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1091 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1096 <!-- _______________________________________________________________________ -->
1097 <div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1099 <div class="doc_text">
1101 <p>The void type does not represent any value and has no size.</p>
1110 <!-- _______________________________________________________________________ -->
1111 <div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1113 <div class="doc_text">
1115 <p>The label type represents code labels.</p>
1125 <!-- ======================================================================= -->
1126 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1128 <div class="doc_text">
1130 <p>The real power in LLVM comes from the derived types in the system.
1131 This is what allows a programmer to represent arrays, functions,
1132 pointers, and other useful types. Note that these derived types may be
1133 recursive: For example, it is possible to have a two dimensional array.</p>
1137 <!-- _______________________________________________________________________ -->
1138 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1140 <div class="doc_text">
1143 <p>The integer type is a very simple derived type that simply specifies an
1144 arbitrary bit width for the integer type desired. Any bit width from 1 bit to
1145 2^23-1 (about 8 million) can be specified.</p>
1153 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1157 <table class="layout">
1160 <td><tt>i1</tt></td>
1161 <td>a single-bit integer.</td>
1163 <td><tt>i32</tt></td>
1164 <td>a 32-bit integer.</td>
1166 <td><tt>i1942652</tt></td>
1167 <td>a really big integer of over 1 million bits.</td>
1173 <!-- _______________________________________________________________________ -->
1174 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1176 <div class="doc_text">
1180 <p>The array type is a very simple derived type that arranges elements
1181 sequentially in memory. The array type requires a size (number of
1182 elements) and an underlying data type.</p>
1187 [<# elements> x <elementtype>]
1190 <p>The number of elements is a constant integer value; elementtype may
1191 be any type with a size.</p>
1194 <table class="layout">
1196 <td class="left"><tt>[40 x i32]</tt></td>
1197 <td class="left">Array of 40 32-bit integer values.</td>
1200 <td class="left"><tt>[41 x i32]</tt></td>
1201 <td class="left">Array of 41 32-bit integer values.</td>
1204 <td class="left"><tt>[4 x i8]</tt></td>
1205 <td class="left">Array of 4 8-bit integer values.</td>
1208 <p>Here are some examples of multidimensional arrays:</p>
1209 <table class="layout">
1211 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1212 <td class="left">3x4 array of 32-bit integer values.</td>
1215 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1216 <td class="left">12x10 array of single precision floating point values.</td>
1219 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1220 <td class="left">2x3x4 array of 16-bit integer values.</td>
1224 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1225 length array. Normally, accesses past the end of an array are undefined in
1226 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
1227 As a special case, however, zero length arrays are recognized to be variable
1228 length. This allows implementation of 'pascal style arrays' with the LLVM
1229 type "{ i32, [0 x float]}", for example.</p>
1233 <!-- _______________________________________________________________________ -->
1234 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1235 <div class="doc_text">
1239 <p>The function type can be thought of as a function signature. It
1240 consists of a return type and a list of formal parameter types. The
1241 return type of a function type is a scalar type, a void type, or a struct type.
1242 If the return type is a struct type then all struct elements must be of first
1243 class types, and the struct must have at least one element.</p>
1248 <returntype list> (<parameter list>)
1251 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1252 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1253 which indicates that the function takes a variable number of arguments.
1254 Variable argument functions can access their arguments with the <a
1255 href="#int_varargs">variable argument handling intrinsic</a> functions.
1256 '<tt><returntype list></tt>' is a comma-separated list of
1257 <a href="#t_firstclass">first class</a> type specifiers.</p>
1260 <table class="layout">
1262 <td class="left"><tt>i32 (i32)</tt></td>
1263 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1265 </tr><tr class="layout">
1266 <td class="left"><tt>float (i16 signext, i32 *) *
1268 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1269 an <tt>i16</tt> that should be sign extended and a
1270 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1273 </tr><tr class="layout">
1274 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1275 <td class="left">A vararg function that takes at least one
1276 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1277 which returns an integer. This is the signature for <tt>printf</tt> in
1280 </tr><tr class="layout">
1281 <td class="left"><tt>{i32, i32} (i32)</tt></td>
1282 <td class="left">A function taking an <tt>i32></tt>, returning two
1283 <tt> i32 </tt> values as an aggregate of type <tt>{ i32, i32 }</tt>
1289 <!-- _______________________________________________________________________ -->
1290 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1291 <div class="doc_text">
1293 <p>The structure type is used to represent a collection of data members
1294 together in memory. The packing of the field types is defined to match
1295 the ABI of the underlying processor. The elements of a structure may
1296 be any type that has a size.</p>
1297 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1298 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1299 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1302 <pre> { <type list> }<br></pre>
1304 <table class="layout">
1306 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1307 <td class="left">A triple of three <tt>i32</tt> values</td>
1308 </tr><tr class="layout">
1309 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1310 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1311 second element is a <a href="#t_pointer">pointer</a> to a
1312 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1313 an <tt>i32</tt>.</td>
1318 <!-- _______________________________________________________________________ -->
1319 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1321 <div class="doc_text">
1323 <p>The packed structure type is used to represent a collection of data members
1324 together in memory. There is no padding between fields. Further, the alignment
1325 of a packed structure is 1 byte. The elements of a packed structure may
1326 be any type that has a size.</p>
1327 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1328 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1329 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1332 <pre> < { <type list> } > <br></pre>
1334 <table class="layout">
1336 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1337 <td class="left">A triple of three <tt>i32</tt> values</td>
1338 </tr><tr class="layout">
1339 <td class="left"><tt>< { float, i32 (i32)* } ></tt></td>
1340 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1341 second element is a <a href="#t_pointer">pointer</a> to a
1342 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1343 an <tt>i32</tt>.</td>
1348 <!-- _______________________________________________________________________ -->
1349 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1350 <div class="doc_text">
1352 <p>As in many languages, the pointer type represents a pointer or
1353 reference to another object, which must live in memory. Pointer types may have
1354 an optional address space attribute defining the target-specific numbered
1355 address space where the pointed-to object resides. The default address space is
1358 <pre> <type> *<br></pre>
1360 <table class="layout">
1362 <td class="left"><tt>[4x i32]*</tt></td>
1363 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1364 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1367 <td class="left"><tt>i32 (i32 *) *</tt></td>
1368 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
1369 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1373 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1374 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1375 that resides in address space #5.</td>
1380 <!-- _______________________________________________________________________ -->
1381 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1382 <div class="doc_text">
1386 <p>A vector type is a simple derived type that represents a vector
1387 of elements. Vector types are used when multiple primitive data
1388 are operated in parallel using a single instruction (SIMD).
1389 A vector type requires a size (number of
1390 elements) and an underlying primitive data type. Vectors must have a power
1391 of two length (1, 2, 4, 8, 16 ...). Vector types are
1392 considered <a href="#t_firstclass">first class</a>.</p>
1397 < <# elements> x <elementtype> >
1400 <p>The number of elements is a constant integer value; elementtype may
1401 be any integer or floating point type.</p>
1405 <table class="layout">
1407 <td class="left"><tt><4 x i32></tt></td>
1408 <td class="left">Vector of 4 32-bit integer values.</td>
1411 <td class="left"><tt><8 x float></tt></td>
1412 <td class="left">Vector of 8 32-bit floating-point values.</td>
1415 <td class="left"><tt><2 x i64></tt></td>
1416 <td class="left">Vector of 2 64-bit integer values.</td>
1421 <!-- _______________________________________________________________________ -->
1422 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1423 <div class="doc_text">
1427 <p>Opaque types are used to represent unknown types in the system. This
1428 corresponds (for example) to the C notion of a forward declared structure type.
1429 In LLVM, opaque types can eventually be resolved to any type (not just a
1430 structure type).</p>
1440 <table class="layout">
1442 <td class="left"><tt>opaque</tt></td>
1443 <td class="left">An opaque type.</td>
1449 <!-- *********************************************************************** -->
1450 <div class="doc_section"> <a name="constants">Constants</a> </div>
1451 <!-- *********************************************************************** -->
1453 <div class="doc_text">
1455 <p>LLVM has several different basic types of constants. This section describes
1456 them all and their syntax.</p>
1460 <!-- ======================================================================= -->
1461 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1463 <div class="doc_text">
1466 <dt><b>Boolean constants</b></dt>
1468 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1469 constants of the <tt><a href="#t_primitive">i1</a></tt> type.
1472 <dt><b>Integer constants</b></dt>
1474 <dd>Standard integers (such as '4') are constants of the <a
1475 href="#t_integer">integer</a> type. Negative numbers may be used with
1479 <dt><b>Floating point constants</b></dt>
1481 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1482 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1483 notation (see below). The assembler requires the exact decimal value of
1484 a floating-point constant. For example, the assembler accepts 1.25 but
1485 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
1486 constants must have a <a href="#t_floating">floating point</a> type. </dd>
1488 <dt><b>Null pointer constants</b></dt>
1490 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1491 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1495 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1496 of floating point constants. For example, the form '<tt>double
1497 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1498 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1499 (and the only time that they are generated by the disassembler) is when a
1500 floating point constant must be emitted but it cannot be represented as a
1501 decimal floating point number. For example, NaN's, infinities, and other
1502 special values are represented in their IEEE hexadecimal format so that
1503 assembly and disassembly do not cause any bits to change in the constants.</p>
1507 <!-- ======================================================================= -->
1508 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1511 <div class="doc_text">
1512 <p>Aggregate constants arise from aggregation of simple constants
1513 and smaller aggregate constants.</p>
1516 <dt><b>Structure constants</b></dt>
1518 <dd>Structure constants are represented with notation similar to structure
1519 type definitions (a comma separated list of elements, surrounded by braces
1520 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
1521 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>". Structure constants
1522 must have <a href="#t_struct">structure type</a>, and the number and
1523 types of elements must match those specified by the type.
1526 <dt><b>Array constants</b></dt>
1528 <dd>Array constants are represented with notation similar to array type
1529 definitions (a comma separated list of elements, surrounded by square brackets
1530 (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74 ]</tt>". Array
1531 constants must have <a href="#t_array">array type</a>, and the number and
1532 types of elements must match those specified by the type.
1535 <dt><b>Vector constants</b></dt>
1537 <dd>Vector constants are represented with notation similar to vector type
1538 definitions (a comma separated list of elements, surrounded by
1539 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32 42,
1540 i32 11, i32 74, i32 100 ></tt>". Vector constants must have <a
1541 href="#t_vector">vector type</a>, and the number and types of elements must
1542 match those specified by the type.
1545 <dt><b>Zero initialization</b></dt>
1547 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1548 value to zero of <em>any</em> type, including scalar and aggregate types.
1549 This is often used to avoid having to print large zero initializers (e.g. for
1550 large arrays) and is always exactly equivalent to using explicit zero
1557 <!-- ======================================================================= -->
1558 <div class="doc_subsection">
1559 <a name="globalconstants">Global Variable and Function Addresses</a>
1562 <div class="doc_text">
1564 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1565 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1566 constants. These constants are explicitly referenced when the <a
1567 href="#identifiers">identifier for the global</a> is used and always have <a
1568 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1571 <div class="doc_code">
1575 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1581 <!-- ======================================================================= -->
1582 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1583 <div class="doc_text">
1584 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1585 no specific value. Undefined values may be of any type and be used anywhere
1586 a constant is permitted.</p>
1588 <p>Undefined values indicate to the compiler that the program is well defined
1589 no matter what value is used, giving the compiler more freedom to optimize.
1593 <!-- ======================================================================= -->
1594 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1597 <div class="doc_text">
1599 <p>Constant expressions are used to allow expressions involving other constants
1600 to be used as constants. Constant expressions may be of any <a
1601 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1602 that does not have side effects (e.g. load and call are not supported). The
1603 following is the syntax for constant expressions:</p>
1606 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1607 <dd>Truncate a constant to another type. The bit size of CST must be larger
1608 than the bit size of TYPE. Both types must be integers.</dd>
1610 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1611 <dd>Zero extend a constant to another type. The bit size of CST must be
1612 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1614 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1615 <dd>Sign extend a constant to another type. The bit size of CST must be
1616 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1618 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1619 <dd>Truncate a floating point constant to another floating point type. The
1620 size of CST must be larger than the size of TYPE. Both types must be
1621 floating point.</dd>
1623 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1624 <dd>Floating point extend a constant to another type. The size of CST must be
1625 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1627 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
1628 <dd>Convert a floating point constant to the corresponding unsigned integer
1629 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1630 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1631 of the same number of elements. If the value won't fit in the integer type,
1632 the results are undefined.</dd>
1634 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
1635 <dd>Convert a floating point constant to the corresponding signed integer
1636 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1637 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1638 of the same number of elements. If the value won't fit in the integer type,
1639 the results are undefined.</dd>
1641 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
1642 <dd>Convert an unsigned integer constant to the corresponding floating point
1643 constant. TYPE must be a scalar or vector floating point type. CST must be of
1644 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1645 of the same number of elements. If the value won't fit in the floating point
1646 type, the results are undefined.</dd>
1648 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
1649 <dd>Convert a signed integer constant to the corresponding floating point
1650 constant. TYPE must be a scalar or vector floating point type. CST must be of
1651 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1652 of the same number of elements. If the value won't fit in the floating point
1653 type, the results are undefined.</dd>
1655 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
1656 <dd>Convert a pointer typed constant to the corresponding integer constant
1657 TYPE must be an integer type. CST must be of pointer type. The CST value is
1658 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1660 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
1661 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1662 pointer type. CST must be of integer type. The CST value is zero extended,
1663 truncated, or unchanged to make it fit in a pointer size. This one is
1664 <i>really</i> dangerous!</dd>
1666 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
1667 <dd>Convert a constant, CST, to another TYPE. The size of CST and TYPE must be
1668 identical (same number of bits). The conversion is done as if the CST value
1669 was stored to memory and read back as TYPE. In other words, no bits change
1670 with this operator, just the type. This can be used for conversion of
1671 vector types to any other type, as long as they have the same bit width. For
1672 pointers it is only valid to cast to another pointer type.
1675 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1677 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1678 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1679 instruction, the index list may have zero or more indexes, which are required
1680 to make sense for the type of "CSTPTR".</dd>
1682 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1684 <dd>Perform the <a href="#i_select">select operation</a> on
1687 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
1688 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
1690 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
1691 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
1693 <dt><b><tt>vicmp COND ( VAL1, VAL2 )</tt></b></dt>
1694 <dd>Performs the <a href="#i_vicmp">vicmp operation</a> on constants.</dd>
1696 <dt><b><tt>vfcmp COND ( VAL1, VAL2 )</tt></b></dt>
1697 <dd>Performs the <a href="#i_vfcmp">vfcmp operation</a> on constants.</dd>
1699 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1701 <dd>Perform the <a href="#i_extractelement">extractelement
1702 operation</a> on constants.
1704 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1706 <dd>Perform the <a href="#i_insertelement">insertelement
1707 operation</a> on constants.</dd>
1710 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
1712 <dd>Perform the <a href="#i_shufflevector">shufflevector
1713 operation</a> on constants.</dd>
1715 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1717 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1718 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1719 binary</a> operations. The constraints on operands are the same as those for
1720 the corresponding instruction (e.g. no bitwise operations on floating point
1721 values are allowed).</dd>
1725 <!-- *********************************************************************** -->
1726 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1727 <!-- *********************************************************************** -->
1729 <!-- ======================================================================= -->
1730 <div class="doc_subsection">
1731 <a name="inlineasm">Inline Assembler Expressions</a>
1734 <div class="doc_text">
1737 LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
1738 Module-Level Inline Assembly</a>) through the use of a special value. This
1739 value represents the inline assembler as a string (containing the instructions
1740 to emit), a list of operand constraints (stored as a string), and a flag that
1741 indicates whether or not the inline asm expression has side effects. An example
1742 inline assembler expression is:
1745 <div class="doc_code">
1747 i32 (i32) asm "bswap $0", "=r,r"
1752 Inline assembler expressions may <b>only</b> be used as the callee operand of
1753 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
1756 <div class="doc_code">
1758 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
1763 Inline asms with side effects not visible in the constraint list must be marked
1764 as having side effects. This is done through the use of the
1765 '<tt>sideeffect</tt>' keyword, like so:
1768 <div class="doc_code">
1770 call void asm sideeffect "eieio", ""()
1774 <p>TODO: The format of the asm and constraints string still need to be
1775 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1776 need to be documented).
1781 <!-- *********************************************************************** -->
1782 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1783 <!-- *********************************************************************** -->
1785 <div class="doc_text">
1787 <p>The LLVM instruction set consists of several different
1788 classifications of instructions: <a href="#terminators">terminator
1789 instructions</a>, <a href="#binaryops">binary instructions</a>,
1790 <a href="#bitwiseops">bitwise binary instructions</a>, <a
1791 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1792 instructions</a>.</p>
1796 <!-- ======================================================================= -->
1797 <div class="doc_subsection"> <a name="terminators">Terminator
1798 Instructions</a> </div>
1800 <div class="doc_text">
1802 <p>As mentioned <a href="#functionstructure">previously</a>, every
1803 basic block in a program ends with a "Terminator" instruction, which
1804 indicates which block should be executed after the current block is
1805 finished. These terminator instructions typically yield a '<tt>void</tt>'
1806 value: they produce control flow, not values (the one exception being
1807 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1808 <p>There are six different terminator instructions: the '<a
1809 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1810 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1811 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1812 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1813 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1817 <!-- _______________________________________________________________________ -->
1818 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1819 Instruction</a> </div>
1820 <div class="doc_text">
1822 <pre> ret <type> <value> <i>; Return a value from a non-void function</i>
1823 ret void <i>; Return from void function</i>
1824 ret <type> <value>, <type> <value> <i>; Return two values from a non-void function </i>
1829 <p>The '<tt>ret</tt>' instruction is used to return control flow (and a
1830 value) from a function back to the caller.</p>
1831 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
1832 returns value(s) and then causes control flow, and one that just causes
1833 control flow to occur.</p>
1837 <p>The '<tt>ret</tt>' instruction may return zero, one or multiple values.
1838 The type of each return value must be a '<a href="#t_firstclass">first
1839 class</a>' type. Note that a function is not <a href="#wellformed">well
1840 formed</a> if there exists a '<tt>ret</tt>' instruction inside of the
1841 function that returns values that do not match the return type of the
1846 <p>When the '<tt>ret</tt>' instruction is executed, control flow
1847 returns back to the calling function's context. If the caller is a "<a
1848 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1849 the instruction after the call. If the caller was an "<a
1850 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1851 at the beginning of the "normal" destination block. If the instruction
1852 returns a value, that value shall set the call or invoke instruction's
1853 return value. If the instruction returns multiple values then these
1854 values can only be accessed through a '<a href="#i_getresult"><tt>getresult</tt>
1855 </a>' instruction.</p>
1860 ret i32 5 <i>; Return an integer value of 5</i>
1861 ret void <i>; Return from a void function</i>
1862 ret i32 4, i8 2 <i>; Return two values 4 and 2 </i>
1865 <!-- _______________________________________________________________________ -->
1866 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1867 <div class="doc_text">
1869 <pre> br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1872 <p>The '<tt>br</tt>' instruction is used to cause control flow to
1873 transfer to a different basic block in the current function. There are
1874 two forms of this instruction, corresponding to a conditional branch
1875 and an unconditional branch.</p>
1877 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1878 single '<tt>i1</tt>' value and two '<tt>label</tt>' values. The
1879 unconditional form of the '<tt>br</tt>' instruction takes a single
1880 '<tt>label</tt>' value as a target.</p>
1882 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
1883 argument is evaluated. If the value is <tt>true</tt>, control flows
1884 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1885 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1887 <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
1888 href="#i_ret">ret</a> i32 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> i32 0<br></pre>
1890 <!-- _______________________________________________________________________ -->
1891 <div class="doc_subsubsection">
1892 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
1895 <div class="doc_text">
1899 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
1904 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
1905 several different places. It is a generalization of the '<tt>br</tt>'
1906 instruction, allowing a branch to occur to one of many possible
1912 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
1913 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
1914 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
1915 table is not allowed to contain duplicate constant entries.</p>
1919 <p>The <tt>switch</tt> instruction specifies a table of values and
1920 destinations. When the '<tt>switch</tt>' instruction is executed, this
1921 table is searched for the given value. If the value is found, control flow is
1922 transfered to the corresponding destination; otherwise, control flow is
1923 transfered to the default destination.</p>
1925 <h5>Implementation:</h5>
1927 <p>Depending on properties of the target machine and the particular
1928 <tt>switch</tt> instruction, this instruction may be code generated in different
1929 ways. For example, it could be generated as a series of chained conditional
1930 branches or with a lookup table.</p>
1935 <i>; Emulate a conditional br instruction</i>
1936 %Val = <a href="#i_zext">zext</a> i1 %value to i32
1937 switch i32 %Val, label %truedest [i32 0, label %falsedest ]
1939 <i>; Emulate an unconditional br instruction</i>
1940 switch i32 0, label %dest [ ]
1942 <i>; Implement a jump table:</i>
1943 switch i32 %val, label %otherwise [ i32 0, label %onzero
1945 i32 2, label %ontwo ]
1949 <!-- _______________________________________________________________________ -->
1950 <div class="doc_subsubsection">
1951 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
1954 <div class="doc_text">
1959 <result> = invoke [<a href="#callingconv">cconv</a>] <ptr to function ty> <function ptr val>(<function args>)
1960 to label <normal label> unwind label <exception label>
1965 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
1966 function, with the possibility of control flow transfer to either the
1967 '<tt>normal</tt>' label or the
1968 '<tt>exception</tt>' label. If the callee function returns with the
1969 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
1970 "normal" label. If the callee (or any indirect callees) returns with the "<a
1971 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
1972 continued at the dynamically nearest "exception" label. If the callee function
1973 returns multiple values then individual return values are only accessible through
1974 a '<tt><a href="#i_getresult">getresult</a></tt>' instruction.</p>
1978 <p>This instruction requires several arguments:</p>
1982 The optional "cconv" marker indicates which <a href="#callingconv">calling
1983 convention</a> the call should use. If none is specified, the call defaults
1984 to using C calling conventions.
1986 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
1987 function value being invoked. In most cases, this is a direct function
1988 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
1989 an arbitrary pointer to function value.
1992 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
1993 function to be invoked. </li>
1995 <li>'<tt>function args</tt>': argument list whose types match the function
1996 signature argument types. If the function signature indicates the function
1997 accepts a variable number of arguments, the extra arguments can be
2000 <li>'<tt>normal label</tt>': the label reached when the called function
2001 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
2003 <li>'<tt>exception label</tt>': the label reached when a callee returns with
2004 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
2010 <p>This instruction is designed to operate as a standard '<tt><a
2011 href="#i_call">call</a></tt>' instruction in most regards. The primary
2012 difference is that it establishes an association with a label, which is used by
2013 the runtime library to unwind the stack.</p>
2015 <p>This instruction is used in languages with destructors to ensure that proper
2016 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
2017 exception. Additionally, this is important for implementation of
2018 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
2022 %retval = invoke i32 @Test(i32 15) to label %Continue
2023 unwind label %TestCleanup <i>; {i32}:retval set</i>
2024 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
2025 unwind label %TestCleanup <i>; {i32}:retval set</i>
2030 <!-- _______________________________________________________________________ -->
2032 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
2033 Instruction</a> </div>
2035 <div class="doc_text">
2044 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
2045 at the first callee in the dynamic call stack which used an <a
2046 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
2047 primarily used to implement exception handling.</p>
2051 <p>The '<tt>unwind</tt>' instruction causes execution of the current function to
2052 immediately halt. The dynamic call stack is then searched for the first <a
2053 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
2054 execution continues at the "exceptional" destination block specified by the
2055 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
2056 dynamic call chain, undefined behavior results.</p>
2059 <!-- _______________________________________________________________________ -->
2061 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
2062 Instruction</a> </div>
2064 <div class="doc_text">
2073 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
2074 instruction is used to inform the optimizer that a particular portion of the
2075 code is not reachable. This can be used to indicate that the code after a
2076 no-return function cannot be reached, and other facts.</p>
2080 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
2085 <!-- ======================================================================= -->
2086 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
2087 <div class="doc_text">
2088 <p>Binary operators are used to do most of the computation in a
2089 program. They require two operands of the same type, execute an operation on them, and
2090 produce a single value. The operands might represent
2091 multiple data, as is the case with the <a href="#t_vector">vector</a> data type.
2092 The result value has the same type as its operands.</p>
2093 <p>There are several different binary operators:</p>
2095 <!-- _______________________________________________________________________ -->
2096 <div class="doc_subsubsection"> <a name="i_add">'<tt>add</tt>'
2097 Instruction</a> </div>
2098 <div class="doc_text">
2100 <pre> <result> = add <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2103 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
2105 <p>The two arguments to the '<tt>add</tt>' instruction must be either <a
2106 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a> values.
2107 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2108 Both arguments must have identical types.</p>
2110 <p>The value produced is the integer or floating point sum of the two
2112 <p>If an integer sum has unsigned overflow, the result returned is the
2113 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2115 <p>Because LLVM integers use a two's complement representation, this
2116 instruction is appropriate for both signed and unsigned integers.</p>
2118 <pre> <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
2121 <!-- _______________________________________________________________________ -->
2122 <div class="doc_subsubsection"> <a name="i_sub">'<tt>sub</tt>'
2123 Instruction</a> </div>
2124 <div class="doc_text">
2126 <pre> <result> = sub <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2129 <p>The '<tt>sub</tt>' instruction returns the difference of its two
2131 <p>Note that the '<tt>sub</tt>' instruction is used to represent the '<tt>neg</tt>'
2132 instruction present in most other intermediate representations.</p>
2134 <p>The two arguments to the '<tt>sub</tt>' instruction must be either <a
2135 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
2137 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2138 Both arguments must have identical types.</p>
2140 <p>The value produced is the integer or floating point difference of
2141 the two operands.</p>
2142 <p>If an integer difference has unsigned overflow, the result returned is the
2143 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2145 <p>Because LLVM integers use a two's complement representation, this
2146 instruction is appropriate for both signed and unsigned integers.</p>
2149 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
2150 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
2153 <!-- _______________________________________________________________________ -->
2154 <div class="doc_subsubsection"> <a name="i_mul">'<tt>mul</tt>'
2155 Instruction</a> </div>
2156 <div class="doc_text">
2158 <pre> <result> = mul <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2161 <p>The '<tt>mul</tt>' instruction returns the product of its two
2164 <p>The two arguments to the '<tt>mul</tt>' instruction must be either <a
2165 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
2167 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2168 Both arguments must have identical types.</p>
2170 <p>The value produced is the integer or floating point product of the
2172 <p>If the result of an integer multiplication has unsigned overflow,
2173 the result returned is the mathematical result modulo
2174 2<sup>n</sup>, where n is the bit width of the result.</p>
2175 <p>Because LLVM integers use a two's complement representation, and the
2176 result is the same width as the operands, this instruction returns the
2177 correct result for both signed and unsigned integers. If a full product
2178 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands
2179 should be sign-extended or zero-extended as appropriate to the
2180 width of the full product.</p>
2182 <pre> <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
2185 <!-- _______________________________________________________________________ -->
2186 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
2188 <div class="doc_text">
2190 <pre> <result> = udiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2193 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two
2196 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
2197 <a href="#t_integer">integer</a> values. Both arguments must have identical
2198 types. This instruction can also take <a href="#t_vector">vector</a> versions
2199 of the values in which case the elements must be integers.</p>
2201 <p>The value produced is the unsigned integer quotient of the two operands.</p>
2202 <p>Note that unsigned integer division and signed integer division are distinct
2203 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
2204 <p>Division by zero leads to undefined behavior.</p>
2206 <pre> <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2209 <!-- _______________________________________________________________________ -->
2210 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
2212 <div class="doc_text">
2214 <pre> <result> = sdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2217 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
2220 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
2221 <a href="#t_integer">integer</a> values. Both arguments must have identical
2222 types. This instruction can also take <a href="#t_vector">vector</a> versions
2223 of the values in which case the elements must be integers.</p>
2225 <p>The value produced is the signed integer quotient of the two operands rounded towards zero.</p>
2226 <p>Note that signed integer division and unsigned integer division are distinct
2227 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
2228 <p>Division by zero leads to undefined behavior. Overflow also leads to
2229 undefined behavior; this is a rare case, but can occur, for example,
2230 by doing a 32-bit division of -2147483648 by -1.</p>
2232 <pre> <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2235 <!-- _______________________________________________________________________ -->
2236 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
2237 Instruction</a> </div>
2238 <div class="doc_text">
2240 <pre> <result> = fdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2243 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
2246 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
2247 <a href="#t_floating">floating point</a> values. Both arguments must have
2248 identical types. This instruction can also take <a href="#t_vector">vector</a>
2249 versions of floating point values.</p>
2251 <p>The value produced is the floating point quotient of the two operands.</p>
2253 <pre> <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
2256 <!-- _______________________________________________________________________ -->
2257 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
2259 <div class="doc_text">
2261 <pre> <result> = urem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2264 <p>The '<tt>urem</tt>' instruction returns the remainder from the
2265 unsigned division of its two arguments.</p>
2267 <p>The two arguments to the '<tt>urem</tt>' instruction must be
2268 <a href="#t_integer">integer</a> values. Both arguments must have identical
2269 types. This instruction can also take <a href="#t_vector">vector</a> versions
2270 of the values in which case the elements must be integers.</p>
2272 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
2273 This instruction always performs an unsigned division to get the remainder.</p>
2274 <p>Note that unsigned integer remainder and signed integer remainder are
2275 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
2276 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
2278 <pre> <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2282 <!-- _______________________________________________________________________ -->
2283 <div class="doc_subsubsection"> <a name="i_srem">'<tt>srem</tt>'
2284 Instruction</a> </div>
2285 <div class="doc_text">
2287 <pre> <result> = srem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2290 <p>The '<tt>srem</tt>' instruction returns the remainder from the
2291 signed division of its two operands. This instruction can also take
2292 <a href="#t_vector">vector</a> versions of the values in which case
2293 the elements must be integers.</p>
2296 <p>The two arguments to the '<tt>srem</tt>' instruction must be
2297 <a href="#t_integer">integer</a> values. Both arguments must have identical
2300 <p>This instruction returns the <i>remainder</i> of a division (where the result
2301 has the same sign as the dividend, <tt>var1</tt>), not the <i>modulo</i>
2302 operator (where the result has the same sign as the divisor, <tt>var2</tt>) of
2303 a value. For more information about the difference, see <a
2304 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
2305 Math Forum</a>. For a table of how this is implemented in various languages,
2306 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
2307 Wikipedia: modulo operation</a>.</p>
2308 <p>Note that signed integer remainder and unsigned integer remainder are
2309 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
2310 <p>Taking the remainder of a division by zero leads to undefined behavior.
2311 Overflow also leads to undefined behavior; this is a rare case, but can occur,
2312 for example, by taking the remainder of a 32-bit division of -2147483648 by -1.
2313 (The remainder doesn't actually overflow, but this rule lets srem be
2314 implemented using instructions that return both the result of the division
2315 and the remainder.)</p>
2317 <pre> <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2321 <!-- _______________________________________________________________________ -->
2322 <div class="doc_subsubsection"> <a name="i_frem">'<tt>frem</tt>'
2323 Instruction</a> </div>
2324 <div class="doc_text">
2326 <pre> <result> = frem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2329 <p>The '<tt>frem</tt>' instruction returns the remainder from the
2330 division of its two operands.</p>
2332 <p>The two arguments to the '<tt>frem</tt>' instruction must be
2333 <a href="#t_floating">floating point</a> values. Both arguments must have
2334 identical types. This instruction can also take <a href="#t_vector">vector</a>
2335 versions of floating point values.</p>
2337 <p>This instruction returns the <i>remainder</i> of a division.
2338 The remainder has the same sign as the dividend.</p>
2340 <pre> <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
2344 <!-- ======================================================================= -->
2345 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
2346 Operations</a> </div>
2347 <div class="doc_text">
2348 <p>Bitwise binary operators are used to do various forms of
2349 bit-twiddling in a program. They are generally very efficient
2350 instructions and can commonly be strength reduced from other
2351 instructions. They require two operands of the same type, execute an operation on them,
2352 and produce a single value. The resulting value is the same type as its operands.</p>
2355 <!-- _______________________________________________________________________ -->
2356 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
2357 Instruction</a> </div>
2358 <div class="doc_text">
2360 <pre> <result> = shl <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2365 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2366 the left a specified number of bits.</p>
2370 <p>Both arguments to the '<tt>shl</tt>' instruction must be the same <a
2371 href="#t_integer">integer</a> type. '<tt>var2</tt>' is treated as an
2376 <p>The value produced is <tt>var1</tt> * 2<sup><tt>var2</tt></sup> mod 2<sup>n</sup>,
2377 where n is the width of the result. If <tt>var2</tt> is (statically or dynamically) negative or
2378 equal to or larger than the number of bits in <tt>var1</tt>, the result is undefined.</p>
2380 <h5>Example:</h5><pre>
2381 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
2382 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
2383 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
2384 <result> = shl i32 1, 32 <i>; undefined</i>
2387 <!-- _______________________________________________________________________ -->
2388 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
2389 Instruction</a> </div>
2390 <div class="doc_text">
2392 <pre> <result> = lshr <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2396 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
2397 operand shifted to the right a specified number of bits with zero fill.</p>
2400 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
2401 <a href="#t_integer">integer</a> type. '<tt>var2</tt>' is treated as an
2406 <p>This instruction always performs a logical shift right operation. The most
2407 significant bits of the result will be filled with zero bits after the
2408 shift. If <tt>var2</tt> is (statically or dynamically) equal to or larger than
2409 the number of bits in <tt>var1</tt>, the result is undefined.</p>
2413 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
2414 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
2415 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
2416 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
2417 <result> = lshr i32 1, 32 <i>; undefined</i>
2421 <!-- _______________________________________________________________________ -->
2422 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
2423 Instruction</a> </div>
2424 <div class="doc_text">
2427 <pre> <result> = ashr <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2431 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
2432 operand shifted to the right a specified number of bits with sign extension.</p>
2435 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
2436 <a href="#t_integer">integer</a> type. '<tt>var2</tt>' is treated as an
2440 <p>This instruction always performs an arithmetic shift right operation,
2441 The most significant bits of the result will be filled with the sign bit
2442 of <tt>var1</tt>. If <tt>var2</tt> is (statically or dynamically) equal to or
2443 larger than the number of bits in <tt>var1</tt>, the result is undefined.
2448 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
2449 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
2450 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
2451 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
2452 <result> = ashr i32 1, 32 <i>; undefined</i>
2456 <!-- _______________________________________________________________________ -->
2457 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
2458 Instruction</a> </div>
2459 <div class="doc_text">
2461 <pre> <result> = and <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2464 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
2465 its two operands.</p>
2467 <p>The two arguments to the '<tt>and</tt>' instruction must be <a
2468 href="#t_integer">integer</a> values. Both arguments must have
2469 identical types.</p>
2471 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
2473 <div style="align: center">
2474 <table border="1" cellspacing="0" cellpadding="4">
2505 <pre> <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
2506 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
2507 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
2510 <!-- _______________________________________________________________________ -->
2511 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
2512 <div class="doc_text">
2514 <pre> <result> = or <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2517 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
2518 or of its two operands.</p>
2520 <p>The two arguments to the '<tt>or</tt>' instruction must be <a
2521 href="#t_integer">integer</a> values. Both arguments must have
2522 identical types.</p>
2524 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
2526 <div style="align: center">
2527 <table border="1" cellspacing="0" cellpadding="4">
2558 <pre> <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
2559 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
2560 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
2563 <!-- _______________________________________________________________________ -->
2564 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
2565 Instruction</a> </div>
2566 <div class="doc_text">
2568 <pre> <result> = xor <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2571 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
2572 or of its two operands. The <tt>xor</tt> is used to implement the
2573 "one's complement" operation, which is the "~" operator in C.</p>
2575 <p>The two arguments to the '<tt>xor</tt>' instruction must be <a
2576 href="#t_integer">integer</a> values. Both arguments must have
2577 identical types.</p>
2579 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
2581 <div style="align: center">
2582 <table border="1" cellspacing="0" cellpadding="4">
2614 <pre> <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
2615 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
2616 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
2617 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
2621 <!-- ======================================================================= -->
2622 <div class="doc_subsection">
2623 <a name="vectorops">Vector Operations</a>
2626 <div class="doc_text">
2628 <p>LLVM supports several instructions to represent vector operations in a
2629 target-independent manner. These instructions cover the element-access and
2630 vector-specific operations needed to process vectors effectively. While LLVM
2631 does directly support these vector operations, many sophisticated algorithms
2632 will want to use target-specific intrinsics to take full advantage of a specific
2637 <!-- _______________________________________________________________________ -->
2638 <div class="doc_subsubsection">
2639 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2642 <div class="doc_text">
2647 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
2653 The '<tt>extractelement</tt>' instruction extracts a single scalar
2654 element from a vector at a specified index.
2661 The first operand of an '<tt>extractelement</tt>' instruction is a
2662 value of <a href="#t_vector">vector</a> type. The second operand is
2663 an index indicating the position from which to extract the element.
2664 The index may be a variable.</p>
2669 The result is a scalar of the same type as the element type of
2670 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2671 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2672 results are undefined.
2678 %result = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
2683 <!-- _______________________________________________________________________ -->
2684 <div class="doc_subsubsection">
2685 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
2688 <div class="doc_text">
2693 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
2699 The '<tt>insertelement</tt>' instruction inserts a scalar
2700 element into a vector at a specified index.
2707 The first operand of an '<tt>insertelement</tt>' instruction is a
2708 value of <a href="#t_vector">vector</a> type. The second operand is a
2709 scalar value whose type must equal the element type of the first
2710 operand. The third operand is an index indicating the position at
2711 which to insert the value. The index may be a variable.</p>
2716 The result is a vector of the same type as <tt>val</tt>. Its
2717 element values are those of <tt>val</tt> except at position
2718 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
2719 exceeds the length of <tt>val</tt>, the results are undefined.
2725 %result = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
2729 <!-- _______________________________________________________________________ -->
2730 <div class="doc_subsubsection">
2731 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
2734 <div class="doc_text">
2739 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <n x i32> <mask> <i>; yields <n x <ty>></i>
2745 The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
2746 from two input vectors, returning a vector of the same type.
2752 The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
2753 with types that match each other and types that match the result of the
2754 instruction. The third argument is a shuffle mask, which has the same number
2755 of elements as the other vector type, but whose element type is always 'i32'.
2759 The shuffle mask operand is required to be a constant vector with either
2760 constant integer or undef values.
2766 The elements of the two input vectors are numbered from left to right across
2767 both of the vectors. The shuffle mask operand specifies, for each element of
2768 the result vector, which element of the two input registers the result element
2769 gets. The element selector may be undef (meaning "don't care") and the second
2770 operand may be undef if performing a shuffle from only one vector.
2776 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
2777 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
2778 %result = shufflevector <4 x i32> %v1, <4 x i32> undef,
2779 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
2784 <!-- ======================================================================= -->
2785 <div class="doc_subsection">
2786 <a name="aggregateops">Aggregate Operations</a>
2789 <div class="doc_text">
2791 <p>LLVM supports several instructions for working with aggregate values.
2796 <!-- _______________________________________________________________________ -->
2797 <div class="doc_subsubsection">
2798 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
2801 <div class="doc_text">
2806 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
2812 The '<tt>extractvalue</tt>' instruction extracts the value of a struct field
2813 or array element from an aggregate value.
2820 The first operand of an '<tt>extractvalue</tt>' instruction is a
2821 value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a>
2822 type. The operands are constant indices to specify which value to extract
2823 in the same manner as indices in a
2824 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
2830 The result is the value at the position in the aggregate specified by
2837 %result = extractvalue {i32, float} %agg, i32 0 <i>; yields i32</i>
2842 <!-- _______________________________________________________________________ -->
2843 <div class="doc_subsubsection">
2844 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
2847 <div class="doc_text">
2852 <result> = insertvalue <aggregate type> <val>, <ty> <val>, i32 <idx> <i>; yields <n x <ty>></i>
2858 The '<tt>insertvalue</tt>' instruction inserts a value
2859 into a struct field or array element in an aggregate.
2866 The first operand of an '<tt>insertvalue</tt>' instruction is a
2867 value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a> type.
2868 The second operand is a first-class value to insert.
2869 type of the first operand. The following operands are constant indices
2870 indicating the position at which to insert the value in the same manner as
2872 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
2873 The value to insert must have the same type as the value identified
2879 The result is an aggregate of the same type as <tt>val</tt>. Its
2880 value is that of <tt>val</tt> except that the value at the position
2881 specified by the indices is that of <tt>elt</tt>.
2887 %result = insertvalue {i32, float} %agg, i32 1, i32 0 <i>; yields {i32, float}</i>
2892 <!-- ======================================================================= -->
2893 <div class="doc_subsection">
2894 <a name="memoryops">Memory Access and Addressing Operations</a>
2897 <div class="doc_text">
2899 <p>A key design point of an SSA-based representation is how it
2900 represents memory. In LLVM, no memory locations are in SSA form, which
2901 makes things very simple. This section describes how to read, write,
2902 allocate, and free memory in LLVM.</p>
2906 <!-- _______________________________________________________________________ -->
2907 <div class="doc_subsubsection">
2908 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
2911 <div class="doc_text">
2916 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2921 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
2922 heap and returns a pointer to it. The object is always allocated in the generic
2923 address space (address space zero).</p>
2927 <p>The '<tt>malloc</tt>' instruction allocates
2928 <tt>sizeof(<type>)*NumElements</tt>
2929 bytes of memory from the operating system and returns a pointer of the
2930 appropriate type to the program. If "NumElements" is specified, it is the
2931 number of elements allocated, otherwise "NumElements" is defaulted to be one.
2932 If a constant alignment is specified, the value result of the allocation is guaranteed to
2933 be aligned to at least that boundary. If not specified, or if zero, the target can
2934 choose to align the allocation on any convenient boundary.</p>
2936 <p>'<tt>type</tt>' must be a sized type.</p>
2940 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
2941 a pointer is returned. The result of a zero byte allocattion is undefined. The
2942 result is null if there is insufficient memory available.</p>
2947 %array = malloc [4 x i8 ] <i>; yields {[%4 x i8]*}:array</i>
2949 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
2950 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
2951 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
2952 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
2953 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
2957 <!-- _______________________________________________________________________ -->
2958 <div class="doc_subsubsection">
2959 <a name="i_free">'<tt>free</tt>' Instruction</a>
2962 <div class="doc_text">
2967 free <type> <value> <i>; yields {void}</i>
2972 <p>The '<tt>free</tt>' instruction returns memory back to the unused
2973 memory heap to be reallocated in the future.</p>
2977 <p>'<tt>value</tt>' shall be a pointer value that points to a value
2978 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
2983 <p>Access to the memory pointed to by the pointer is no longer defined
2984 after this instruction executes. If the pointer is null, the operation
2990 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
2991 free [4 x i8]* %array
2995 <!-- _______________________________________________________________________ -->
2996 <div class="doc_subsubsection">
2997 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
3000 <div class="doc_text">
3005 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3010 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
3011 currently executing function, to be automatically released when this function
3012 returns to its caller. The object is always allocated in the generic address
3013 space (address space zero).</p>
3017 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
3018 bytes of memory on the runtime stack, returning a pointer of the
3019 appropriate type to the program. If "NumElements" is specified, it is the
3020 number of elements allocated, otherwise "NumElements" is defaulted to be one.
3021 If a constant alignment is specified, the value result of the allocation is guaranteed
3022 to be aligned to at least that boundary. If not specified, or if zero, the target
3023 can choose to align the allocation on any convenient boundary.</p>
3025 <p>'<tt>type</tt>' may be any sized type.</p>
3029 <p>Memory is allocated; a pointer is returned. The operation is undefiend if
3030 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
3031 memory is automatically released when the function returns. The '<tt>alloca</tt>'
3032 instruction is commonly used to represent automatic variables that must
3033 have an address available. When the function returns (either with the <tt><a
3034 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
3035 instructions), the memory is reclaimed. Allocating zero bytes
3036 is legal, but the result is undefined.</p>
3041 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
3042 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
3043 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
3044 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
3048 <!-- _______________________________________________________________________ -->
3049 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
3050 Instruction</a> </div>
3051 <div class="doc_text">
3053 <pre> <result> = load <ty>* <pointer>[, align <alignment>]<br> <result> = volatile load <ty>* <pointer>[, align <alignment>]<br></pre>
3055 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
3057 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
3058 address from which to load. The pointer must point to a <a
3059 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
3060 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
3061 the number or order of execution of this <tt>load</tt> with other
3062 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
3065 The optional constant "align" argument specifies the alignment of the operation
3066 (that is, the alignment of the memory address). A value of 0 or an
3067 omitted "align" argument means that the operation has the preferential
3068 alignment for the target. It is the responsibility of the code emitter
3069 to ensure that the alignment information is correct. Overestimating
3070 the alignment results in an undefined behavior. Underestimating the
3071 alignment may produce less efficient code. An alignment of 1 is always
3075 <p>The location of memory pointed to is loaded.</p>
3077 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3079 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
3080 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
3083 <!-- _______________________________________________________________________ -->
3084 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
3085 Instruction</a> </div>
3086 <div class="doc_text">
3088 <pre> store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3089 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3092 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
3094 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
3095 to store and an address at which to store it. The type of the '<tt><pointer></tt>'
3096 operand must be a pointer to the <a href="#t_firstclass">first class</a> type
3097 of the '<tt><value></tt>'
3098 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
3099 optimizer is not allowed to modify the number or order of execution of
3100 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
3101 href="#i_store">store</a></tt> instructions.</p>
3103 The optional constant "align" argument specifies the alignment of the operation
3104 (that is, the alignment of the memory address). A value of 0 or an
3105 omitted "align" argument means that the operation has the preferential
3106 alignment for the target. It is the responsibility of the code emitter
3107 to ensure that the alignment information is correct. Overestimating
3108 the alignment results in an undefined behavior. Underestimating the
3109 alignment may produce less efficient code. An alignment of 1 is always
3113 <p>The contents of memory are updated to contain '<tt><value></tt>'
3114 at the location specified by the '<tt><pointer></tt>' operand.</p>
3116 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3117 store i32 3, i32* %ptr <i>; yields {void}</i>
3118 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
3122 <!-- _______________________________________________________________________ -->
3123 <div class="doc_subsubsection">
3124 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
3127 <div class="doc_text">
3130 <result> = getelementptr <ty>* <ptrval>{, <ty> <idx>}*
3136 The '<tt>getelementptr</tt>' instruction is used to get the address of a
3137 subelement of an aggregate data structure.</p>
3141 <p>This instruction takes a list of integer operands that indicate what
3142 elements of the aggregate object to index to. The actual types of the arguments
3143 provided depend on the type of the first pointer argument. The
3144 '<tt>getelementptr</tt>' instruction is used to index down through the type
3145 levels of a structure or to a specific index in an array. When indexing into a
3146 structure, only <tt>i32</tt> integer constants are allowed. When indexing
3147 into an array or pointer, only integers of 32 or 64 bits are allowed; 32-bit
3148 values will be sign extended to 64-bits if required.</p>
3150 <p>For example, let's consider a C code fragment and how it gets
3151 compiled to LLVM:</p>
3153 <div class="doc_code">
3166 int *foo(struct ST *s) {
3167 return &s[1].Z.B[5][13];
3172 <p>The LLVM code generated by the GCC frontend is:</p>
3174 <div class="doc_code">
3176 %RT = type { i8 , [10 x [20 x i32]], i8 }
3177 %ST = type { i32, double, %RT }
3179 define i32* %foo(%ST* %s) {
3181 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
3189 <p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
3190 on the pointer type that is being indexed into. <a href="#t_pointer">Pointer</a>
3191 and <a href="#t_array">array</a> types can use a 32-bit or 64-bit
3192 <a href="#t_integer">integer</a> type but the value will always be sign extended
3193 to 64-bits. <a href="#t_struct">Structure</a> and <a href="#t_pstruct">packed
3194 structure</a> types require <tt>i32</tt> <b>constants</b>.</p>
3196 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
3197 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
3198 }</tt>' type, a structure. The second index indexes into the third element of
3199 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
3200 i8 }</tt>' type, another structure. The third index indexes into the second
3201 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
3202 array. The two dimensions of the array are subscripted into, yielding an
3203 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
3204 to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
3206 <p>Note that it is perfectly legal to index partially through a
3207 structure, returning a pointer to an inner element. Because of this,
3208 the LLVM code for the given testcase is equivalent to:</p>
3211 define i32* %foo(%ST* %s) {
3212 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
3213 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
3214 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
3215 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
3216 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
3221 <p>Note that it is undefined to access an array out of bounds: array and
3222 pointer indexes must always be within the defined bounds of the array type.
3223 The one exception for this rule is zero length arrays. These arrays are
3224 defined to be accessible as variable length arrays, which requires access
3225 beyond the zero'th element.</p>
3227 <p>The getelementptr instruction is often confusing. For some more insight
3228 into how it works, see <a href="GetElementPtr.html">the getelementptr
3234 <i>; yields [12 x i8]*:aptr</i>
3235 %aptr = getelementptr {i32, [12 x i8]}* %sptr, i64 0, i32 1
3239 <!-- ======================================================================= -->
3240 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
3242 <div class="doc_text">
3243 <p>The instructions in this category are the conversion instructions (casting)
3244 which all take a single operand and a type. They perform various bit conversions
3248 <!-- _______________________________________________________________________ -->
3249 <div class="doc_subsubsection">
3250 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
3252 <div class="doc_text">
3256 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
3261 The '<tt>trunc</tt>' instruction truncates its operand to the type <tt>ty2</tt>.
3266 The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
3267 be an <a href="#t_integer">integer</a> type, and a type that specifies the size
3268 and type of the result, which must be an <a href="#t_integer">integer</a>
3269 type. The bit size of <tt>value</tt> must be larger than the bit size of
3270 <tt>ty2</tt>. Equal sized types are not allowed.</p>
3274 The '<tt>trunc</tt>' instruction truncates the high order bits in <tt>value</tt>
3275 and converts the remaining bits to <tt>ty2</tt>. Since the source size must be
3276 larger than the destination size, <tt>trunc</tt> cannot be a <i>no-op cast</i>.
3277 It will always truncate bits.</p>
3281 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
3282 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
3283 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
3287 <!-- _______________________________________________________________________ -->
3288 <div class="doc_subsubsection">
3289 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
3291 <div class="doc_text">
3295 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
3299 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
3304 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
3305 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3306 also be of <a href="#t_integer">integer</a> type. The bit size of the
3307 <tt>value</tt> must be smaller than the bit size of the destination type,
3311 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
3312 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
3314 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
3318 %X = zext i32 257 to i64 <i>; yields i64:257</i>
3319 %Y = zext i1 true to i32 <i>; yields i32:1</i>
3323 <!-- _______________________________________________________________________ -->
3324 <div class="doc_subsubsection">
3325 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
3327 <div class="doc_text">
3331 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
3335 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
3339 The '<tt>sext</tt>' instruction takes a value to cast, which must be of
3340 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3341 also be of <a href="#t_integer">integer</a> type. The bit size of the
3342 <tt>value</tt> must be smaller than the bit size of the destination type,
3347 The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
3348 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size of
3349 the type <tt>ty2</tt>.</p>
3351 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
3355 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
3356 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
3360 <!-- _______________________________________________________________________ -->
3361 <div class="doc_subsubsection">
3362 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
3365 <div class="doc_text">
3370 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
3374 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
3379 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
3380 point</a> value to cast and a <a href="#t_floating">floating point</a> type to
3381 cast it to. The size of <tt>value</tt> must be larger than the size of
3382 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
3383 <i>no-op cast</i>.</p>
3386 <p> The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
3387 <a href="#t_floating">floating point</a> type to a smaller
3388 <a href="#t_floating">floating point</a> type. If the value cannot fit within
3389 the destination type, <tt>ty2</tt>, then the results are undefined.</p>
3393 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
3394 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
3398 <!-- _______________________________________________________________________ -->
3399 <div class="doc_subsubsection">
3400 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
3402 <div class="doc_text">
3406 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
3410 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
3411 floating point value.</p>
3414 <p>The '<tt>fpext</tt>' instruction takes a
3415 <a href="#t_floating">floating point</a> <tt>value</tt> to cast,
3416 and a <a href="#t_floating">floating point</a> type to cast it to. The source
3417 type must be smaller than the destination type.</p>
3420 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
3421 <a href="#t_floating">floating point</a> type to a larger
3422 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
3423 used to make a <i>no-op cast</i> because it always changes bits. Use
3424 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
3428 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
3429 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
3433 <!-- _______________________________________________________________________ -->
3434 <div class="doc_subsubsection">
3435 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
3437 <div class="doc_text">
3441 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
3445 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
3446 unsigned integer equivalent of type <tt>ty2</tt>.
3450 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
3451 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3452 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3453 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3454 vector integer type with the same number of elements as <tt>ty</tt></p>
3457 <p> The '<tt>fptoui</tt>' instruction converts its
3458 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3459 towards zero) unsigned integer value. If the value cannot fit in <tt>ty2</tt>,
3460 the results are undefined.</p>
3464 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
3465 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
3466 %X = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
3470 <!-- _______________________________________________________________________ -->
3471 <div class="doc_subsubsection">
3472 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
3474 <div class="doc_text">
3478 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
3482 <p>The '<tt>fptosi</tt>' instruction converts
3483 <a href="#t_floating">floating point</a> <tt>value</tt> to type <tt>ty2</tt>.
3487 <p> The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
3488 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3489 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3490 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3491 vector integer type with the same number of elements as <tt>ty</tt></p>
3494 <p>The '<tt>fptosi</tt>' instruction converts its
3495 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3496 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
3497 the results are undefined.</p>
3501 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
3502 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
3503 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
3507 <!-- _______________________________________________________________________ -->
3508 <div class="doc_subsubsection">
3509 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
3511 <div class="doc_text">
3515 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3519 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
3520 integer and converts that value to the <tt>ty2</tt> type.</p>
3523 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
3524 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3525 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3526 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3527 floating point type with the same number of elements as <tt>ty</tt></p>
3530 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
3531 integer quantity and converts it to the corresponding floating point value. If
3532 the value cannot fit in the floating point value, the results are undefined.</p>
3536 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
3537 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
3541 <!-- _______________________________________________________________________ -->
3542 <div class="doc_subsubsection">
3543 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
3545 <div class="doc_text">
3549 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3553 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed
3554 integer and converts that value to the <tt>ty2</tt> type.</p>
3557 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
3558 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3559 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3560 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3561 floating point type with the same number of elements as <tt>ty</tt></p>
3564 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed
3565 integer quantity and converts it to the corresponding floating point value. If
3566 the value cannot fit in the floating point value, the results are undefined.</p>
3570 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
3571 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
3575 <!-- _______________________________________________________________________ -->
3576 <div class="doc_subsubsection">
3577 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
3579 <div class="doc_text">
3583 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
3587 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
3588 the integer type <tt>ty2</tt>.</p>
3591 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
3592 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
3593 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.
3596 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
3597 <tt>ty2</tt> by interpreting the pointer value as an integer and either
3598 truncating or zero extending that value to the size of the integer type. If
3599 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
3600 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
3601 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
3606 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
3607 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
3611 <!-- _______________________________________________________________________ -->
3612 <div class="doc_subsubsection">
3613 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
3615 <div class="doc_text">
3619 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
3623 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to
3624 a pointer type, <tt>ty2</tt>.</p>
3627 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
3628 value to cast, and a type to cast it to, which must be a
3629 <a href="#t_pointer">pointer</a> type.
3632 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
3633 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
3634 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
3635 size of a pointer then a truncation is done. If <tt>value</tt> is smaller than
3636 the size of a pointer then a zero extension is done. If they are the same size,
3637 nothing is done (<i>no-op cast</i>).</p>
3641 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
3642 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
3643 %Y = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
3647 <!-- _______________________________________________________________________ -->
3648 <div class="doc_subsubsection">
3649 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
3651 <div class="doc_text">
3655 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
3659 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3660 <tt>ty2</tt> without changing any bits.</p>
3663 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be
3664 a first class value, and a type to cast it to, which must also be a <a
3665 href="#t_firstclass">first class</a> type. The bit sizes of <tt>value</tt>
3666 and the destination type, <tt>ty2</tt>, must be identical. If the source
3667 type is a pointer, the destination type must also be a pointer.</p>
3670 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3671 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
3672 this conversion. The conversion is done as if the <tt>value</tt> had been
3673 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only be
3674 converted to other pointer types with this instruction. To convert pointers to
3675 other types, use the <a href="#i_inttoptr">inttoptr</a> or
3676 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
3680 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
3681 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
3682 %Z = bitcast <2xint> %V to i64; <i>; yields i64: %V</i>
3686 <!-- ======================================================================= -->
3687 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
3688 <div class="doc_text">
3689 <p>The instructions in this category are the "miscellaneous"
3690 instructions, which defy better classification.</p>
3693 <!-- _______________________________________________________________________ -->
3694 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
3696 <div class="doc_text">
3698 <pre> <result> = icmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3701 <p>The '<tt>icmp</tt>' instruction returns a boolean value based on comparison
3702 of its two integer or pointer operands.</p>
3704 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
3705 the condition code indicating the kind of comparison to perform. It is not
3706 a value, just a keyword. The possible condition code are:
3708 <li><tt>eq</tt>: equal</li>
3709 <li><tt>ne</tt>: not equal </li>
3710 <li><tt>ugt</tt>: unsigned greater than</li>
3711 <li><tt>uge</tt>: unsigned greater or equal</li>
3712 <li><tt>ult</tt>: unsigned less than</li>
3713 <li><tt>ule</tt>: unsigned less or equal</li>
3714 <li><tt>sgt</tt>: signed greater than</li>
3715 <li><tt>sge</tt>: signed greater or equal</li>
3716 <li><tt>slt</tt>: signed less than</li>
3717 <li><tt>sle</tt>: signed less or equal</li>
3719 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
3720 <a href="#t_pointer">pointer</a> typed. They must also be identical types.</p>
3722 <p>The '<tt>icmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3723 the condition code given as <tt>cond</tt>. The comparison performed always
3724 yields a <a href="#t_primitive">i1</a> result, as follows:
3726 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
3727 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3729 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
3730 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3731 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
3732 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3733 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
3734 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3735 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
3736 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3737 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
3738 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3739 <li><tt>sgt</tt>: interprets the operands as signed values and yields
3740 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3741 <li><tt>sge</tt>: interprets the operands as signed values and yields
3742 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3743 <li><tt>slt</tt>: interprets the operands as signed values and yields
3744 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3745 <li><tt>sle</tt>: interprets the operands as signed values and yields
3746 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3748 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
3749 values are compared as if they were integers.</p>
3752 <pre> <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
3753 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
3754 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
3755 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
3756 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
3757 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
3761 <!-- _______________________________________________________________________ -->
3762 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
3764 <div class="doc_text">
3766 <pre> <result> = fcmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3769 <p>The '<tt>fcmp</tt>' instruction returns a boolean value based on comparison
3770 of its floating point operands.</p>
3772 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
3773 the condition code indicating the kind of comparison to perform. It is not
3774 a value, just a keyword. The possible condition code are:
3776 <li><tt>false</tt>: no comparison, always returns false</li>
3777 <li><tt>oeq</tt>: ordered and equal</li>
3778 <li><tt>ogt</tt>: ordered and greater than </li>
3779 <li><tt>oge</tt>: ordered and greater than or equal</li>
3780 <li><tt>olt</tt>: ordered and less than </li>
3781 <li><tt>ole</tt>: ordered and less than or equal</li>
3782 <li><tt>one</tt>: ordered and not equal</li>
3783 <li><tt>ord</tt>: ordered (no nans)</li>
3784 <li><tt>ueq</tt>: unordered or equal</li>
3785 <li><tt>ugt</tt>: unordered or greater than </li>
3786 <li><tt>uge</tt>: unordered or greater than or equal</li>
3787 <li><tt>ult</tt>: unordered or less than </li>
3788 <li><tt>ule</tt>: unordered or less than or equal</li>
3789 <li><tt>une</tt>: unordered or not equal</li>
3790 <li><tt>uno</tt>: unordered (either nans)</li>
3791 <li><tt>true</tt>: no comparison, always returns true</li>
3793 <p><i>Ordered</i> means that neither operand is a QNAN while
3794 <i>unordered</i> means that either operand may be a QNAN.</p>
3795 <p>The <tt>val1</tt> and <tt>val2</tt> arguments must be
3796 <a href="#t_floating">floating point</a> typed. They must have identical
3799 <p>The '<tt>fcmp</tt>' instruction compares <tt>var1</tt> and <tt>var2</tt>
3800 according to the condition code given as <tt>cond</tt>. The comparison performed
3801 always yields a <a href="#t_primitive">i1</a> result, as follows:
3803 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
3804 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3805 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3806 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3807 <tt>var1</tt> is greather than <tt>var2</tt>.</li>
3808 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3809 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3810 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3811 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3812 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3813 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3814 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3815 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3816 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
3817 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
3818 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3819 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
3820 <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3821 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
3822 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3823 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
3824 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3825 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
3826 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3827 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
3828 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3829 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
3830 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
3834 <pre> <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
3835 <result> = icmp one float 4.0, 5.0 <i>; yields: result=true</i>
3836 <result> = icmp olt float 4.0, 5.0 <i>; yields: result=true</i>
3837 <result> = icmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
3841 <!-- _______________________________________________________________________ -->
3842 <div class="doc_subsubsection">
3843 <a name="i_vicmp">'<tt>vicmp</tt>' Instruction</a>
3845 <div class="doc_text">
3847 <pre> <result> = vicmp <cond> <ty> <var1>, <var2> <i>; yields {ty}:result</i>
3850 <p>The '<tt>vicmp</tt>' instruction returns an integer vector value based on
3851 element-wise comparison of its two integer vector operands.</p>
3853 <p>The '<tt>vicmp</tt>' instruction takes three operands. The first operand is
3854 the condition code indicating the kind of comparison to perform. It is not
3855 a value, just a keyword. The possible condition code are:
3857 <li><tt>eq</tt>: equal</li>
3858 <li><tt>ne</tt>: not equal </li>
3859 <li><tt>ugt</tt>: unsigned greater than</li>
3860 <li><tt>uge</tt>: unsigned greater or equal</li>
3861 <li><tt>ult</tt>: unsigned less than</li>
3862 <li><tt>ule</tt>: unsigned less or equal</li>
3863 <li><tt>sgt</tt>: signed greater than</li>
3864 <li><tt>sge</tt>: signed greater or equal</li>
3865 <li><tt>slt</tt>: signed less than</li>
3866 <li><tt>sle</tt>: signed less or equal</li>
3868 <p>The remaining two arguments must be <a href="#t_vector">vector</a> of
3869 <a href="#t_integer">integer</a> typed. They must also be identical types.</p>
3871 <p>The '<tt>vicmp</tt>' instruction compares <tt>var1</tt> and <tt>var2</tt>
3872 according to the condition code given as <tt>cond</tt>. The comparison yields a
3873 <a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, of
3874 identical type as the values being compared. The most significant bit in each
3875 element is 1 if the element-wise comparison evaluates to true, and is 0
3876 otherwise. All other bits of the result are undefined. The condition codes
3877 are evaluated identically to the <a href="#i_icmp">'<tt>icmp</tt>'
3882 <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>
3883 <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>
3887 <!-- _______________________________________________________________________ -->
3888 <div class="doc_subsubsection">
3889 <a name="i_vfcmp">'<tt>vfcmp</tt>' Instruction</a>
3891 <div class="doc_text">
3893 <pre> <result> = vfcmp <cond> <ty> <var1>, <var2></pre>
3895 <p>The '<tt>vfcmp</tt>' instruction returns an integer vector value based on
3896 element-wise comparison of its two floating point vector operands. The output
3897 elements have the same width as the input elements.</p>
3899 <p>The '<tt>vfcmp</tt>' instruction takes three operands. The first operand is
3900 the condition code indicating the kind of comparison to perform. It is not
3901 a value, just a keyword. The possible condition code are:
3903 <li><tt>false</tt>: no comparison, always returns false</li>
3904 <li><tt>oeq</tt>: ordered and equal</li>
3905 <li><tt>ogt</tt>: ordered and greater than </li>
3906 <li><tt>oge</tt>: ordered and greater than or equal</li>
3907 <li><tt>olt</tt>: ordered and less than </li>
3908 <li><tt>ole</tt>: ordered and less than or equal</li>
3909 <li><tt>one</tt>: ordered and not equal</li>
3910 <li><tt>ord</tt>: ordered (no nans)</li>
3911 <li><tt>ueq</tt>: unordered or equal</li>
3912 <li><tt>ugt</tt>: unordered or greater than </li>
3913 <li><tt>uge</tt>: unordered or greater than or equal</li>
3914 <li><tt>ult</tt>: unordered or less than </li>
3915 <li><tt>ule</tt>: unordered or less than or equal</li>
3916 <li><tt>une</tt>: unordered or not equal</li>
3917 <li><tt>uno</tt>: unordered (either nans)</li>
3918 <li><tt>true</tt>: no comparison, always returns true</li>
3920 <p>The remaining two arguments must be <a href="#t_vector">vector</a> of
3921 <a href="#t_floating">floating point</a> typed. They must also be identical
3924 <p>The '<tt>vfcmp</tt>' instruction compares <tt>var1</tt> and <tt>var2</tt>
3925 according to the condition code given as <tt>cond</tt>. The comparison yields a
3926 <a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, with
3927 an identical number of elements as the values being compared, and each element
3928 having identical with to the width of the floating point elements. The most
3929 significant bit in each element is 1 if the element-wise comparison evaluates to
3930 true, and is 0 otherwise. All other bits of the result are undefined. The
3931 condition codes are evaluated identically to the
3932 <a href="#i_fcmp">'<tt>fcmp</tt>' instruction</a>.
3936 <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>
3937 <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>
3941 <!-- _______________________________________________________________________ -->
3942 <div class="doc_subsubsection"> <a name="i_phi">'<tt>phi</tt>'
3943 Instruction</a> </div>
3944 <div class="doc_text">
3946 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
3948 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
3949 the SSA graph representing the function.</p>
3951 <p>The type of the incoming values is specified with the first type
3952 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
3953 as arguments, with one pair for each predecessor basic block of the
3954 current block. Only values of <a href="#t_firstclass">first class</a>
3955 type may be used as the value arguments to the PHI node. Only labels
3956 may be used as the label arguments.</p>
3957 <p>There must be no non-phi instructions between the start of a basic
3958 block and the PHI instructions: i.e. PHI instructions must be first in
3961 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
3962 specified by the pair corresponding to the predecessor basic block that executed
3963 just prior to the current block.</p>
3965 <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>
3968 <!-- _______________________________________________________________________ -->
3969 <div class="doc_subsubsection">
3970 <a name="i_select">'<tt>select</tt>' Instruction</a>
3973 <div class="doc_text">
3978 <result> = select i1 <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
3984 The '<tt>select</tt>' instruction is used to choose one value based on a
3985 condition, without branching.
3992 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.
3998 If the boolean condition evaluates to true, the instruction returns the first
3999 value argument; otherwise, it returns the second value argument.
4005 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
4010 <!-- _______________________________________________________________________ -->
4011 <div class="doc_subsubsection">
4012 <a name="i_call">'<tt>call</tt>' Instruction</a>
4015 <div class="doc_text">
4019 <result> = [tail] call [<a href="#callingconv">cconv</a>] <ty> [<fnty>*] <fnptrval>(<param list>)
4024 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
4028 <p>This instruction requires several arguments:</p>
4032 <p>The optional "tail" marker indicates whether the callee function accesses
4033 any allocas or varargs in the caller. If the "tail" marker is present, the
4034 function call is eligible for tail call optimization. Note that calls may
4035 be marked "tail" even if they do not occur before a <a
4036 href="#i_ret"><tt>ret</tt></a> instruction.
4039 <p>The optional "cconv" marker indicates which <a href="#callingconv">calling
4040 convention</a> the call should use. If none is specified, the call defaults
4041 to using C calling conventions.
4044 <p>'<tt>ty</tt>': the type of the call instruction itself which is also
4045 the type of the return value. Functions that return no value are marked
4046 <tt><a href="#t_void">void</a></tt>.</p>
4049 <p>'<tt>fnty</tt>': shall be the signature of the pointer to function
4050 value being invoked. The argument types must match the types implied by
4051 this signature. This type can be omitted if the function is not varargs
4052 and if the function type does not return a pointer to a function.</p>
4055 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
4056 be invoked. In most cases, this is a direct function invocation, but
4057 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
4058 to function value.</p>
4061 <p>'<tt>function args</tt>': argument list whose types match the
4062 function signature argument types. All arguments must be of
4063 <a href="#t_firstclass">first class</a> type. If the function signature
4064 indicates the function accepts a variable number of arguments, the extra
4065 arguments can be specified.</p>
4071 <p>The '<tt>call</tt>' instruction is used to cause control flow to
4072 transfer to a specified function, with its incoming arguments bound to
4073 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
4074 instruction in the called function, control flow continues with the
4075 instruction after the function call, and the return value of the
4076 function is bound to the result argument. If the callee returns multiple
4077 values then the return values of the function are only accessible through
4078 the '<tt><a href="#i_getresult">getresult</a></tt>' instruction.</p>
4083 %retval = call i32 @test(i32 %argc)
4084 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42) <i>; yields i32</i>
4085 %X = tail call i32 @foo() <i>; yields i32</i>
4086 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
4087 call void %foo(i8 97 signext)
4089 %struct.A = type { i32, i8 }
4090 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
4091 %gr = getresult %struct.A %r, 0 <i>; yields i32</i>
4092 %gr1 = getresult %struct.A %r, 1 <i>; yields i8</i>
4097 <!-- _______________________________________________________________________ -->
4098 <div class="doc_subsubsection">
4099 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
4102 <div class="doc_text">
4107 <resultval> = va_arg <va_list*> <arglist>, <argty>
4112 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
4113 the "variable argument" area of a function call. It is used to implement the
4114 <tt>va_arg</tt> macro in C.</p>
4118 <p>This instruction takes a <tt>va_list*</tt> value and the type of
4119 the argument. It returns a value of the specified argument type and
4120 increments the <tt>va_list</tt> to point to the next argument. The
4121 actual type of <tt>va_list</tt> is target specific.</p>
4125 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
4126 type from the specified <tt>va_list</tt> and causes the
4127 <tt>va_list</tt> to point to the next argument. For more information,
4128 see the variable argument handling <a href="#int_varargs">Intrinsic
4131 <p>It is legal for this instruction to be called in a function which does not
4132 take a variable number of arguments, for example, the <tt>vfprintf</tt>
4135 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
4136 href="#intrinsics">intrinsic function</a> because it takes a type as an
4141 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
4145 <!-- _______________________________________________________________________ -->
4146 <div class="doc_subsubsection">
4147 <a name="i_getresult">'<tt>getresult</tt>' Instruction</a>
4150 <div class="doc_text">
4154 <resultval> = getresult <type> <retval>, <index>
4159 <p> The '<tt>getresult</tt>' instruction is used to extract individual values
4160 from a '<tt><a href="#i_call">call</a></tt>'
4161 or '<tt><a href="#i_invoke">invoke</a></tt>' instruction that returns multiple
4166 <p>The '<tt>getresult</tt>' instruction takes a call or invoke value as its
4167 first argument, or an undef value. The value must have <a
4168 href="#t_struct">structure type</a>. The second argument is a constant
4169 unsigned index value which must be in range for the number of values returned
4174 <p>The '<tt>getresult</tt>' instruction extracts the element identified by
4175 '<tt>index</tt>' from the aggregate value.</p>
4180 %struct.A = type { i32, i8 }
4182 %r = call %struct.A @foo()
4183 %gr = getresult %struct.A %r, 0 <i>; yields i32:%gr</i>
4184 %gr1 = getresult %struct.A %r, 1 <i>; yields i8:%gr1</i>
4191 <!-- *********************************************************************** -->
4192 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
4193 <!-- *********************************************************************** -->
4195 <div class="doc_text">
4197 <p>LLVM supports the notion of an "intrinsic function". These functions have
4198 well known names and semantics and are required to follow certain restrictions.
4199 Overall, these intrinsics represent an extension mechanism for the LLVM
4200 language that does not require changing all of the transformations in LLVM when
4201 adding to the language (or the bitcode reader/writer, the parser, etc...).</p>
4203 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
4204 prefix is reserved in LLVM for intrinsic names; thus, function names may not
4205 begin with this prefix. Intrinsic functions must always be external functions:
4206 you cannot define the body of intrinsic functions. Intrinsic functions may
4207 only be used in call or invoke instructions: it is illegal to take the address
4208 of an intrinsic function. Additionally, because intrinsic functions are part
4209 of the LLVM language, it is required if any are added that they be documented
4212 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
4213 a family of functions that perform the same operation but on different data
4214 types. Because LLVM can represent over 8 million different integer types,
4215 overloading is used commonly to allow an intrinsic function to operate on any
4216 integer type. One or more of the argument types or the result type can be
4217 overloaded to accept any integer type. Argument types may also be defined as
4218 exactly matching a previous argument's type or the result type. This allows an
4219 intrinsic function which accepts multiple arguments, but needs all of them to
4220 be of the same type, to only be overloaded with respect to a single argument or
4223 <p>Overloaded intrinsics will have the names of its overloaded argument types
4224 encoded into its function name, each preceded by a period. Only those types
4225 which are overloaded result in a name suffix. Arguments whose type is matched
4226 against another type do not. For example, the <tt>llvm.ctpop</tt> function can
4227 take an integer of any width and returns an integer of exactly the same integer
4228 width. This leads to a family of functions such as
4229 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29 %val)</tt>.
4230 Only one type, the return type, is overloaded, and only one type suffix is
4231 required. Because the argument's type is matched against the return type, it
4232 does not require its own name suffix.</p>
4234 <p>To learn how to add an intrinsic function, please see the
4235 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.
4240 <!-- ======================================================================= -->
4241 <div class="doc_subsection">
4242 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
4245 <div class="doc_text">
4247 <p>Variable argument support is defined in LLVM with the <a
4248 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
4249 intrinsic functions. These functions are related to the similarly
4250 named macros defined in the <tt><stdarg.h></tt> header file.</p>
4252 <p>All of these functions operate on arguments that use a
4253 target-specific value type "<tt>va_list</tt>". The LLVM assembly
4254 language reference manual does not define what this type is, so all
4255 transformations should be prepared to handle these functions regardless of
4258 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
4259 instruction and the variable argument handling intrinsic functions are
4262 <div class="doc_code">
4264 define i32 @test(i32 %X, ...) {
4265 ; Initialize variable argument processing
4267 %ap2 = bitcast i8** %ap to i8*
4268 call void @llvm.va_start(i8* %ap2)
4270 ; Read a single integer argument
4271 %tmp = va_arg i8** %ap, i32
4273 ; Demonstrate usage of llvm.va_copy and llvm.va_end
4275 %aq2 = bitcast i8** %aq to i8*
4276 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
4277 call void @llvm.va_end(i8* %aq2)
4279 ; Stop processing of arguments.
4280 call void @llvm.va_end(i8* %ap2)
4284 declare void @llvm.va_start(i8*)
4285 declare void @llvm.va_copy(i8*, i8*)
4286 declare void @llvm.va_end(i8*)
4292 <!-- _______________________________________________________________________ -->
4293 <div class="doc_subsubsection">
4294 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
4298 <div class="doc_text">
4300 <pre> declare void %llvm.va_start(i8* <arglist>)<br></pre>
4302 <P>The '<tt>llvm.va_start</tt>' intrinsic initializes
4303 <tt>*<arglist></tt> for subsequent use by <tt><a
4304 href="#i_va_arg">va_arg</a></tt>.</p>
4308 <P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
4312 <P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
4313 macro available in C. In a target-dependent way, it initializes the
4314 <tt>va_list</tt> element to which the argument points, so that the next call to
4315 <tt>va_arg</tt> will produce the first variable argument passed to the function.
4316 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
4317 last argument of the function as the compiler can figure that out.</p>
4321 <!-- _______________________________________________________________________ -->
4322 <div class="doc_subsubsection">
4323 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
4326 <div class="doc_text">
4328 <pre> declare void @llvm.va_end(i8* <arglist>)<br></pre>
4331 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
4332 which has been initialized previously with <tt><a href="#int_va_start">llvm.va_start</a></tt>
4333 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
4337 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
4341 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
4342 macro available in C. In a target-dependent way, it destroys the
4343 <tt>va_list</tt> element to which the argument points. Calls to <a
4344 href="#int_va_start"><tt>llvm.va_start</tt></a> and <a href="#int_va_copy">
4345 <tt>llvm.va_copy</tt></a> must be matched exactly with calls to
4346 <tt>llvm.va_end</tt>.</p>
4350 <!-- _______________________________________________________________________ -->
4351 <div class="doc_subsubsection">
4352 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
4355 <div class="doc_text">
4360 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
4365 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
4366 from the source argument list to the destination argument list.</p>
4370 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
4371 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
4376 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
4377 macro available in C. In a target-dependent way, it copies the source
4378 <tt>va_list</tt> element into the destination <tt>va_list</tt> element. This
4379 intrinsic is necessary because the <tt><a href="#int_va_start">
4380 llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
4381 example, memory allocation.</p>
4385 <!-- ======================================================================= -->
4386 <div class="doc_subsection">
4387 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
4390 <div class="doc_text">
4393 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
4394 Collection</a> requires the implementation and generation of these intrinsics.
4395 These intrinsics allow identification of <a href="#int_gcroot">GC roots on the
4396 stack</a>, as well as garbage collector implementations that require <a
4397 href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a> barriers.
4398 Front-ends for type-safe garbage collected languages should generate these
4399 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
4400 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
4403 <p>The garbage collection intrinsics only operate on objects in the generic
4404 address space (address space zero).</p>
4408 <!-- _______________________________________________________________________ -->
4409 <div class="doc_subsubsection">
4410 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
4413 <div class="doc_text">
4418 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
4423 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
4424 the code generator, and allows some metadata to be associated with it.</p>
4428 <p>The first argument specifies the address of a stack object that contains the
4429 root pointer. The second pointer (which must be either a constant or a global
4430 value address) contains the meta-data to be associated with the root.</p>
4434 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
4435 location. At compile-time, the code generator generates information to allow
4436 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
4437 intrinsic may only be used in a function which <a href="#gc">specifies a GC
4443 <!-- _______________________________________________________________________ -->
4444 <div class="doc_subsubsection">
4445 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
4448 <div class="doc_text">
4453 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
4458 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
4459 locations, allowing garbage collector implementations that require read
4464 <p>The second argument is the address to read from, which should be an address
4465 allocated from the garbage collector. The first object is a pointer to the
4466 start of the referenced object, if needed by the language runtime (otherwise
4471 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
4472 instruction, but may be replaced with substantially more complex code by the
4473 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
4474 may only be used in a function which <a href="#gc">specifies a GC
4480 <!-- _______________________________________________________________________ -->
4481 <div class="doc_subsubsection">
4482 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
4485 <div class="doc_text">
4490 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
4495 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
4496 locations, allowing garbage collector implementations that require write
4497 barriers (such as generational or reference counting collectors).</p>
4501 <p>The first argument is the reference to store, the second is the start of the
4502 object to store it to, and the third is the address of the field of Obj to
4503 store to. If the runtime does not require a pointer to the object, Obj may be
4508 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
4509 instruction, but may be replaced with substantially more complex code by the
4510 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
4511 may only be used in a function which <a href="#gc">specifies a GC
4518 <!-- ======================================================================= -->
4519 <div class="doc_subsection">
4520 <a name="int_codegen">Code Generator Intrinsics</a>
4523 <div class="doc_text">
4525 These intrinsics are provided by LLVM to expose special features that may only
4526 be implemented with code generator support.
4531 <!-- _______________________________________________________________________ -->
4532 <div class="doc_subsubsection">
4533 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
4536 <div class="doc_text">
4540 declare i8 *@llvm.returnaddress(i32 <level>)
4546 The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
4547 target-specific value indicating the return address of the current function
4548 or one of its callers.
4554 The argument to this intrinsic indicates which function to return the address
4555 for. Zero indicates the calling function, one indicates its caller, etc. The
4556 argument is <b>required</b> to be a constant integer value.
4562 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
4563 the return address of the specified call frame, or zero if it cannot be
4564 identified. The value returned by this intrinsic is likely to be incorrect or 0
4565 for arguments other than zero, so it should only be used for debugging purposes.
4569 Note that calling this intrinsic does not prevent function inlining or other
4570 aggressive transformations, so the value returned may not be that of the obvious
4571 source-language caller.
4576 <!-- _______________________________________________________________________ -->
4577 <div class="doc_subsubsection">
4578 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
4581 <div class="doc_text">
4585 declare i8 *@llvm.frameaddress(i32 <level>)
4591 The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
4592 target-specific frame pointer value for the specified stack frame.
4598 The argument to this intrinsic indicates which function to return the frame
4599 pointer for. Zero indicates the calling function, one indicates its caller,
4600 etc. The argument is <b>required</b> to be a constant integer value.
4606 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
4607 the frame address of the specified call frame, or zero if it cannot be
4608 identified. The value returned by this intrinsic is likely to be incorrect or 0
4609 for arguments other than zero, so it should only be used for debugging purposes.
4613 Note that calling this intrinsic does not prevent function inlining or other
4614 aggressive transformations, so the value returned may not be that of the obvious
4615 source-language caller.
4619 <!-- _______________________________________________________________________ -->
4620 <div class="doc_subsubsection">
4621 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
4624 <div class="doc_text">
4628 declare i8 *@llvm.stacksave()
4634 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
4635 the function stack, for use with <a href="#int_stackrestore">
4636 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
4637 features like scoped automatic variable sized arrays in C99.
4643 This intrinsic returns a opaque pointer value that can be passed to <a
4644 href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
4645 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
4646 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
4647 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
4648 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
4649 that were allocated after the <tt>llvm.stacksave</tt> was executed.
4654 <!-- _______________________________________________________________________ -->
4655 <div class="doc_subsubsection">
4656 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
4659 <div class="doc_text">
4663 declare void @llvm.stackrestore(i8 * %ptr)
4669 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
4670 the function stack to the state it was in when the corresponding <a
4671 href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
4672 useful for implementing language features like scoped automatic variable sized
4679 See the description for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.
4685 <!-- _______________________________________________________________________ -->
4686 <div class="doc_subsubsection">
4687 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
4690 <div class="doc_text">
4694 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>)
4701 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
4702 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
4704 effect on the behavior of the program but can change its performance
4711 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
4712 determining if the fetch should be for a read (0) or write (1), and
4713 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
4714 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
4715 <tt>locality</tt> arguments must be constant integers.
4721 This intrinsic does not modify the behavior of the program. In particular,
4722 prefetches cannot trap and do not produce a value. On targets that support this
4723 intrinsic, the prefetch can provide hints to the processor cache for better
4729 <!-- _______________________________________________________________________ -->
4730 <div class="doc_subsubsection">
4731 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
4734 <div class="doc_text">
4738 declare void @llvm.pcmarker(i32 <id>)
4745 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
4747 code to simulators and other tools. The method is target specific, but it is
4748 expected that the marker will use exported symbols to transmit the PC of the marker.
4749 The marker makes no guarantees that it will remain with any specific instruction
4750 after optimizations. It is possible that the presence of a marker will inhibit
4751 optimizations. The intended use is to be inserted after optimizations to allow
4752 correlations of simulation runs.
4758 <tt>id</tt> is a numerical id identifying the marker.
4764 This intrinsic does not modify the behavior of the program. Backends that do not
4765 support this intrinisic may ignore it.
4770 <!-- _______________________________________________________________________ -->
4771 <div class="doc_subsubsection">
4772 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
4775 <div class="doc_text">
4779 declare i64 @llvm.readcyclecounter( )
4786 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
4787 counter register (or similar low latency, high accuracy clocks) on those targets
4788 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
4789 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
4790 should only be used for small timings.
4796 When directly supported, reading the cycle counter should not modify any memory.
4797 Implementations are allowed to either return a application specific value or a
4798 system wide value. On backends without support, this is lowered to a constant 0.
4803 <!-- ======================================================================= -->
4804 <div class="doc_subsection">
4805 <a name="int_libc">Standard C Library Intrinsics</a>
4808 <div class="doc_text">
4810 LLVM provides intrinsics for a few important standard C library functions.
4811 These intrinsics allow source-language front-ends to pass information about the
4812 alignment of the pointer arguments to the code generator, providing opportunity
4813 for more efficient code generation.
4818 <!-- _______________________________________________________________________ -->
4819 <div class="doc_subsubsection">
4820 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
4823 <div class="doc_text">
4827 declare void @llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
4828 i32 <len>, i32 <align>)
4829 declare void @llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
4830 i64 <len>, i32 <align>)
4836 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4837 location to the destination location.
4841 Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
4842 intrinsics do not return a value, and takes an extra alignment argument.
4848 The first argument is a pointer to the destination, the second is a pointer to
4849 the source. The third argument is an integer argument
4850 specifying the number of bytes to copy, and the fourth argument is the alignment
4851 of the source and destination locations.
4855 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4856 the caller guarantees that both the source and destination pointers are aligned
4863 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4864 location to the destination location, which are not allowed to overlap. It
4865 copies "len" bytes of memory over. If the argument is known to be aligned to
4866 some boundary, this can be specified as the fourth argument, otherwise it should
4872 <!-- _______________________________________________________________________ -->
4873 <div class="doc_subsubsection">
4874 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
4877 <div class="doc_text">
4881 declare void @llvm.memmove.i32(i8 * <dest>, i8 * <src>,
4882 i32 <len>, i32 <align>)
4883 declare void @llvm.memmove.i64(i8 * <dest>, i8 * <src>,
4884 i64 <len>, i32 <align>)
4890 The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
4891 location to the destination location. It is similar to the
4892 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to overlap.
4896 Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
4897 intrinsics do not return a value, and takes an extra alignment argument.
4903 The first argument is a pointer to the destination, the second is a pointer to
4904 the source. The third argument is an integer argument
4905 specifying the number of bytes to copy, and the fourth argument is the alignment
4906 of the source and destination locations.
4910 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4911 the caller guarantees that the source and destination pointers are aligned to
4918 The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
4919 location to the destination location, which may overlap. It
4920 copies "len" bytes of memory over. If the argument is known to be aligned to
4921 some boundary, this can be specified as the fourth argument, otherwise it should
4927 <!-- _______________________________________________________________________ -->
4928 <div class="doc_subsubsection">
4929 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
4932 <div class="doc_text">
4936 declare void @llvm.memset.i32(i8 * <dest>, i8 <val>,
4937 i32 <len>, i32 <align>)
4938 declare void @llvm.memset.i64(i8 * <dest>, i8 <val>,
4939 i64 <len>, i32 <align>)
4945 The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
4950 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
4951 does not return a value, and takes an extra alignment argument.
4957 The first argument is a pointer to the destination to fill, the second is the
4958 byte value to fill it with, the third argument is an integer
4959 argument specifying the number of bytes to fill, and the fourth argument is the
4960 known alignment of destination location.
4964 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4965 the caller guarantees that the destination pointer is aligned to that boundary.
4971 The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
4973 destination location. If the argument is known to be aligned to some boundary,
4974 this can be specified as the fourth argument, otherwise it should be set to 0 or
4980 <!-- _______________________________________________________________________ -->
4981 <div class="doc_subsubsection">
4982 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
4985 <div class="doc_text">
4988 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
4989 floating point or vector of floating point type. Not all targets support all
4992 declare float @llvm.sqrt.f32(float %Val)
4993 declare double @llvm.sqrt.f64(double %Val)
4994 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
4995 declare fp128 @llvm.sqrt.f128(fp128 %Val)
4996 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
5002 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
5003 returning the same value as the libm '<tt>sqrt</tt>' functions would. Unlike
5004 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
5005 negative numbers other than -0.0 (which allows for better optimization, because
5006 there is no need to worry about errno being set). <tt>llvm.sqrt(-0.0)</tt> is
5007 defined to return -0.0 like IEEE sqrt.
5013 The argument and return value are floating point numbers of the same type.
5019 This function returns the sqrt of the specified operand if it is a nonnegative
5020 floating point number.
5024 <!-- _______________________________________________________________________ -->
5025 <div class="doc_subsubsection">
5026 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
5029 <div class="doc_text">
5032 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
5033 floating point or vector of floating point type. Not all targets support all
5036 declare float @llvm.powi.f32(float %Val, i32 %power)
5037 declare double @llvm.powi.f64(double %Val, i32 %power)
5038 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
5039 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
5040 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
5046 The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
5047 specified (positive or negative) power. The order of evaluation of
5048 multiplications is not defined. When a vector of floating point type is
5049 used, the second argument remains a scalar integer value.
5055 The second argument is an integer power, and the first is a value to raise to
5062 This function returns the first value raised to the second power with an
5063 unspecified sequence of rounding operations.</p>
5066 <!-- _______________________________________________________________________ -->
5067 <div class="doc_subsubsection">
5068 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
5071 <div class="doc_text">
5074 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
5075 floating point or vector of floating point type. Not all targets support all
5078 declare float @llvm.sin.f32(float %Val)
5079 declare double @llvm.sin.f64(double %Val)
5080 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
5081 declare fp128 @llvm.sin.f128(fp128 %Val)
5082 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
5088 The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.
5094 The argument and return value are floating point numbers of the same type.
5100 This function returns the sine of the specified operand, returning the
5101 same values as the libm <tt>sin</tt> functions would, and handles error
5102 conditions in the same way.</p>
5105 <!-- _______________________________________________________________________ -->
5106 <div class="doc_subsubsection">
5107 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
5110 <div class="doc_text">
5113 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
5114 floating point or vector of floating point type. Not all targets support all
5117 declare float @llvm.cos.f32(float %Val)
5118 declare double @llvm.cos.f64(double %Val)
5119 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
5120 declare fp128 @llvm.cos.f128(fp128 %Val)
5121 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
5127 The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.
5133 The argument and return value are floating point numbers of the same type.
5139 This function returns the cosine of the specified operand, returning the
5140 same values as the libm <tt>cos</tt> functions would, and handles error
5141 conditions in the same way.</p>
5144 <!-- _______________________________________________________________________ -->
5145 <div class="doc_subsubsection">
5146 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
5149 <div class="doc_text">
5152 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
5153 floating point or vector of floating point type. Not all targets support all
5156 declare float @llvm.pow.f32(float %Val, float %Power)
5157 declare double @llvm.pow.f64(double %Val, double %Power)
5158 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
5159 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
5160 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
5166 The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
5167 specified (positive or negative) power.
5173 The second argument is a floating point power, and the first is a value to
5174 raise to that power.
5180 This function returns the first value raised to the second power,
5182 same values as the libm <tt>pow</tt> functions would, and handles error
5183 conditions in the same way.</p>
5187 <!-- ======================================================================= -->
5188 <div class="doc_subsection">
5189 <a name="int_manip">Bit Manipulation Intrinsics</a>
5192 <div class="doc_text">
5194 LLVM provides intrinsics for a few important bit manipulation operations.
5195 These allow efficient code generation for some algorithms.
5200 <!-- _______________________________________________________________________ -->
5201 <div class="doc_subsubsection">
5202 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
5205 <div class="doc_text">
5208 <p>This is an overloaded intrinsic function. You can use bswap on any integer
5209 type that is an even number of bytes (i.e. BitWidth % 16 == 0).
5211 declare i16 @llvm.bswap.i16(i16 <id>)
5212 declare i32 @llvm.bswap.i32(i32 <id>)
5213 declare i64 @llvm.bswap.i64(i64 <id>)
5219 The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
5220 values with an even number of bytes (positive multiple of 16 bits). These are
5221 useful for performing operations on data that is not in the target's native
5228 The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
5229 and low byte of the input i16 swapped. Similarly, the <tt>llvm.bswap.i32</tt>
5230 intrinsic returns an i32 value that has the four bytes of the input i32
5231 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the returned
5232 i32 will have its bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i48</tt>,
5233 <tt>llvm.bswap.i64</tt> and other intrinsics extend this concept to
5234 additional even-byte lengths (6 bytes, 8 bytes and more, respectively).
5239 <!-- _______________________________________________________________________ -->
5240 <div class="doc_subsubsection">
5241 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
5244 <div class="doc_text">
5247 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
5248 width. Not all targets support all bit widths however.
5250 declare i8 @llvm.ctpop.i8 (i8 <src>)
5251 declare i16 @llvm.ctpop.i16(i16 <src>)
5252 declare i32 @llvm.ctpop.i32(i32 <src>)
5253 declare i64 @llvm.ctpop.i64(i64 <src>)
5254 declare i256 @llvm.ctpop.i256(i256 <src>)
5260 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
5267 The only argument is the value to be counted. The argument may be of any
5268 integer type. The return type must match the argument type.
5274 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
5278 <!-- _______________________________________________________________________ -->
5279 <div class="doc_subsubsection">
5280 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
5283 <div class="doc_text">
5286 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
5287 integer bit width. Not all targets support all bit widths however.
5289 declare i8 @llvm.ctlz.i8 (i8 <src>)
5290 declare i16 @llvm.ctlz.i16(i16 <src>)
5291 declare i32 @llvm.ctlz.i32(i32 <src>)
5292 declare i64 @llvm.ctlz.i64(i64 <src>)
5293 declare i256 @llvm.ctlz.i256(i256 <src>)
5299 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
5300 leading zeros in a variable.
5306 The only argument is the value to be counted. The argument may be of any
5307 integer type. The return type must match the argument type.
5313 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
5314 in a variable. If the src == 0 then the result is the size in bits of the type
5315 of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.
5321 <!-- _______________________________________________________________________ -->
5322 <div class="doc_subsubsection">
5323 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
5326 <div class="doc_text">
5329 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
5330 integer bit width. Not all targets support all bit widths however.
5332 declare i8 @llvm.cttz.i8 (i8 <src>)
5333 declare i16 @llvm.cttz.i16(i16 <src>)
5334 declare i32 @llvm.cttz.i32(i32 <src>)
5335 declare i64 @llvm.cttz.i64(i64 <src>)
5336 declare i256 @llvm.cttz.i256(i256 <src>)
5342 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
5349 The only argument is the value to be counted. The argument may be of any
5350 integer type. The return type must match the argument type.
5356 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
5357 in a variable. If the src == 0 then the result is the size in bits of the type
5358 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
5362 <!-- _______________________________________________________________________ -->
5363 <div class="doc_subsubsection">
5364 <a name="int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic</a>
5367 <div class="doc_text">
5370 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.select</tt>
5371 on any integer bit width.
5373 declare i17 @llvm.part.select.i17 (i17 %val, i32 %loBit, i32 %hiBit)
5374 declare i29 @llvm.part.select.i29 (i29 %val, i32 %loBit, i32 %hiBit)
5378 <p>The '<tt>llvm.part.select</tt>' family of intrinsic functions selects a
5379 range of bits from an integer value and returns them in the same bit width as
5380 the original value.</p>
5383 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5384 any bit width but they must have the same bit width. The second and third
5385 arguments must be <tt>i32</tt> type since they specify only a bit index.</p>
5388 <p>The operation of the '<tt>llvm.part.select</tt>' intrinsic has two modes
5389 of operation: forwards and reverse. If <tt>%loBit</tt> is greater than
5390 <tt>%hiBits</tt> then the intrinsic operates in reverse mode. Otherwise it
5391 operates in forward mode.</p>
5392 <p>In forward mode, this intrinsic is the equivalent of shifting <tt>%val</tt>
5393 right by <tt>%loBit</tt> bits and then ANDing it with a mask with
5394 only the <tt>%hiBit - %loBit</tt> bits set, as follows:</p>
5396 <li>The <tt>%val</tt> is shifted right (LSHR) by the number of bits specified
5397 by <tt>%loBits</tt>. This normalizes the value to the low order bits.</li>
5398 <li>The <tt>%loBits</tt> value is subtracted from the <tt>%hiBits</tt> value
5399 to determine the number of bits to retain.</li>
5400 <li>A mask of the retained bits is created by shifting a -1 value.</li>
5401 <li>The mask is ANDed with <tt>%val</tt> to produce the result.
5403 <p>In reverse mode, a similar computation is made except that the bits are
5404 returned in the reverse order. So, for example, if <tt>X</tt> has the value
5405 <tt>i16 0x0ACF (101011001111)</tt> and we apply
5406 <tt>part.select(i16 X, 8, 3)</tt> to it, we get back the value
5407 <tt>i16 0x0026 (000000100110)</tt>.</p>
5410 <div class="doc_subsubsection">
5411 <a name="int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic</a>
5414 <div class="doc_text">
5417 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.set</tt>
5418 on any integer bit width.
5420 declare i17 @llvm.part.set.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
5421 declare i29 @llvm.part.set.i29.i9 (i29 %val, i9 %repl, i32 %lo, i32 %hi)
5425 <p>The '<tt>llvm.part.set</tt>' family of intrinsic functions replaces a range
5426 of bits in an integer value with another integer value. It returns the integer
5427 with the replaced bits.</p>
5430 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5431 any bit width but they must have the same bit width. <tt>%val</tt> is the value
5432 whose bits will be replaced. The second argument, <tt>%repl</tt> may be an
5433 integer of any bit width. The third and fourth arguments must be <tt>i32</tt>
5434 type since they specify only a bit index.</p>
5437 <p>The operation of the '<tt>llvm.part.set</tt>' intrinsic has two modes
5438 of operation: forwards and reverse. If <tt>%lo</tt> is greater than
5439 <tt>%hi</tt> then the intrinsic operates in reverse mode. Otherwise it
5440 operates in forward mode.</p>
5441 <p>For both modes, the <tt>%repl</tt> value is prepared for use by either
5442 truncating it down to the size of the replacement area or zero extending it
5443 up to that size.</p>
5444 <p>In forward mode, the bits between <tt>%lo</tt> and <tt>%hi</tt> (inclusive)
5445 are replaced with corresponding bits from <tt>%repl</tt>. That is the 0th bit
5446 in <tt>%repl</tt> replaces the <tt>%lo</tt>th bit in <tt>%val</tt> and etc. up
5447 to the <tt>%hi</tt>th bit.
5448 <p>In reverse mode, a similar computation is made except that the bits are
5449 reversed. That is, the <tt>0</tt>th bit in <tt>%repl</tt> replaces the
5450 <tt>%hi</tt> bit in <tt>%val</tt> and etc. down to the <tt>%lo</tt>th bit.
5453 llvm.part.set(0xFFFF, 0, 4, 7) -> 0xFF0F
5454 llvm.part.set(0xFFFF, 0, 7, 4) -> 0xFF0F
5455 llvm.part.set(0xFFFF, 1, 7, 4) -> 0xFF8F
5456 llvm.part.set(0xFFFF, F, 8, 3) -> 0xFFE7
5457 llvm.part.set(0xFFFF, 0, 3, 8) -> 0xFE07
5461 <!-- ======================================================================= -->
5462 <div class="doc_subsection">
5463 <a name="int_debugger">Debugger Intrinsics</a>
5466 <div class="doc_text">
5468 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
5469 are described in the <a
5470 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
5471 Debugging</a> document.
5476 <!-- ======================================================================= -->
5477 <div class="doc_subsection">
5478 <a name="int_eh">Exception Handling Intrinsics</a>
5481 <div class="doc_text">
5482 <p> The LLVM exception handling intrinsics (which all start with
5483 <tt>llvm.eh.</tt> prefix), are described in the <a
5484 href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
5485 Handling</a> document. </p>
5488 <!-- ======================================================================= -->
5489 <div class="doc_subsection">
5490 <a name="int_trampoline">Trampoline Intrinsic</a>
5493 <div class="doc_text">
5495 This intrinsic makes it possible to excise one parameter, marked with
5496 the <tt>nest</tt> attribute, from a function. The result is a callable
5497 function pointer lacking the nest parameter - the caller does not need
5498 to provide a value for it. Instead, the value to use is stored in
5499 advance in a "trampoline", a block of memory usually allocated
5500 on the stack, which also contains code to splice the nest value into the
5501 argument list. This is used to implement the GCC nested function address
5505 For example, if the function is
5506 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
5507 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as follows:</p>
5509 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
5510 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
5511 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
5512 %fp = bitcast i8* %p to i32 (i32, i32)*
5514 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
5515 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
5518 <!-- _______________________________________________________________________ -->
5519 <div class="doc_subsubsection">
5520 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
5522 <div class="doc_text">
5525 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
5529 This fills the memory pointed to by <tt>tramp</tt> with code
5530 and returns a function pointer suitable for executing it.
5534 The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
5535 pointers. The <tt>tramp</tt> argument must point to a sufficiently large
5536 and sufficiently aligned block of memory; this memory is written to by the
5537 intrinsic. Note that the size and the alignment are target-specific - LLVM
5538 currently provides no portable way of determining them, so a front-end that
5539 generates this intrinsic needs to have some target-specific knowledge.
5540 The <tt>func</tt> argument must hold a function bitcast to an <tt>i8*</tt>.
5544 The block of memory pointed to by <tt>tramp</tt> is filled with target
5545 dependent code, turning it into a function. A pointer to this function is
5546 returned, but needs to be bitcast to an
5547 <a href="#int_trampoline">appropriate function pointer type</a>
5548 before being called. The new function's signature is the same as that of
5549 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
5550 removed. At most one such <tt>nest</tt> argument is allowed, and it must be
5551 of pointer type. Calling the new function is equivalent to calling
5552 <tt>func</tt> with the same argument list, but with <tt>nval</tt> used for the
5553 missing <tt>nest</tt> argument. If, after calling
5554 <tt>llvm.init.trampoline</tt>, the memory pointed to by <tt>tramp</tt> is
5555 modified, then the effect of any later call to the returned function pointer is
5560 <!-- ======================================================================= -->
5561 <div class="doc_subsection">
5562 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
5565 <div class="doc_text">
5567 These intrinsic functions expand the "universal IR" of LLVM to represent
5568 hardware constructs for atomic operations and memory synchronization. This
5569 provides an interface to the hardware, not an interface to the programmer. It
5570 is aimed at a low enough level to allow any programming models or APIs which
5571 need atomic behaviors to map cleanly onto it. It is also modeled primarily on
5572 hardware behavior. Just as hardware provides a "universal IR" for source
5573 languages, it also provides a starting point for developing a "universal"
5574 atomic operation and synchronization IR.
5577 These do <em>not</em> form an API such as high-level threading libraries,
5578 software transaction memory systems, atomic primitives, and intrinsic
5579 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
5580 application libraries. The hardware interface provided by LLVM should allow
5581 a clean implementation of all of these APIs and parallel programming models.
5582 No one model or paradigm should be selected above others unless the hardware
5583 itself ubiquitously does so.
5588 <!-- _______________________________________________________________________ -->
5589 <div class="doc_subsubsection">
5590 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
5592 <div class="doc_text">
5595 declare void @llvm.memory.barrier( i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>,
5601 The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
5602 specific pairs of memory access types.
5606 The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
5607 The first four arguments enables a specific barrier as listed below. The fith
5608 argument specifies that the barrier applies to io or device or uncached memory.
5612 <li><tt>ll</tt>: load-load barrier</li>
5613 <li><tt>ls</tt>: load-store barrier</li>
5614 <li><tt>sl</tt>: store-load barrier</li>
5615 <li><tt>ss</tt>: store-store barrier</li>
5616 <li><tt>device</tt>: barrier applies to device and uncached memory also.
5620 This intrinsic causes the system to enforce some ordering constraints upon
5621 the loads and stores of the program. This barrier does not indicate
5622 <em>when</em> any events will occur, it only enforces an <em>order</em> in
5623 which they occur. For any of the specified pairs of load and store operations
5624 (f.ex. load-load, or store-load), all of the first operations preceding the
5625 barrier will complete before any of the second operations succeeding the
5626 barrier begin. Specifically the semantics for each pairing is as follows:
5629 <li><tt>ll</tt>: All loads before the barrier must complete before any load
5630 after the barrier begins.</li>
5632 <li><tt>ls</tt>: All loads before the barrier must complete before any
5633 store after the barrier begins.</li>
5634 <li><tt>ss</tt>: All stores before the barrier must complete before any
5635 store after the barrier begins.</li>
5636 <li><tt>sl</tt>: All stores before the barrier must complete before any
5637 load after the barrier begins.</li>
5640 These semantics are applied with a logical "and" behavior when more than one
5641 is enabled in a single memory barrier intrinsic.
5644 Backends may implement stronger barriers than those requested when they do not
5645 support as fine grained a barrier as requested. Some architectures do not
5646 need all types of barriers and on such architectures, these become noops.
5653 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
5654 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
5655 <i>; guarantee the above finishes</i>
5656 store i32 8, %ptr <i>; before this begins</i>
5660 <!-- _______________________________________________________________________ -->
5661 <div class="doc_subsubsection">
5662 <a name="int_atomic_lcs">'<tt>llvm.atomic.lcs.*</tt>' Intrinsic</a>
5664 <div class="doc_text">
5667 This is an overloaded intrinsic. You can use <tt>llvm.atomic.lcs</tt> on any
5668 integer bit width. Not all targets support all bit widths however.</p>
5671 declare i8 @llvm.atomic.lcs.i8( i8* <ptr>, i8 <cmp>, i8 <val> )
5672 declare i16 @llvm.atomic.lcs.i16( i16* <ptr>, i16 <cmp>, i16 <val> )
5673 declare i32 @llvm.atomic.lcs.i32( i32* <ptr>, i32 <cmp>, i32 <val> )
5674 declare i64 @llvm.atomic.lcs.i64( i64* <ptr>, i64 <cmp>, i64 <val> )
5679 This loads a value in memory and compares it to a given value. If they are
5680 equal, it stores a new value into the memory.
5684 The <tt>llvm.atomic.lcs</tt> intrinsic takes three arguments. The result as
5685 well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
5686 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
5687 this integer type. While any bit width integer may be used, targets may only
5688 lower representations they support in hardware.
5693 This entire intrinsic must be executed atomically. It first loads the value
5694 in memory pointed to by <tt>ptr</tt> and compares it with the value
5695 <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the memory. The
5696 loaded value is yielded in all cases. This provides the equivalent of an
5697 atomic compare-and-swap operation within the SSA framework.
5705 %val1 = add i32 4, 4
5706 %result1 = call i32 @llvm.atomic.lcs.i32( i32* %ptr, i32 4, %val1 )
5707 <i>; yields {i32}:result1 = 4</i>
5708 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
5709 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
5711 %val2 = add i32 1, 1
5712 %result2 = call i32 @llvm.atomic.lcs.i32( i32* %ptr, i32 5, %val2 )
5713 <i>; yields {i32}:result2 = 8</i>
5714 %stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
5716 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
5720 <!-- _______________________________________________________________________ -->
5721 <div class="doc_subsubsection">
5722 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
5724 <div class="doc_text">
5728 This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
5729 integer bit width. Not all targets support all bit widths however.</p>
5731 declare i8 @llvm.atomic.swap.i8( i8* <ptr>, i8 <val> )
5732 declare i16 @llvm.atomic.swap.i16( i16* <ptr>, i16 <val> )
5733 declare i32 @llvm.atomic.swap.i32( i32* <ptr>, i32 <val> )
5734 declare i64 @llvm.atomic.swap.i64( i64* <ptr>, i64 <val> )
5739 This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
5740 the value from memory. It then stores the value in <tt>val</tt> in the memory
5746 The <tt>llvm.atomic.ls</tt> intrinsic takes two arguments. Both the
5747 <tt>val</tt> argument and the result must be integers of the same bit width.
5748 The first argument, <tt>ptr</tt>, must be a pointer to a value of this
5749 integer type. The targets may only lower integer representations they
5754 This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
5755 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
5756 equivalent of an atomic swap operation within the SSA framework.
5764 %val1 = add i32 4, 4
5765 %result1 = call i32 @llvm.atomic.swap.i32( i32* %ptr, i32 %val1 )
5766 <i>; yields {i32}:result1 = 4</i>
5767 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
5768 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
5770 %val2 = add i32 1, 1
5771 %result2 = call i32 @llvm.atomic.swap.i32( i32* %ptr, i32 %val2 )
5772 <i>; yields {i32}:result2 = 8</i>
5774 %stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
5775 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
5779 <!-- _______________________________________________________________________ -->
5780 <div class="doc_subsubsection">
5781 <a name="int_atomic_las">'<tt>llvm.atomic.las.*</tt>' Intrinsic</a>
5784 <div class="doc_text">
5787 This is an overloaded intrinsic. You can use <tt>llvm.atomic.las</tt> on any
5788 integer bit width. Not all targets support all bit widths however.</p>
5790 declare i8 @llvm.atomic.las.i8.( i8* <ptr>, i8 <delta> )
5791 declare i16 @llvm.atomic.las.i16.( i16* <ptr>, i16 <delta> )
5792 declare i32 @llvm.atomic.las.i32.( i32* <ptr>, i32 <delta> )
5793 declare i64 @llvm.atomic.las.i64.( i64* <ptr>, i64 <delta> )
5798 This intrinsic adds <tt>delta</tt> to the value stored in memory at
5799 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
5804 The intrinsic takes two arguments, the first a pointer to an integer value
5805 and the second an integer value. The result is also an integer value. These
5806 integer types can have any bit width, but they must all have the same bit
5807 width. The targets may only lower integer representations they support.
5811 This intrinsic does a series of operations atomically. It first loads the
5812 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
5813 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
5820 %result1 = call i32 @llvm.atomic.las.i32( i32* %ptr, i32 4 )
5821 <i>; yields {i32}:result1 = 4</i>
5822 %result2 = call i32 @llvm.atomic.las.i32( i32* %ptr, i32 2 )
5823 <i>; yields {i32}:result2 = 8</i>
5824 %result3 = call i32 @llvm.atomic.las.i32( i32* %ptr, i32 5 )
5825 <i>; yields {i32}:result3 = 10</i>
5826 %memval = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
5831 <!-- ======================================================================= -->
5832 <div class="doc_subsection">
5833 <a name="int_general">General Intrinsics</a>
5836 <div class="doc_text">
5837 <p> This class of intrinsics is designed to be generic and has
5838 no specific purpose. </p>
5841 <!-- _______________________________________________________________________ -->
5842 <div class="doc_subsubsection">
5843 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
5846 <div class="doc_text">
5850 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
5856 The '<tt>llvm.var.annotation</tt>' intrinsic
5862 The first argument is a pointer to a value, the second is a pointer to a
5863 global string, the third is a pointer to a global string which is the source
5864 file name, and the last argument is the line number.
5870 This intrinsic allows annotation of local variables with arbitrary strings.
5871 This can be useful for special purpose optimizations that want to look for these
5872 annotations. These have no other defined use, they are ignored by code
5873 generation and optimization.
5877 <!-- _______________________________________________________________________ -->
5878 <div class="doc_subsubsection">
5879 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
5882 <div class="doc_text">
5885 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
5886 any integer bit width.
5889 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
5890 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
5891 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
5892 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
5893 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
5899 The '<tt>llvm.annotation</tt>' intrinsic.
5905 The first argument is an integer value (result of some expression),
5906 the second is a pointer to a global string, the third is a pointer to a global
5907 string which is the source file name, and the last argument is the line number.
5908 It returns the value of the first argument.
5914 This intrinsic allows annotations to be put on arbitrary expressions
5915 with arbitrary strings. This can be useful for special purpose optimizations
5916 that want to look for these annotations. These have no other defined use, they
5917 are ignored by code generation and optimization.
5920 <!-- _______________________________________________________________________ -->
5921 <div class="doc_subsubsection">
5922 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
5925 <div class="doc_text">
5929 declare void @llvm.trap()
5935 The '<tt>llvm.trap</tt>' intrinsic
5947 This intrinsics is lowered to the target dependent trap instruction. If the
5948 target does not have a trap instruction, this intrinsic will be lowered to the
5949 call of the abort() function.
5953 <!-- *********************************************************************** -->
5956 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
5957 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
5958 <a href="http://validator.w3.org/check/referer"><img
5959 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
5961 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
5962 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
5963 Last modified: $Date$