1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <title>LLVM Assembly Language Reference Manual</title>
6 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
7 <meta name="author" content="Chris Lattner">
8 <meta name="description"
9 content="LLVM Assembly Language Reference Manual.">
10 <link rel="stylesheet" href="llvm.css" type="text/css">
15 <div class="doc_title"> LLVM Language Reference Manual </div>
17 <li><a href="#abstract">Abstract</a></li>
18 <li><a href="#introduction">Introduction</a></li>
19 <li><a href="#identifiers">Identifiers</a></li>
20 <li><a href="#highlevel">High Level Structure</a>
22 <li><a href="#modulestructure">Module Structure</a></li>
23 <li><a href="#linkage">Linkage Types</a></li>
24 <li><a href="#callingconv">Calling Conventions</a></li>
25 <li><a href="#globalvars">Global Variables</a></li>
26 <li><a href="#functionstructure">Functions</a></li>
27 <li><a href="#aliasstructure">Aliases</a>
28 <li><a href="#paramattrs">Parameter Attributes</a></li>
29 <li><a href="#gc">Garbage Collector Names</a></li>
30 <li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
31 <li><a href="#datalayout">Data Layout</a></li>
34 <li><a href="#typesystem">Type System</a>
36 <li><a href="#t_classifications">Type Classifications</a></li>
37 <li><a href="#t_primitive">Primitive Types</a>
39 <li><a href="#t_floating">Floating Point Types</a></li>
40 <li><a href="#t_void">Void Type</a></li>
41 <li><a href="#t_label">Label Type</a></li>
44 <li><a href="#t_derived">Derived Types</a>
46 <li><a href="#t_integer">Integer Type</a></li>
47 <li><a href="#t_array">Array Type</a></li>
48 <li><a href="#t_function">Function Type</a></li>
49 <li><a href="#t_pointer">Pointer Type</a></li>
50 <li><a href="#t_struct">Structure Type</a></li>
51 <li><a href="#t_pstruct">Packed Structure Type</a></li>
52 <li><a href="#t_vector">Vector Type</a></li>
53 <li><a href="#t_opaque">Opaque Type</a></li>
58 <li><a href="#constants">Constants</a>
60 <li><a href="#simpleconstants">Simple Constants</a>
61 <li><a href="#aggregateconstants">Aggregate Constants</a>
62 <li><a href="#globalconstants">Global Variable and Function Addresses</a>
63 <li><a href="#undefvalues">Undefined Values</a>
64 <li><a href="#constantexprs">Constant Expressions</a>
67 <li><a href="#othervalues">Other Values</a>
69 <li><a href="#inlineasm">Inline Assembler Expressions</a>
72 <li><a href="#instref">Instruction Reference</a>
74 <li><a href="#terminators">Terminator Instructions</a>
76 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
77 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
78 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
79 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
80 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
81 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
84 <li><a href="#binaryops">Binary Operations</a>
86 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
87 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
88 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
89 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
90 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
91 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
92 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
93 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
94 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
97 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
99 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
100 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
101 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
102 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
103 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
104 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
107 <li><a href="#vectorops">Vector Operations</a>
109 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
110 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
111 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
114 <li><a href="#aggregateops">Aggregate Operations</a>
116 <li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li>
117 <li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li>
120 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
122 <li><a href="#i_malloc">'<tt>malloc</tt>' Instruction</a></li>
123 <li><a href="#i_free">'<tt>free</tt>' Instruction</a></li>
124 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
125 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
126 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
127 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
130 <li><a href="#convertops">Conversion Operations</a>
132 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
133 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
134 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
135 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
136 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
137 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
138 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
139 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
140 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
141 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
142 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
143 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
145 <li><a href="#otherops">Other Operations</a>
147 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
148 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
149 <li><a href="#i_vicmp">'<tt>vicmp</tt>' Instruction</a></li>
150 <li><a href="#i_vfcmp">'<tt>vfcmp</tt>' Instruction</a></li>
151 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
152 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
153 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
154 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
155 <li><a href="#i_getresult">'<tt>getresult</tt>' Instruction</a></li>
160 <li><a href="#intrinsics">Intrinsic Functions</a>
162 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
164 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
165 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
166 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
169 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
171 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
172 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
173 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
176 <li><a href="#int_codegen">Code Generator Intrinsics</a>
178 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
179 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
180 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
181 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
182 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
183 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
184 <li><a href="#int_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
187 <li><a href="#int_libc">Standard C Library Intrinsics</a>
189 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
190 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
191 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
192 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
193 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
194 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
195 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
196 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
199 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
201 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
202 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
203 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
204 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
205 <li><a href="#int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic </a></li>
206 <li><a href="#int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic </a></li>
209 <li><a href="#int_debugger">Debugger intrinsics</a></li>
210 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
211 <li><a href="#int_trampoline">Trampoline Intrinsic</a>
213 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
216 <li><a href="#int_atomics">Atomic intrinsics</a>
218 <li><a href="#int_memory_barrier"><tt>llvm.memory_barrier</tt></a></li>
219 <li><a href="#int_atomic_lcs"><tt>llvm.atomic.lcs</tt></a></li>
220 <li><a href="#int_atomic_las"><tt>llvm.atomic.las</tt></a></li>
221 <li><a href="#int_atomic_swap"><tt>llvm.atomic.swap</tt></a></li>
224 <li><a href="#int_general">General intrinsics</a>
226 <li><a href="#int_var_annotation">
227 <tt>llvm.var.annotation</tt>' Intrinsic</a></li>
228 <li><a href="#int_annotation">
229 <tt>llvm.annotation.*</tt>' Intrinsic</a></li>
230 <li><a href="#int_trap">
231 <tt>llvm.trap</tt>' Intrinsic</a></li>
238 <div class="doc_author">
239 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
240 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
243 <!-- *********************************************************************** -->
244 <div class="doc_section"> <a name="abstract">Abstract </a></div>
245 <!-- *********************************************************************** -->
247 <div class="doc_text">
248 <p>This document is a reference manual for the LLVM assembly language.
249 LLVM is an SSA based representation that provides type safety,
250 low-level operations, flexibility, and the capability of representing
251 'all' high-level languages cleanly. It is the common code
252 representation used throughout all phases of the LLVM compilation
256 <!-- *********************************************************************** -->
257 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
258 <!-- *********************************************************************** -->
260 <div class="doc_text">
262 <p>The LLVM code representation is designed to be used in three
263 different forms: as an in-memory compiler IR, as an on-disk bitcode
264 representation (suitable for fast loading by a Just-In-Time compiler),
265 and as a human readable assembly language representation. This allows
266 LLVM to provide a powerful intermediate representation for efficient
267 compiler transformations and analysis, while providing a natural means
268 to debug and visualize the transformations. The three different forms
269 of LLVM are all equivalent. This document describes the human readable
270 representation and notation.</p>
272 <p>The LLVM representation aims to be light-weight and low-level
273 while being expressive, typed, and extensible at the same time. It
274 aims to be a "universal IR" of sorts, by being at a low enough level
275 that high-level ideas may be cleanly mapped to it (similar to how
276 microprocessors are "universal IR's", allowing many source languages to
277 be mapped to them). By providing type information, LLVM can be used as
278 the target of optimizations: for example, through pointer analysis, it
279 can be proven that a C automatic variable is never accessed outside of
280 the current function... allowing it to be promoted to a simple SSA
281 value instead of a memory location.</p>
285 <!-- _______________________________________________________________________ -->
286 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
288 <div class="doc_text">
290 <p>It is important to note that this document describes 'well formed'
291 LLVM assembly language. There is a difference between what the parser
292 accepts and what is considered 'well formed'. For example, the
293 following instruction is syntactically okay, but not well formed:</p>
295 <div class="doc_code">
297 %x = <a href="#i_add">add</a> i32 1, %x
301 <p>...because the definition of <tt>%x</tt> does not dominate all of
302 its uses. The LLVM infrastructure provides a verification pass that may
303 be used to verify that an LLVM module is well formed. This pass is
304 automatically run by the parser after parsing input assembly and by
305 the optimizer before it outputs bitcode. The violations pointed out
306 by the verifier pass indicate bugs in transformation passes or input to
310 <!-- Describe the typesetting conventions here. -->
312 <!-- *********************************************************************** -->
313 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
314 <!-- *********************************************************************** -->
316 <div class="doc_text">
318 <p>LLVM identifiers come in two basic types: global and local. Global
319 identifiers (functions, global variables) begin with the @ character. Local
320 identifiers (register names, types) begin with the % character. Additionally,
321 there are three different formats for identifiers, for different purposes:
324 <li>Named values are represented as a string of characters with their prefix.
325 For example, %foo, @DivisionByZero, %a.really.long.identifier. The actual
326 regular expression used is '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
327 Identifiers which require other characters in their names can be surrounded
328 with quotes. In this way, anything except a <tt>"</tt> character can
329 be used in a named value.</li>
331 <li>Unnamed values are represented as an unsigned numeric value with their
332 prefix. For example, %12, @2, %44.</li>
334 <li>Constants, which are described in a <a href="#constants">section about
335 constants</a>, below.</li>
338 <p>LLVM requires that values start with a prefix for two reasons: Compilers
339 don't need to worry about name clashes with reserved words, and the set of
340 reserved words may be expanded in the future without penalty. Additionally,
341 unnamed identifiers allow a compiler to quickly come up with a temporary
342 variable without having to avoid symbol table conflicts.</p>
344 <p>Reserved words in LLVM are very similar to reserved words in other
345 languages. There are keywords for different opcodes
346 ('<tt><a href="#i_add">add</a></tt>',
347 '<tt><a href="#i_bitcast">bitcast</a></tt>',
348 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
349 href="#t_void">void</a></tt>', '<tt><a href="#t_primitive">i32</a></tt>', etc...),
350 and others. These reserved words cannot conflict with variable names, because
351 none of them start with a prefix character ('%' or '@').</p>
353 <p>Here is an example of LLVM code to multiply the integer variable
354 '<tt>%X</tt>' by 8:</p>
358 <div class="doc_code">
360 %result = <a href="#i_mul">mul</a> i32 %X, 8
364 <p>After strength reduction:</p>
366 <div class="doc_code">
368 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
372 <p>And the hard way:</p>
374 <div class="doc_code">
376 <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
377 <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
378 %result = <a href="#i_add">add</a> i32 %1, %1
382 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
383 important lexical features of LLVM:</p>
387 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
390 <li>Unnamed temporaries are created when the result of a computation is not
391 assigned to a named value.</li>
393 <li>Unnamed temporaries are numbered sequentially</li>
397 <p>...and it also shows a convention that we follow in this document. When
398 demonstrating instructions, we will follow an instruction with a comment that
399 defines the type and name of value produced. Comments are shown in italic
404 <!-- *********************************************************************** -->
405 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
406 <!-- *********************************************************************** -->
408 <!-- ======================================================================= -->
409 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
412 <div class="doc_text">
414 <p>LLVM programs are composed of "Module"s, each of which is a
415 translation unit of the input programs. Each module consists of
416 functions, global variables, and symbol table entries. Modules may be
417 combined together with the LLVM linker, which merges function (and
418 global variable) definitions, resolves forward declarations, and merges
419 symbol table entries. Here is an example of the "hello world" module:</p>
421 <div class="doc_code">
422 <pre><i>; Declare the string constant as a global constant...</i>
423 <a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a
424 href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
426 <i>; External declaration of the puts function</i>
427 <a href="#functionstructure">declare</a> i32 @puts(i8 *) <i>; i32(i8 *)* </i>
429 <i>; Definition of main function</i>
430 define i32 @main() { <i>; i32()* </i>
431 <i>; Convert [13x i8 ]* to i8 *...</i>
433 href="#i_getelementptr">getelementptr</a> [13 x i8 ]* @.LC0, i64 0, i64 0 <i>; i8 *</i>
435 <i>; Call puts function to write out the string to stdout...</i>
437 href="#i_call">call</a> i32 @puts(i8 * %cast210) <i>; i32</i>
439 href="#i_ret">ret</a> i32 0<br>}<br>
443 <p>This example is made up of a <a href="#globalvars">global variable</a>
444 named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
445 function, and a <a href="#functionstructure">function definition</a>
446 for "<tt>main</tt>".</p>
448 <p>In general, a module is made up of a list of global values,
449 where both functions and global variables are global values. Global values are
450 represented by a pointer to a memory location (in this case, a pointer to an
451 array of char, and a pointer to a function), and have one of the following <a
452 href="#linkage">linkage types</a>.</p>
456 <!-- ======================================================================= -->
457 <div class="doc_subsection">
458 <a name="linkage">Linkage Types</a>
461 <div class="doc_text">
464 All Global Variables and Functions have one of the following types of linkage:
469 <dt><tt><b><a name="linkage_internal">internal</a></b></tt> </dt>
471 <dd>Global values with internal linkage are only directly accessible by
472 objects in the current module. In particular, linking code into a module with
473 an internal global value may cause the internal to be renamed as necessary to
474 avoid collisions. Because the symbol is internal to the module, all
475 references can be updated. This corresponds to the notion of the
476 '<tt>static</tt>' keyword in C.
479 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
481 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
482 the same name when linkage occurs. This is typically used to implement
483 inline functions, templates, or other code which must be generated in each
484 translation unit that uses it. Unreferenced <tt>linkonce</tt> globals are
485 allowed to be discarded.
488 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
490 <dd>"<tt>weak</tt>" linkage is exactly the same as <tt>linkonce</tt> linkage,
491 except that unreferenced <tt>weak</tt> globals may not be discarded. This is
492 used for globals that may be emitted in multiple translation units, but that
493 are not guaranteed to be emitted into every translation unit that uses them.
494 One example of this are common globals in C, such as "<tt>int X;</tt>" at
498 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
500 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
501 pointer to array type. When two global variables with appending linkage are
502 linked together, the two global arrays are appended together. This is the
503 LLVM, typesafe, equivalent of having the system linker append together
504 "sections" with identical names when .o files are linked.
507 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt>: </dt>
508 <dd>The semantics of this linkage follow the ELF model: the symbol is weak
509 until linked, if not linked, the symbol becomes null instead of being an
513 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
515 <dd>If none of the above identifiers are used, the global is externally
516 visible, meaning that it participates in linkage and can be used to resolve
517 external symbol references.
522 The next two types of linkage are targeted for Microsoft Windows platform
523 only. They are designed to support importing (exporting) symbols from (to)
528 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt>: </dt>
530 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
531 or variable via a global pointer to a pointer that is set up by the DLL
532 exporting the symbol. On Microsoft Windows targets, the pointer name is
533 formed by combining <code>_imp__</code> and the function or variable name.
536 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt>: </dt>
538 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
539 pointer to a pointer in a DLL, so that it can be referenced with the
540 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
541 name is formed by combining <code>_imp__</code> and the function or variable
547 <p><a name="linkage_external"></a>For example, since the "<tt>.LC0</tt>"
548 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
549 variable and was linked with this one, one of the two would be renamed,
550 preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
551 external (i.e., lacking any linkage declarations), they are accessible
552 outside of the current module.</p>
553 <p>It is illegal for a function <i>declaration</i>
554 to have any linkage type other than "externally visible", <tt>dllimport</tt>,
555 or <tt>extern_weak</tt>.</p>
556 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt> and <tt>weak</tt>
560 <!-- ======================================================================= -->
561 <div class="doc_subsection">
562 <a name="callingconv">Calling Conventions</a>
565 <div class="doc_text">
567 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
568 and <a href="#i_invoke">invokes</a> can all have an optional calling convention
569 specified for the call. The calling convention of any pair of dynamic
570 caller/callee must match, or the behavior of the program is undefined. The
571 following calling conventions are supported by LLVM, and more may be added in
575 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
577 <dd>This calling convention (the default if no other calling convention is
578 specified) matches the target C calling conventions. This calling convention
579 supports varargs function calls and tolerates some mismatch in the declared
580 prototype and implemented declaration of the function (as does normal C).
583 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
585 <dd>This calling convention attempts to make calls as fast as possible
586 (e.g. by passing things in registers). This calling convention allows the
587 target to use whatever tricks it wants to produce fast code for the target,
588 without having to conform to an externally specified ABI. Implementations of
589 this convention should allow arbitrary
590 <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> to be
591 supported. This calling convention does not support varargs and requires the
592 prototype of all callees to exactly match the prototype of the function
596 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
598 <dd>This calling convention attempts to make code in the caller as efficient
599 as possible under the assumption that the call is not commonly executed. As
600 such, these calls often preserve all registers so that the call does not break
601 any live ranges in the caller side. This calling convention does not support
602 varargs and requires the prototype of all callees to exactly match the
603 prototype of the function definition.
606 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
608 <dd>Any calling convention may be specified by number, allowing
609 target-specific calling conventions to be used. Target specific calling
610 conventions start at 64.
614 <p>More calling conventions can be added/defined on an as-needed basis, to
615 support pascal conventions or any other well-known target-independent
620 <!-- ======================================================================= -->
621 <div class="doc_subsection">
622 <a name="visibility">Visibility Styles</a>
625 <div class="doc_text">
628 All Global Variables and Functions have one of the following visibility styles:
632 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
634 <dd>On ELF, default visibility means that the declaration is visible to other
635 modules and, in shared libraries, means that the declared entity may be
636 overridden. On Darwin, default visibility means that the declaration is
637 visible to other modules. Default visibility corresponds to "external
638 linkage" in the language.
641 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
643 <dd>Two declarations of an object with hidden visibility refer to the same
644 object if they are in the same shared object. Usually, hidden visibility
645 indicates that the symbol will not be placed into the dynamic symbol table,
646 so no other module (executable or shared library) can reference it
650 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
652 <dd>On ELF, protected visibility indicates that the symbol will be placed in
653 the dynamic symbol table, but that references within the defining module will
654 bind to the local symbol. That is, the symbol cannot be overridden by another
661 <!-- ======================================================================= -->
662 <div class="doc_subsection">
663 <a name="globalvars">Global Variables</a>
666 <div class="doc_text">
668 <p>Global variables define regions of memory allocated at compilation time
669 instead of run-time. Global variables may optionally be initialized, may have
670 an explicit section to be placed in, and may have an optional explicit alignment
671 specified. A variable may be defined as "thread_local", which means that it
672 will not be shared by threads (each thread will have a separated copy of the
673 variable). A variable may be defined as a global "constant," which indicates
674 that the contents of the variable will <b>never</b> be modified (enabling better
675 optimization, allowing the global data to be placed in the read-only section of
676 an executable, etc). Note that variables that need runtime initialization
677 cannot be marked "constant" as there is a store to the variable.</p>
680 LLVM explicitly allows <em>declarations</em> of global variables to be marked
681 constant, even if the final definition of the global is not. This capability
682 can be used to enable slightly better optimization of the program, but requires
683 the language definition to guarantee that optimizations based on the
684 'constantness' are valid for the translation units that do not include the
688 <p>As SSA values, global variables define pointer values that are in
689 scope (i.e. they dominate) all basic blocks in the program. Global
690 variables always define a pointer to their "content" type because they
691 describe a region of memory, and all memory objects in LLVM are
692 accessed through pointers.</p>
694 <p>A global variable may be declared to reside in a target-specifc numbered
695 address space. For targets that support them, address spaces may affect how
696 optimizations are performed and/or what target instructions are used to access
697 the variable. The default address space is zero. The address space qualifier
698 must precede any other attributes.</p>
700 <p>LLVM allows an explicit section to be specified for globals. If the target
701 supports it, it will emit globals to the section specified.</p>
703 <p>An explicit alignment may be specified for a global. If not present, or if
704 the alignment is set to zero, the alignment of the global is set by the target
705 to whatever it feels convenient. If an explicit alignment is specified, the
706 global is forced to have at least that much alignment. All alignments must be
709 <p>For example, the following defines a global in a numbered address space with
710 an initializer, section, and alignment:</p>
712 <div class="doc_code">
714 @G = constant float 1.0 addrspace(5), section "foo", align 4
721 <!-- ======================================================================= -->
722 <div class="doc_subsection">
723 <a name="functionstructure">Functions</a>
726 <div class="doc_text">
728 <p>LLVM function definitions consist of the "<tt>define</tt>" keyord,
729 an optional <a href="#linkage">linkage type</a>, an optional
730 <a href="#visibility">visibility style</a>, an optional
731 <a href="#callingconv">calling convention</a>, a return type, an optional
732 <a href="#paramattrs">parameter attribute</a> for the return type, a function
733 name, a (possibly empty) argument list (each with optional
734 <a href="#paramattrs">parameter attributes</a>), an optional section, an
735 optional alignment, an optional <a href="#gc">garbage collector name</a>, an
736 opening curly brace, a list of basic blocks, and a closing curly brace.
738 LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
739 optional <a href="#linkage">linkage type</a>, an optional
740 <a href="#visibility">visibility style</a>, an optional
741 <a href="#callingconv">calling convention</a>, a return type, an optional
742 <a href="#paramattrs">parameter attribute</a> for the return type, a function
743 name, a possibly empty list of arguments, an optional alignment, and an optional
744 <a href="#gc">garbage collector name</a>.</p>
746 <p>A function definition contains a list of basic blocks, forming the CFG for
747 the function. Each basic block may optionally start with a label (giving the
748 basic block a symbol table entry), contains a list of instructions, and ends
749 with a <a href="#terminators">terminator</a> instruction (such as a branch or
750 function return).</p>
752 <p>The first basic block in a function is special in two ways: it is immediately
753 executed on entrance to the function, and it is not allowed to have predecessor
754 basic blocks (i.e. there can not be any branches to the entry block of a
755 function). Because the block can have no predecessors, it also cannot have any
756 <a href="#i_phi">PHI nodes</a>.</p>
758 <p>LLVM allows an explicit section to be specified for functions. If the target
759 supports it, it will emit functions to the section specified.</p>
761 <p>An explicit alignment may be specified for a function. If not present, or if
762 the alignment is set to zero, the alignment of the function is set by the target
763 to whatever it feels convenient. If an explicit alignment is specified, the
764 function is forced to have at least that much alignment. All alignments must be
770 <!-- ======================================================================= -->
771 <div class="doc_subsection">
772 <a name="aliasstructure">Aliases</a>
774 <div class="doc_text">
775 <p>Aliases act as "second name" for the aliasee value (which can be either
776 function, global variable, another alias or bitcast of global value). Aliases
777 may have an optional <a href="#linkage">linkage type</a>, and an
778 optional <a href="#visibility">visibility style</a>.</p>
782 <div class="doc_code">
784 @<Name> = [Linkage] [Visibility] alias <AliaseeTy> @<Aliasee>
792 <!-- ======================================================================= -->
793 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
794 <div class="doc_text">
795 <p>The return type and each parameter of a function type may have a set of
796 <i>parameter attributes</i> associated with them. Parameter attributes are
797 used to communicate additional information about the result or parameters of
798 a function. Parameter attributes are considered to be part of the function,
799 not of the function type, so functions with different parameter attributes
800 can have the same function type.</p>
802 <p>Parameter attributes are simple keywords that follow the type specified. If
803 multiple parameter attributes are needed, they are space separated. For
806 <div class="doc_code">
808 declare i32 @printf(i8* noalias , ...) nounwind
809 declare i32 @atoi(i8*) nounwind readonly
813 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
814 <tt>readonly</tt>) come immediately after the argument list.</p>
816 <p>Currently, only the following parameter attributes are defined:</p>
818 <dt><tt>zeroext</tt></dt>
819 <dd>This indicates that the parameter should be zero extended just before
820 a call to this function.</dd>
822 <dt><tt>signext</tt></dt>
823 <dd>This indicates that the parameter should be sign extended just before
824 a call to this function.</dd>
826 <dt><tt>inreg</tt></dt>
827 <dd>This indicates that the parameter should be placed in register (if
828 possible) during assembling function call. Support for this attribute is
831 <dt><tt>byval</tt></dt>
832 <dd>This indicates that the pointer parameter should really be passed by
833 value to the function. The attribute implies that a hidden copy of the
834 pointee is made between the caller and the callee, so the callee is unable
835 to modify the value in the callee. This attribute is only valid on llvm
836 pointer arguments. It is generally used to pass structs and arrays by
837 value, but is also valid on scalars (even though this is silly).</dd>
839 <dt><tt>sret</tt></dt>
840 <dd>This indicates that the pointer parameter specifies the address of a
841 structure that is the return value of the function in the source program.
842 Loads and stores to the structure are assumed not to trap.
843 May only be applied to the first parameter.</dd>
845 <dt><tt>noalias</tt></dt>
846 <dd>This indicates that the parameter does not alias any global or any other
847 parameter. The caller is responsible for ensuring that this is the case,
848 usually by placing the value in a stack allocation.</dd>
850 <dt><tt>noreturn</tt></dt>
851 <dd>This function attribute indicates that the function never returns. This
852 indicates to LLVM that every call to this function should be treated as if
853 an <tt>unreachable</tt> instruction immediately followed the call.</dd>
855 <dt><tt>nounwind</tt></dt>
856 <dd>This function attribute indicates that no exceptions unwind out of the
857 function. Usually this is because the function makes no use of exceptions,
858 but it may also be that the function catches any exceptions thrown when
861 <dt><tt>nest</tt></dt>
862 <dd>This indicates that the parameter can be excised using the
863 <a href="#int_trampoline">trampoline intrinsics</a>.</dd>
864 <dt><tt>readonly</tt></dt>
865 <dd>This function attribute indicates that the function has no side-effects
866 except for producing a return value or throwing an exception. The value
867 returned must only depend on the function arguments and/or global variables.
868 It may use values obtained by dereferencing pointers.</dd>
869 <dt><tt>readnone</tt></dt>
870 <dd>A <tt>readnone</tt> function has the same restrictions as a <tt>readonly</tt>
871 function, but in addition it is not allowed to dereference any pointer arguments
877 <!-- ======================================================================= -->
878 <div class="doc_subsection">
879 <a name="gc">Garbage Collector Names</a>
882 <div class="doc_text">
883 <p>Each function may specify a garbage collector name, which is simply a
886 <div class="doc_code"><pre
887 >define void @f() gc "name" { ...</pre></div>
889 <p>The compiler declares the supported values of <i>name</i>. Specifying a
890 collector which will cause the compiler to alter its output in order to support
891 the named garbage collection algorithm.</p>
894 <!-- ======================================================================= -->
895 <div class="doc_subsection">
896 <a name="moduleasm">Module-Level Inline Assembly</a>
899 <div class="doc_text">
901 Modules may contain "module-level inline asm" blocks, which corresponds to the
902 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
903 LLVM and treated as a single unit, but may be separated in the .ll file if
904 desired. The syntax is very simple:
907 <div class="doc_code">
909 module asm "inline asm code goes here"
910 module asm "more can go here"
914 <p>The strings can contain any character by escaping non-printable characters.
915 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
920 The inline asm code is simply printed to the machine code .s file when
921 assembly code is generated.
925 <!-- ======================================================================= -->
926 <div class="doc_subsection">
927 <a name="datalayout">Data Layout</a>
930 <div class="doc_text">
931 <p>A module may specify a target specific data layout string that specifies how
932 data is to be laid out in memory. The syntax for the data layout is simply:</p>
933 <pre> target datalayout = "<i>layout specification</i>"</pre>
934 <p>The <i>layout specification</i> consists of a list of specifications
935 separated by the minus sign character ('-'). Each specification starts with a
936 letter and may include other information after the letter to define some
937 aspect of the data layout. The specifications accepted are as follows: </p>
940 <dd>Specifies that the target lays out data in big-endian form. That is, the
941 bits with the most significance have the lowest address location.</dd>
943 <dd>Specifies that hte target lays out data in little-endian form. That is,
944 the bits with the least significance have the lowest address location.</dd>
945 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
946 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
947 <i>preferred</i> alignments. All sizes are in bits. Specifying the <i>pref</i>
948 alignment is optional. If omitted, the preceding <tt>:</tt> should be omitted
950 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
951 <dd>This specifies the alignment for an integer type of a given bit
952 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
953 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
954 <dd>This specifies the alignment for a vector type of a given bit
956 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
957 <dd>This specifies the alignment for a floating point type of a given bit
958 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
960 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
961 <dd>This specifies the alignment for an aggregate type of a given bit
964 <p>When constructing the data layout for a given target, LLVM starts with a
965 default set of specifications which are then (possibly) overriden by the
966 specifications in the <tt>datalayout</tt> keyword. The default specifications
967 are given in this list:</p>
969 <li><tt>E</tt> - big endian</li>
970 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
971 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
972 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
973 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
974 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
975 <li><tt>i64:32:64</tt> - i64 has abi alignment of 32-bits but preferred
976 alignment of 64-bits</li>
977 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
978 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
979 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
980 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
981 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
983 <p>When llvm is determining the alignment for a given type, it uses the
986 <li>If the type sought is an exact match for one of the specifications, that
987 specification is used.</li>
988 <li>If no match is found, and the type sought is an integer type, then the
989 smallest integer type that is larger than the bitwidth of the sought type is
990 used. If none of the specifications are larger than the bitwidth then the the
991 largest integer type is used. For example, given the default specifications
992 above, the i7 type will use the alignment of i8 (next largest) while both
993 i65 and i256 will use the alignment of i64 (largest specified).</li>
994 <li>If no match is found, and the type sought is a vector type, then the
995 largest vector type that is smaller than the sought vector type will be used
996 as a fall back. This happens because <128 x double> can be implemented in
997 terms of 64 <2 x double>, for example.</li>
1001 <!-- *********************************************************************** -->
1002 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
1003 <!-- *********************************************************************** -->
1005 <div class="doc_text">
1007 <p>The LLVM type system is one of the most important features of the
1008 intermediate representation. Being typed enables a number of
1009 optimizations to be performed on the IR directly, without having to do
1010 extra analyses on the side before the transformation. A strong type
1011 system makes it easier to read the generated code and enables novel
1012 analyses and transformations that are not feasible to perform on normal
1013 three address code representations.</p>
1017 <!-- ======================================================================= -->
1018 <div class="doc_subsection"> <a name="t_classifications">Type
1019 Classifications</a> </div>
1020 <div class="doc_text">
1021 <p>The types fall into a few useful
1022 classifications:</p>
1024 <table border="1" cellspacing="0" cellpadding="4">
1026 <tr><th>Classification</th><th>Types</th></tr>
1028 <td><a href="#t_integer">integer</a></td>
1029 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1032 <td><a href="#t_floating">floating point</a></td>
1033 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1036 <td><a name="t_firstclass">first class</a></td>
1037 <td><a href="#t_integer">integer</a>,
1038 <a href="#t_floating">floating point</a>,
1039 <a href="#t_pointer">pointer</a>,
1040 <a href="#t_vector">vector</a>
1041 <a href="#t_struct">structure</a>,
1042 <a href="#t_array">array</a>,
1046 <td><a href="#t_primitive">primitive</a></td>
1047 <td><a href="#t_label">label</a>,
1048 <a href="#t_void">void</a>,
1049 <a href="#t_integer">integer</a>,
1050 <a href="#t_floating">floating point</a>.</td>
1053 <td><a href="#t_derived">derived</a></td>
1054 <td><a href="#t_integer">integer</a>,
1055 <a href="#t_array">array</a>,
1056 <a href="#t_function">function</a>,
1057 <a href="#t_pointer">pointer</a>,
1058 <a href="#t_struct">structure</a>,
1059 <a href="#t_pstruct">packed structure</a>,
1060 <a href="#t_vector">vector</a>,
1061 <a href="#t_opaque">opaque</a>.
1066 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
1067 most important. Values of these types are the only ones which can be
1068 produced by instructions, passed as arguments, or used as operands to
1072 <!-- ======================================================================= -->
1073 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
1075 <div class="doc_text">
1076 <p>The primitive types are the fundamental building blocks of the LLVM
1081 <!-- _______________________________________________________________________ -->
1082 <div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1084 <div class="doc_text">
1087 <tr><th>Type</th><th>Description</th></tr>
1088 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1089 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1090 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1091 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1092 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1097 <!-- _______________________________________________________________________ -->
1098 <div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1100 <div class="doc_text">
1102 <p>The void type does not represent any value and has no size.</p>
1111 <!-- _______________________________________________________________________ -->
1112 <div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1114 <div class="doc_text">
1116 <p>The label type represents code labels.</p>
1126 <!-- ======================================================================= -->
1127 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1129 <div class="doc_text">
1131 <p>The real power in LLVM comes from the derived types in the system.
1132 This is what allows a programmer to represent arrays, functions,
1133 pointers, and other useful types. Note that these derived types may be
1134 recursive: For example, it is possible to have a two dimensional array.</p>
1138 <!-- _______________________________________________________________________ -->
1139 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1141 <div class="doc_text">
1144 <p>The integer type is a very simple derived type that simply specifies an
1145 arbitrary bit width for the integer type desired. Any bit width from 1 bit to
1146 2^23-1 (about 8 million) can be specified.</p>
1154 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1158 <table class="layout">
1161 <td><tt>i1</tt></td>
1162 <td>a single-bit integer.</td>
1164 <td><tt>i32</tt></td>
1165 <td>a 32-bit integer.</td>
1167 <td><tt>i1942652</tt></td>
1168 <td>a really big integer of over 1 million bits.</td>
1174 <!-- _______________________________________________________________________ -->
1175 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1177 <div class="doc_text">
1181 <p>The array type is a very simple derived type that arranges elements
1182 sequentially in memory. The array type requires a size (number of
1183 elements) and an underlying data type.</p>
1188 [<# elements> x <elementtype>]
1191 <p>The number of elements is a constant integer value; elementtype may
1192 be any type with a size.</p>
1195 <table class="layout">
1197 <td class="left"><tt>[40 x i32]</tt></td>
1198 <td class="left">Array of 40 32-bit integer values.</td>
1201 <td class="left"><tt>[41 x i32]</tt></td>
1202 <td class="left">Array of 41 32-bit integer values.</td>
1205 <td class="left"><tt>[4 x i8]</tt></td>
1206 <td class="left">Array of 4 8-bit integer values.</td>
1209 <p>Here are some examples of multidimensional arrays:</p>
1210 <table class="layout">
1212 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1213 <td class="left">3x4 array of 32-bit integer values.</td>
1216 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1217 <td class="left">12x10 array of single precision floating point values.</td>
1220 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1221 <td class="left">2x3x4 array of 16-bit integer values.</td>
1225 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1226 length array. Normally, accesses past the end of an array are undefined in
1227 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
1228 As a special case, however, zero length arrays are recognized to be variable
1229 length. This allows implementation of 'pascal style arrays' with the LLVM
1230 type "{ i32, [0 x float]}", for example.</p>
1234 <!-- _______________________________________________________________________ -->
1235 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1236 <div class="doc_text">
1240 <p>The function type can be thought of as a function signature. It
1241 consists of a return type and a list of formal parameter types. The
1242 return type of a function type is a scalar type, a void type, or a struct type.
1243 If the return type is a struct type then all struct elements must be of first
1244 class types, and the struct must have at least one element.</p>
1249 <returntype list> (<parameter list>)
1252 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1253 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1254 which indicates that the function takes a variable number of arguments.
1255 Variable argument functions can access their arguments with the <a
1256 href="#int_varargs">variable argument handling intrinsic</a> functions.
1257 '<tt><returntype list></tt>' is a comma-separated list of
1258 <a href="#t_firstclass">first class</a> type specifiers.</p>
1261 <table class="layout">
1263 <td class="left"><tt>i32 (i32)</tt></td>
1264 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1266 </tr><tr class="layout">
1267 <td class="left"><tt>float (i16 signext, i32 *) *
1269 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1270 an <tt>i16</tt> that should be sign extended and a
1271 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1274 </tr><tr class="layout">
1275 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1276 <td class="left">A vararg function that takes at least one
1277 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1278 which returns an integer. This is the signature for <tt>printf</tt> in
1281 </tr><tr class="layout">
1282 <td class="left"><tt>{i32, i32} (i32)</tt></td>
1283 <td class="left">A function taking an <tt>i32></tt>, returning two
1284 <tt> i32 </tt> values as an aggregate of type <tt>{ i32, i32 }</tt>
1290 <!-- _______________________________________________________________________ -->
1291 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1292 <div class="doc_text">
1294 <p>The structure type is used to represent a collection of data members
1295 together in memory. The packing of the field types is defined to match
1296 the ABI of the underlying processor. The elements of a structure may
1297 be any type that has a size.</p>
1298 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1299 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1300 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1303 <pre> { <type list> }<br></pre>
1305 <table class="layout">
1307 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1308 <td class="left">A triple of three <tt>i32</tt> values</td>
1309 </tr><tr class="layout">
1310 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1311 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1312 second element is a <a href="#t_pointer">pointer</a> to a
1313 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1314 an <tt>i32</tt>.</td>
1319 <!-- _______________________________________________________________________ -->
1320 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1322 <div class="doc_text">
1324 <p>The packed structure type is used to represent a collection of data members
1325 together in memory. There is no padding between fields. Further, the alignment
1326 of a packed structure is 1 byte. The elements of a packed structure may
1327 be any type that has a size.</p>
1328 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1329 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1330 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1333 <pre> < { <type list> } > <br></pre>
1335 <table class="layout">
1337 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1338 <td class="left">A triple of three <tt>i32</tt> values</td>
1339 </tr><tr class="layout">
1340 <td class="left"><tt>< { float, i32 (i32)* } ></tt></td>
1341 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1342 second element is a <a href="#t_pointer">pointer</a> to a
1343 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1344 an <tt>i32</tt>.</td>
1349 <!-- _______________________________________________________________________ -->
1350 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1351 <div class="doc_text">
1353 <p>As in many languages, the pointer type represents a pointer or
1354 reference to another object, which must live in memory. Pointer types may have
1355 an optional address space attribute defining the target-specific numbered
1356 address space where the pointed-to object resides. The default address space is
1359 <pre> <type> *<br></pre>
1361 <table class="layout">
1363 <td class="left"><tt>[4x i32]*</tt></td>
1364 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1365 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1368 <td class="left"><tt>i32 (i32 *) *</tt></td>
1369 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
1370 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1374 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1375 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1376 that resides in address space #5.</td>
1381 <!-- _______________________________________________________________________ -->
1382 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1383 <div class="doc_text">
1387 <p>A vector type is a simple derived type that represents a vector
1388 of elements. Vector types are used when multiple primitive data
1389 are operated in parallel using a single instruction (SIMD).
1390 A vector type requires a size (number of
1391 elements) and an underlying primitive data type. Vectors must have a power
1392 of two length (1, 2, 4, 8, 16 ...). Vector types are
1393 considered <a href="#t_firstclass">first class</a>.</p>
1398 < <# elements> x <elementtype> >
1401 <p>The number of elements is a constant integer value; elementtype may
1402 be any integer or floating point type.</p>
1406 <table class="layout">
1408 <td class="left"><tt><4 x i32></tt></td>
1409 <td class="left">Vector of 4 32-bit integer values.</td>
1412 <td class="left"><tt><8 x float></tt></td>
1413 <td class="left">Vector of 8 32-bit floating-point values.</td>
1416 <td class="left"><tt><2 x i64></tt></td>
1417 <td class="left">Vector of 2 64-bit integer values.</td>
1422 <!-- _______________________________________________________________________ -->
1423 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1424 <div class="doc_text">
1428 <p>Opaque types are used to represent unknown types in the system. This
1429 corresponds (for example) to the C notion of a forward declared structure type.
1430 In LLVM, opaque types can eventually be resolved to any type (not just a
1431 structure type).</p>
1441 <table class="layout">
1443 <td class="left"><tt>opaque</tt></td>
1444 <td class="left">An opaque type.</td>
1450 <!-- *********************************************************************** -->
1451 <div class="doc_section"> <a name="constants">Constants</a> </div>
1452 <!-- *********************************************************************** -->
1454 <div class="doc_text">
1456 <p>LLVM has several different basic types of constants. This section describes
1457 them all and their syntax.</p>
1461 <!-- ======================================================================= -->
1462 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1464 <div class="doc_text">
1467 <dt><b>Boolean constants</b></dt>
1469 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1470 constants of the <tt><a href="#t_primitive">i1</a></tt> type.
1473 <dt><b>Integer constants</b></dt>
1475 <dd>Standard integers (such as '4') are constants of the <a
1476 href="#t_integer">integer</a> type. Negative numbers may be used with
1480 <dt><b>Floating point constants</b></dt>
1482 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1483 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1484 notation (see below). The assembler requires the exact decimal value of
1485 a floating-point constant. For example, the assembler accepts 1.25 but
1486 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
1487 constants must have a <a href="#t_floating">floating point</a> type. </dd>
1489 <dt><b>Null pointer constants</b></dt>
1491 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1492 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1496 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1497 of floating point constants. For example, the form '<tt>double
1498 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1499 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1500 (and the only time that they are generated by the disassembler) is when a
1501 floating point constant must be emitted but it cannot be represented as a
1502 decimal floating point number. For example, NaN's, infinities, and other
1503 special values are represented in their IEEE hexadecimal format so that
1504 assembly and disassembly do not cause any bits to change in the constants.</p>
1508 <!-- ======================================================================= -->
1509 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1512 <div class="doc_text">
1513 <p>Aggregate constants arise from aggregation of simple constants
1514 and smaller aggregate constants.</p>
1517 <dt><b>Structure constants</b></dt>
1519 <dd>Structure constants are represented with notation similar to structure
1520 type definitions (a comma separated list of elements, surrounded by braces
1521 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
1522 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>". Structure constants
1523 must have <a href="#t_struct">structure type</a>, and the number and
1524 types of elements must match those specified by the type.
1527 <dt><b>Array constants</b></dt>
1529 <dd>Array constants are represented with notation similar to array type
1530 definitions (a comma separated list of elements, surrounded by square brackets
1531 (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74 ]</tt>". Array
1532 constants must have <a href="#t_array">array type</a>, and the number and
1533 types of elements must match those specified by the type.
1536 <dt><b>Vector constants</b></dt>
1538 <dd>Vector constants are represented with notation similar to vector type
1539 definitions (a comma separated list of elements, surrounded by
1540 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32 42,
1541 i32 11, i32 74, i32 100 ></tt>". Vector constants must have <a
1542 href="#t_vector">vector type</a>, and the number and types of elements must
1543 match those specified by the type.
1546 <dt><b>Zero initialization</b></dt>
1548 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1549 value to zero of <em>any</em> type, including scalar and aggregate types.
1550 This is often used to avoid having to print large zero initializers (e.g. for
1551 large arrays) and is always exactly equivalent to using explicit zero
1558 <!-- ======================================================================= -->
1559 <div class="doc_subsection">
1560 <a name="globalconstants">Global Variable and Function Addresses</a>
1563 <div class="doc_text">
1565 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1566 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1567 constants. These constants are explicitly referenced when the <a
1568 href="#identifiers">identifier for the global</a> is used and always have <a
1569 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1572 <div class="doc_code">
1576 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1582 <!-- ======================================================================= -->
1583 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1584 <div class="doc_text">
1585 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1586 no specific value. Undefined values may be of any type and be used anywhere
1587 a constant is permitted.</p>
1589 <p>Undefined values indicate to the compiler that the program is well defined
1590 no matter what value is used, giving the compiler more freedom to optimize.
1594 <!-- ======================================================================= -->
1595 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1598 <div class="doc_text">
1600 <p>Constant expressions are used to allow expressions involving other constants
1601 to be used as constants. Constant expressions may be of any <a
1602 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1603 that does not have side effects (e.g. load and call are not supported). The
1604 following is the syntax for constant expressions:</p>
1607 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1608 <dd>Truncate a constant to another type. The bit size of CST must be larger
1609 than the bit size of TYPE. Both types must be integers.</dd>
1611 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1612 <dd>Zero extend a constant to another type. The bit size of CST must be
1613 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1615 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1616 <dd>Sign extend a constant to another type. The bit size of CST must be
1617 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1619 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1620 <dd>Truncate a floating point constant to another floating point type. The
1621 size of CST must be larger than the size of TYPE. Both types must be
1622 floating point.</dd>
1624 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1625 <dd>Floating point extend a constant to another type. The size of CST must be
1626 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1628 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
1629 <dd>Convert a floating point constant to the corresponding unsigned integer
1630 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1631 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1632 of the same number of elements. If the value won't fit in the integer type,
1633 the results are undefined.</dd>
1635 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
1636 <dd>Convert a floating point constant to the corresponding signed integer
1637 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1638 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1639 of the same number of elements. If the value won't fit in the integer type,
1640 the results are undefined.</dd>
1642 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
1643 <dd>Convert an unsigned integer constant to the corresponding floating point
1644 constant. TYPE must be a scalar or vector floating point type. CST must be of
1645 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1646 of the same number of elements. If the value won't fit in the floating point
1647 type, the results are undefined.</dd>
1649 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
1650 <dd>Convert a signed integer constant to the corresponding floating point
1651 constant. TYPE must be a scalar or vector floating point type. CST must be of
1652 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1653 of the same number of elements. If the value won't fit in the floating point
1654 type, the results are undefined.</dd>
1656 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
1657 <dd>Convert a pointer typed constant to the corresponding integer constant
1658 TYPE must be an integer type. CST must be of pointer type. The CST value is
1659 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1661 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
1662 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1663 pointer type. CST must be of integer type. The CST value is zero extended,
1664 truncated, or unchanged to make it fit in a pointer size. This one is
1665 <i>really</i> dangerous!</dd>
1667 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
1668 <dd>Convert a constant, CST, to another TYPE. The size of CST and TYPE must be
1669 identical (same number of bits). The conversion is done as if the CST value
1670 was stored to memory and read back as TYPE. In other words, no bits change
1671 with this operator, just the type. This can be used for conversion of
1672 vector types to any other type, as long as they have the same bit width. For
1673 pointers it is only valid to cast to another pointer type.
1676 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1678 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1679 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1680 instruction, the index list may have zero or more indexes, which are required
1681 to make sense for the type of "CSTPTR".</dd>
1683 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1685 <dd>Perform the <a href="#i_select">select operation</a> on
1688 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
1689 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
1691 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
1692 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
1694 <dt><b><tt>vicmp COND ( VAL1, VAL2 )</tt></b></dt>
1695 <dd>Performs the <a href="#i_vicmp">vicmp operation</a> on constants.</dd>
1697 <dt><b><tt>vfcmp COND ( VAL1, VAL2 )</tt></b></dt>
1698 <dd>Performs the <a href="#i_vfcmp">vfcmp operation</a> on constants.</dd>
1700 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1702 <dd>Perform the <a href="#i_extractelement">extractelement
1703 operation</a> on constants.
1705 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1707 <dd>Perform the <a href="#i_insertelement">insertelement
1708 operation</a> on constants.</dd>
1711 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
1713 <dd>Perform the <a href="#i_shufflevector">shufflevector
1714 operation</a> on constants.</dd>
1716 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1718 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1719 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1720 binary</a> operations. The constraints on operands are the same as those for
1721 the corresponding instruction (e.g. no bitwise operations on floating point
1722 values are allowed).</dd>
1726 <!-- *********************************************************************** -->
1727 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1728 <!-- *********************************************************************** -->
1730 <!-- ======================================================================= -->
1731 <div class="doc_subsection">
1732 <a name="inlineasm">Inline Assembler Expressions</a>
1735 <div class="doc_text">
1738 LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
1739 Module-Level Inline Assembly</a>) through the use of a special value. This
1740 value represents the inline assembler as a string (containing the instructions
1741 to emit), a list of operand constraints (stored as a string), and a flag that
1742 indicates whether or not the inline asm expression has side effects. An example
1743 inline assembler expression is:
1746 <div class="doc_code">
1748 i32 (i32) asm "bswap $0", "=r,r"
1753 Inline assembler expressions may <b>only</b> be used as the callee operand of
1754 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
1757 <div class="doc_code">
1759 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
1764 Inline asms with side effects not visible in the constraint list must be marked
1765 as having side effects. This is done through the use of the
1766 '<tt>sideeffect</tt>' keyword, like so:
1769 <div class="doc_code">
1771 call void asm sideeffect "eieio", ""()
1775 <p>TODO: The format of the asm and constraints string still need to be
1776 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1777 need to be documented).
1782 <!-- *********************************************************************** -->
1783 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1784 <!-- *********************************************************************** -->
1786 <div class="doc_text">
1788 <p>The LLVM instruction set consists of several different
1789 classifications of instructions: <a href="#terminators">terminator
1790 instructions</a>, <a href="#binaryops">binary instructions</a>,
1791 <a href="#bitwiseops">bitwise binary instructions</a>, <a
1792 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1793 instructions</a>.</p>
1797 <!-- ======================================================================= -->
1798 <div class="doc_subsection"> <a name="terminators">Terminator
1799 Instructions</a> </div>
1801 <div class="doc_text">
1803 <p>As mentioned <a href="#functionstructure">previously</a>, every
1804 basic block in a program ends with a "Terminator" instruction, which
1805 indicates which block should be executed after the current block is
1806 finished. These terminator instructions typically yield a '<tt>void</tt>'
1807 value: they produce control flow, not values (the one exception being
1808 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1809 <p>There are six different terminator instructions: the '<a
1810 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1811 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1812 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1813 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1814 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1818 <!-- _______________________________________________________________________ -->
1819 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1820 Instruction</a> </div>
1821 <div class="doc_text">
1823 <pre> ret <type> <value> <i>; Return a value from a non-void function</i>
1824 ret void <i>; Return from void function</i>
1825 ret <type> <value>, <type> <value> <i>; Return two values from a non-void function </i>
1830 <p>The '<tt>ret</tt>' instruction is used to return control flow (and a
1831 value) from a function back to the caller.</p>
1832 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
1833 returns value(s) and then causes control flow, and one that just causes
1834 control flow to occur.</p>
1838 <p>The '<tt>ret</tt>' instruction may return zero, one or multiple values.
1839 The type of each return value must be a '<a href="#t_firstclass">first
1840 class</a>' type. Note that a function is not <a href="#wellformed">well
1841 formed</a> if there exists a '<tt>ret</tt>' instruction inside of the
1842 function that returns values that do not match the return type of the
1847 <p>When the '<tt>ret</tt>' instruction is executed, control flow
1848 returns back to the calling function's context. If the caller is a "<a
1849 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1850 the instruction after the call. If the caller was an "<a
1851 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1852 at the beginning of the "normal" destination block. If the instruction
1853 returns a value, that value shall set the call or invoke instruction's
1854 return value. If the instruction returns multiple values then these
1855 values can only be accessed through a '<a href="#i_getresult"><tt>getresult</tt>
1856 </a>' instruction.</p>
1861 ret i32 5 <i>; Return an integer value of 5</i>
1862 ret void <i>; Return from a void function</i>
1863 ret i32 4, i8 2 <i>; Return two values 4 and 2 </i>
1866 <!-- _______________________________________________________________________ -->
1867 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1868 <div class="doc_text">
1870 <pre> br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1873 <p>The '<tt>br</tt>' instruction is used to cause control flow to
1874 transfer to a different basic block in the current function. There are
1875 two forms of this instruction, corresponding to a conditional branch
1876 and an unconditional branch.</p>
1878 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1879 single '<tt>i1</tt>' value and two '<tt>label</tt>' values. The
1880 unconditional form of the '<tt>br</tt>' instruction takes a single
1881 '<tt>label</tt>' value as a target.</p>
1883 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
1884 argument is evaluated. If the value is <tt>true</tt>, control flows
1885 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1886 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1888 <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
1889 href="#i_ret">ret</a> i32 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> i32 0<br></pre>
1891 <!-- _______________________________________________________________________ -->
1892 <div class="doc_subsubsection">
1893 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
1896 <div class="doc_text">
1900 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
1905 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
1906 several different places. It is a generalization of the '<tt>br</tt>'
1907 instruction, allowing a branch to occur to one of many possible
1913 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
1914 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
1915 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
1916 table is not allowed to contain duplicate constant entries.</p>
1920 <p>The <tt>switch</tt> instruction specifies a table of values and
1921 destinations. When the '<tt>switch</tt>' instruction is executed, this
1922 table is searched for the given value. If the value is found, control flow is
1923 transfered to the corresponding destination; otherwise, control flow is
1924 transfered to the default destination.</p>
1926 <h5>Implementation:</h5>
1928 <p>Depending on properties of the target machine and the particular
1929 <tt>switch</tt> instruction, this instruction may be code generated in different
1930 ways. For example, it could be generated as a series of chained conditional
1931 branches or with a lookup table.</p>
1936 <i>; Emulate a conditional br instruction</i>
1937 %Val = <a href="#i_zext">zext</a> i1 %value to i32
1938 switch i32 %Val, label %truedest [i32 0, label %falsedest ]
1940 <i>; Emulate an unconditional br instruction</i>
1941 switch i32 0, label %dest [ ]
1943 <i>; Implement a jump table:</i>
1944 switch i32 %val, label %otherwise [ i32 0, label %onzero
1946 i32 2, label %ontwo ]
1950 <!-- _______________________________________________________________________ -->
1951 <div class="doc_subsubsection">
1952 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
1955 <div class="doc_text">
1960 <result> = invoke [<a href="#callingconv">cconv</a>] <ptr to function ty> <function ptr val>(<function args>)
1961 to label <normal label> unwind label <exception label>
1966 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
1967 function, with the possibility of control flow transfer to either the
1968 '<tt>normal</tt>' label or the
1969 '<tt>exception</tt>' label. If the callee function returns with the
1970 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
1971 "normal" label. If the callee (or any indirect callees) returns with the "<a
1972 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
1973 continued at the dynamically nearest "exception" label. If the callee function
1974 returns multiple values then individual return values are only accessible through
1975 a '<tt><a href="#i_getresult">getresult</a></tt>' instruction.</p>
1979 <p>This instruction requires several arguments:</p>
1983 The optional "cconv" marker indicates which <a href="#callingconv">calling
1984 convention</a> the call should use. If none is specified, the call defaults
1985 to using C calling conventions.
1987 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
1988 function value being invoked. In most cases, this is a direct function
1989 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
1990 an arbitrary pointer to function value.
1993 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
1994 function to be invoked. </li>
1996 <li>'<tt>function args</tt>': argument list whose types match the function
1997 signature argument types. If the function signature indicates the function
1998 accepts a variable number of arguments, the extra arguments can be
2001 <li>'<tt>normal label</tt>': the label reached when the called function
2002 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
2004 <li>'<tt>exception label</tt>': the label reached when a callee returns with
2005 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
2011 <p>This instruction is designed to operate as a standard '<tt><a
2012 href="#i_call">call</a></tt>' instruction in most regards. The primary
2013 difference is that it establishes an association with a label, which is used by
2014 the runtime library to unwind the stack.</p>
2016 <p>This instruction is used in languages with destructors to ensure that proper
2017 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
2018 exception. Additionally, this is important for implementation of
2019 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
2023 %retval = invoke i32 @Test(i32 15) to label %Continue
2024 unwind label %TestCleanup <i>; {i32}:retval set</i>
2025 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
2026 unwind label %TestCleanup <i>; {i32}:retval set</i>
2031 <!-- _______________________________________________________________________ -->
2033 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
2034 Instruction</a> </div>
2036 <div class="doc_text">
2045 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
2046 at the first callee in the dynamic call stack which used an <a
2047 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
2048 primarily used to implement exception handling.</p>
2052 <p>The '<tt>unwind</tt>' instruction causes execution of the current function to
2053 immediately halt. The dynamic call stack is then searched for the first <a
2054 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
2055 execution continues at the "exceptional" destination block specified by the
2056 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
2057 dynamic call chain, undefined behavior results.</p>
2060 <!-- _______________________________________________________________________ -->
2062 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
2063 Instruction</a> </div>
2065 <div class="doc_text">
2074 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
2075 instruction is used to inform the optimizer that a particular portion of the
2076 code is not reachable. This can be used to indicate that the code after a
2077 no-return function cannot be reached, and other facts.</p>
2081 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
2086 <!-- ======================================================================= -->
2087 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
2088 <div class="doc_text">
2089 <p>Binary operators are used to do most of the computation in a
2090 program. They require two operands of the same type, execute an operation on them, and
2091 produce a single value. The operands might represent
2092 multiple data, as is the case with the <a href="#t_vector">vector</a> data type.
2093 The result value has the same type as its operands.</p>
2094 <p>There are several different binary operators:</p>
2096 <!-- _______________________________________________________________________ -->
2097 <div class="doc_subsubsection">
2098 <a name="i_add">'<tt>add</tt>' Instruction</a>
2101 <div class="doc_text">
2106 <result> = add <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2111 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
2115 <p>The two arguments to the '<tt>add</tt>' instruction must be <a
2116 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>, or
2117 <a href="#t_vector">vector</a> values. Both arguments must have identical
2122 <p>The value produced is the integer or floating point sum of the two
2125 <p>If an integer sum has unsigned overflow, the result returned is the
2126 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2129 <p>Because LLVM integers use a two's complement representation, this
2130 instruction is appropriate for both signed and unsigned integers.</p>
2135 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
2138 <!-- _______________________________________________________________________ -->
2139 <div class="doc_subsubsection">
2140 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
2143 <div class="doc_text">
2148 <result> = sub <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2153 <p>The '<tt>sub</tt>' instruction returns the difference of its two
2156 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
2157 '<tt>neg</tt>' instruction present in most other intermediate
2158 representations.</p>
2162 <p>The two arguments to the '<tt>sub</tt>' instruction must be <a
2163 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>,
2164 or <a href="#t_vector">vector</a> values. Both arguments must have identical
2169 <p>The value produced is the integer or floating point difference of
2170 the two operands.</p>
2172 <p>If an integer difference has unsigned overflow, the result returned is the
2173 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2176 <p>Because LLVM integers use a two's complement representation, this
2177 instruction is appropriate for both signed and unsigned integers.</p>
2181 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
2182 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
2186 <!-- _______________________________________________________________________ -->
2187 <div class="doc_subsubsection">
2188 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
2191 <div class="doc_text">
2194 <pre> <result> = mul <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2197 <p>The '<tt>mul</tt>' instruction returns the product of its two
2202 <p>The two arguments to the '<tt>mul</tt>' instruction must be <a
2203 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>,
2204 or <a href="#t_vector">vector</a> values. Both arguments must have identical
2209 <p>The value produced is the integer or floating point product of the
2212 <p>If the result of an integer multiplication has unsigned overflow,
2213 the result returned is the mathematical result modulo
2214 2<sup>n</sup>, where n is the bit width of the result.</p>
2215 <p>Because LLVM integers use a two's complement representation, and the
2216 result is the same width as the operands, this instruction returns the
2217 correct result for both signed and unsigned integers. If a full product
2218 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands
2219 should be sign-extended or zero-extended as appropriate to the
2220 width of the full product.</p>
2222 <pre> <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
2226 <!-- _______________________________________________________________________ -->
2227 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
2229 <div class="doc_text">
2231 <pre> <result> = udiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2234 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two
2239 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
2240 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2241 values. Both arguments must have identical types.</p>
2245 <p>The value produced is the unsigned integer quotient of the two operands.</p>
2246 <p>Note that unsigned integer division and signed integer division are distinct
2247 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
2248 <p>Division by zero leads to undefined behavior.</p>
2250 <pre> <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2253 <!-- _______________________________________________________________________ -->
2254 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
2256 <div class="doc_text">
2259 <result> = sdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2264 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
2269 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
2270 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2271 values. Both arguments must have identical types.</p>
2274 <p>The value produced is the signed integer quotient of the two operands rounded towards zero.</p>
2275 <p>Note that signed integer division and unsigned integer division are distinct
2276 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
2277 <p>Division by zero leads to undefined behavior. Overflow also leads to
2278 undefined behavior; this is a rare case, but can occur, for example,
2279 by doing a 32-bit division of -2147483648 by -1.</p>
2281 <pre> <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2284 <!-- _______________________________________________________________________ -->
2285 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
2286 Instruction</a> </div>
2287 <div class="doc_text">
2290 <result> = fdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2294 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
2299 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
2300 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a>
2301 of floating point values. Both arguments must have identical types.</p>
2305 <p>The value produced is the floating point quotient of the two operands.</p>
2310 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
2314 <!-- _______________________________________________________________________ -->
2315 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
2317 <div class="doc_text">
2319 <pre> <result> = urem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2322 <p>The '<tt>urem</tt>' instruction returns the remainder from the
2323 unsigned division of its two arguments.</p>
2325 <p>The two arguments to the '<tt>urem</tt>' instruction must be
2326 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2327 values. Both arguments must have identical types.</p>
2329 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
2330 This instruction always performs an unsigned division to get the remainder.</p>
2331 <p>Note that unsigned integer remainder and signed integer remainder are
2332 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
2333 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
2335 <pre> <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2339 <!-- _______________________________________________________________________ -->
2340 <div class="doc_subsubsection">
2341 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
2344 <div class="doc_text">
2349 <result> = srem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2354 <p>The '<tt>srem</tt>' instruction returns the remainder from the
2355 signed division of its two operands. This instruction can also take
2356 <a href="#t_vector">vector</a> versions of the values in which case
2357 the elements must be integers.</p>
2361 <p>The two arguments to the '<tt>srem</tt>' instruction must be
2362 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2363 values. Both arguments must have identical types.</p>
2367 <p>This instruction returns the <i>remainder</i> of a division (where the result
2368 has the same sign as the dividend, <tt>var1</tt>), not the <i>modulo</i>
2369 operator (where the result has the same sign as the divisor, <tt>var2</tt>) of
2370 a value. For more information about the difference, see <a
2371 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
2372 Math Forum</a>. For a table of how this is implemented in various languages,
2373 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
2374 Wikipedia: modulo operation</a>.</p>
2375 <p>Note that signed integer remainder and unsigned integer remainder are
2376 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
2377 <p>Taking the remainder of a division by zero leads to undefined behavior.
2378 Overflow also leads to undefined behavior; this is a rare case, but can occur,
2379 for example, by taking the remainder of a 32-bit division of -2147483648 by -1.
2380 (The remainder doesn't actually overflow, but this rule lets srem be
2381 implemented using instructions that return both the result of the division
2382 and the remainder.)</p>
2384 <pre> <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2388 <!-- _______________________________________________________________________ -->
2389 <div class="doc_subsubsection">
2390 <a name="i_frem">'<tt>frem</tt>' Instruction</a> </div>
2392 <div class="doc_text">
2395 <pre> <result> = frem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2398 <p>The '<tt>frem</tt>' instruction returns the remainder from the
2399 division of its two operands.</p>
2401 <p>The two arguments to the '<tt>frem</tt>' instruction must be
2402 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a>
2403 of floating point values. Both arguments must have identical types.</p>
2407 <p>This instruction returns the <i>remainder</i> of a division.
2408 The remainder has the same sign as the dividend.</p>
2413 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
2417 <!-- ======================================================================= -->
2418 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
2419 Operations</a> </div>
2420 <div class="doc_text">
2421 <p>Bitwise binary operators are used to do various forms of
2422 bit-twiddling in a program. They are generally very efficient
2423 instructions and can commonly be strength reduced from other
2424 instructions. They require two operands of the same type, execute an operation on them,
2425 and produce a single value. The resulting value is the same type as its operands.</p>
2428 <!-- _______________________________________________________________________ -->
2429 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
2430 Instruction</a> </div>
2431 <div class="doc_text">
2433 <pre> <result> = shl <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2438 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2439 the left a specified number of bits.</p>
2443 <p>Both arguments to the '<tt>shl</tt>' instruction must be the same <a
2444 href="#t_integer">integer</a> type. '<tt>var2</tt>' is treated as an
2445 unsigned value. This instruction does not support
2446 <a href="#t_vector">vector</a> operands.</p>
2450 <p>The value produced is <tt>var1</tt> * 2<sup><tt>var2</tt></sup> mod 2<sup>n</sup>,
2451 where n is the width of the result. If <tt>var2</tt> is (statically or dynamically) negative or
2452 equal to or larger than the number of bits in <tt>var1</tt>, the result is undefined.</p>
2454 <h5>Example:</h5><pre>
2455 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
2456 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
2457 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
2458 <result> = shl i32 1, 32 <i>; undefined</i>
2461 <!-- _______________________________________________________________________ -->
2462 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
2463 Instruction</a> </div>
2464 <div class="doc_text">
2466 <pre> <result> = lshr <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2470 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
2471 operand shifted to the right a specified number of bits with zero fill.</p>
2474 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
2475 <a href="#t_integer">integer</a> type. '<tt>var2</tt>' is treated as an
2476 unsigned value. This instruction does not support
2477 <a href="#t_vector">vector</a> operands.</p>
2481 <p>This instruction always performs a logical shift right operation. The most
2482 significant bits of the result will be filled with zero bits after the
2483 shift. If <tt>var2</tt> is (statically or dynamically) equal to or larger than
2484 the number of bits in <tt>var1</tt>, the result is undefined.</p>
2488 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
2489 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
2490 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
2491 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
2492 <result> = lshr i32 1, 32 <i>; undefined</i>
2496 <!-- _______________________________________________________________________ -->
2497 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
2498 Instruction</a> </div>
2499 <div class="doc_text">
2502 <pre> <result> = ashr <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2506 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
2507 operand shifted to the right a specified number of bits with sign extension.</p>
2510 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
2511 <a href="#t_integer">integer</a> type. '<tt>var2</tt>' is treated as an
2512 unsigned value. This instruction does not support
2513 <a href="#t_vector">vector</a> operands.</p>
2516 <p>This instruction always performs an arithmetic shift right operation,
2517 The most significant bits of the result will be filled with the sign bit
2518 of <tt>var1</tt>. If <tt>var2</tt> is (statically or dynamically) equal to or
2519 larger than the number of bits in <tt>var1</tt>, the result is undefined.
2524 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
2525 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
2526 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
2527 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
2528 <result> = ashr i32 1, 32 <i>; undefined</i>
2532 <!-- _______________________________________________________________________ -->
2533 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
2534 Instruction</a> </div>
2536 <div class="doc_text">
2541 <result> = and <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2546 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
2547 its two operands.</p>
2551 <p>The two arguments to the '<tt>and</tt>' instruction must be
2552 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2553 values. Both arguments must have identical types.</p>
2556 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
2558 <div style="align: center">
2559 <table border="1" cellspacing="0" cellpadding="4">
2591 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
2592 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
2593 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
2596 <!-- _______________________________________________________________________ -->
2597 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
2598 <div class="doc_text">
2600 <pre> <result> = or <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2603 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
2604 or of its two operands.</p>
2607 <p>The two arguments to the '<tt>or</tt>' instruction must be
2608 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2609 values. Both arguments must have identical types.</p>
2611 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
2613 <div style="align: center">
2614 <table border="1" cellspacing="0" cellpadding="4">
2645 <pre> <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
2646 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
2647 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
2650 <!-- _______________________________________________________________________ -->
2651 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
2652 Instruction</a> </div>
2653 <div class="doc_text">
2655 <pre> <result> = xor <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2658 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
2659 or of its two operands. The <tt>xor</tt> is used to implement the
2660 "one's complement" operation, which is the "~" operator in C.</p>
2662 <p>The two arguments to the '<tt>xor</tt>' instruction must be
2663 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2664 values. Both arguments must have identical types.</p>
2668 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
2670 <div style="align: center">
2671 <table border="1" cellspacing="0" cellpadding="4">
2703 <pre> <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
2704 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
2705 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
2706 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
2710 <!-- ======================================================================= -->
2711 <div class="doc_subsection">
2712 <a name="vectorops">Vector Operations</a>
2715 <div class="doc_text">
2717 <p>LLVM supports several instructions to represent vector operations in a
2718 target-independent manner. These instructions cover the element-access and
2719 vector-specific operations needed to process vectors effectively. While LLVM
2720 does directly support these vector operations, many sophisticated algorithms
2721 will want to use target-specific intrinsics to take full advantage of a specific
2726 <!-- _______________________________________________________________________ -->
2727 <div class="doc_subsubsection">
2728 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2731 <div class="doc_text">
2736 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
2742 The '<tt>extractelement</tt>' instruction extracts a single scalar
2743 element from a vector at a specified index.
2750 The first operand of an '<tt>extractelement</tt>' instruction is a
2751 value of <a href="#t_vector">vector</a> type. The second operand is
2752 an index indicating the position from which to extract the element.
2753 The index may be a variable.</p>
2758 The result is a scalar of the same type as the element type of
2759 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2760 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2761 results are undefined.
2767 %result = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
2772 <!-- _______________________________________________________________________ -->
2773 <div class="doc_subsubsection">
2774 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
2777 <div class="doc_text">
2782 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
2788 The '<tt>insertelement</tt>' instruction inserts a scalar
2789 element into a vector at a specified index.
2796 The first operand of an '<tt>insertelement</tt>' instruction is a
2797 value of <a href="#t_vector">vector</a> type. The second operand is a
2798 scalar value whose type must equal the element type of the first
2799 operand. The third operand is an index indicating the position at
2800 which to insert the value. The index may be a variable.</p>
2805 The result is a vector of the same type as <tt>val</tt>. Its
2806 element values are those of <tt>val</tt> except at position
2807 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
2808 exceeds the length of <tt>val</tt>, the results are undefined.
2814 %result = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
2818 <!-- _______________________________________________________________________ -->
2819 <div class="doc_subsubsection">
2820 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
2823 <div class="doc_text">
2828 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <n x i32> <mask> <i>; yields <n x <ty>></i>
2834 The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
2835 from two input vectors, returning a vector of the same type.
2841 The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
2842 with types that match each other and types that match the result of the
2843 instruction. The third argument is a shuffle mask, which has the same number
2844 of elements as the other vector type, but whose element type is always 'i32'.
2848 The shuffle mask operand is required to be a constant vector with either
2849 constant integer or undef values.
2855 The elements of the two input vectors are numbered from left to right across
2856 both of the vectors. The shuffle mask operand specifies, for each element of
2857 the result vector, which element of the two input registers the result element
2858 gets. The element selector may be undef (meaning "don't care") and the second
2859 operand may be undef if performing a shuffle from only one vector.
2865 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
2866 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
2867 %result = shufflevector <4 x i32> %v1, <4 x i32> undef,
2868 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
2873 <!-- ======================================================================= -->
2874 <div class="doc_subsection">
2875 <a name="aggregateops">Aggregate Operations</a>
2878 <div class="doc_text">
2880 <p>LLVM supports several instructions for working with aggregate values.
2885 <!-- _______________________________________________________________________ -->
2886 <div class="doc_subsubsection">
2887 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
2890 <div class="doc_text">
2895 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
2901 The '<tt>extractvalue</tt>' instruction extracts the value of a struct field
2902 or array element from an aggregate value.
2909 The first operand of an '<tt>extractvalue</tt>' instruction is a
2910 value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a>
2911 type. The operands are constant indices to specify which value to extract
2912 in the same manner as indices in a
2913 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
2919 The result is the value at the position in the aggregate specified by
2926 %result = extractvalue {i32, float} %agg, i32 0 <i>; yields i32</i>
2931 <!-- _______________________________________________________________________ -->
2932 <div class="doc_subsubsection">
2933 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
2936 <div class="doc_text">
2941 <result> = insertvalue <aggregate type> <val>, <ty> <val>, i32 <idx> <i>; yields <n x <ty>></i>
2947 The '<tt>insertvalue</tt>' instruction inserts a value
2948 into a struct field or array element in an aggregate.
2955 The first operand of an '<tt>insertvalue</tt>' instruction is a
2956 value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a> type.
2957 The second operand is a first-class value to insert.
2958 The following operands are constant indices
2959 indicating the position at which to insert the value in the same manner as
2961 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
2962 The value to insert must have the same type as the value identified
2968 The result is an aggregate of the same type as <tt>val</tt>. Its
2969 value is that of <tt>val</tt> except that the value at the position
2970 specified by the indices is that of <tt>elt</tt>.
2976 %result = insertvalue {i32, float} %agg, i32 1, i32 0 <i>; yields {i32, float}</i>
2981 <!-- ======================================================================= -->
2982 <div class="doc_subsection">
2983 <a name="memoryops">Memory Access and Addressing Operations</a>
2986 <div class="doc_text">
2988 <p>A key design point of an SSA-based representation is how it
2989 represents memory. In LLVM, no memory locations are in SSA form, which
2990 makes things very simple. This section describes how to read, write,
2991 allocate, and free memory in LLVM.</p>
2995 <!-- _______________________________________________________________________ -->
2996 <div class="doc_subsubsection">
2997 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
3000 <div class="doc_text">
3005 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3010 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
3011 heap and returns a pointer to it. The object is always allocated in the generic
3012 address space (address space zero).</p>
3016 <p>The '<tt>malloc</tt>' instruction allocates
3017 <tt>sizeof(<type>)*NumElements</tt>
3018 bytes of memory from the operating system and returns 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 to
3022 be aligned to at least that boundary. If not specified, or if zero, the target can
3023 choose to align the allocation on any convenient boundary.</p>
3025 <p>'<tt>type</tt>' must be a sized type.</p>
3029 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
3030 a pointer is returned. The result of a zero byte allocattion is undefined. The
3031 result is null if there is insufficient memory available.</p>
3036 %array = malloc [4 x i8 ] <i>; yields {[%4 x i8]*}:array</i>
3038 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
3039 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
3040 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
3041 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
3042 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
3046 <!-- _______________________________________________________________________ -->
3047 <div class="doc_subsubsection">
3048 <a name="i_free">'<tt>free</tt>' Instruction</a>
3051 <div class="doc_text">
3056 free <type> <value> <i>; yields {void}</i>
3061 <p>The '<tt>free</tt>' instruction returns memory back to the unused
3062 memory heap to be reallocated in the future.</p>
3066 <p>'<tt>value</tt>' shall be a pointer value that points to a value
3067 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
3072 <p>Access to the memory pointed to by the pointer is no longer defined
3073 after this instruction executes. If the pointer is null, the operation
3079 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
3080 free [4 x i8]* %array
3084 <!-- _______________________________________________________________________ -->
3085 <div class="doc_subsubsection">
3086 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
3089 <div class="doc_text">
3094 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3099 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
3100 currently executing function, to be automatically released when this function
3101 returns to its caller. The object is always allocated in the generic address
3102 space (address space zero).</p>
3106 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
3107 bytes of memory on the runtime stack, returning a pointer of the
3108 appropriate type to the program. If "NumElements" is specified, it is the
3109 number of elements allocated, otherwise "NumElements" is defaulted to be one.
3110 If a constant alignment is specified, the value result of the allocation is guaranteed
3111 to be aligned to at least that boundary. If not specified, or if zero, the target
3112 can choose to align the allocation on any convenient boundary.</p>
3114 <p>'<tt>type</tt>' may be any sized type.</p>
3118 <p>Memory is allocated; a pointer is returned. The operation is undefiend if
3119 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
3120 memory is automatically released when the function returns. The '<tt>alloca</tt>'
3121 instruction is commonly used to represent automatic variables that must
3122 have an address available. When the function returns (either with the <tt><a
3123 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
3124 instructions), the memory is reclaimed. Allocating zero bytes
3125 is legal, but the result is undefined.</p>
3130 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
3131 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
3132 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
3133 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
3137 <!-- _______________________________________________________________________ -->
3138 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
3139 Instruction</a> </div>
3140 <div class="doc_text">
3142 <pre> <result> = load <ty>* <pointer>[, align <alignment>]<br> <result> = volatile load <ty>* <pointer>[, align <alignment>]<br></pre>
3144 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
3146 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
3147 address from which to load. The pointer must point to a <a
3148 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
3149 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
3150 the number or order of execution of this <tt>load</tt> with other
3151 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
3154 The optional constant "align" argument specifies the alignment of the operation
3155 (that is, the alignment of the memory address). A value of 0 or an
3156 omitted "align" argument means that the operation has the preferential
3157 alignment for the target. It is the responsibility of the code emitter
3158 to ensure that the alignment information is correct. Overestimating
3159 the alignment results in an undefined behavior. Underestimating the
3160 alignment may produce less efficient code. An alignment of 1 is always
3164 <p>The location of memory pointed to is loaded.</p>
3166 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3168 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
3169 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
3172 <!-- _______________________________________________________________________ -->
3173 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
3174 Instruction</a> </div>
3175 <div class="doc_text">
3177 <pre> store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3178 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3181 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
3183 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
3184 to store and an address at which to store it. The type of the '<tt><pointer></tt>'
3185 operand must be a pointer to the <a href="#t_firstclass">first class</a> type
3186 of the '<tt><value></tt>'
3187 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
3188 optimizer is not allowed to modify the number or order of execution of
3189 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
3190 href="#i_store">store</a></tt> instructions.</p>
3192 The optional constant "align" argument specifies the alignment of the operation
3193 (that is, the alignment of the memory address). A value of 0 or an
3194 omitted "align" argument means that the operation has the preferential
3195 alignment for the target. It is the responsibility of the code emitter
3196 to ensure that the alignment information is correct. Overestimating
3197 the alignment results in an undefined behavior. Underestimating the
3198 alignment may produce less efficient code. An alignment of 1 is always
3202 <p>The contents of memory are updated to contain '<tt><value></tt>'
3203 at the location specified by the '<tt><pointer></tt>' operand.</p>
3205 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3206 store i32 3, i32* %ptr <i>; yields {void}</i>
3207 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
3211 <!-- _______________________________________________________________________ -->
3212 <div class="doc_subsubsection">
3213 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
3216 <div class="doc_text">
3219 <result> = getelementptr <ty>* <ptrval>{, <ty> <idx>}*
3225 The '<tt>getelementptr</tt>' instruction is used to get the address of a
3226 subelement of an aggregate data structure.</p>
3230 <p>This instruction takes a list of integer operands that indicate what
3231 elements of the aggregate object to index to. The actual types of the arguments
3232 provided depend on the type of the first pointer argument. The
3233 '<tt>getelementptr</tt>' instruction is used to index down through the type
3234 levels of a structure or to a specific index in an array. When indexing into a
3235 structure, only <tt>i32</tt> integer constants are allowed. When indexing
3236 into an array or pointer, only integers of 32 or 64 bits are allowed; 32-bit
3237 values will be sign extended to 64-bits if required.</p>
3239 <p>For example, let's consider a C code fragment and how it gets
3240 compiled to LLVM:</p>
3242 <div class="doc_code">
3255 int *foo(struct ST *s) {
3256 return &s[1].Z.B[5][13];
3261 <p>The LLVM code generated by the GCC frontend is:</p>
3263 <div class="doc_code">
3265 %RT = type { i8 , [10 x [20 x i32]], i8 }
3266 %ST = type { i32, double, %RT }
3268 define i32* %foo(%ST* %s) {
3270 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
3278 <p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
3279 on the pointer type that is being indexed into. <a href="#t_pointer">Pointer</a>
3280 and <a href="#t_array">array</a> types can use a 32-bit or 64-bit
3281 <a href="#t_integer">integer</a> type but the value will always be sign extended
3282 to 64-bits. <a href="#t_struct">Structure</a> and <a href="#t_pstruct">packed
3283 structure</a> types require <tt>i32</tt> <b>constants</b>.</p>
3285 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
3286 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
3287 }</tt>' type, a structure. The second index indexes into the third element of
3288 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
3289 i8 }</tt>' type, another structure. The third index indexes into the second
3290 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
3291 array. The two dimensions of the array are subscripted into, yielding an
3292 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
3293 to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
3295 <p>Note that it is perfectly legal to index partially through a
3296 structure, returning a pointer to an inner element. Because of this,
3297 the LLVM code for the given testcase is equivalent to:</p>
3300 define i32* %foo(%ST* %s) {
3301 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
3302 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
3303 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
3304 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
3305 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
3310 <p>Note that it is undefined to access an array out of bounds: array and
3311 pointer indexes must always be within the defined bounds of the array type.
3312 The one exception for this rule is zero length arrays. These arrays are
3313 defined to be accessible as variable length arrays, which requires access
3314 beyond the zero'th element.</p>
3316 <p>The getelementptr instruction is often confusing. For some more insight
3317 into how it works, see <a href="GetElementPtr.html">the getelementptr
3323 <i>; yields [12 x i8]*:aptr</i>
3324 %aptr = getelementptr {i32, [12 x i8]}* %sptr, i64 0, i32 1
3328 <!-- ======================================================================= -->
3329 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
3331 <div class="doc_text">
3332 <p>The instructions in this category are the conversion instructions (casting)
3333 which all take a single operand and a type. They perform various bit conversions
3337 <!-- _______________________________________________________________________ -->
3338 <div class="doc_subsubsection">
3339 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
3341 <div class="doc_text">
3345 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
3350 The '<tt>trunc</tt>' instruction truncates its operand to the type <tt>ty2</tt>.
3355 The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
3356 be an <a href="#t_integer">integer</a> type, and a type that specifies the size
3357 and type of the result, which must be an <a href="#t_integer">integer</a>
3358 type. The bit size of <tt>value</tt> must be larger than the bit size of
3359 <tt>ty2</tt>. Equal sized types are not allowed.</p>
3363 The '<tt>trunc</tt>' instruction truncates the high order bits in <tt>value</tt>
3364 and converts the remaining bits to <tt>ty2</tt>. Since the source size must be
3365 larger than the destination size, <tt>trunc</tt> cannot be a <i>no-op cast</i>.
3366 It will always truncate bits.</p>
3370 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
3371 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
3372 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
3376 <!-- _______________________________________________________________________ -->
3377 <div class="doc_subsubsection">
3378 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
3380 <div class="doc_text">
3384 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
3388 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
3393 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
3394 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3395 also be of <a href="#t_integer">integer</a> type. The bit size of the
3396 <tt>value</tt> must be smaller than the bit size of the destination type,
3400 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
3401 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
3403 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
3407 %X = zext i32 257 to i64 <i>; yields i64:257</i>
3408 %Y = zext i1 true to i32 <i>; yields i32:1</i>
3412 <!-- _______________________________________________________________________ -->
3413 <div class="doc_subsubsection">
3414 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
3416 <div class="doc_text">
3420 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
3424 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
3428 The '<tt>sext</tt>' instruction takes a value to cast, which must be of
3429 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3430 also be of <a href="#t_integer">integer</a> type. The bit size of the
3431 <tt>value</tt> must be smaller than the bit size of the destination type,
3436 The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
3437 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size of
3438 the type <tt>ty2</tt>.</p>
3440 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
3444 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
3445 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
3449 <!-- _______________________________________________________________________ -->
3450 <div class="doc_subsubsection">
3451 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
3454 <div class="doc_text">
3459 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
3463 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
3468 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
3469 point</a> value to cast and a <a href="#t_floating">floating point</a> type to
3470 cast it to. The size of <tt>value</tt> must be larger than the size of
3471 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
3472 <i>no-op cast</i>.</p>
3475 <p> The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
3476 <a href="#t_floating">floating point</a> type to a smaller
3477 <a href="#t_floating">floating point</a> type. If the value cannot fit within
3478 the destination type, <tt>ty2</tt>, then the results are undefined.</p>
3482 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
3483 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
3487 <!-- _______________________________________________________________________ -->
3488 <div class="doc_subsubsection">
3489 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
3491 <div class="doc_text">
3495 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
3499 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
3500 floating point value.</p>
3503 <p>The '<tt>fpext</tt>' instruction takes a
3504 <a href="#t_floating">floating point</a> <tt>value</tt> to cast,
3505 and a <a href="#t_floating">floating point</a> type to cast it to. The source
3506 type must be smaller than the destination type.</p>
3509 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
3510 <a href="#t_floating">floating point</a> type to a larger
3511 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
3512 used to make a <i>no-op cast</i> because it always changes bits. Use
3513 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
3517 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
3518 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
3522 <!-- _______________________________________________________________________ -->
3523 <div class="doc_subsubsection">
3524 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
3526 <div class="doc_text">
3530 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
3534 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
3535 unsigned integer equivalent of type <tt>ty2</tt>.
3539 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
3540 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3541 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3542 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3543 vector integer type with the same number of elements as <tt>ty</tt></p>
3546 <p> The '<tt>fptoui</tt>' instruction converts its
3547 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3548 towards zero) unsigned integer value. If the value cannot fit in <tt>ty2</tt>,
3549 the results are undefined.</p>
3553 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
3554 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
3555 %X = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
3559 <!-- _______________________________________________________________________ -->
3560 <div class="doc_subsubsection">
3561 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
3563 <div class="doc_text">
3567 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
3571 <p>The '<tt>fptosi</tt>' instruction converts
3572 <a href="#t_floating">floating point</a> <tt>value</tt> to type <tt>ty2</tt>.
3576 <p> The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
3577 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3578 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3579 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3580 vector integer type with the same number of elements as <tt>ty</tt></p>
3583 <p>The '<tt>fptosi</tt>' instruction converts its
3584 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3585 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
3586 the results are undefined.</p>
3590 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
3591 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
3592 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
3596 <!-- _______________________________________________________________________ -->
3597 <div class="doc_subsubsection">
3598 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
3600 <div class="doc_text">
3604 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3608 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
3609 integer and converts that value to the <tt>ty2</tt> type.</p>
3612 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
3613 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3614 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3615 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3616 floating point type with the same number of elements as <tt>ty</tt></p>
3619 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
3620 integer quantity and converts it to the corresponding floating point value. If
3621 the value cannot fit in the floating point value, the results are undefined.</p>
3625 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
3626 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
3630 <!-- _______________________________________________________________________ -->
3631 <div class="doc_subsubsection">
3632 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
3634 <div class="doc_text">
3638 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3642 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed
3643 integer and converts that value to the <tt>ty2</tt> type.</p>
3646 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
3647 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3648 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3649 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3650 floating point type with the same number of elements as <tt>ty</tt></p>
3653 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed
3654 integer quantity and converts it to the corresponding floating point value. If
3655 the value cannot fit in the floating point value, the results are undefined.</p>
3659 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
3660 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
3664 <!-- _______________________________________________________________________ -->
3665 <div class="doc_subsubsection">
3666 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
3668 <div class="doc_text">
3672 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
3676 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
3677 the integer type <tt>ty2</tt>.</p>
3680 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
3681 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
3682 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.
3685 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
3686 <tt>ty2</tt> by interpreting the pointer value as an integer and either
3687 truncating or zero extending that value to the size of the integer type. If
3688 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
3689 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
3690 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
3695 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
3696 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
3700 <!-- _______________________________________________________________________ -->
3701 <div class="doc_subsubsection">
3702 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
3704 <div class="doc_text">
3708 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
3712 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to
3713 a pointer type, <tt>ty2</tt>.</p>
3716 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
3717 value to cast, and a type to cast it to, which must be a
3718 <a href="#t_pointer">pointer</a> type.
3721 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
3722 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
3723 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
3724 size of a pointer then a truncation is done. If <tt>value</tt> is smaller than
3725 the size of a pointer then a zero extension is done. If they are the same size,
3726 nothing is done (<i>no-op cast</i>).</p>
3730 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
3731 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
3732 %Y = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
3736 <!-- _______________________________________________________________________ -->
3737 <div class="doc_subsubsection">
3738 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
3740 <div class="doc_text">
3744 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
3749 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3750 <tt>ty2</tt> without changing any bits.</p>
3754 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be
3755 a first class value, and a type to cast it to, which must also be a <a
3756 href="#t_firstclass">first class</a> type. The bit sizes of <tt>value</tt>
3757 and the destination type, <tt>ty2</tt>, must be identical. If the source
3758 type is a pointer, the destination type must also be a pointer. This
3759 instruction supports bitwise conversion of vectors to integers and to vectors
3760 of other types (as long as they have the same size).</p>
3763 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3764 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
3765 this conversion. The conversion is done as if the <tt>value</tt> had been
3766 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only be
3767 converted to other pointer types with this instruction. To convert pointers to
3768 other types, use the <a href="#i_inttoptr">inttoptr</a> or
3769 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
3773 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
3774 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
3775 %Z = bitcast <2xint> %V to i64; <i>; yields i64: %V</i>
3779 <!-- ======================================================================= -->
3780 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
3781 <div class="doc_text">
3782 <p>The instructions in this category are the "miscellaneous"
3783 instructions, which defy better classification.</p>
3786 <!-- _______________________________________________________________________ -->
3787 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
3789 <div class="doc_text">
3791 <pre> <result> = icmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3794 <p>The '<tt>icmp</tt>' instruction returns a boolean value based on comparison
3795 of its two integer or pointer operands.</p>
3797 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
3798 the condition code indicating the kind of comparison to perform. It is not
3799 a value, just a keyword. The possible condition code are:
3801 <li><tt>eq</tt>: equal</li>
3802 <li><tt>ne</tt>: not equal </li>
3803 <li><tt>ugt</tt>: unsigned greater than</li>
3804 <li><tt>uge</tt>: unsigned greater or equal</li>
3805 <li><tt>ult</tt>: unsigned less than</li>
3806 <li><tt>ule</tt>: unsigned less or equal</li>
3807 <li><tt>sgt</tt>: signed greater than</li>
3808 <li><tt>sge</tt>: signed greater or equal</li>
3809 <li><tt>slt</tt>: signed less than</li>
3810 <li><tt>sle</tt>: signed less or equal</li>
3812 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
3813 <a href="#t_pointer">pointer</a> typed. They must also be identical types.</p>
3815 <p>The '<tt>icmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3816 the condition code given as <tt>cond</tt>. The comparison performed always
3817 yields a <a href="#t_primitive">i1</a> result, as follows:
3819 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
3820 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3822 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
3823 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3824 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
3825 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3826 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
3827 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3828 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
3829 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3830 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
3831 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3832 <li><tt>sgt</tt>: interprets the operands as signed values and yields
3833 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3834 <li><tt>sge</tt>: interprets the operands as signed values and yields
3835 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3836 <li><tt>slt</tt>: interprets the operands as signed values and yields
3837 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3838 <li><tt>sle</tt>: interprets the operands as signed values and yields
3839 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3841 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
3842 values are compared as if they were integers.</p>
3845 <pre> <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
3846 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
3847 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
3848 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
3849 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
3850 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
3854 <!-- _______________________________________________________________________ -->
3855 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
3857 <div class="doc_text">
3859 <pre> <result> = fcmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3862 <p>The '<tt>fcmp</tt>' instruction returns a boolean value based on comparison
3863 of its floating point operands.</p>
3865 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
3866 the condition code indicating the kind of comparison to perform. It is not
3867 a value, just a keyword. The possible condition code are:
3869 <li><tt>false</tt>: no comparison, always returns false</li>
3870 <li><tt>oeq</tt>: ordered and equal</li>
3871 <li><tt>ogt</tt>: ordered and greater than </li>
3872 <li><tt>oge</tt>: ordered and greater than or equal</li>
3873 <li><tt>olt</tt>: ordered and less than </li>
3874 <li><tt>ole</tt>: ordered and less than or equal</li>
3875 <li><tt>one</tt>: ordered and not equal</li>
3876 <li><tt>ord</tt>: ordered (no nans)</li>
3877 <li><tt>ueq</tt>: unordered or equal</li>
3878 <li><tt>ugt</tt>: unordered or greater than </li>
3879 <li><tt>uge</tt>: unordered or greater than or equal</li>
3880 <li><tt>ult</tt>: unordered or less than </li>
3881 <li><tt>ule</tt>: unordered or less than or equal</li>
3882 <li><tt>une</tt>: unordered or not equal</li>
3883 <li><tt>uno</tt>: unordered (either nans)</li>
3884 <li><tt>true</tt>: no comparison, always returns true</li>
3886 <p><i>Ordered</i> means that neither operand is a QNAN while
3887 <i>unordered</i> means that either operand may be a QNAN.</p>
3888 <p>The <tt>val1</tt> and <tt>val2</tt> arguments must be
3889 <a href="#t_floating">floating point</a> typed. They must have identical
3892 <p>The '<tt>fcmp</tt>' instruction compares <tt>var1</tt> and <tt>var2</tt>
3893 according to the condition code given as <tt>cond</tt>. The comparison performed
3894 always yields a <a href="#t_primitive">i1</a> result, as follows:
3896 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
3897 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3898 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3899 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3900 <tt>var1</tt> is greather than <tt>var2</tt>.</li>
3901 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3902 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3903 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3904 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3905 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3906 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3907 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3908 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3909 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
3910 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
3911 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3912 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
3913 <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3914 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
3915 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3916 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
3917 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3918 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
3919 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3920 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
3921 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3922 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
3923 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
3927 <pre> <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
3928 <result> = icmp one float 4.0, 5.0 <i>; yields: result=true</i>
3929 <result> = icmp olt float 4.0, 5.0 <i>; yields: result=true</i>
3930 <result> = icmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
3934 <!-- _______________________________________________________________________ -->
3935 <div class="doc_subsubsection">
3936 <a name="i_vicmp">'<tt>vicmp</tt>' Instruction</a>
3938 <div class="doc_text">
3940 <pre> <result> = vicmp <cond> <ty> <var1>, <var2> <i>; yields {ty}:result</i>
3943 <p>The '<tt>vicmp</tt>' instruction returns an integer vector value based on
3944 element-wise comparison of its two integer vector operands.</p>
3946 <p>The '<tt>vicmp</tt>' instruction takes three operands. The first operand is
3947 the condition code indicating the kind of comparison to perform. It is not
3948 a value, just a keyword. The possible condition code are:
3950 <li><tt>eq</tt>: equal</li>
3951 <li><tt>ne</tt>: not equal </li>
3952 <li><tt>ugt</tt>: unsigned greater than</li>
3953 <li><tt>uge</tt>: unsigned greater or equal</li>
3954 <li><tt>ult</tt>: unsigned less than</li>
3955 <li><tt>ule</tt>: unsigned less or equal</li>
3956 <li><tt>sgt</tt>: signed greater than</li>
3957 <li><tt>sge</tt>: signed greater or equal</li>
3958 <li><tt>slt</tt>: signed less than</li>
3959 <li><tt>sle</tt>: signed less or equal</li>
3961 <p>The remaining two arguments must be <a href="#t_vector">vector</a> of
3962 <a href="#t_integer">integer</a> typed. They must also be identical types.</p>
3964 <p>The '<tt>vicmp</tt>' instruction compares <tt>var1</tt> and <tt>var2</tt>
3965 according to the condition code given as <tt>cond</tt>. The comparison yields a
3966 <a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, of
3967 identical type as the values being compared. The most significant bit in each
3968 element is 1 if the element-wise comparison evaluates to true, and is 0
3969 otherwise. All other bits of the result are undefined. The condition codes
3970 are evaluated identically to the <a href="#i_icmp">'<tt>icmp</tt>'
3975 <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>
3976 <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>
3980 <!-- _______________________________________________________________________ -->
3981 <div class="doc_subsubsection">
3982 <a name="i_vfcmp">'<tt>vfcmp</tt>' Instruction</a>
3984 <div class="doc_text">
3986 <pre> <result> = vfcmp <cond> <ty> <var1>, <var2></pre>
3988 <p>The '<tt>vfcmp</tt>' instruction returns an integer vector value based on
3989 element-wise comparison of its two floating point vector operands. The output
3990 elements have the same width as the input elements.</p>
3992 <p>The '<tt>vfcmp</tt>' instruction takes three operands. The first operand is
3993 the condition code indicating the kind of comparison to perform. It is not
3994 a value, just a keyword. The possible condition code are:
3996 <li><tt>false</tt>: no comparison, always returns false</li>
3997 <li><tt>oeq</tt>: ordered and equal</li>
3998 <li><tt>ogt</tt>: ordered and greater than </li>
3999 <li><tt>oge</tt>: ordered and greater than or equal</li>
4000 <li><tt>olt</tt>: ordered and less than </li>
4001 <li><tt>ole</tt>: ordered and less than or equal</li>
4002 <li><tt>one</tt>: ordered and not equal</li>
4003 <li><tt>ord</tt>: ordered (no nans)</li>
4004 <li><tt>ueq</tt>: unordered or equal</li>
4005 <li><tt>ugt</tt>: unordered or greater than </li>
4006 <li><tt>uge</tt>: unordered or greater than or equal</li>
4007 <li><tt>ult</tt>: unordered or less than </li>
4008 <li><tt>ule</tt>: unordered or less than or equal</li>
4009 <li><tt>une</tt>: unordered or not equal</li>
4010 <li><tt>uno</tt>: unordered (either nans)</li>
4011 <li><tt>true</tt>: no comparison, always returns true</li>
4013 <p>The remaining two arguments must be <a href="#t_vector">vector</a> of
4014 <a href="#t_floating">floating point</a> typed. They must also be identical
4017 <p>The '<tt>vfcmp</tt>' instruction compares <tt>var1</tt> and <tt>var2</tt>
4018 according to the condition code given as <tt>cond</tt>. The comparison yields a
4019 <a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, with
4020 an identical number of elements as the values being compared, and each element
4021 having identical with to the width of the floating point elements. The most
4022 significant bit in each element is 1 if the element-wise comparison evaluates to
4023 true, and is 0 otherwise. All other bits of the result are undefined. The
4024 condition codes are evaluated identically to the
4025 <a href="#i_fcmp">'<tt>fcmp</tt>' instruction</a>.
4029 <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>
4030 <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>
4034 <!-- _______________________________________________________________________ -->
4035 <div class="doc_subsubsection">
4036 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
4039 <div class="doc_text">
4043 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
4045 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
4046 the SSA graph representing the function.</p>
4049 <p>The type of the incoming values is specified with the first type
4050 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
4051 as arguments, with one pair for each predecessor basic block of the
4052 current block. Only values of <a href="#t_firstclass">first class</a>
4053 type may be used as the value arguments to the PHI node. Only labels
4054 may be used as the label arguments.</p>
4056 <p>There must be no non-phi instructions between the start of a basic
4057 block and the PHI instructions: i.e. PHI instructions must be first in
4062 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
4063 specified by the pair corresponding to the predecessor basic block that executed
4064 just prior to the current block.</p>
4068 Loop: ; Infinite loop that counts from 0 on up...
4069 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
4070 %nextindvar = add i32 %indvar, 1
4075 <!-- _______________________________________________________________________ -->
4076 <div class="doc_subsubsection">
4077 <a name="i_select">'<tt>select</tt>' Instruction</a>
4080 <div class="doc_text">
4085 <result> = select i1 <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
4091 The '<tt>select</tt>' instruction is used to choose one value based on a
4092 condition, without branching.
4099 The '<tt>select</tt>' instruction requires an 'i1' value indicating the
4100 condition, and two values of the same <a href="#t_firstclass">first class</a>
4101 type. If the val1/val2 are vectors, the entire vectors are selected, not
4102 individual elements.
4108 If the i1 condition evaluates is 1, the instruction returns the first
4109 value argument; otherwise, it returns the second value argument.
4115 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
4120 <!-- _______________________________________________________________________ -->
4121 <div class="doc_subsubsection">
4122 <a name="i_call">'<tt>call</tt>' Instruction</a>
4125 <div class="doc_text">
4129 <result> = [tail] call [<a href="#callingconv">cconv</a>] <ty> [<fnty>*] <fnptrval>(<param list>)
4134 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
4138 <p>This instruction requires several arguments:</p>
4142 <p>The optional "tail" marker indicates whether the callee function accesses
4143 any allocas or varargs in the caller. If the "tail" marker is present, the
4144 function call is eligible for tail call optimization. Note that calls may
4145 be marked "tail" even if they do not occur before a <a
4146 href="#i_ret"><tt>ret</tt></a> instruction.
4149 <p>The optional "cconv" marker indicates which <a href="#callingconv">calling
4150 convention</a> the call should use. If none is specified, the call defaults
4151 to using C calling conventions.
4154 <p>'<tt>ty</tt>': the type of the call instruction itself which is also
4155 the type of the return value. Functions that return no value are marked
4156 <tt><a href="#t_void">void</a></tt>.</p>
4159 <p>'<tt>fnty</tt>': shall be the signature of the pointer to function
4160 value being invoked. The argument types must match the types implied by
4161 this signature. This type can be omitted if the function is not varargs
4162 and if the function type does not return a pointer to a function.</p>
4165 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
4166 be invoked. In most cases, this is a direct function invocation, but
4167 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
4168 to function value.</p>
4171 <p>'<tt>function args</tt>': argument list whose types match the
4172 function signature argument types. All arguments must be of
4173 <a href="#t_firstclass">first class</a> type. If the function signature
4174 indicates the function accepts a variable number of arguments, the extra
4175 arguments can be specified.</p>
4181 <p>The '<tt>call</tt>' instruction is used to cause control flow to
4182 transfer to a specified function, with its incoming arguments bound to
4183 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
4184 instruction in the called function, control flow continues with the
4185 instruction after the function call, and the return value of the
4186 function is bound to the result argument. If the callee returns multiple
4187 values then the return values of the function are only accessible through
4188 the '<tt><a href="#i_getresult">getresult</a></tt>' instruction.</p>
4193 %retval = call i32 @test(i32 %argc)
4194 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42) <i>; yields i32</i>
4195 %X = tail call i32 @foo() <i>; yields i32</i>
4196 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
4197 call void %foo(i8 97 signext)
4199 %struct.A = type { i32, i8 }
4200 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
4201 %gr = getresult %struct.A %r, 0 <i>; yields i32</i>
4202 %gr1 = getresult %struct.A %r, 1 <i>; yields i8</i>
4207 <!-- _______________________________________________________________________ -->
4208 <div class="doc_subsubsection">
4209 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
4212 <div class="doc_text">
4217 <resultval> = va_arg <va_list*> <arglist>, <argty>
4222 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
4223 the "variable argument" area of a function call. It is used to implement the
4224 <tt>va_arg</tt> macro in C.</p>
4228 <p>This instruction takes a <tt>va_list*</tt> value and the type of
4229 the argument. It returns a value of the specified argument type and
4230 increments the <tt>va_list</tt> to point to the next argument. The
4231 actual type of <tt>va_list</tt> is target specific.</p>
4235 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
4236 type from the specified <tt>va_list</tt> and causes the
4237 <tt>va_list</tt> to point to the next argument. For more information,
4238 see the variable argument handling <a href="#int_varargs">Intrinsic
4241 <p>It is legal for this instruction to be called in a function which does not
4242 take a variable number of arguments, for example, the <tt>vfprintf</tt>
4245 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
4246 href="#intrinsics">intrinsic function</a> because it takes a type as an
4251 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
4255 <!-- _______________________________________________________________________ -->
4256 <div class="doc_subsubsection">
4257 <a name="i_getresult">'<tt>getresult</tt>' Instruction</a>
4260 <div class="doc_text">
4264 <resultval> = getresult <type> <retval>, <index>
4269 <p> The '<tt>getresult</tt>' instruction is used to extract individual values
4270 from a '<tt><a href="#i_call">call</a></tt>'
4271 or '<tt><a href="#i_invoke">invoke</a></tt>' instruction that returns multiple
4276 <p>The '<tt>getresult</tt>' instruction takes a call or invoke value as its
4277 first argument, or an undef value. The value must have <a
4278 href="#t_struct">structure type</a>. The second argument is a constant
4279 unsigned index value which must be in range for the number of values returned
4284 <p>The '<tt>getresult</tt>' instruction extracts the element identified by
4285 '<tt>index</tt>' from the aggregate value.</p>
4290 %struct.A = type { i32, i8 }
4292 %r = call %struct.A @foo()
4293 %gr = getresult %struct.A %r, 0 <i>; yields i32:%gr</i>
4294 %gr1 = getresult %struct.A %r, 1 <i>; yields i8:%gr1</i>
4301 <!-- *********************************************************************** -->
4302 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
4303 <!-- *********************************************************************** -->
4305 <div class="doc_text">
4307 <p>LLVM supports the notion of an "intrinsic function". These functions have
4308 well known names and semantics and are required to follow certain restrictions.
4309 Overall, these intrinsics represent an extension mechanism for the LLVM
4310 language that does not require changing all of the transformations in LLVM when
4311 adding to the language (or the bitcode reader/writer, the parser, etc...).</p>
4313 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
4314 prefix is reserved in LLVM for intrinsic names; thus, function names may not
4315 begin with this prefix. Intrinsic functions must always be external functions:
4316 you cannot define the body of intrinsic functions. Intrinsic functions may
4317 only be used in call or invoke instructions: it is illegal to take the address
4318 of an intrinsic function. Additionally, because intrinsic functions are part
4319 of the LLVM language, it is required if any are added that they be documented
4322 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
4323 a family of functions that perform the same operation but on different data
4324 types. Because LLVM can represent over 8 million different integer types,
4325 overloading is used commonly to allow an intrinsic function to operate on any
4326 integer type. One or more of the argument types or the result type can be
4327 overloaded to accept any integer type. Argument types may also be defined as
4328 exactly matching a previous argument's type or the result type. This allows an
4329 intrinsic function which accepts multiple arguments, but needs all of them to
4330 be of the same type, to only be overloaded with respect to a single argument or
4333 <p>Overloaded intrinsics will have the names of its overloaded argument types
4334 encoded into its function name, each preceded by a period. Only those types
4335 which are overloaded result in a name suffix. Arguments whose type is matched
4336 against another type do not. For example, the <tt>llvm.ctpop</tt> function can
4337 take an integer of any width and returns an integer of exactly the same integer
4338 width. This leads to a family of functions such as
4339 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29 %val)</tt>.
4340 Only one type, the return type, is overloaded, and only one type suffix is
4341 required. Because the argument's type is matched against the return type, it
4342 does not require its own name suffix.</p>
4344 <p>To learn how to add an intrinsic function, please see the
4345 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.
4350 <!-- ======================================================================= -->
4351 <div class="doc_subsection">
4352 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
4355 <div class="doc_text">
4357 <p>Variable argument support is defined in LLVM with the <a
4358 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
4359 intrinsic functions. These functions are related to the similarly
4360 named macros defined in the <tt><stdarg.h></tt> header file.</p>
4362 <p>All of these functions operate on arguments that use a
4363 target-specific value type "<tt>va_list</tt>". The LLVM assembly
4364 language reference manual does not define what this type is, so all
4365 transformations should be prepared to handle these functions regardless of
4368 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
4369 instruction and the variable argument handling intrinsic functions are
4372 <div class="doc_code">
4374 define i32 @test(i32 %X, ...) {
4375 ; Initialize variable argument processing
4377 %ap2 = bitcast i8** %ap to i8*
4378 call void @llvm.va_start(i8* %ap2)
4380 ; Read a single integer argument
4381 %tmp = va_arg i8** %ap, i32
4383 ; Demonstrate usage of llvm.va_copy and llvm.va_end
4385 %aq2 = bitcast i8** %aq to i8*
4386 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
4387 call void @llvm.va_end(i8* %aq2)
4389 ; Stop processing of arguments.
4390 call void @llvm.va_end(i8* %ap2)
4394 declare void @llvm.va_start(i8*)
4395 declare void @llvm.va_copy(i8*, i8*)
4396 declare void @llvm.va_end(i8*)
4402 <!-- _______________________________________________________________________ -->
4403 <div class="doc_subsubsection">
4404 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
4408 <div class="doc_text">
4410 <pre> declare void %llvm.va_start(i8* <arglist>)<br></pre>
4412 <P>The '<tt>llvm.va_start</tt>' intrinsic initializes
4413 <tt>*<arglist></tt> for subsequent use by <tt><a
4414 href="#i_va_arg">va_arg</a></tt>.</p>
4418 <P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
4422 <P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
4423 macro available in C. In a target-dependent way, it initializes the
4424 <tt>va_list</tt> element to which the argument points, so that the next call to
4425 <tt>va_arg</tt> will produce the first variable argument passed to the function.
4426 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
4427 last argument of the function as the compiler can figure that out.</p>
4431 <!-- _______________________________________________________________________ -->
4432 <div class="doc_subsubsection">
4433 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
4436 <div class="doc_text">
4438 <pre> declare void @llvm.va_end(i8* <arglist>)<br></pre>
4441 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
4442 which has been initialized previously with <tt><a href="#int_va_start">llvm.va_start</a></tt>
4443 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
4447 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
4451 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
4452 macro available in C. In a target-dependent way, it destroys the
4453 <tt>va_list</tt> element to which the argument points. Calls to <a
4454 href="#int_va_start"><tt>llvm.va_start</tt></a> and <a href="#int_va_copy">
4455 <tt>llvm.va_copy</tt></a> must be matched exactly with calls to
4456 <tt>llvm.va_end</tt>.</p>
4460 <!-- _______________________________________________________________________ -->
4461 <div class="doc_subsubsection">
4462 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
4465 <div class="doc_text">
4470 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
4475 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
4476 from the source argument list to the destination argument list.</p>
4480 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
4481 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
4486 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
4487 macro available in C. In a target-dependent way, it copies the source
4488 <tt>va_list</tt> element into the destination <tt>va_list</tt> element. This
4489 intrinsic is necessary because the <tt><a href="#int_va_start">
4490 llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
4491 example, memory allocation.</p>
4495 <!-- ======================================================================= -->
4496 <div class="doc_subsection">
4497 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
4500 <div class="doc_text">
4503 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
4504 Collection</a> requires the implementation and generation of these intrinsics.
4505 These intrinsics allow identification of <a href="#int_gcroot">GC roots on the
4506 stack</a>, as well as garbage collector implementations that require <a
4507 href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a> barriers.
4508 Front-ends for type-safe garbage collected languages should generate these
4509 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
4510 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
4513 <p>The garbage collection intrinsics only operate on objects in the generic
4514 address space (address space zero).</p>
4518 <!-- _______________________________________________________________________ -->
4519 <div class="doc_subsubsection">
4520 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
4523 <div class="doc_text">
4528 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
4533 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
4534 the code generator, and allows some metadata to be associated with it.</p>
4538 <p>The first argument specifies the address of a stack object that contains the
4539 root pointer. The second pointer (which must be either a constant or a global
4540 value address) contains the meta-data to be associated with the root.</p>
4544 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
4545 location. At compile-time, the code generator generates information to allow
4546 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
4547 intrinsic may only be used in a function which <a href="#gc">specifies a GC
4553 <!-- _______________________________________________________________________ -->
4554 <div class="doc_subsubsection">
4555 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
4558 <div class="doc_text">
4563 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
4568 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
4569 locations, allowing garbage collector implementations that require read
4574 <p>The second argument is the address to read from, which should be an address
4575 allocated from the garbage collector. The first object is a pointer to the
4576 start of the referenced object, if needed by the language runtime (otherwise
4581 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
4582 instruction, but may be replaced with substantially more complex code by the
4583 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
4584 may only be used in a function which <a href="#gc">specifies a GC
4590 <!-- _______________________________________________________________________ -->
4591 <div class="doc_subsubsection">
4592 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
4595 <div class="doc_text">
4600 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
4605 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
4606 locations, allowing garbage collector implementations that require write
4607 barriers (such as generational or reference counting collectors).</p>
4611 <p>The first argument is the reference to store, the second is the start of the
4612 object to store it to, and the third is the address of the field of Obj to
4613 store to. If the runtime does not require a pointer to the object, Obj may be
4618 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
4619 instruction, but may be replaced with substantially more complex code by the
4620 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
4621 may only be used in a function which <a href="#gc">specifies a GC
4628 <!-- ======================================================================= -->
4629 <div class="doc_subsection">
4630 <a name="int_codegen">Code Generator Intrinsics</a>
4633 <div class="doc_text">
4635 These intrinsics are provided by LLVM to expose special features that may only
4636 be implemented with code generator support.
4641 <!-- _______________________________________________________________________ -->
4642 <div class="doc_subsubsection">
4643 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
4646 <div class="doc_text">
4650 declare i8 *@llvm.returnaddress(i32 <level>)
4656 The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
4657 target-specific value indicating the return address of the current function
4658 or one of its callers.
4664 The argument to this intrinsic indicates which function to return the address
4665 for. Zero indicates the calling function, one indicates its caller, etc. The
4666 argument is <b>required</b> to be a constant integer value.
4672 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
4673 the return address of the specified call frame, or zero if it cannot be
4674 identified. The value returned by this intrinsic is likely to be incorrect or 0
4675 for arguments other than zero, so it should only be used for debugging purposes.
4679 Note that calling this intrinsic does not prevent function inlining or other
4680 aggressive transformations, so the value returned may not be that of the obvious
4681 source-language caller.
4686 <!-- _______________________________________________________________________ -->
4687 <div class="doc_subsubsection">
4688 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
4691 <div class="doc_text">
4695 declare i8 *@llvm.frameaddress(i32 <level>)
4701 The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
4702 target-specific frame pointer value for the specified stack frame.
4708 The argument to this intrinsic indicates which function to return the frame
4709 pointer for. Zero indicates the calling function, one indicates its caller,
4710 etc. The argument is <b>required</b> to be a constant integer value.
4716 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
4717 the frame address of the specified call frame, or zero if it cannot be
4718 identified. The value returned by this intrinsic is likely to be incorrect or 0
4719 for arguments other than zero, so it should only be used for debugging purposes.
4723 Note that calling this intrinsic does not prevent function inlining or other
4724 aggressive transformations, so the value returned may not be that of the obvious
4725 source-language caller.
4729 <!-- _______________________________________________________________________ -->
4730 <div class="doc_subsubsection">
4731 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
4734 <div class="doc_text">
4738 declare i8 *@llvm.stacksave()
4744 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
4745 the function stack, for use with <a href="#int_stackrestore">
4746 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
4747 features like scoped automatic variable sized arrays in C99.
4753 This intrinsic returns a opaque pointer value that can be passed to <a
4754 href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
4755 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
4756 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
4757 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
4758 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
4759 that were allocated after the <tt>llvm.stacksave</tt> was executed.
4764 <!-- _______________________________________________________________________ -->
4765 <div class="doc_subsubsection">
4766 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
4769 <div class="doc_text">
4773 declare void @llvm.stackrestore(i8 * %ptr)
4779 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
4780 the function stack to the state it was in when the corresponding <a
4781 href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
4782 useful for implementing language features like scoped automatic variable sized
4789 See the description for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.
4795 <!-- _______________________________________________________________________ -->
4796 <div class="doc_subsubsection">
4797 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
4800 <div class="doc_text">
4804 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>)
4811 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
4812 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
4814 effect on the behavior of the program but can change its performance
4821 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
4822 determining if the fetch should be for a read (0) or write (1), and
4823 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
4824 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
4825 <tt>locality</tt> arguments must be constant integers.
4831 This intrinsic does not modify the behavior of the program. In particular,
4832 prefetches cannot trap and do not produce a value. On targets that support this
4833 intrinsic, the prefetch can provide hints to the processor cache for better
4839 <!-- _______________________________________________________________________ -->
4840 <div class="doc_subsubsection">
4841 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
4844 <div class="doc_text">
4848 declare void @llvm.pcmarker(i32 <id>)
4855 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
4857 code to simulators and other tools. The method is target specific, but it is
4858 expected that the marker will use exported symbols to transmit the PC of the marker.
4859 The marker makes no guarantees that it will remain with any specific instruction
4860 after optimizations. It is possible that the presence of a marker will inhibit
4861 optimizations. The intended use is to be inserted after optimizations to allow
4862 correlations of simulation runs.
4868 <tt>id</tt> is a numerical id identifying the marker.
4874 This intrinsic does not modify the behavior of the program. Backends that do not
4875 support this intrinisic may ignore it.
4880 <!-- _______________________________________________________________________ -->
4881 <div class="doc_subsubsection">
4882 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
4885 <div class="doc_text">
4889 declare i64 @llvm.readcyclecounter( )
4896 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
4897 counter register (or similar low latency, high accuracy clocks) on those targets
4898 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
4899 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
4900 should only be used for small timings.
4906 When directly supported, reading the cycle counter should not modify any memory.
4907 Implementations are allowed to either return a application specific value or a
4908 system wide value. On backends without support, this is lowered to a constant 0.
4913 <!-- ======================================================================= -->
4914 <div class="doc_subsection">
4915 <a name="int_libc">Standard C Library Intrinsics</a>
4918 <div class="doc_text">
4920 LLVM provides intrinsics for a few important standard C library functions.
4921 These intrinsics allow source-language front-ends to pass information about the
4922 alignment of the pointer arguments to the code generator, providing opportunity
4923 for more efficient code generation.
4928 <!-- _______________________________________________________________________ -->
4929 <div class="doc_subsubsection">
4930 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
4933 <div class="doc_text">
4937 declare void @llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
4938 i32 <len>, i32 <align>)
4939 declare void @llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
4940 i64 <len>, i32 <align>)
4946 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4947 location to the destination location.
4951 Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
4952 intrinsics do not return a value, and takes an extra alignment argument.
4958 The first argument is a pointer to the destination, the second is a pointer to
4959 the source. The third argument is an integer argument
4960 specifying the number of bytes to copy, and the fourth argument is the alignment
4961 of the source and destination locations.
4965 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4966 the caller guarantees that both the source and destination pointers are aligned
4973 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4974 location to the destination location, which are not allowed to overlap. It
4975 copies "len" bytes of memory over. If the argument is known to be aligned to
4976 some boundary, this can be specified as the fourth argument, otherwise it should
4982 <!-- _______________________________________________________________________ -->
4983 <div class="doc_subsubsection">
4984 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
4987 <div class="doc_text">
4991 declare void @llvm.memmove.i32(i8 * <dest>, i8 * <src>,
4992 i32 <len>, i32 <align>)
4993 declare void @llvm.memmove.i64(i8 * <dest>, i8 * <src>,
4994 i64 <len>, i32 <align>)
5000 The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
5001 location to the destination location. It is similar to the
5002 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to overlap.
5006 Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
5007 intrinsics do not return a value, and takes an extra alignment argument.
5013 The first argument is a pointer to the destination, the second is a pointer to
5014 the source. The third argument is an integer argument
5015 specifying the number of bytes to copy, and the fourth argument is the alignment
5016 of the source and destination locations.
5020 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5021 the caller guarantees that the source and destination pointers are aligned to
5028 The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
5029 location to the destination location, which may overlap. It
5030 copies "len" bytes of memory over. If the argument is known to be aligned to
5031 some boundary, this can be specified as the fourth argument, otherwise it should
5037 <!-- _______________________________________________________________________ -->
5038 <div class="doc_subsubsection">
5039 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
5042 <div class="doc_text">
5046 declare void @llvm.memset.i32(i8 * <dest>, i8 <val>,
5047 i32 <len>, i32 <align>)
5048 declare void @llvm.memset.i64(i8 * <dest>, i8 <val>,
5049 i64 <len>, i32 <align>)
5055 The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
5060 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
5061 does not return a value, and takes an extra alignment argument.
5067 The first argument is a pointer to the destination to fill, the second is the
5068 byte value to fill it with, the third argument is an integer
5069 argument specifying the number of bytes to fill, and the fourth argument is the
5070 known alignment of destination location.
5074 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5075 the caller guarantees that the destination pointer is aligned to that boundary.
5081 The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
5083 destination location. If the argument is known to be aligned to some boundary,
5084 this can be specified as the fourth argument, otherwise it should be set to 0 or
5090 <!-- _______________________________________________________________________ -->
5091 <div class="doc_subsubsection">
5092 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
5095 <div class="doc_text">
5098 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
5099 floating point or vector of floating point type. Not all targets support all
5102 declare float @llvm.sqrt.f32(float %Val)
5103 declare double @llvm.sqrt.f64(double %Val)
5104 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
5105 declare fp128 @llvm.sqrt.f128(fp128 %Val)
5106 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
5112 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
5113 returning the same value as the libm '<tt>sqrt</tt>' functions would. Unlike
5114 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
5115 negative numbers other than -0.0 (which allows for better optimization, because
5116 there is no need to worry about errno being set). <tt>llvm.sqrt(-0.0)</tt> is
5117 defined to return -0.0 like IEEE sqrt.
5123 The argument and return value are floating point numbers of the same type.
5129 This function returns the sqrt of the specified operand if it is a nonnegative
5130 floating point number.
5134 <!-- _______________________________________________________________________ -->
5135 <div class="doc_subsubsection">
5136 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
5139 <div class="doc_text">
5142 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
5143 floating point or vector of floating point type. Not all targets support all
5146 declare float @llvm.powi.f32(float %Val, i32 %power)
5147 declare double @llvm.powi.f64(double %Val, i32 %power)
5148 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
5149 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
5150 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
5156 The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
5157 specified (positive or negative) power. The order of evaluation of
5158 multiplications is not defined. When a vector of floating point type is
5159 used, the second argument remains a scalar integer value.
5165 The second argument is an integer power, and the first is a value to raise to
5172 This function returns the first value raised to the second power with an
5173 unspecified sequence of rounding operations.</p>
5176 <!-- _______________________________________________________________________ -->
5177 <div class="doc_subsubsection">
5178 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
5181 <div class="doc_text">
5184 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
5185 floating point or vector of floating point type. Not all targets support all
5188 declare float @llvm.sin.f32(float %Val)
5189 declare double @llvm.sin.f64(double %Val)
5190 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
5191 declare fp128 @llvm.sin.f128(fp128 %Val)
5192 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
5198 The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.
5204 The argument and return value are floating point numbers of the same type.
5210 This function returns the sine of the specified operand, returning the
5211 same values as the libm <tt>sin</tt> functions would, and handles error
5212 conditions in the same way.</p>
5215 <!-- _______________________________________________________________________ -->
5216 <div class="doc_subsubsection">
5217 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
5220 <div class="doc_text">
5223 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
5224 floating point or vector of floating point type. Not all targets support all
5227 declare float @llvm.cos.f32(float %Val)
5228 declare double @llvm.cos.f64(double %Val)
5229 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
5230 declare fp128 @llvm.cos.f128(fp128 %Val)
5231 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
5237 The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.
5243 The argument and return value are floating point numbers of the same type.
5249 This function returns the cosine of the specified operand, returning the
5250 same values as the libm <tt>cos</tt> functions would, and handles error
5251 conditions in the same way.</p>
5254 <!-- _______________________________________________________________________ -->
5255 <div class="doc_subsubsection">
5256 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
5259 <div class="doc_text">
5262 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
5263 floating point or vector of floating point type. Not all targets support all
5266 declare float @llvm.pow.f32(float %Val, float %Power)
5267 declare double @llvm.pow.f64(double %Val, double %Power)
5268 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
5269 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
5270 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
5276 The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
5277 specified (positive or negative) power.
5283 The second argument is a floating point power, and the first is a value to
5284 raise to that power.
5290 This function returns the first value raised to the second power,
5292 same values as the libm <tt>pow</tt> functions would, and handles error
5293 conditions in the same way.</p>
5297 <!-- ======================================================================= -->
5298 <div class="doc_subsection">
5299 <a name="int_manip">Bit Manipulation Intrinsics</a>
5302 <div class="doc_text">
5304 LLVM provides intrinsics for a few important bit manipulation operations.
5305 These allow efficient code generation for some algorithms.
5310 <!-- _______________________________________________________________________ -->
5311 <div class="doc_subsubsection">
5312 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
5315 <div class="doc_text">
5318 <p>This is an overloaded intrinsic function. You can use bswap on any integer
5319 type that is an even number of bytes (i.e. BitWidth % 16 == 0).
5321 declare i16 @llvm.bswap.i16(i16 <id>)
5322 declare i32 @llvm.bswap.i32(i32 <id>)
5323 declare i64 @llvm.bswap.i64(i64 <id>)
5329 The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
5330 values with an even number of bytes (positive multiple of 16 bits). These are
5331 useful for performing operations on data that is not in the target's native
5338 The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
5339 and low byte of the input i16 swapped. Similarly, the <tt>llvm.bswap.i32</tt>
5340 intrinsic returns an i32 value that has the four bytes of the input i32
5341 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the returned
5342 i32 will have its bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i48</tt>,
5343 <tt>llvm.bswap.i64</tt> and other intrinsics extend this concept to
5344 additional even-byte lengths (6 bytes, 8 bytes and more, respectively).
5349 <!-- _______________________________________________________________________ -->
5350 <div class="doc_subsubsection">
5351 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
5354 <div class="doc_text">
5357 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
5358 width. Not all targets support all bit widths however.
5360 declare i8 @llvm.ctpop.i8 (i8 <src>)
5361 declare i16 @llvm.ctpop.i16(i16 <src>)
5362 declare i32 @llvm.ctpop.i32(i32 <src>)
5363 declare i64 @llvm.ctpop.i64(i64 <src>)
5364 declare i256 @llvm.ctpop.i256(i256 <src>)
5370 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
5377 The only argument is the value to be counted. The argument may be of any
5378 integer type. The return type must match the argument type.
5384 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
5388 <!-- _______________________________________________________________________ -->
5389 <div class="doc_subsubsection">
5390 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
5393 <div class="doc_text">
5396 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
5397 integer bit width. Not all targets support all bit widths however.
5399 declare i8 @llvm.ctlz.i8 (i8 <src>)
5400 declare i16 @llvm.ctlz.i16(i16 <src>)
5401 declare i32 @llvm.ctlz.i32(i32 <src>)
5402 declare i64 @llvm.ctlz.i64(i64 <src>)
5403 declare i256 @llvm.ctlz.i256(i256 <src>)
5409 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
5410 leading zeros in a variable.
5416 The only argument is the value to be counted. The argument may be of any
5417 integer type. The return type must match the argument type.
5423 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
5424 in a variable. If the src == 0 then the result is the size in bits of the type
5425 of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.
5431 <!-- _______________________________________________________________________ -->
5432 <div class="doc_subsubsection">
5433 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
5436 <div class="doc_text">
5439 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
5440 integer bit width. Not all targets support all bit widths however.
5442 declare i8 @llvm.cttz.i8 (i8 <src>)
5443 declare i16 @llvm.cttz.i16(i16 <src>)
5444 declare i32 @llvm.cttz.i32(i32 <src>)
5445 declare i64 @llvm.cttz.i64(i64 <src>)
5446 declare i256 @llvm.cttz.i256(i256 <src>)
5452 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
5459 The only argument is the value to be counted. The argument may be of any
5460 integer type. The return type must match the argument type.
5466 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
5467 in a variable. If the src == 0 then the result is the size in bits of the type
5468 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
5472 <!-- _______________________________________________________________________ -->
5473 <div class="doc_subsubsection">
5474 <a name="int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic</a>
5477 <div class="doc_text">
5480 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.select</tt>
5481 on any integer bit width.
5483 declare i17 @llvm.part.select.i17 (i17 %val, i32 %loBit, i32 %hiBit)
5484 declare i29 @llvm.part.select.i29 (i29 %val, i32 %loBit, i32 %hiBit)
5488 <p>The '<tt>llvm.part.select</tt>' family of intrinsic functions selects a
5489 range of bits from an integer value and returns them in the same bit width as
5490 the original value.</p>
5493 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5494 any bit width but they must have the same bit width. The second and third
5495 arguments must be <tt>i32</tt> type since they specify only a bit index.</p>
5498 <p>The operation of the '<tt>llvm.part.select</tt>' intrinsic has two modes
5499 of operation: forwards and reverse. If <tt>%loBit</tt> is greater than
5500 <tt>%hiBits</tt> then the intrinsic operates in reverse mode. Otherwise it
5501 operates in forward mode.</p>
5502 <p>In forward mode, this intrinsic is the equivalent of shifting <tt>%val</tt>
5503 right by <tt>%loBit</tt> bits and then ANDing it with a mask with
5504 only the <tt>%hiBit - %loBit</tt> bits set, as follows:</p>
5506 <li>The <tt>%val</tt> is shifted right (LSHR) by the number of bits specified
5507 by <tt>%loBits</tt>. This normalizes the value to the low order bits.</li>
5508 <li>The <tt>%loBits</tt> value is subtracted from the <tt>%hiBits</tt> value
5509 to determine the number of bits to retain.</li>
5510 <li>A mask of the retained bits is created by shifting a -1 value.</li>
5511 <li>The mask is ANDed with <tt>%val</tt> to produce the result.
5513 <p>In reverse mode, a similar computation is made except that the bits are
5514 returned in the reverse order. So, for example, if <tt>X</tt> has the value
5515 <tt>i16 0x0ACF (101011001111)</tt> and we apply
5516 <tt>part.select(i16 X, 8, 3)</tt> to it, we get back the value
5517 <tt>i16 0x0026 (000000100110)</tt>.</p>
5520 <div class="doc_subsubsection">
5521 <a name="int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic</a>
5524 <div class="doc_text">
5527 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.set</tt>
5528 on any integer bit width.
5530 declare i17 @llvm.part.set.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
5531 declare i29 @llvm.part.set.i29.i9 (i29 %val, i9 %repl, i32 %lo, i32 %hi)
5535 <p>The '<tt>llvm.part.set</tt>' family of intrinsic functions replaces a range
5536 of bits in an integer value with another integer value. It returns the integer
5537 with the replaced bits.</p>
5540 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5541 any bit width but they must have the same bit width. <tt>%val</tt> is the value
5542 whose bits will be replaced. The second argument, <tt>%repl</tt> may be an
5543 integer of any bit width. The third and fourth arguments must be <tt>i32</tt>
5544 type since they specify only a bit index.</p>
5547 <p>The operation of the '<tt>llvm.part.set</tt>' intrinsic has two modes
5548 of operation: forwards and reverse. If <tt>%lo</tt> is greater than
5549 <tt>%hi</tt> then the intrinsic operates in reverse mode. Otherwise it
5550 operates in forward mode.</p>
5551 <p>For both modes, the <tt>%repl</tt> value is prepared for use by either
5552 truncating it down to the size of the replacement area or zero extending it
5553 up to that size.</p>
5554 <p>In forward mode, the bits between <tt>%lo</tt> and <tt>%hi</tt> (inclusive)
5555 are replaced with corresponding bits from <tt>%repl</tt>. That is the 0th bit
5556 in <tt>%repl</tt> replaces the <tt>%lo</tt>th bit in <tt>%val</tt> and etc. up
5557 to the <tt>%hi</tt>th bit.
5558 <p>In reverse mode, a similar computation is made except that the bits are
5559 reversed. That is, the <tt>0</tt>th bit in <tt>%repl</tt> replaces the
5560 <tt>%hi</tt> bit in <tt>%val</tt> and etc. down to the <tt>%lo</tt>th bit.
5563 llvm.part.set(0xFFFF, 0, 4, 7) -> 0xFF0F
5564 llvm.part.set(0xFFFF, 0, 7, 4) -> 0xFF0F
5565 llvm.part.set(0xFFFF, 1, 7, 4) -> 0xFF8F
5566 llvm.part.set(0xFFFF, F, 8, 3) -> 0xFFE7
5567 llvm.part.set(0xFFFF, 0, 3, 8) -> 0xFE07
5571 <!-- ======================================================================= -->
5572 <div class="doc_subsection">
5573 <a name="int_debugger">Debugger Intrinsics</a>
5576 <div class="doc_text">
5578 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
5579 are described in the <a
5580 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
5581 Debugging</a> document.
5586 <!-- ======================================================================= -->
5587 <div class="doc_subsection">
5588 <a name="int_eh">Exception Handling Intrinsics</a>
5591 <div class="doc_text">
5592 <p> The LLVM exception handling intrinsics (which all start with
5593 <tt>llvm.eh.</tt> prefix), are described in the <a
5594 href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
5595 Handling</a> document. </p>
5598 <!-- ======================================================================= -->
5599 <div class="doc_subsection">
5600 <a name="int_trampoline">Trampoline Intrinsic</a>
5603 <div class="doc_text">
5605 This intrinsic makes it possible to excise one parameter, marked with
5606 the <tt>nest</tt> attribute, from a function. The result is a callable
5607 function pointer lacking the nest parameter - the caller does not need
5608 to provide a value for it. Instead, the value to use is stored in
5609 advance in a "trampoline", a block of memory usually allocated
5610 on the stack, which also contains code to splice the nest value into the
5611 argument list. This is used to implement the GCC nested function address
5615 For example, if the function is
5616 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
5617 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as follows:</p>
5619 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
5620 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
5621 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
5622 %fp = bitcast i8* %p to i32 (i32, i32)*
5624 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
5625 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
5628 <!-- _______________________________________________________________________ -->
5629 <div class="doc_subsubsection">
5630 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
5632 <div class="doc_text">
5635 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
5639 This fills the memory pointed to by <tt>tramp</tt> with code
5640 and returns a function pointer suitable for executing it.
5644 The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
5645 pointers. The <tt>tramp</tt> argument must point to a sufficiently large
5646 and sufficiently aligned block of memory; this memory is written to by the
5647 intrinsic. Note that the size and the alignment are target-specific - LLVM
5648 currently provides no portable way of determining them, so a front-end that
5649 generates this intrinsic needs to have some target-specific knowledge.
5650 The <tt>func</tt> argument must hold a function bitcast to an <tt>i8*</tt>.
5654 The block of memory pointed to by <tt>tramp</tt> is filled with target
5655 dependent code, turning it into a function. A pointer to this function is
5656 returned, but needs to be bitcast to an
5657 <a href="#int_trampoline">appropriate function pointer type</a>
5658 before being called. The new function's signature is the same as that of
5659 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
5660 removed. At most one such <tt>nest</tt> argument is allowed, and it must be
5661 of pointer type. Calling the new function is equivalent to calling
5662 <tt>func</tt> with the same argument list, but with <tt>nval</tt> used for the
5663 missing <tt>nest</tt> argument. If, after calling
5664 <tt>llvm.init.trampoline</tt>, the memory pointed to by <tt>tramp</tt> is
5665 modified, then the effect of any later call to the returned function pointer is
5670 <!-- ======================================================================= -->
5671 <div class="doc_subsection">
5672 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
5675 <div class="doc_text">
5677 These intrinsic functions expand the "universal IR" of LLVM to represent
5678 hardware constructs for atomic operations and memory synchronization. This
5679 provides an interface to the hardware, not an interface to the programmer. It
5680 is aimed at a low enough level to allow any programming models or APIs which
5681 need atomic behaviors to map cleanly onto it. It is also modeled primarily on
5682 hardware behavior. Just as hardware provides a "universal IR" for source
5683 languages, it also provides a starting point for developing a "universal"
5684 atomic operation and synchronization IR.
5687 These do <em>not</em> form an API such as high-level threading libraries,
5688 software transaction memory systems, atomic primitives, and intrinsic
5689 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
5690 application libraries. The hardware interface provided by LLVM should allow
5691 a clean implementation of all of these APIs and parallel programming models.
5692 No one model or paradigm should be selected above others unless the hardware
5693 itself ubiquitously does so.
5698 <!-- _______________________________________________________________________ -->
5699 <div class="doc_subsubsection">
5700 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
5702 <div class="doc_text">
5705 declare void @llvm.memory.barrier( i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>,
5711 The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
5712 specific pairs of memory access types.
5716 The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
5717 The first four arguments enables a specific barrier as listed below. The fith
5718 argument specifies that the barrier applies to io or device or uncached memory.
5722 <li><tt>ll</tt>: load-load barrier</li>
5723 <li><tt>ls</tt>: load-store barrier</li>
5724 <li><tt>sl</tt>: store-load barrier</li>
5725 <li><tt>ss</tt>: store-store barrier</li>
5726 <li><tt>device</tt>: barrier applies to device and uncached memory also.
5730 This intrinsic causes the system to enforce some ordering constraints upon
5731 the loads and stores of the program. This barrier does not indicate
5732 <em>when</em> any events will occur, it only enforces an <em>order</em> in
5733 which they occur. For any of the specified pairs of load and store operations
5734 (f.ex. load-load, or store-load), all of the first operations preceding the
5735 barrier will complete before any of the second operations succeeding the
5736 barrier begin. Specifically the semantics for each pairing is as follows:
5739 <li><tt>ll</tt>: All loads before the barrier must complete before any load
5740 after the barrier begins.</li>
5742 <li><tt>ls</tt>: All loads before the barrier must complete before any
5743 store after the barrier begins.</li>
5744 <li><tt>ss</tt>: All stores before the barrier must complete before any
5745 store after the barrier begins.</li>
5746 <li><tt>sl</tt>: All stores before the barrier must complete before any
5747 load after the barrier begins.</li>
5750 These semantics are applied with a logical "and" behavior when more than one
5751 is enabled in a single memory barrier intrinsic.
5754 Backends may implement stronger barriers than those requested when they do not
5755 support as fine grained a barrier as requested. Some architectures do not
5756 need all types of barriers and on such architectures, these become noops.
5763 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
5764 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
5765 <i>; guarantee the above finishes</i>
5766 store i32 8, %ptr <i>; before this begins</i>
5770 <!-- _______________________________________________________________________ -->
5771 <div class="doc_subsubsection">
5772 <a name="int_atomic_lcs">'<tt>llvm.atomic.lcs.*</tt>' Intrinsic</a>
5774 <div class="doc_text">
5777 This is an overloaded intrinsic. You can use <tt>llvm.atomic.lcs</tt> on any
5778 integer bit width. Not all targets support all bit widths however.</p>
5781 declare i8 @llvm.atomic.lcs.i8( i8* <ptr>, i8 <cmp>, i8 <val> )
5782 declare i16 @llvm.atomic.lcs.i16( i16* <ptr>, i16 <cmp>, i16 <val> )
5783 declare i32 @llvm.atomic.lcs.i32( i32* <ptr>, i32 <cmp>, i32 <val> )
5784 declare i64 @llvm.atomic.lcs.i64( i64* <ptr>, i64 <cmp>, i64 <val> )
5789 This loads a value in memory and compares it to a given value. If they are
5790 equal, it stores a new value into the memory.
5794 The <tt>llvm.atomic.lcs</tt> intrinsic takes three arguments. The result as
5795 well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
5796 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
5797 this integer type. While any bit width integer may be used, targets may only
5798 lower representations they support in hardware.
5803 This entire intrinsic must be executed atomically. It first loads the value
5804 in memory pointed to by <tt>ptr</tt> and compares it with the value
5805 <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the memory. The
5806 loaded value is yielded in all cases. This provides the equivalent of an
5807 atomic compare-and-swap operation within the SSA framework.
5815 %val1 = add i32 4, 4
5816 %result1 = call i32 @llvm.atomic.lcs.i32( i32* %ptr, i32 4, %val1 )
5817 <i>; yields {i32}:result1 = 4</i>
5818 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
5819 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
5821 %val2 = add i32 1, 1
5822 %result2 = call i32 @llvm.atomic.lcs.i32( i32* %ptr, i32 5, %val2 )
5823 <i>; yields {i32}:result2 = 8</i>
5824 %stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
5826 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
5830 <!-- _______________________________________________________________________ -->
5831 <div class="doc_subsubsection">
5832 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
5834 <div class="doc_text">
5838 This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
5839 integer bit width. Not all targets support all bit widths however.</p>
5841 declare i8 @llvm.atomic.swap.i8( i8* <ptr>, i8 <val> )
5842 declare i16 @llvm.atomic.swap.i16( i16* <ptr>, i16 <val> )
5843 declare i32 @llvm.atomic.swap.i32( i32* <ptr>, i32 <val> )
5844 declare i64 @llvm.atomic.swap.i64( i64* <ptr>, i64 <val> )
5849 This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
5850 the value from memory. It then stores the value in <tt>val</tt> in the memory
5856 The <tt>llvm.atomic.ls</tt> intrinsic takes two arguments. Both the
5857 <tt>val</tt> argument and the result must be integers of the same bit width.
5858 The first argument, <tt>ptr</tt>, must be a pointer to a value of this
5859 integer type. The targets may only lower integer representations they
5864 This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
5865 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
5866 equivalent of an atomic swap operation within the SSA framework.
5874 %val1 = add i32 4, 4
5875 %result1 = call i32 @llvm.atomic.swap.i32( i32* %ptr, i32 %val1 )
5876 <i>; yields {i32}:result1 = 4</i>
5877 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
5878 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
5880 %val2 = add i32 1, 1
5881 %result2 = call i32 @llvm.atomic.swap.i32( i32* %ptr, i32 %val2 )
5882 <i>; yields {i32}:result2 = 8</i>
5884 %stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
5885 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
5889 <!-- _______________________________________________________________________ -->
5890 <div class="doc_subsubsection">
5891 <a name="int_atomic_las">'<tt>llvm.atomic.las.*</tt>' Intrinsic</a>
5894 <div class="doc_text">
5897 This is an overloaded intrinsic. You can use <tt>llvm.atomic.las</tt> on any
5898 integer bit width. Not all targets support all bit widths however.</p>
5900 declare i8 @llvm.atomic.las.i8.( i8* <ptr>, i8 <delta> )
5901 declare i16 @llvm.atomic.las.i16.( i16* <ptr>, i16 <delta> )
5902 declare i32 @llvm.atomic.las.i32.( i32* <ptr>, i32 <delta> )
5903 declare i64 @llvm.atomic.las.i64.( i64* <ptr>, i64 <delta> )
5908 This intrinsic adds <tt>delta</tt> to the value stored in memory at
5909 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
5914 The intrinsic takes two arguments, the first a pointer to an integer value
5915 and the second an integer value. The result is also an integer value. These
5916 integer types can have any bit width, but they must all have the same bit
5917 width. The targets may only lower integer representations they support.
5921 This intrinsic does a series of operations atomically. It first loads the
5922 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
5923 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
5930 %result1 = call i32 @llvm.atomic.las.i32( i32* %ptr, i32 4 )
5931 <i>; yields {i32}:result1 = 4</i>
5932 %result2 = call i32 @llvm.atomic.las.i32( i32* %ptr, i32 2 )
5933 <i>; yields {i32}:result2 = 8</i>
5934 %result3 = call i32 @llvm.atomic.las.i32( i32* %ptr, i32 5 )
5935 <i>; yields {i32}:result3 = 10</i>
5936 %memval = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
5941 <!-- ======================================================================= -->
5942 <div class="doc_subsection">
5943 <a name="int_general">General Intrinsics</a>
5946 <div class="doc_text">
5947 <p> This class of intrinsics is designed to be generic and has
5948 no specific purpose. </p>
5951 <!-- _______________________________________________________________________ -->
5952 <div class="doc_subsubsection">
5953 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
5956 <div class="doc_text">
5960 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
5966 The '<tt>llvm.var.annotation</tt>' intrinsic
5972 The first argument is a pointer to a value, the second is a pointer to a
5973 global string, the third is a pointer to a global string which is the source
5974 file name, and the last argument is the line number.
5980 This intrinsic allows annotation of local variables with arbitrary strings.
5981 This can be useful for special purpose optimizations that want to look for these
5982 annotations. These have no other defined use, they are ignored by code
5983 generation and optimization.
5987 <!-- _______________________________________________________________________ -->
5988 <div class="doc_subsubsection">
5989 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
5992 <div class="doc_text">
5995 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
5996 any integer bit width.
5999 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
6000 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
6001 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
6002 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
6003 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
6009 The '<tt>llvm.annotation</tt>' intrinsic.
6015 The first argument is an integer value (result of some expression),
6016 the second is a pointer to a global string, the third is a pointer to a global
6017 string which is the source file name, and the last argument is the line number.
6018 It returns the value of the first argument.
6024 This intrinsic allows annotations to be put on arbitrary expressions
6025 with arbitrary strings. This can be useful for special purpose optimizations
6026 that want to look for these annotations. These have no other defined use, they
6027 are ignored by code generation and optimization.
6030 <!-- _______________________________________________________________________ -->
6031 <div class="doc_subsubsection">
6032 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
6035 <div class="doc_text">
6039 declare void @llvm.trap()
6045 The '<tt>llvm.trap</tt>' intrinsic
6057 This intrinsics is lowered to the target dependent trap instruction. If the
6058 target does not have a trap instruction, this intrinsic will be lowered to the
6059 call of the abort() function.
6063 <!-- *********************************************************************** -->
6066 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
6067 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
6068 <a href="http://validator.w3.org/check/referer"><img
6069 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
6071 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
6072 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
6073 Last modified: $Date$