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="#memoryops">Memory Access and Addressing Operations</a>
116 <li><a href="#i_malloc">'<tt>malloc</tt>' Instruction</a></li>
117 <li><a href="#i_free">'<tt>free</tt>' Instruction</a></li>
118 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
119 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
120 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
121 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
124 <li><a href="#convertops">Conversion Operations</a>
126 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
127 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
128 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
129 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
130 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
131 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
132 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
133 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
134 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
135 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
136 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
137 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
139 <li><a href="#otherops">Other Operations</a>
141 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
142 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
143 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
144 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
145 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
146 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
151 <li><a href="#intrinsics">Intrinsic Functions</a>
153 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
155 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
156 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
157 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
160 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
162 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
163 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
164 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
167 <li><a href="#int_codegen">Code Generator Intrinsics</a>
169 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
170 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
171 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
172 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
173 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
174 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
175 <li><a href="#int_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
178 <li><a href="#int_libc">Standard C Library Intrinsics</a>
180 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
181 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
182 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
183 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
184 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
185 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
186 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
187 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
190 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
192 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
193 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
194 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
195 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
196 <li><a href="#int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic </a></li>
197 <li><a href="#int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic </a></li>
200 <li><a href="#int_debugger">Debugger intrinsics</a></li>
201 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
202 <li><a href="#int_trampoline">Trampoline Intrinsic</a>
204 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
207 <li><a href="#int_atomics">Atomic intrinsics</a>
209 <li><a href="#int_memory_barrier"><tt>llvm.memory_barrier</tt></a></li>
210 <li><a href="#int_atomic_lcs"><tt>llvm.atomic.lcs</tt></a></li>
211 <li><a href="#int_atomic_las"><tt>llvm.atomic.las</tt></a></li>
212 <li><a href="#int_atomic_swap"><tt>llvm.atomic.swap</tt></a></li>
215 <li><a href="#int_general">General intrinsics</a>
217 <li><a href="#int_var_annotation">
218 <tt>llvm.var.annotation</tt>' Intrinsic</a></li>
219 <li><a href="#int_annotation">
220 <tt>llvm.annotation.*</tt>' Intrinsic</a></li>
221 <li><a href="#int_trap">
222 <tt>llvm.trap</tt>' Intrinsic</a></li>
229 <div class="doc_author">
230 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
231 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
234 <!-- *********************************************************************** -->
235 <div class="doc_section"> <a name="abstract">Abstract </a></div>
236 <!-- *********************************************************************** -->
238 <div class="doc_text">
239 <p>This document is a reference manual for the LLVM assembly language.
240 LLVM is an SSA based representation that provides type safety,
241 low-level operations, flexibility, and the capability of representing
242 'all' high-level languages cleanly. It is the common code
243 representation used throughout all phases of the LLVM compilation
247 <!-- *********************************************************************** -->
248 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
249 <!-- *********************************************************************** -->
251 <div class="doc_text">
253 <p>The LLVM code representation is designed to be used in three
254 different forms: as an in-memory compiler IR, as an on-disk bitcode
255 representation (suitable for fast loading by a Just-In-Time compiler),
256 and as a human readable assembly language representation. This allows
257 LLVM to provide a powerful intermediate representation for efficient
258 compiler transformations and analysis, while providing a natural means
259 to debug and visualize the transformations. The three different forms
260 of LLVM are all equivalent. This document describes the human readable
261 representation and notation.</p>
263 <p>The LLVM representation aims to be light-weight and low-level
264 while being expressive, typed, and extensible at the same time. It
265 aims to be a "universal IR" of sorts, by being at a low enough level
266 that high-level ideas may be cleanly mapped to it (similar to how
267 microprocessors are "universal IR's", allowing many source languages to
268 be mapped to them). By providing type information, LLVM can be used as
269 the target of optimizations: for example, through pointer analysis, it
270 can be proven that a C automatic variable is never accessed outside of
271 the current function... allowing it to be promoted to a simple SSA
272 value instead of a memory location.</p>
276 <!-- _______________________________________________________________________ -->
277 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
279 <div class="doc_text">
281 <p>It is important to note that this document describes 'well formed'
282 LLVM assembly language. There is a difference between what the parser
283 accepts and what is considered 'well formed'. For example, the
284 following instruction is syntactically okay, but not well formed:</p>
286 <div class="doc_code">
288 %x = <a href="#i_add">add</a> i32 1, %x
292 <p>...because the definition of <tt>%x</tt> does not dominate all of
293 its uses. The LLVM infrastructure provides a verification pass that may
294 be used to verify that an LLVM module is well formed. This pass is
295 automatically run by the parser after parsing input assembly and by
296 the optimizer before it outputs bitcode. The violations pointed out
297 by the verifier pass indicate bugs in transformation passes or input to
301 <!-- Describe the typesetting conventions here. -->
303 <!-- *********************************************************************** -->
304 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
305 <!-- *********************************************************************** -->
307 <div class="doc_text">
309 <p>LLVM identifiers come in two basic types: global and local. Global
310 identifiers (functions, global variables) begin with the @ character. Local
311 identifiers (register names, types) begin with the % character. Additionally,
312 there are three different formats for identifiers, for different purposes:
315 <li>Named values are represented as a string of characters with their prefix.
316 For example, %foo, @DivisionByZero, %a.really.long.identifier. The actual
317 regular expression used is '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
318 Identifiers which require other characters in their names can be surrounded
319 with quotes. In this way, anything except a <tt>"</tt> character can
320 be used in a named value.</li>
322 <li>Unnamed values are represented as an unsigned numeric value with their
323 prefix. For example, %12, @2, %44.</li>
325 <li>Constants, which are described in a <a href="#constants">section about
326 constants</a>, below.</li>
329 <p>LLVM requires that values start with a prefix for two reasons: Compilers
330 don't need to worry about name clashes with reserved words, and the set of
331 reserved words may be expanded in the future without penalty. Additionally,
332 unnamed identifiers allow a compiler to quickly come up with a temporary
333 variable without having to avoid symbol table conflicts.</p>
335 <p>Reserved words in LLVM are very similar to reserved words in other
336 languages. There are keywords for different opcodes
337 ('<tt><a href="#i_add">add</a></tt>',
338 '<tt><a href="#i_bitcast">bitcast</a></tt>',
339 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
340 href="#t_void">void</a></tt>', '<tt><a href="#t_primitive">i32</a></tt>', etc...),
341 and others. These reserved words cannot conflict with variable names, because
342 none of them start with a prefix character ('%' or '@').</p>
344 <p>Here is an example of LLVM code to multiply the integer variable
345 '<tt>%X</tt>' by 8:</p>
349 <div class="doc_code">
351 %result = <a href="#i_mul">mul</a> i32 %X, 8
355 <p>After strength reduction:</p>
357 <div class="doc_code">
359 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
363 <p>And the hard way:</p>
365 <div class="doc_code">
367 <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
368 <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
369 %result = <a href="#i_add">add</a> i32 %1, %1
373 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
374 important lexical features of LLVM:</p>
378 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
381 <li>Unnamed temporaries are created when the result of a computation is not
382 assigned to a named value.</li>
384 <li>Unnamed temporaries are numbered sequentially</li>
388 <p>...and it also shows a convention that we follow in this document. When
389 demonstrating instructions, we will follow an instruction with a comment that
390 defines the type and name of value produced. Comments are shown in italic
395 <!-- *********************************************************************** -->
396 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
397 <!-- *********************************************************************** -->
399 <!-- ======================================================================= -->
400 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
403 <div class="doc_text">
405 <p>LLVM programs are composed of "Module"s, each of which is a
406 translation unit of the input programs. Each module consists of
407 functions, global variables, and symbol table entries. Modules may be
408 combined together with the LLVM linker, which merges function (and
409 global variable) definitions, resolves forward declarations, and merges
410 symbol table entries. Here is an example of the "hello world" module:</p>
412 <div class="doc_code">
413 <pre><i>; Declare the string constant as a global constant...</i>
414 <a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a
415 href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
417 <i>; External declaration of the puts function</i>
418 <a href="#functionstructure">declare</a> i32 @puts(i8 *) <i>; i32(i8 *)* </i>
420 <i>; Definition of main function</i>
421 define i32 @main() { <i>; i32()* </i>
422 <i>; Convert [13x i8 ]* to i8 *...</i>
424 href="#i_getelementptr">getelementptr</a> [13 x i8 ]* @.LC0, i64 0, i64 0 <i>; i8 *</i>
426 <i>; Call puts function to write out the string to stdout...</i>
428 href="#i_call">call</a> i32 @puts(i8 * %cast210) <i>; i32</i>
430 href="#i_ret">ret</a> i32 0<br>}<br>
434 <p>This example is made up of a <a href="#globalvars">global variable</a>
435 named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
436 function, and a <a href="#functionstructure">function definition</a>
437 for "<tt>main</tt>".</p>
439 <p>In general, a module is made up of a list of global values,
440 where both functions and global variables are global values. Global values are
441 represented by a pointer to a memory location (in this case, a pointer to an
442 array of char, and a pointer to a function), and have one of the following <a
443 href="#linkage">linkage types</a>.</p>
447 <!-- ======================================================================= -->
448 <div class="doc_subsection">
449 <a name="linkage">Linkage Types</a>
452 <div class="doc_text">
455 All Global Variables and Functions have one of the following types of linkage:
460 <dt><tt><b><a name="linkage_internal">internal</a></b></tt> </dt>
462 <dd>Global values with internal linkage are only directly accessible by
463 objects in the current module. In particular, linking code into a module with
464 an internal global value may cause the internal to be renamed as necessary to
465 avoid collisions. Because the symbol is internal to the module, all
466 references can be updated. This corresponds to the notion of the
467 '<tt>static</tt>' keyword in C.
470 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
472 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
473 the same name when linkage occurs. This is typically used to implement
474 inline functions, templates, or other code which must be generated in each
475 translation unit that uses it. Unreferenced <tt>linkonce</tt> globals are
476 allowed to be discarded.
479 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
481 <dd>"<tt>weak</tt>" linkage is exactly the same as <tt>linkonce</tt> linkage,
482 except that unreferenced <tt>weak</tt> globals may not be discarded. This is
483 used for globals that may be emitted in multiple translation units, but that
484 are not guaranteed to be emitted into every translation unit that uses them.
485 One example of this are common globals in C, such as "<tt>int X;</tt>" at
489 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
491 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
492 pointer to array type. When two global variables with appending linkage are
493 linked together, the two global arrays are appended together. This is the
494 LLVM, typesafe, equivalent of having the system linker append together
495 "sections" with identical names when .o files are linked.
498 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt>: </dt>
499 <dd>The semantics of this linkage follow the ELF model: the symbol is weak
500 until linked, if not linked, the symbol becomes null instead of being an
504 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
506 <dd>If none of the above identifiers are used, the global is externally
507 visible, meaning that it participates in linkage and can be used to resolve
508 external symbol references.
513 The next two types of linkage are targeted for Microsoft Windows platform
514 only. They are designed to support importing (exporting) symbols from (to)
519 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt>: </dt>
521 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
522 or variable via a global pointer to a pointer that is set up by the DLL
523 exporting the symbol. On Microsoft Windows targets, the pointer name is
524 formed by combining <code>_imp__</code> and the function or variable name.
527 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt>: </dt>
529 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
530 pointer to a pointer in a DLL, so that it can be referenced with the
531 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
532 name is formed by combining <code>_imp__</code> and the function or variable
538 <p><a name="linkage_external"></a>For example, since the "<tt>.LC0</tt>"
539 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
540 variable and was linked with this one, one of the two would be renamed,
541 preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
542 external (i.e., lacking any linkage declarations), they are accessible
543 outside of the current module.</p>
544 <p>It is illegal for a function <i>declaration</i>
545 to have any linkage type other than "externally visible", <tt>dllimport</tt>,
546 or <tt>extern_weak</tt>.</p>
547 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt> and <tt>weak</tt>
551 <!-- ======================================================================= -->
552 <div class="doc_subsection">
553 <a name="callingconv">Calling Conventions</a>
556 <div class="doc_text">
558 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
559 and <a href="#i_invoke">invokes</a> can all have an optional calling convention
560 specified for the call. The calling convention of any pair of dynamic
561 caller/callee must match, or the behavior of the program is undefined. The
562 following calling conventions are supported by LLVM, and more may be added in
566 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
568 <dd>This calling convention (the default if no other calling convention is
569 specified) matches the target C calling conventions. This calling convention
570 supports varargs function calls and tolerates some mismatch in the declared
571 prototype and implemented declaration of the function (as does normal C).
574 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
576 <dd>This calling convention attempts to make calls as fast as possible
577 (e.g. by passing things in registers). This calling convention allows the
578 target to use whatever tricks it wants to produce fast code for the target,
579 without having to conform to an externally specified ABI. Implementations of
580 this convention should allow arbitrary tail call optimization to be supported.
581 This calling convention does not support varargs and requires the prototype of
582 all callees to exactly match the prototype of the function definition.
585 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
587 <dd>This calling convention attempts to make code in the caller as efficient
588 as possible under the assumption that the call is not commonly executed. As
589 such, these calls often preserve all registers so that the call does not break
590 any live ranges in the caller side. This calling convention does not support
591 varargs and requires the prototype of all callees to exactly match the
592 prototype of the function definition.
595 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
597 <dd>Any calling convention may be specified by number, allowing
598 target-specific calling conventions to be used. Target specific calling
599 conventions start at 64.
603 <p>More calling conventions can be added/defined on an as-needed basis, to
604 support pascal conventions or any other well-known target-independent
609 <!-- ======================================================================= -->
610 <div class="doc_subsection">
611 <a name="visibility">Visibility Styles</a>
614 <div class="doc_text">
617 All Global Variables and Functions have one of the following visibility styles:
621 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
623 <dd>On ELF, default visibility means that the declaration is visible to other
624 modules and, in shared libraries, means that the declared entity may be
625 overridden. On Darwin, default visibility means that the declaration is
626 visible to other modules. Default visibility corresponds to "external
627 linkage" in the language.
630 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
632 <dd>Two declarations of an object with hidden visibility refer to the same
633 object if they are in the same shared object. Usually, hidden visibility
634 indicates that the symbol will not be placed into the dynamic symbol table,
635 so no other module (executable or shared library) can reference it
639 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
641 <dd>On ELF, protected visibility indicates that the symbol will be placed in
642 the dynamic symbol table, but that references within the defining module will
643 bind to the local symbol. That is, the symbol cannot be overridden by another
650 <!-- ======================================================================= -->
651 <div class="doc_subsection">
652 <a name="globalvars">Global Variables</a>
655 <div class="doc_text">
657 <p>Global variables define regions of memory allocated at compilation time
658 instead of run-time. Global variables may optionally be initialized, may have
659 an explicit section to be placed in, and may have an optional explicit alignment
660 specified. A variable may be defined as "thread_local", which means that it
661 will not be shared by threads (each thread will have a separated copy of the
662 variable). A variable may be defined as a global "constant," which indicates
663 that the contents of the variable will <b>never</b> be modified (enabling better
664 optimization, allowing the global data to be placed in the read-only section of
665 an executable, etc). Note that variables that need runtime initialization
666 cannot be marked "constant" as there is a store to the variable.</p>
669 LLVM explicitly allows <em>declarations</em> of global variables to be marked
670 constant, even if the final definition of the global is not. This capability
671 can be used to enable slightly better optimization of the program, but requires
672 the language definition to guarantee that optimizations based on the
673 'constantness' are valid for the translation units that do not include the
677 <p>As SSA values, global variables define pointer values that are in
678 scope (i.e. they dominate) all basic blocks in the program. Global
679 variables always define a pointer to their "content" type because they
680 describe a region of memory, and all memory objects in LLVM are
681 accessed through pointers.</p>
683 <p>A global variable may be declared to reside in a target-specifc numbered
684 address space. For targets that support them, address spaces may affect how
685 optimizations are performed and/or what target instructions are used to access
686 the variable. The default address space is zero. The address space qualifier
687 must precede any other attributes.</p>
689 <p>LLVM allows an explicit section to be specified for globals. If the target
690 supports it, it will emit globals to the section specified.</p>
692 <p>An explicit alignment may be specified for a global. If not present, or if
693 the alignment is set to zero, the alignment of the global is set by the target
694 to whatever it feels convenient. If an explicit alignment is specified, the
695 global is forced to have at least that much alignment. All alignments must be
698 <p>For example, the following defines a global in a numbered address space with
699 an initializer, section, and alignment:</p>
701 <div class="doc_code">
703 @G = constant float 1.0 addrspace(5), section "foo", align 4
710 <!-- ======================================================================= -->
711 <div class="doc_subsection">
712 <a name="functionstructure">Functions</a>
715 <div class="doc_text">
717 <p>LLVM function definitions consist of the "<tt>define</tt>" keyord,
718 an optional <a href="#linkage">linkage type</a>, an optional
719 <a href="#visibility">visibility style</a>, an optional
720 <a href="#callingconv">calling convention</a>, a return type, an optional
721 <a href="#paramattrs">parameter attribute</a> for the return type, a function
722 name, a (possibly empty) argument list (each with optional
723 <a href="#paramattrs">parameter attributes</a>), an optional section, an
724 optional alignment, an optional <a href="#gc">garbage collector name</a>, an
725 opening curly brace, a list of basic blocks, and a closing curly brace.
727 LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
728 optional <a href="#linkage">linkage type</a>, an optional
729 <a href="#visibility">visibility style</a>, an optional
730 <a href="#callingconv">calling convention</a>, a return type, an optional
731 <a href="#paramattrs">parameter attribute</a> for the return type, a function
732 name, a possibly empty list of arguments, an optional alignment, and an optional
733 <a href="#gc">garbage collector name</a>.</p>
735 <p>A function definition contains a list of basic blocks, forming the CFG for
736 the function. Each basic block may optionally start with a label (giving the
737 basic block a symbol table entry), contains a list of instructions, and ends
738 with a <a href="#terminators">terminator</a> instruction (such as a branch or
739 function return).</p>
741 <p>The first basic block in a function is special in two ways: it is immediately
742 executed on entrance to the function, and it is not allowed to have predecessor
743 basic blocks (i.e. there can not be any branches to the entry block of a
744 function). Because the block can have no predecessors, it also cannot have any
745 <a href="#i_phi">PHI nodes</a>.</p>
747 <p>LLVM allows an explicit section to be specified for functions. If the target
748 supports it, it will emit functions to the section specified.</p>
750 <p>An explicit alignment may be specified for a function. If not present, or if
751 the alignment is set to zero, the alignment of the function is set by the target
752 to whatever it feels convenient. If an explicit alignment is specified, the
753 function is forced to have at least that much alignment. All alignments must be
759 <!-- ======================================================================= -->
760 <div class="doc_subsection">
761 <a name="aliasstructure">Aliases</a>
763 <div class="doc_text">
764 <p>Aliases act as "second name" for the aliasee value (which can be either
765 function or global variable or bitcast of global value). Aliases may have an
766 optional <a href="#linkage">linkage type</a>, and an
767 optional <a href="#visibility">visibility style</a>.</p>
771 <div class="doc_code">
773 @<Name> = [Linkage] [Visibility] alias <AliaseeTy> @<Aliasee>
781 <!-- ======================================================================= -->
782 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
783 <div class="doc_text">
784 <p>The return type and each parameter of a function type may have a set of
785 <i>parameter attributes</i> associated with them. Parameter attributes are
786 used to communicate additional information about the result or parameters of
787 a function. Parameter attributes are considered to be part of the function,
788 not of the function type, so functions with different parameter attributes
789 can have the same function type.</p>
791 <p>Parameter attributes are simple keywords that follow the type specified. If
792 multiple parameter attributes are needed, they are space separated. For
795 <div class="doc_code">
797 declare i32 @printf(i8* noalias , ...) nounwind
798 declare i32 @atoi(i8*) nounwind readonly
802 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
803 <tt>readonly</tt>) come immediately after the argument list.</p>
805 <p>Currently, only the following parameter attributes are defined:</p>
807 <dt><tt>zeroext</tt></dt>
808 <dd>This indicates that the parameter should be zero extended just before
809 a call to this function.</dd>
811 <dt><tt>signext</tt></dt>
812 <dd>This indicates that the parameter should be sign extended just before
813 a call to this function.</dd>
815 <dt><tt>inreg</tt></dt>
816 <dd>This indicates that the parameter should be placed in register (if
817 possible) during assembling function call. Support for this attribute is
820 <dt><tt>byval</tt></dt>
821 <dd>This indicates that the pointer parameter should really be passed by
822 value to the function. The attribute implies that a hidden copy of the
823 pointee is made between the caller and the callee, so the callee is unable
824 to modify the value in the callee. This attribute is only valid on llvm
825 pointer arguments. It is generally used to pass structs and arrays by
826 value, but is also valid on scalars (even though this is silly).</dd>
828 <dt><tt>sret</tt></dt>
829 <dd>This indicates that the pointer parameter specifies the address of a
830 structure that is the return value of the function in the source program.
831 May only be applied to the first parameter.</dd>
833 <dt><tt>noalias</tt></dt>
834 <dd>This indicates that the parameter does not alias any global or any other
835 parameter. The caller is responsible for ensuring that this is the case,
836 usually by placing the value in a stack allocation.</dd>
838 <dt><tt>noreturn</tt></dt>
839 <dd>This function attribute indicates that the function never returns. This
840 indicates to LLVM that every call to this function should be treated as if
841 an <tt>unreachable</tt> instruction immediately followed the call.</dd>
843 <dt><tt>nounwind</tt></dt>
844 <dd>This function attribute indicates that the function type does not use
845 the unwind instruction and does not allow stack unwinding to propagate
848 <dt><tt>nest</tt></dt>
849 <dd>This indicates that the parameter can be excised using the
850 <a href="#int_trampoline">trampoline intrinsics</a>.</dd>
851 <dt><tt>readonly</tt></dt>
852 <dd>This function attribute indicates that the function has no side-effects
853 except for producing a return value or throwing an exception. The value
854 returned must only depend on the function arguments and/or global variables.
855 It may use values obtained by dereferencing pointers.</dd>
856 <dt><tt>readnone</tt></dt>
857 <dd>A <tt>readnone</tt> function has the same restrictions as a <tt>readonly</tt>
858 function, but in addition it is not allowed to dereference any pointer arguments
864 <!-- ======================================================================= -->
865 <div class="doc_subsection">
866 <a name="gc">Garbage Collector Names</a>
869 <div class="doc_text">
870 <p>Each function may specify a garbage collector name, which is simply a
873 <div class="doc_code"><pre
874 >define void @f() gc "name" { ...</pre></div>
876 <p>The compiler declares the supported values of <i>name</i>. Specifying a
877 collector which will cause the compiler to alter its output in order to support
878 the named garbage collection algorithm.</p>
881 <!-- ======================================================================= -->
882 <div class="doc_subsection">
883 <a name="moduleasm">Module-Level Inline Assembly</a>
886 <div class="doc_text">
888 Modules may contain "module-level inline asm" blocks, which corresponds to the
889 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
890 LLVM and treated as a single unit, but may be separated in the .ll file if
891 desired. The syntax is very simple:
894 <div class="doc_code">
896 module asm "inline asm code goes here"
897 module asm "more can go here"
901 <p>The strings can contain any character by escaping non-printable characters.
902 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
907 The inline asm code is simply printed to the machine code .s file when
908 assembly code is generated.
912 <!-- ======================================================================= -->
913 <div class="doc_subsection">
914 <a name="datalayout">Data Layout</a>
917 <div class="doc_text">
918 <p>A module may specify a target specific data layout string that specifies how
919 data is to be laid out in memory. The syntax for the data layout is simply:</p>
920 <pre> target datalayout = "<i>layout specification</i>"</pre>
921 <p>The <i>layout specification</i> consists of a list of specifications
922 separated by the minus sign character ('-'). Each specification starts with a
923 letter and may include other information after the letter to define some
924 aspect of the data layout. The specifications accepted are as follows: </p>
927 <dd>Specifies that the target lays out data in big-endian form. That is, the
928 bits with the most significance have the lowest address location.</dd>
930 <dd>Specifies that hte target lays out data in little-endian form. That is,
931 the bits with the least significance have the lowest address location.</dd>
932 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
933 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
934 <i>preferred</i> alignments. All sizes are in bits. Specifying the <i>pref</i>
935 alignment is optional. If omitted, the preceding <tt>:</tt> should be omitted
937 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
938 <dd>This specifies the alignment for an integer type of a given bit
939 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
940 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
941 <dd>This specifies the alignment for a vector type of a given bit
943 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
944 <dd>This specifies the alignment for a floating point type of a given bit
945 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
947 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
948 <dd>This specifies the alignment for an aggregate type of a given bit
951 <p>When constructing the data layout for a given target, LLVM starts with a
952 default set of specifications which are then (possibly) overriden by the
953 specifications in the <tt>datalayout</tt> keyword. The default specifications
954 are given in this list:</p>
956 <li><tt>E</tt> - big endian</li>
957 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
958 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
959 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
960 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
961 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
962 <li><tt>i64:32:64</tt> - i64 has abi alignment of 32-bits but preferred
963 alignment of 64-bits</li>
964 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
965 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
966 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
967 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
968 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
970 <p>When llvm is determining the alignment for a given type, it uses the
973 <li>If the type sought is an exact match for one of the specifications, that
974 specification is used.</li>
975 <li>If no match is found, and the type sought is an integer type, then the
976 smallest integer type that is larger than the bitwidth of the sought type is
977 used. If none of the specifications are larger than the bitwidth then the the
978 largest integer type is used. For example, given the default specifications
979 above, the i7 type will use the alignment of i8 (next largest) while both
980 i65 and i256 will use the alignment of i64 (largest specified).</li>
981 <li>If no match is found, and the type sought is a vector type, then the
982 largest vector type that is smaller than the sought vector type will be used
983 as a fall back. This happens because <128 x double> can be implemented in
984 terms of 64 <2 x double>, for example.</li>
988 <!-- *********************************************************************** -->
989 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
990 <!-- *********************************************************************** -->
992 <div class="doc_text">
994 <p>The LLVM type system is one of the most important features of the
995 intermediate representation. Being typed enables a number of
996 optimizations to be performed on the IR directly, without having to do
997 extra analyses on the side before the transformation. A strong type
998 system makes it easier to read the generated code and enables novel
999 analyses and transformations that are not feasible to perform on normal
1000 three address code representations.</p>
1004 <!-- ======================================================================= -->
1005 <div class="doc_subsection"> <a name="t_classifications">Type
1006 Classifications</a> </div>
1007 <div class="doc_text">
1008 <p>The types fall into a few useful
1009 classifications:</p>
1011 <table border="1" cellspacing="0" cellpadding="4">
1013 <tr><th>Classification</th><th>Types</th></tr>
1015 <td><a href="#t_integer">integer</a></td>
1016 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1019 <td><a href="#t_floating">floating point</a></td>
1020 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1023 <td><a name="t_firstclass">first class</a></td>
1024 <td><a href="#t_integer">integer</a>,
1025 <a href="#t_floating">floating point</a>,
1026 <a href="#t_pointer">pointer</a>,
1027 <a href="#t_vector">vector</a>
1031 <td><a href="#t_primitive">primitive</a></td>
1032 <td><a href="#t_label">label</a>,
1033 <a href="#t_void">void</a>,
1034 <a href="#t_integer">integer</a>,
1035 <a href="#t_floating">floating point</a>.</td>
1038 <td><a href="#t_derived">derived</a></td>
1039 <td><a href="#t_integer">integer</a>,
1040 <a href="#t_array">array</a>,
1041 <a href="#t_function">function</a>,
1042 <a href="#t_pointer">pointer</a>,
1043 <a href="#t_struct">structure</a>,
1044 <a href="#t_pstruct">packed structure</a>,
1045 <a href="#t_vector">vector</a>,
1046 <a href="#t_opaque">opaque</a>.
1051 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
1052 most important. Values of these types are the only ones which can be
1053 produced by instructions, passed as arguments, or used as operands to
1054 instructions. This means that all structures and arrays must be
1055 manipulated either by pointer or by component.</p>
1058 <!-- ======================================================================= -->
1059 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
1061 <div class="doc_text">
1062 <p>The primitive types are the fundamental building blocks of the LLVM
1067 <!-- _______________________________________________________________________ -->
1068 <div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1070 <div class="doc_text">
1073 <tr><th>Type</th><th>Description</th></tr>
1074 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1075 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1076 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1077 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1078 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1083 <!-- _______________________________________________________________________ -->
1084 <div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1086 <div class="doc_text">
1088 <p>The void type does not represent any value and has no size.</p>
1097 <!-- _______________________________________________________________________ -->
1098 <div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1100 <div class="doc_text">
1102 <p>The label type represents code labels.</p>
1112 <!-- ======================================================================= -->
1113 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1115 <div class="doc_text">
1117 <p>The real power in LLVM comes from the derived types in the system.
1118 This is what allows a programmer to represent arrays, functions,
1119 pointers, and other useful types. Note that these derived types may be
1120 recursive: For example, it is possible to have a two dimensional array.</p>
1124 <!-- _______________________________________________________________________ -->
1125 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1127 <div class="doc_text">
1130 <p>The integer type is a very simple derived type that simply specifies an
1131 arbitrary bit width for the integer type desired. Any bit width from 1 bit to
1132 2^23-1 (about 8 million) can be specified.</p>
1140 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1144 <table class="layout">
1147 <td><tt>i1</tt></td>
1148 <td>a single-bit integer.</td>
1150 <td><tt>i32</tt></td>
1151 <td>a 32-bit integer.</td>
1153 <td><tt>i1942652</tt></td>
1154 <td>a really big integer of over 1 million bits.</td>
1160 <!-- _______________________________________________________________________ -->
1161 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1163 <div class="doc_text">
1167 <p>The array type is a very simple derived type that arranges elements
1168 sequentially in memory. The array type requires a size (number of
1169 elements) and an underlying data type.</p>
1174 [<# elements> x <elementtype>]
1177 <p>The number of elements is a constant integer value; elementtype may
1178 be any type with a size.</p>
1181 <table class="layout">
1183 <td class="left"><tt>[40 x i32]</tt></td>
1184 <td class="left">Array of 40 32-bit integer values.</td>
1187 <td class="left"><tt>[41 x i32]</tt></td>
1188 <td class="left">Array of 41 32-bit integer values.</td>
1191 <td class="left"><tt>[4 x i8]</tt></td>
1192 <td class="left">Array of 4 8-bit integer values.</td>
1195 <p>Here are some examples of multidimensional arrays:</p>
1196 <table class="layout">
1198 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1199 <td class="left">3x4 array of 32-bit integer values.</td>
1202 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1203 <td class="left">12x10 array of single precision floating point values.</td>
1206 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1207 <td class="left">2x3x4 array of 16-bit integer values.</td>
1211 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1212 length array. Normally, accesses past the end of an array are undefined in
1213 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
1214 As a special case, however, zero length arrays are recognized to be variable
1215 length. This allows implementation of 'pascal style arrays' with the LLVM
1216 type "{ i32, [0 x float]}", for example.</p>
1220 <!-- _______________________________________________________________________ -->
1221 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1222 <div class="doc_text">
1224 <p>The function type can be thought of as a function signature. It
1225 consists of a return type and a list of formal parameter types.
1226 Function types are usually used to build virtual function tables
1227 (which are structures of pointers to functions), for indirect function
1228 calls, and when defining a function.</p>
1230 The return type of a function type cannot be an aggregate type.
1233 <pre> <returntype> (<parameter list>)<br></pre>
1234 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1235 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1236 which indicates that the function takes a variable number of arguments.
1237 Variable argument functions can access their arguments with the <a
1238 href="#int_varargs">variable argument handling intrinsic</a> functions.</p>
1240 <table class="layout">
1242 <td class="left"><tt>i32 (i32)</tt></td>
1243 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1245 </tr><tr class="layout">
1246 <td class="left"><tt>float (i16 signext, i32 *) *
1248 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1249 an <tt>i16</tt> that should be sign extended and a
1250 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1253 </tr><tr class="layout">
1254 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1255 <td class="left">A vararg function that takes at least one
1256 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1257 which returns an integer. This is the signature for <tt>printf</tt> in
1264 <!-- _______________________________________________________________________ -->
1265 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1266 <div class="doc_text">
1268 <p>The structure type is used to represent a collection of data members
1269 together in memory. The packing of the field types is defined to match
1270 the ABI of the underlying processor. The elements of a structure may
1271 be any type that has a size.</p>
1272 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1273 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1274 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1277 <pre> { <type list> }<br></pre>
1279 <table class="layout">
1281 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1282 <td class="left">A triple of three <tt>i32</tt> values</td>
1283 </tr><tr class="layout">
1284 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1285 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1286 second element is a <a href="#t_pointer">pointer</a> to a
1287 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1288 an <tt>i32</tt>.</td>
1293 <!-- _______________________________________________________________________ -->
1294 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1296 <div class="doc_text">
1298 <p>The packed structure type is used to represent a collection of data members
1299 together in memory. There is no padding between fields. Further, the alignment
1300 of a packed structure is 1 byte. The elements of a packed structure may
1301 be any type that has a size.</p>
1302 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1303 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1304 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1307 <pre> < { <type list> } > <br></pre>
1309 <table class="layout">
1311 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1312 <td class="left">A triple of three <tt>i32</tt> values</td>
1313 </tr><tr class="layout">
1314 <td class="left"><tt>< { float, i32 (i32)* } ></tt></td>
1315 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1316 second element is a <a href="#t_pointer">pointer</a> to a
1317 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1318 an <tt>i32</tt>.</td>
1323 <!-- _______________________________________________________________________ -->
1324 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1325 <div class="doc_text">
1327 <p>As in many languages, the pointer type represents a pointer or
1328 reference to another object, which must live in memory. Pointer types may have
1329 an optional address space attribute defining the target-specific numbered
1330 address space where the pointed-to object resides. The default address space is
1333 <pre> <type> *<br></pre>
1335 <table class="layout">
1337 <td class="left"><tt>[4x i32]*</tt></td>
1338 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1339 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1342 <td class="left"><tt>i32 (i32 *) *</tt></td>
1343 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
1344 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1348 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1349 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1350 that resides in address space #5.</td>
1355 <!-- _______________________________________________________________________ -->
1356 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1357 <div class="doc_text">
1361 <p>A vector type is a simple derived type that represents a vector
1362 of elements. Vector types are used when multiple primitive data
1363 are operated in parallel using a single instruction (SIMD).
1364 A vector type requires a size (number of
1365 elements) and an underlying primitive data type. Vectors must have a power
1366 of two length (1, 2, 4, 8, 16 ...). Vector types are
1367 considered <a href="#t_firstclass">first class</a>.</p>
1372 < <# elements> x <elementtype> >
1375 <p>The number of elements is a constant integer value; elementtype may
1376 be any integer or floating point type.</p>
1380 <table class="layout">
1382 <td class="left"><tt><4 x i32></tt></td>
1383 <td class="left">Vector of 4 32-bit integer values.</td>
1386 <td class="left"><tt><8 x float></tt></td>
1387 <td class="left">Vector of 8 32-bit floating-point values.</td>
1390 <td class="left"><tt><2 x i64></tt></td>
1391 <td class="left">Vector of 2 64-bit integer values.</td>
1396 <!-- _______________________________________________________________________ -->
1397 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1398 <div class="doc_text">
1402 <p>Opaque types are used to represent unknown types in the system. This
1403 corresponds (for example) to the C notion of a forward declared structure type.
1404 In LLVM, opaque types can eventually be resolved to any type (not just a
1405 structure type).</p>
1415 <table class="layout">
1417 <td class="left"><tt>opaque</tt></td>
1418 <td class="left">An opaque type.</td>
1424 <!-- *********************************************************************** -->
1425 <div class="doc_section"> <a name="constants">Constants</a> </div>
1426 <!-- *********************************************************************** -->
1428 <div class="doc_text">
1430 <p>LLVM has several different basic types of constants. This section describes
1431 them all and their syntax.</p>
1435 <!-- ======================================================================= -->
1436 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1438 <div class="doc_text">
1441 <dt><b>Boolean constants</b></dt>
1443 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1444 constants of the <tt><a href="#t_primitive">i1</a></tt> type.
1447 <dt><b>Integer constants</b></dt>
1449 <dd>Standard integers (such as '4') are constants of the <a
1450 href="#t_integer">integer</a> type. Negative numbers may be used with
1454 <dt><b>Floating point constants</b></dt>
1456 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1457 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1458 notation (see below). Floating point constants must have a <a
1459 href="#t_floating">floating point</a> type. </dd>
1461 <dt><b>Null pointer constants</b></dt>
1463 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1464 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1468 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1469 of floating point constants. For example, the form '<tt>double
1470 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1471 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1472 (and the only time that they are generated by the disassembler) is when a
1473 floating point constant must be emitted but it cannot be represented as a
1474 decimal floating point number. For example, NaN's, infinities, and other
1475 special values are represented in their IEEE hexadecimal format so that
1476 assembly and disassembly do not cause any bits to change in the constants.</p>
1480 <!-- ======================================================================= -->
1481 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1484 <div class="doc_text">
1485 <p>Aggregate constants arise from aggregation of simple constants
1486 and smaller aggregate constants.</p>
1489 <dt><b>Structure constants</b></dt>
1491 <dd>Structure constants are represented with notation similar to structure
1492 type definitions (a comma separated list of elements, surrounded by braces
1493 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
1494 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>". Structure constants
1495 must have <a href="#t_struct">structure type</a>, and the number and
1496 types of elements must match those specified by the type.
1499 <dt><b>Array constants</b></dt>
1501 <dd>Array constants are represented with notation similar to array type
1502 definitions (a comma separated list of elements, surrounded by square brackets
1503 (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74 ]</tt>". Array
1504 constants must have <a href="#t_array">array type</a>, and the number and
1505 types of elements must match those specified by the type.
1508 <dt><b>Vector constants</b></dt>
1510 <dd>Vector constants are represented with notation similar to vector type
1511 definitions (a comma separated list of elements, surrounded by
1512 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32 42,
1513 i32 11, i32 74, i32 100 ></tt>". Vector constants must have <a
1514 href="#t_vector">vector type</a>, and the number and types of elements must
1515 match those specified by the type.
1518 <dt><b>Zero initialization</b></dt>
1520 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1521 value to zero of <em>any</em> type, including scalar and aggregate types.
1522 This is often used to avoid having to print large zero initializers (e.g. for
1523 large arrays) and is always exactly equivalent to using explicit zero
1530 <!-- ======================================================================= -->
1531 <div class="doc_subsection">
1532 <a name="globalconstants">Global Variable and Function Addresses</a>
1535 <div class="doc_text">
1537 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1538 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1539 constants. These constants are explicitly referenced when the <a
1540 href="#identifiers">identifier for the global</a> is used and always have <a
1541 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1544 <div class="doc_code">
1548 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1554 <!-- ======================================================================= -->
1555 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1556 <div class="doc_text">
1557 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1558 no specific value. Undefined values may be of any type and be used anywhere
1559 a constant is permitted.</p>
1561 <p>Undefined values indicate to the compiler that the program is well defined
1562 no matter what value is used, giving the compiler more freedom to optimize.
1566 <!-- ======================================================================= -->
1567 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1570 <div class="doc_text">
1572 <p>Constant expressions are used to allow expressions involving other constants
1573 to be used as constants. Constant expressions may be of any <a
1574 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1575 that does not have side effects (e.g. load and call are not supported). The
1576 following is the syntax for constant expressions:</p>
1579 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1580 <dd>Truncate a constant to another type. The bit size of CST must be larger
1581 than the bit size of TYPE. Both types must be integers.</dd>
1583 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1584 <dd>Zero extend a constant to another type. The bit size of CST must be
1585 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1587 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1588 <dd>Sign extend a constant to another type. The bit size of CST must be
1589 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1591 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1592 <dd>Truncate a floating point constant to another floating point type. The
1593 size of CST must be larger than the size of TYPE. Both types must be
1594 floating point.</dd>
1596 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1597 <dd>Floating point extend a constant to another type. The size of CST must be
1598 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1600 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
1601 <dd>Convert a floating point constant to the corresponding unsigned integer
1602 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1603 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1604 of the same number of elements. If the value won't fit in the integer type,
1605 the results are undefined.</dd>
1607 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
1608 <dd>Convert a floating point constant to the corresponding signed integer
1609 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1610 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1611 of the same number of elements. If the value won't fit in the integer type,
1612 the results are undefined.</dd>
1614 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
1615 <dd>Convert an unsigned integer constant to the corresponding floating point
1616 constant. TYPE must be a scalar or vector floating point type. CST must be of
1617 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1618 of the same number of elements. If the value won't fit in the floating point
1619 type, the results are undefined.</dd>
1621 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
1622 <dd>Convert a signed integer constant to the corresponding floating point
1623 constant. TYPE must be a scalar or vector floating point type. CST must be of
1624 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1625 of the same number of elements. If the value won't fit in the floating point
1626 type, the results are undefined.</dd>
1628 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
1629 <dd>Convert a pointer typed constant to the corresponding integer constant
1630 TYPE must be an integer type. CST must be of pointer type. The CST value is
1631 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1633 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
1634 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1635 pointer type. CST must be of integer type. The CST value is zero extended,
1636 truncated, or unchanged to make it fit in a pointer size. This one is
1637 <i>really</i> dangerous!</dd>
1639 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
1640 <dd>Convert a constant, CST, to another TYPE. The size of CST and TYPE must be
1641 identical (same number of bits). The conversion is done as if the CST value
1642 was stored to memory and read back as TYPE. In other words, no bits change
1643 with this operator, just the type. This can be used for conversion of
1644 vector types to any other type, as long as they have the same bit width. For
1645 pointers it is only valid to cast to another pointer type.
1648 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1650 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1651 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1652 instruction, the index list may have zero or more indexes, which are required
1653 to make sense for the type of "CSTPTR".</dd>
1655 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1657 <dd>Perform the <a href="#i_select">select operation</a> on
1660 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
1661 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
1663 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
1664 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
1666 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1668 <dd>Perform the <a href="#i_extractelement">extractelement
1669 operation</a> on constants.
1671 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1673 <dd>Perform the <a href="#i_insertelement">insertelement
1674 operation</a> on constants.</dd>
1677 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
1679 <dd>Perform the <a href="#i_shufflevector">shufflevector
1680 operation</a> on constants.</dd>
1682 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1684 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1685 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1686 binary</a> operations. The constraints on operands are the same as those for
1687 the corresponding instruction (e.g. no bitwise operations on floating point
1688 values are allowed).</dd>
1692 <!-- *********************************************************************** -->
1693 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1694 <!-- *********************************************************************** -->
1696 <!-- ======================================================================= -->
1697 <div class="doc_subsection">
1698 <a name="inlineasm">Inline Assembler Expressions</a>
1701 <div class="doc_text">
1704 LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
1705 Module-Level Inline Assembly</a>) through the use of a special value. This
1706 value represents the inline assembler as a string (containing the instructions
1707 to emit), a list of operand constraints (stored as a string), and a flag that
1708 indicates whether or not the inline asm expression has side effects. An example
1709 inline assembler expression is:
1712 <div class="doc_code">
1714 i32 (i32) asm "bswap $0", "=r,r"
1719 Inline assembler expressions may <b>only</b> be used as the callee operand of
1720 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
1723 <div class="doc_code">
1725 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
1730 Inline asms with side effects not visible in the constraint list must be marked
1731 as having side effects. This is done through the use of the
1732 '<tt>sideeffect</tt>' keyword, like so:
1735 <div class="doc_code">
1737 call void asm sideeffect "eieio", ""()
1741 <p>TODO: The format of the asm and constraints string still need to be
1742 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1743 need to be documented).
1748 <!-- *********************************************************************** -->
1749 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1750 <!-- *********************************************************************** -->
1752 <div class="doc_text">
1754 <p>The LLVM instruction set consists of several different
1755 classifications of instructions: <a href="#terminators">terminator
1756 instructions</a>, <a href="#binaryops">binary instructions</a>,
1757 <a href="#bitwiseops">bitwise binary instructions</a>, <a
1758 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1759 instructions</a>.</p>
1763 <!-- ======================================================================= -->
1764 <div class="doc_subsection"> <a name="terminators">Terminator
1765 Instructions</a> </div>
1767 <div class="doc_text">
1769 <p>As mentioned <a href="#functionstructure">previously</a>, every
1770 basic block in a program ends with a "Terminator" instruction, which
1771 indicates which block should be executed after the current block is
1772 finished. These terminator instructions typically yield a '<tt>void</tt>'
1773 value: they produce control flow, not values (the one exception being
1774 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1775 <p>There are six different terminator instructions: the '<a
1776 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1777 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1778 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1779 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1780 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1784 <!-- _______________________________________________________________________ -->
1785 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1786 Instruction</a> </div>
1787 <div class="doc_text">
1789 <pre> ret <type> <value> <i>; Return a value from a non-void function</i>
1790 ret void <i>; Return from void function</i>
1793 <p>The '<tt>ret</tt>' instruction is used to return control flow (and a
1794 value) from a function back to the caller.</p>
1795 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
1796 returns a value and then causes control flow, and one that just causes
1797 control flow to occur.</p>
1799 <p>The '<tt>ret</tt>' instruction may return any '<a
1800 href="#t_firstclass">first class</a>' type. Notice that a function is
1801 not <a href="#wellformed">well formed</a> if there exists a '<tt>ret</tt>'
1802 instruction inside of the function that returns a value that does not
1803 match the return type of the function.</p>
1805 <p>When the '<tt>ret</tt>' instruction is executed, control flow
1806 returns back to the calling function's context. If the caller is a "<a
1807 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1808 the instruction after the call. If the caller was an "<a
1809 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1810 at the beginning of the "normal" destination block. If the instruction
1811 returns a value, that value shall set the call or invoke instruction's
1814 <pre> ret i32 5 <i>; Return an integer value of 5</i>
1815 ret void <i>; Return from a void function</i>
1818 <!-- _______________________________________________________________________ -->
1819 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1820 <div class="doc_text">
1822 <pre> br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1825 <p>The '<tt>br</tt>' instruction is used to cause control flow to
1826 transfer to a different basic block in the current function. There are
1827 two forms of this instruction, corresponding to a conditional branch
1828 and an unconditional branch.</p>
1830 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1831 single '<tt>i1</tt>' value and two '<tt>label</tt>' values. The
1832 unconditional form of the '<tt>br</tt>' instruction takes a single
1833 '<tt>label</tt>' value as a target.</p>
1835 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
1836 argument is evaluated. If the value is <tt>true</tt>, control flows
1837 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1838 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1840 <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
1841 href="#i_ret">ret</a> i32 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> i32 0<br></pre>
1843 <!-- _______________________________________________________________________ -->
1844 <div class="doc_subsubsection">
1845 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
1848 <div class="doc_text">
1852 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
1857 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
1858 several different places. It is a generalization of the '<tt>br</tt>'
1859 instruction, allowing a branch to occur to one of many possible
1865 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
1866 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
1867 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
1868 table is not allowed to contain duplicate constant entries.</p>
1872 <p>The <tt>switch</tt> instruction specifies a table of values and
1873 destinations. When the '<tt>switch</tt>' instruction is executed, this
1874 table is searched for the given value. If the value is found, control flow is
1875 transfered to the corresponding destination; otherwise, control flow is
1876 transfered to the default destination.</p>
1878 <h5>Implementation:</h5>
1880 <p>Depending on properties of the target machine and the particular
1881 <tt>switch</tt> instruction, this instruction may be code generated in different
1882 ways. For example, it could be generated as a series of chained conditional
1883 branches or with a lookup table.</p>
1888 <i>; Emulate a conditional br instruction</i>
1889 %Val = <a href="#i_zext">zext</a> i1 %value to i32
1890 switch i32 %Val, label %truedest [i32 0, label %falsedest ]
1892 <i>; Emulate an unconditional br instruction</i>
1893 switch i32 0, label %dest [ ]
1895 <i>; Implement a jump table:</i>
1896 switch i32 %val, label %otherwise [ i32 0, label %onzero
1898 i32 2, label %ontwo ]
1902 <!-- _______________________________________________________________________ -->
1903 <div class="doc_subsubsection">
1904 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
1907 <div class="doc_text">
1912 <result> = invoke [<a href="#callingconv">cconv</a>] <ptr to function ty> %<function ptr val>(<function args>)
1913 to label <normal label> unwind label <exception label>
1918 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
1919 function, with the possibility of control flow transfer to either the
1920 '<tt>normal</tt>' label or the
1921 '<tt>exception</tt>' label. If the callee function returns with the
1922 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
1923 "normal" label. If the callee (or any indirect callees) returns with the "<a
1924 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
1925 continued at the dynamically nearest "exception" label.</p>
1929 <p>This instruction requires several arguments:</p>
1933 The optional "cconv" marker indicates which <a href="#callingconv">calling
1934 convention</a> the call should use. If none is specified, the call defaults
1935 to using C calling conventions.
1937 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
1938 function value being invoked. In most cases, this is a direct function
1939 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
1940 an arbitrary pointer to function value.
1943 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
1944 function to be invoked. </li>
1946 <li>'<tt>function args</tt>': argument list whose types match the function
1947 signature argument types. If the function signature indicates the function
1948 accepts a variable number of arguments, the extra arguments can be
1951 <li>'<tt>normal label</tt>': the label reached when the called function
1952 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
1954 <li>'<tt>exception label</tt>': the label reached when a callee returns with
1955 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
1961 <p>This instruction is designed to operate as a standard '<tt><a
1962 href="#i_call">call</a></tt>' instruction in most regards. The primary
1963 difference is that it establishes an association with a label, which is used by
1964 the runtime library to unwind the stack.</p>
1966 <p>This instruction is used in languages with destructors to ensure that proper
1967 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
1968 exception. Additionally, this is important for implementation of
1969 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
1973 %retval = invoke i32 %Test(i32 15) to label %Continue
1974 unwind label %TestCleanup <i>; {i32}:retval set</i>
1975 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Test(i32 15) to label %Continue
1976 unwind label %TestCleanup <i>; {i32}:retval set</i>
1981 <!-- _______________________________________________________________________ -->
1983 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
1984 Instruction</a> </div>
1986 <div class="doc_text">
1995 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
1996 at the first callee in the dynamic call stack which used an <a
1997 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
1998 primarily used to implement exception handling.</p>
2002 <p>The '<tt>unwind</tt>' intrinsic causes execution of the current function to
2003 immediately halt. The dynamic call stack is then searched for the first <a
2004 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
2005 execution continues at the "exceptional" destination block specified by the
2006 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
2007 dynamic call chain, undefined behavior results.</p>
2010 <!-- _______________________________________________________________________ -->
2012 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
2013 Instruction</a> </div>
2015 <div class="doc_text">
2024 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
2025 instruction is used to inform the optimizer that a particular portion of the
2026 code is not reachable. This can be used to indicate that the code after a
2027 no-return function cannot be reached, and other facts.</p>
2031 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
2036 <!-- ======================================================================= -->
2037 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
2038 <div class="doc_text">
2039 <p>Binary operators are used to do most of the computation in a
2040 program. They require two operands, execute an operation on them, and
2041 produce a single value. The operands might represent
2042 multiple data, as is the case with the <a href="#t_vector">vector</a> data type.
2043 The result value of a binary operator is not
2044 necessarily the same type as its operands.</p>
2045 <p>There are several different binary operators:</p>
2047 <!-- _______________________________________________________________________ -->
2048 <div class="doc_subsubsection"> <a name="i_add">'<tt>add</tt>'
2049 Instruction</a> </div>
2050 <div class="doc_text">
2052 <pre> <result> = add <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2055 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
2057 <p>The two arguments to the '<tt>add</tt>' instruction must be either <a
2058 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a> values.
2059 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2060 Both arguments must have identical types.</p>
2062 <p>The value produced is the integer or floating point sum of the two
2064 <p>If an integer sum has unsigned overflow, the result returned is the
2065 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2067 <p>Because LLVM integers use a two's complement representation, this
2068 instruction is appropriate for both signed and unsigned integers.</p>
2070 <pre> <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
2073 <!-- _______________________________________________________________________ -->
2074 <div class="doc_subsubsection"> <a name="i_sub">'<tt>sub</tt>'
2075 Instruction</a> </div>
2076 <div class="doc_text">
2078 <pre> <result> = sub <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2081 <p>The '<tt>sub</tt>' instruction returns the difference of its two
2083 <p>Note that the '<tt>sub</tt>' instruction is used to represent the '<tt>neg</tt>'
2084 instruction present in most other intermediate representations.</p>
2086 <p>The two arguments to the '<tt>sub</tt>' instruction must be either <a
2087 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
2089 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2090 Both arguments must have identical types.</p>
2092 <p>The value produced is the integer or floating point difference of
2093 the two operands.</p>
2094 <p>If an integer difference has unsigned overflow, the result returned is the
2095 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2097 <p>Because LLVM integers use a two's complement representation, this
2098 instruction is appropriate for both signed and unsigned integers.</p>
2101 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
2102 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
2105 <!-- _______________________________________________________________________ -->
2106 <div class="doc_subsubsection"> <a name="i_mul">'<tt>mul</tt>'
2107 Instruction</a> </div>
2108 <div class="doc_text">
2110 <pre> <result> = mul <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2113 <p>The '<tt>mul</tt>' instruction returns the product of its two
2116 <p>The two arguments to the '<tt>mul</tt>' instruction must be either <a
2117 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
2119 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2120 Both arguments must have identical types.</p>
2122 <p>The value produced is the integer or floating point product of the
2124 <p>If the result of an integer multiplication has unsigned overflow,
2125 the result returned is the mathematical result modulo
2126 2<sup>n</sup>, where n is the bit width of the result.</p>
2127 <p>Because LLVM integers use a two's complement representation, and the
2128 result is the same width as the operands, this instruction returns the
2129 correct result for both signed and unsigned integers. If a full product
2130 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands
2131 should be sign-extended or zero-extended as appropriate to the
2132 width of the full product.</p>
2134 <pre> <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
2137 <!-- _______________________________________________________________________ -->
2138 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
2140 <div class="doc_text">
2142 <pre> <result> = udiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2145 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two
2148 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
2149 <a href="#t_integer">integer</a> values. Both arguments must have identical
2150 types. This instruction can also take <a href="#t_vector">vector</a> versions
2151 of the values in which case the elements must be integers.</p>
2153 <p>The value produced is the unsigned integer quotient of the two operands.</p>
2154 <p>Note that unsigned integer division and signed integer division are distinct
2155 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
2156 <p>Division by zero leads to undefined behavior.</p>
2158 <pre> <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2161 <!-- _______________________________________________________________________ -->
2162 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
2164 <div class="doc_text">
2166 <pre> <result> = sdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2169 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
2172 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
2173 <a href="#t_integer">integer</a> values. Both arguments must have identical
2174 types. This instruction can also take <a href="#t_vector">vector</a> versions
2175 of the values in which case the elements must be integers.</p>
2177 <p>The value produced is the signed integer quotient of the two operands.</p>
2178 <p>Note that signed integer division and unsigned integer division are distinct
2179 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
2180 <p>Division by zero leads to undefined behavior. Overflow also leads to
2181 undefined behavior; this is a rare case, but can occur, for example,
2182 by doing a 32-bit division of -2147483648 by -1.</p>
2184 <pre> <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2187 <!-- _______________________________________________________________________ -->
2188 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
2189 Instruction</a> </div>
2190 <div class="doc_text">
2192 <pre> <result> = fdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2195 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
2198 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
2199 <a href="#t_floating">floating point</a> values. Both arguments must have
2200 identical types. This instruction can also take <a href="#t_vector">vector</a>
2201 versions of floating point values.</p>
2203 <p>The value produced is the floating point quotient of the two operands.</p>
2205 <pre> <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
2208 <!-- _______________________________________________________________________ -->
2209 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
2211 <div class="doc_text">
2213 <pre> <result> = urem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2216 <p>The '<tt>urem</tt>' instruction returns the remainder from the
2217 unsigned division of its two arguments.</p>
2219 <p>The two arguments to the '<tt>urem</tt>' instruction must be
2220 <a href="#t_integer">integer</a> values. Both arguments must have identical
2221 types. This instruction can also take <a href="#t_vector">vector</a> versions
2222 of the values in which case the elements must be integers.</p>
2224 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
2225 This instruction always performs an unsigned division to get the remainder,
2226 regardless of whether the arguments are unsigned or not.</p>
2227 <p>Note that unsigned integer remainder and signed integer remainder are
2228 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
2229 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
2231 <pre> <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2235 <!-- _______________________________________________________________________ -->
2236 <div class="doc_subsubsection"> <a name="i_srem">'<tt>srem</tt>'
2237 Instruction</a> </div>
2238 <div class="doc_text">
2240 <pre> <result> = srem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2243 <p>The '<tt>srem</tt>' instruction returns the remainder from the
2244 signed division of its two operands. This instruction can also take
2245 <a href="#t_vector">vector</a> versions of the values in which case
2246 the elements must be integers.</p>
2249 <p>The two arguments to the '<tt>srem</tt>' instruction must be
2250 <a href="#t_integer">integer</a> values. Both arguments must have identical
2253 <p>This instruction returns the <i>remainder</i> of a division (where the result
2254 has the same sign as the dividend, <tt>var1</tt>), not the <i>modulo</i>
2255 operator (where the result has the same sign as the divisor, <tt>var2</tt>) of
2256 a value. For more information about the difference, see <a
2257 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
2258 Math Forum</a>. For a table of how this is implemented in various languages,
2259 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
2260 Wikipedia: modulo operation</a>.</p>
2261 <p>Note that signed integer remainder and unsigned integer remainder are
2262 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
2263 <p>Taking the remainder of a division by zero leads to undefined behavior.
2264 Overflow also leads to undefined behavior; this is a rare case, but can occur,
2265 for example, by taking the remainder of a 32-bit division of -2147483648 by -1.
2266 (The remainder doesn't actually overflow, but this rule lets srem be
2267 implemented using instructions that return both the result of the division
2268 and the remainder.)</p>
2270 <pre> <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2274 <!-- _______________________________________________________________________ -->
2275 <div class="doc_subsubsection"> <a name="i_frem">'<tt>frem</tt>'
2276 Instruction</a> </div>
2277 <div class="doc_text">
2279 <pre> <result> = frem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2282 <p>The '<tt>frem</tt>' instruction returns the remainder from the
2283 division of its two operands.</p>
2285 <p>The two arguments to the '<tt>frem</tt>' instruction must be
2286 <a href="#t_floating">floating point</a> values. Both arguments must have
2287 identical types. This instruction can also take <a href="#t_vector">vector</a>
2288 versions of floating point values.</p>
2290 <p>This instruction returns the <i>remainder</i> of a division.</p>
2292 <pre> <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
2296 <!-- ======================================================================= -->
2297 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
2298 Operations</a> </div>
2299 <div class="doc_text">
2300 <p>Bitwise binary operators are used to do various forms of
2301 bit-twiddling in a program. They are generally very efficient
2302 instructions and can commonly be strength reduced from other
2303 instructions. They require two operands, execute an operation on them,
2304 and produce a single value. The resulting value of the bitwise binary
2305 operators is always the same type as its first operand.</p>
2308 <!-- _______________________________________________________________________ -->
2309 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
2310 Instruction</a> </div>
2311 <div class="doc_text">
2313 <pre> <result> = shl <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2318 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2319 the left a specified number of bits.</p>
2323 <p>Both arguments to the '<tt>shl</tt>' instruction must be the same <a
2324 href="#t_integer">integer</a> type.</p>
2328 <p>The value produced is <tt>var1</tt> * 2<sup><tt>var2</tt></sup>. If
2329 <tt>var2</tt> is (statically or dynamically) equal to or larger than the number
2330 of bits in <tt>var1</tt>, the result is undefined.</p>
2332 <h5>Example:</h5><pre>
2333 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
2334 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
2335 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
2336 <result> = shl i32 1, 32 <i>; undefined</i>
2339 <!-- _______________________________________________________________________ -->
2340 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
2341 Instruction</a> </div>
2342 <div class="doc_text">
2344 <pre> <result> = lshr <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2348 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
2349 operand shifted to the right a specified number of bits with zero fill.</p>
2352 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
2353 <a href="#t_integer">integer</a> type.</p>
2357 <p>This instruction always performs a logical shift right operation. The most
2358 significant bits of the result will be filled with zero bits after the
2359 shift. If <tt>var2</tt> is (statically or dynamically) equal to or larger than
2360 the number of bits in <tt>var1</tt>, the result is undefined.</p>
2364 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
2365 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
2366 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
2367 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
2368 <result> = lshr i32 1, 32 <i>; undefined</i>
2372 <!-- _______________________________________________________________________ -->
2373 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
2374 Instruction</a> </div>
2375 <div class="doc_text">
2378 <pre> <result> = ashr <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2382 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
2383 operand shifted to the right a specified number of bits with sign extension.</p>
2386 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
2387 <a href="#t_integer">integer</a> type.</p>
2390 <p>This instruction always performs an arithmetic shift right operation,
2391 The most significant bits of the result will be filled with the sign bit
2392 of <tt>var1</tt>. If <tt>var2</tt> is (statically or dynamically) equal to or
2393 larger than the number of bits in <tt>var1</tt>, the result is undefined.
2398 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
2399 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
2400 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
2401 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
2402 <result> = ashr i32 1, 32 <i>; undefined</i>
2406 <!-- _______________________________________________________________________ -->
2407 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
2408 Instruction</a> </div>
2409 <div class="doc_text">
2411 <pre> <result> = and <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2414 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
2415 its two operands.</p>
2417 <p>The two arguments to the '<tt>and</tt>' instruction must be <a
2418 href="#t_integer">integer</a> values. Both arguments must have
2419 identical types.</p>
2421 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
2423 <div style="align: center">
2424 <table border="1" cellspacing="0" cellpadding="4">
2455 <pre> <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
2456 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
2457 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
2460 <!-- _______________________________________________________________________ -->
2461 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
2462 <div class="doc_text">
2464 <pre> <result> = or <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2467 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
2468 or of its two operands.</p>
2470 <p>The two arguments to the '<tt>or</tt>' instruction must be <a
2471 href="#t_integer">integer</a> values. Both arguments must have
2472 identical types.</p>
2474 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
2476 <div style="align: center">
2477 <table border="1" cellspacing="0" cellpadding="4">
2508 <pre> <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
2509 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
2510 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
2513 <!-- _______________________________________________________________________ -->
2514 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
2515 Instruction</a> </div>
2516 <div class="doc_text">
2518 <pre> <result> = xor <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2521 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
2522 or of its two operands. The <tt>xor</tt> is used to implement the
2523 "one's complement" operation, which is the "~" operator in C.</p>
2525 <p>The two arguments to the '<tt>xor</tt>' instruction must be <a
2526 href="#t_integer">integer</a> values. Both arguments must have
2527 identical types.</p>
2529 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
2531 <div style="align: center">
2532 <table border="1" cellspacing="0" cellpadding="4">
2564 <pre> <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
2565 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
2566 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
2567 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
2571 <!-- ======================================================================= -->
2572 <div class="doc_subsection">
2573 <a name="vectorops">Vector Operations</a>
2576 <div class="doc_text">
2578 <p>LLVM supports several instructions to represent vector operations in a
2579 target-independent manner. These instructions cover the element-access and
2580 vector-specific operations needed to process vectors effectively. While LLVM
2581 does directly support these vector operations, many sophisticated algorithms
2582 will want to use target-specific intrinsics to take full advantage of a specific
2587 <!-- _______________________________________________________________________ -->
2588 <div class="doc_subsubsection">
2589 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2592 <div class="doc_text">
2597 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
2603 The '<tt>extractelement</tt>' instruction extracts a single scalar
2604 element from a vector at a specified index.
2611 The first operand of an '<tt>extractelement</tt>' instruction is a
2612 value of <a href="#t_vector">vector</a> type. The second operand is
2613 an index indicating the position from which to extract the element.
2614 The index may be a variable.</p>
2619 The result is a scalar of the same type as the element type of
2620 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2621 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2622 results are undefined.
2628 %result = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
2633 <!-- _______________________________________________________________________ -->
2634 <div class="doc_subsubsection">
2635 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
2638 <div class="doc_text">
2643 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
2649 The '<tt>insertelement</tt>' instruction inserts a scalar
2650 element into a vector at a specified index.
2657 The first operand of an '<tt>insertelement</tt>' instruction is a
2658 value of <a href="#t_vector">vector</a> type. The second operand is a
2659 scalar value whose type must equal the element type of the first
2660 operand. The third operand is an index indicating the position at
2661 which to insert the value. The index may be a variable.</p>
2666 The result is a vector of the same type as <tt>val</tt>. Its
2667 element values are those of <tt>val</tt> except at position
2668 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
2669 exceeds the length of <tt>val</tt>, the results are undefined.
2675 %result = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
2679 <!-- _______________________________________________________________________ -->
2680 <div class="doc_subsubsection">
2681 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
2684 <div class="doc_text">
2689 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <n x i32> <mask> <i>; yields <n x <ty>></i>
2695 The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
2696 from two input vectors, returning a vector of the same type.
2702 The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
2703 with types that match each other and types that match the result of the
2704 instruction. The third argument is a shuffle mask, which has the same number
2705 of elements as the other vector type, but whose element type is always 'i32'.
2709 The shuffle mask operand is required to be a constant vector with either
2710 constant integer or undef values.
2716 The elements of the two input vectors are numbered from left to right across
2717 both of the vectors. The shuffle mask operand specifies, for each element of
2718 the result vector, which element of the two input registers the result element
2719 gets. The element selector may be undef (meaning "don't care") and the second
2720 operand may be undef if performing a shuffle from only one vector.
2726 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
2727 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
2728 %result = shufflevector <4 x i32> %v1, <4 x i32> undef,
2729 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
2734 <!-- ======================================================================= -->
2735 <div class="doc_subsection">
2736 <a name="memoryops">Memory Access and Addressing Operations</a>
2739 <div class="doc_text">
2741 <p>A key design point of an SSA-based representation is how it
2742 represents memory. In LLVM, no memory locations are in SSA form, which
2743 makes things very simple. This section describes how to read, write,
2744 allocate, and free memory in LLVM.</p>
2748 <!-- _______________________________________________________________________ -->
2749 <div class="doc_subsubsection">
2750 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
2753 <div class="doc_text">
2758 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2763 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
2764 heap and returns a pointer to it. The object is always allocated in the generic
2765 address space (address space zero).</p>
2769 <p>The '<tt>malloc</tt>' instruction allocates
2770 <tt>sizeof(<type>)*NumElements</tt>
2771 bytes of memory from the operating system and returns a pointer of the
2772 appropriate type to the program. If "NumElements" is specified, it is the
2773 number of elements allocated, otherwise "NumElements" is defaulted to be one.
2774 If an alignment is specified, the value result of the allocation is guaranteed to
2775 be aligned to at least that boundary. If not specified, or if zero, the target can
2776 choose to align the allocation on any convenient boundary.</p>
2778 <p>'<tt>type</tt>' must be a sized type.</p>
2782 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
2783 a pointer is returned.</p>
2788 %array = malloc [4 x i8 ] <i>; yields {[%4 x i8]*}:array</i>
2790 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
2791 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
2792 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
2793 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
2794 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
2798 <!-- _______________________________________________________________________ -->
2799 <div class="doc_subsubsection">
2800 <a name="i_free">'<tt>free</tt>' Instruction</a>
2803 <div class="doc_text">
2808 free <type> <value> <i>; yields {void}</i>
2813 <p>The '<tt>free</tt>' instruction returns memory back to the unused
2814 memory heap to be reallocated in the future.</p>
2818 <p>'<tt>value</tt>' shall be a pointer value that points to a value
2819 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
2824 <p>Access to the memory pointed to by the pointer is no longer defined
2825 after this instruction executes.</p>
2830 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
2831 free [4 x i8]* %array
2835 <!-- _______________________________________________________________________ -->
2836 <div class="doc_subsubsection">
2837 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
2840 <div class="doc_text">
2845 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2850 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
2851 currently executing function, to be automatically released when this function
2852 returns to its caller. The object is always allocated in the generic address
2853 space (address space zero).</p>
2857 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
2858 bytes of memory on the runtime stack, returning a pointer of the
2859 appropriate type to the program. If "NumElements" is specified, it is the
2860 number of elements allocated, otherwise "NumElements" is defaulted to be one.
2861 If an alignment is specified, the value result of the allocation is guaranteed
2862 to be aligned to at least that boundary. If not specified, or if zero, the target
2863 can choose to align the allocation on any convenient boundary.</p>
2865 <p>'<tt>type</tt>' may be any sized type.</p>
2869 <p>Memory is allocated; a pointer is returned. '<tt>alloca</tt>'d
2870 memory is automatically released when the function returns. The '<tt>alloca</tt>'
2871 instruction is commonly used to represent automatic variables that must
2872 have an address available. When the function returns (either with the <tt><a
2873 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
2874 instructions), the memory is reclaimed.</p>
2879 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
2880 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
2881 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
2882 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
2886 <!-- _______________________________________________________________________ -->
2887 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
2888 Instruction</a> </div>
2889 <div class="doc_text">
2891 <pre> <result> = load <ty>* <pointer>[, align <alignment>]<br> <result> = volatile load <ty>* <pointer>[, align <alignment>]<br></pre>
2893 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
2895 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
2896 address from which to load. The pointer must point to a <a
2897 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
2898 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
2899 the number or order of execution of this <tt>load</tt> with other
2900 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
2903 The optional "align" argument specifies the alignment of the operation
2904 (that is, the alignment of the memory address). A value of 0 or an
2905 omitted "align" argument means that the operation has the preferential
2906 alignment for the target. It is the responsibility of the code emitter
2907 to ensure that the alignment information is correct. Overestimating
2908 the alignment results in an undefined behavior. Underestimating the
2909 alignment may produce less efficient code. An alignment of 1 is always
2913 <p>The location of memory pointed to is loaded.</p>
2915 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
2917 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
2918 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
2921 <!-- _______________________________________________________________________ -->
2922 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
2923 Instruction</a> </div>
2924 <div class="doc_text">
2926 <pre> store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
2927 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
2930 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
2932 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
2933 to store and an address at which to store it. The type of the '<tt><pointer></tt>'
2934 operand must be a pointer to the type of the '<tt><value></tt>'
2935 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
2936 optimizer is not allowed to modify the number or order of execution of
2937 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
2938 href="#i_store">store</a></tt> instructions.</p>
2940 The optional "align" argument specifies the alignment of the operation
2941 (that is, the alignment of the memory address). A value of 0 or an
2942 omitted "align" argument means that the operation has the preferential
2943 alignment for the target. It is the responsibility of the code emitter
2944 to ensure that the alignment information is correct. Overestimating
2945 the alignment results in an undefined behavior. Underestimating the
2946 alignment may produce less efficient code. An alignment of 1 is always
2950 <p>The contents of memory are updated to contain '<tt><value></tt>'
2951 at the location specified by the '<tt><pointer></tt>' operand.</p>
2953 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
2954 store i32 3, i32* %ptr <i>; yields {void}</i>
2955 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
2959 <!-- _______________________________________________________________________ -->
2960 <div class="doc_subsubsection">
2961 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
2964 <div class="doc_text">
2967 <result> = getelementptr <ty>* <ptrval>{, <ty> <idx>}*
2973 The '<tt>getelementptr</tt>' instruction is used to get the address of a
2974 subelement of an aggregate data structure.</p>
2978 <p>This instruction takes a list of integer operands that indicate what
2979 elements of the aggregate object to index to. The actual types of the arguments
2980 provided depend on the type of the first pointer argument. The
2981 '<tt>getelementptr</tt>' instruction is used to index down through the type
2982 levels of a structure or to a specific index in an array. When indexing into a
2983 structure, only <tt>i32</tt> integer constants are allowed. When indexing
2984 into an array or pointer, only integers of 32 or 64 bits are allowed, and will
2985 be sign extended to 64-bit values.</p>
2987 <p>For example, let's consider a C code fragment and how it gets
2988 compiled to LLVM:</p>
2990 <div class="doc_code">
3003 int *foo(struct ST *s) {
3004 return &s[1].Z.B[5][13];
3009 <p>The LLVM code generated by the GCC frontend is:</p>
3011 <div class="doc_code">
3013 %RT = type { i8 , [10 x [20 x i32]], i8 }
3014 %ST = type { i32, double, %RT }
3016 define i32* %foo(%ST* %s) {
3018 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
3026 <p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
3027 on the pointer type that is being indexed into. <a href="#t_pointer">Pointer</a>
3028 and <a href="#t_array">array</a> types can use a 32-bit or 64-bit
3029 <a href="#t_integer">integer</a> type but the value will always be sign extended
3030 to 64-bits. <a href="#t_struct">Structure</a> types require <tt>i32</tt>
3031 <b>constants</b>.</p>
3033 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
3034 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
3035 }</tt>' type, a structure. The second index indexes into the third element of
3036 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
3037 i8 }</tt>' type, another structure. The third index indexes into the second
3038 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
3039 array. The two dimensions of the array are subscripted into, yielding an
3040 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
3041 to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
3043 <p>Note that it is perfectly legal to index partially through a
3044 structure, returning a pointer to an inner element. Because of this,
3045 the LLVM code for the given testcase is equivalent to:</p>
3048 define i32* %foo(%ST* %s) {
3049 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
3050 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
3051 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
3052 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
3053 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
3058 <p>Note that it is undefined to access an array out of bounds: array and
3059 pointer indexes must always be within the defined bounds of the array type.
3060 The one exception for this rules is zero length arrays. These arrays are
3061 defined to be accessible as variable length arrays, which requires access
3062 beyond the zero'th element.</p>
3064 <p>The getelementptr instruction is often confusing. For some more insight
3065 into how it works, see <a href="GetElementPtr.html">the getelementptr
3071 <i>; yields [12 x i8]*:aptr</i>
3072 %aptr = getelementptr {i32, [12 x i8]}* %sptr, i64 0, i32 1
3076 <!-- ======================================================================= -->
3077 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
3079 <div class="doc_text">
3080 <p>The instructions in this category are the conversion instructions (casting)
3081 which all take a single operand and a type. They perform various bit conversions
3085 <!-- _______________________________________________________________________ -->
3086 <div class="doc_subsubsection">
3087 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
3089 <div class="doc_text">
3093 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
3098 The '<tt>trunc</tt>' instruction truncates its operand to the type <tt>ty2</tt>.
3103 The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
3104 be an <a href="#t_integer">integer</a> type, and a type that specifies the size
3105 and type of the result, which must be an <a href="#t_integer">integer</a>
3106 type. The bit size of <tt>value</tt> must be larger than the bit size of
3107 <tt>ty2</tt>. Equal sized types are not allowed.</p>
3111 The '<tt>trunc</tt>' instruction truncates the high order bits in <tt>value</tt>
3112 and converts the remaining bits to <tt>ty2</tt>. Since the source size must be
3113 larger than the destination size, <tt>trunc</tt> cannot be a <i>no-op cast</i>.
3114 It will always truncate bits.</p>
3118 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
3119 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
3120 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
3124 <!-- _______________________________________________________________________ -->
3125 <div class="doc_subsubsection">
3126 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
3128 <div class="doc_text">
3132 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
3136 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
3141 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
3142 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3143 also be of <a href="#t_integer">integer</a> type. The bit size of the
3144 <tt>value</tt> must be smaller than the bit size of the destination type,
3148 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
3149 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
3151 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
3155 %X = zext i32 257 to i64 <i>; yields i64:257</i>
3156 %Y = zext i1 true to i32 <i>; yields i32:1</i>
3160 <!-- _______________________________________________________________________ -->
3161 <div class="doc_subsubsection">
3162 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
3164 <div class="doc_text">
3168 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
3172 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
3176 The '<tt>sext</tt>' instruction takes a value to cast, which must be of
3177 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3178 also be of <a href="#t_integer">integer</a> type. The bit size of the
3179 <tt>value</tt> must be smaller than the bit size of the destination type,
3184 The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
3185 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size of
3186 the type <tt>ty2</tt>.</p>
3188 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
3192 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
3193 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
3197 <!-- _______________________________________________________________________ -->
3198 <div class="doc_subsubsection">
3199 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
3202 <div class="doc_text">
3207 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
3211 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
3216 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
3217 point</a> value to cast and a <a href="#t_floating">floating point</a> type to
3218 cast it to. The size of <tt>value</tt> must be larger than the size of
3219 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
3220 <i>no-op cast</i>.</p>
3223 <p> The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
3224 <a href="#t_floating">floating point</a> type to a smaller
3225 <a href="#t_floating">floating point</a> type. If the value cannot fit within
3226 the destination type, <tt>ty2</tt>, then the results are undefined.</p>
3230 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
3231 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
3235 <!-- _______________________________________________________________________ -->
3236 <div class="doc_subsubsection">
3237 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
3239 <div class="doc_text">
3243 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
3247 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
3248 floating point value.</p>
3251 <p>The '<tt>fpext</tt>' instruction takes a
3252 <a href="#t_floating">floating point</a> <tt>value</tt> to cast,
3253 and a <a href="#t_floating">floating point</a> type to cast it to. The source
3254 type must be smaller than the destination type.</p>
3257 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
3258 <a href="#t_floating">floating point</a> type to a larger
3259 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
3260 used to make a <i>no-op cast</i> because it always changes bits. Use
3261 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
3265 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
3266 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
3270 <!-- _______________________________________________________________________ -->
3271 <div class="doc_subsubsection">
3272 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
3274 <div class="doc_text">
3278 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
3282 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
3283 unsigned integer equivalent of type <tt>ty2</tt>.
3287 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
3288 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3289 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3290 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3291 vector integer type with the same number of elements as <tt>ty</tt></p>
3294 <p> The '<tt>fptoui</tt>' instruction converts its
3295 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3296 towards zero) unsigned integer value. If the value cannot fit in <tt>ty2</tt>,
3297 the results are undefined.</p>
3301 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
3302 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
3303 %X = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
3307 <!-- _______________________________________________________________________ -->
3308 <div class="doc_subsubsection">
3309 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
3311 <div class="doc_text">
3315 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
3319 <p>The '<tt>fptosi</tt>' instruction converts
3320 <a href="#t_floating">floating point</a> <tt>value</tt> to type <tt>ty2</tt>.
3324 <p> The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
3325 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3326 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3327 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3328 vector integer type with the same number of elements as <tt>ty</tt></p>
3331 <p>The '<tt>fptosi</tt>' instruction converts its
3332 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3333 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
3334 the results are undefined.</p>
3338 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
3339 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
3340 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
3344 <!-- _______________________________________________________________________ -->
3345 <div class="doc_subsubsection">
3346 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
3348 <div class="doc_text">
3352 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3356 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
3357 integer and converts that value to the <tt>ty2</tt> type.</p>
3360 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
3361 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3362 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3363 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3364 floating point type with the same number of elements as <tt>ty</tt></p>
3367 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
3368 integer quantity and converts it to the corresponding floating point value. If
3369 the value cannot fit in the floating point value, the results are undefined.</p>
3373 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
3374 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
3378 <!-- _______________________________________________________________________ -->
3379 <div class="doc_subsubsection">
3380 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
3382 <div class="doc_text">
3386 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3390 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed
3391 integer and converts that value to the <tt>ty2</tt> type.</p>
3394 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
3395 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3396 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3397 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3398 floating point type with the same number of elements as <tt>ty</tt></p>
3401 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed
3402 integer quantity and converts it to the corresponding floating point value. If
3403 the value cannot fit in the floating point value, the results are undefined.</p>
3407 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
3408 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
3412 <!-- _______________________________________________________________________ -->
3413 <div class="doc_subsubsection">
3414 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
3416 <div class="doc_text">
3420 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
3424 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
3425 the integer type <tt>ty2</tt>.</p>
3428 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
3429 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
3430 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.
3433 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
3434 <tt>ty2</tt> by interpreting the pointer value as an integer and either
3435 truncating or zero extending that value to the size of the integer type. If
3436 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
3437 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
3438 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
3443 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
3444 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
3448 <!-- _______________________________________________________________________ -->
3449 <div class="doc_subsubsection">
3450 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
3452 <div class="doc_text">
3456 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
3460 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to
3461 a pointer type, <tt>ty2</tt>.</p>
3464 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
3465 value to cast, and a type to cast it to, which must be a
3466 <a href="#t_pointer">pointer</a> type.
3469 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
3470 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
3471 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
3472 size of a pointer then a truncation is done. If <tt>value</tt> is smaller than
3473 the size of a pointer then a zero extension is done. If they are the same size,
3474 nothing is done (<i>no-op cast</i>).</p>
3478 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
3479 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
3480 %Y = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
3484 <!-- _______________________________________________________________________ -->
3485 <div class="doc_subsubsection">
3486 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
3488 <div class="doc_text">
3492 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
3496 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3497 <tt>ty2</tt> without changing any bits.</p>
3500 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be
3501 a first class value, and a type to cast it to, which must also be a <a
3502 href="#t_firstclass">first class</a> type. The bit sizes of <tt>value</tt>
3503 and the destination type, <tt>ty2</tt>, must be identical. If the source
3504 type is a pointer, the destination type must also be a pointer.</p>
3507 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3508 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
3509 this conversion. The conversion is done as if the <tt>value</tt> had been
3510 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only be
3511 converted to other pointer types with this instruction. To convert pointers to
3512 other types, use the <a href="#i_inttoptr">inttoptr</a> or
3513 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
3517 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
3518 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
3519 %Z = bitcast <2xint> %V to i64; <i>; yields i64: %V</i>
3523 <!-- ======================================================================= -->
3524 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
3525 <div class="doc_text">
3526 <p>The instructions in this category are the "miscellaneous"
3527 instructions, which defy better classification.</p>
3530 <!-- _______________________________________________________________________ -->
3531 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
3533 <div class="doc_text">
3535 <pre> <result> = icmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3538 <p>The '<tt>icmp</tt>' instruction returns a boolean value based on comparison
3539 of its two integer operands.</p>
3541 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
3542 the condition code indicating the kind of comparison to perform. It is not
3543 a value, just a keyword. The possible condition code are:
3545 <li><tt>eq</tt>: equal</li>
3546 <li><tt>ne</tt>: not equal </li>
3547 <li><tt>ugt</tt>: unsigned greater than</li>
3548 <li><tt>uge</tt>: unsigned greater or equal</li>
3549 <li><tt>ult</tt>: unsigned less than</li>
3550 <li><tt>ule</tt>: unsigned less or equal</li>
3551 <li><tt>sgt</tt>: signed greater than</li>
3552 <li><tt>sge</tt>: signed greater or equal</li>
3553 <li><tt>slt</tt>: signed less than</li>
3554 <li><tt>sle</tt>: signed less or equal</li>
3556 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
3557 <a href="#t_pointer">pointer</a> typed. They must also be identical types.</p>
3559 <p>The '<tt>icmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3560 the condition code given as <tt>cond</tt>. The comparison performed always
3561 yields a <a href="#t_primitive">i1</a> result, as follows:
3563 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
3564 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3566 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
3567 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3568 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
3569 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3570 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
3571 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3572 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
3573 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3574 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
3575 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3576 <li><tt>sgt</tt>: interprets the operands as signed values and yields
3577 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3578 <li><tt>sge</tt>: interprets the operands as signed values and yields
3579 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3580 <li><tt>slt</tt>: interprets the operands as signed values and yields
3581 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3582 <li><tt>sle</tt>: interprets the operands as signed values and yields
3583 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3585 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
3586 values are compared as if they were integers.</p>
3589 <pre> <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
3590 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
3591 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
3592 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
3593 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
3594 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
3598 <!-- _______________________________________________________________________ -->
3599 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
3601 <div class="doc_text">
3603 <pre> <result> = fcmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3606 <p>The '<tt>fcmp</tt>' instruction returns a boolean value based on comparison
3607 of its floating point operands.</p>
3609 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
3610 the condition code indicating the kind of comparison to perform. It is not
3611 a value, just a keyword. The possible condition code are:
3613 <li><tt>false</tt>: no comparison, always returns false</li>
3614 <li><tt>oeq</tt>: ordered and equal</li>
3615 <li><tt>ogt</tt>: ordered and greater than </li>
3616 <li><tt>oge</tt>: ordered and greater than or equal</li>
3617 <li><tt>olt</tt>: ordered and less than </li>
3618 <li><tt>ole</tt>: ordered and less than or equal</li>
3619 <li><tt>one</tt>: ordered and not equal</li>
3620 <li><tt>ord</tt>: ordered (no nans)</li>
3621 <li><tt>ueq</tt>: unordered or equal</li>
3622 <li><tt>ugt</tt>: unordered or greater than </li>
3623 <li><tt>uge</tt>: unordered or greater than or equal</li>
3624 <li><tt>ult</tt>: unordered or less than </li>
3625 <li><tt>ule</tt>: unordered or less than or equal</li>
3626 <li><tt>une</tt>: unordered or not equal</li>
3627 <li><tt>uno</tt>: unordered (either nans)</li>
3628 <li><tt>true</tt>: no comparison, always returns true</li>
3630 <p><i>Ordered</i> means that neither operand is a QNAN while
3631 <i>unordered</i> means that either operand may be a QNAN.</p>
3632 <p>The <tt>val1</tt> and <tt>val2</tt> arguments must be
3633 <a href="#t_floating">floating point</a> typed. They must have identical
3636 <p>The '<tt>fcmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3637 the condition code given as <tt>cond</tt>. The comparison performed always
3638 yields a <a href="#t_primitive">i1</a> result, as follows:
3640 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
3641 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3642 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3643 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3644 <tt>var1</tt> is greather than <tt>var2</tt>.</li>
3645 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3646 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3647 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3648 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3649 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3650 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3651 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3652 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3653 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
3654 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
3655 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3656 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
3657 <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3658 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
3659 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3660 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
3661 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3662 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
3663 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3664 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
3665 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3666 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
3667 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
3671 <pre> <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
3672 <result> = icmp one float 4.0, 5.0 <i>; yields: result=true</i>
3673 <result> = icmp olt float 4.0, 5.0 <i>; yields: result=true</i>
3674 <result> = icmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
3678 <!-- _______________________________________________________________________ -->
3679 <div class="doc_subsubsection"> <a name="i_phi">'<tt>phi</tt>'
3680 Instruction</a> </div>
3681 <div class="doc_text">
3683 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
3685 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
3686 the SSA graph representing the function.</p>
3688 <p>The type of the incoming values is specified with the first type
3689 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
3690 as arguments, with one pair for each predecessor basic block of the
3691 current block. Only values of <a href="#t_firstclass">first class</a>
3692 type may be used as the value arguments to the PHI node. Only labels
3693 may be used as the label arguments.</p>
3694 <p>There must be no non-phi instructions between the start of a basic
3695 block and the PHI instructions: i.e. PHI instructions must be first in
3698 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
3699 specified by the pair corresponding to the predecessor basic block that executed
3700 just prior to the current block.</p>
3702 <pre>Loop: ; Infinite loop that counts from 0 on up...<br> %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]<br> %nextindvar = add i32 %indvar, 1<br> br label %Loop<br></pre>
3705 <!-- _______________________________________________________________________ -->
3706 <div class="doc_subsubsection">
3707 <a name="i_select">'<tt>select</tt>' Instruction</a>
3710 <div class="doc_text">
3715 <result> = select i1 <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
3721 The '<tt>select</tt>' instruction is used to choose one value based on a
3722 condition, without branching.
3729 The '<tt>select</tt>' instruction requires a boolean value indicating the condition, and two values of the same <a href="#t_firstclass">first class</a> type.
3735 If the boolean condition evaluates to true, the instruction returns the first
3736 value argument; otherwise, it returns the second value argument.
3742 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
3747 <!-- _______________________________________________________________________ -->
3748 <div class="doc_subsubsection">
3749 <a name="i_call">'<tt>call</tt>' Instruction</a>
3752 <div class="doc_text">
3756 <result> = [tail] call [<a href="#callingconv">cconv</a>] <ty> [<fnty>*] <fnptrval>(<param list>)
3761 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
3765 <p>This instruction requires several arguments:</p>
3769 <p>The optional "tail" marker indicates whether the callee function accesses
3770 any allocas or varargs in the caller. If the "tail" marker is present, the
3771 function call is eligible for tail call optimization. Note that calls may
3772 be marked "tail" even if they do not occur before a <a
3773 href="#i_ret"><tt>ret</tt></a> instruction.
3776 <p>The optional "cconv" marker indicates which <a href="#callingconv">calling
3777 convention</a> the call should use. If none is specified, the call defaults
3778 to using C calling conventions.
3781 <p>'<tt>ty</tt>': the type of the call instruction itself which is also
3782 the type of the return value. Functions that return no value are marked
3783 <tt><a href="#t_void">void</a></tt>.</p>
3786 <p>'<tt>fnty</tt>': shall be the signature of the pointer to function
3787 value being invoked. The argument types must match the types implied by
3788 this signature. This type can be omitted if the function is not varargs
3789 and if the function type does not return a pointer to a function.</p>
3792 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
3793 be invoked. In most cases, this is a direct function invocation, but
3794 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
3795 to function value.</p>
3798 <p>'<tt>function args</tt>': argument list whose types match the
3799 function signature argument types. All arguments must be of
3800 <a href="#t_firstclass">first class</a> type. If the function signature
3801 indicates the function accepts a variable number of arguments, the extra
3802 arguments can be specified.</p>
3808 <p>The '<tt>call</tt>' instruction is used to cause control flow to
3809 transfer to a specified function, with its incoming arguments bound to
3810 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
3811 instruction in the called function, control flow continues with the
3812 instruction after the function call, and the return value of the
3813 function is bound to the result argument. This is a simpler case of
3814 the <a href="#i_invoke">invoke</a> instruction.</p>
3819 %retval = call i32 @test(i32 %argc)
3820 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42);
3821 %X = tail call i32 @foo()
3822 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo()
3823 %Z = call void %foo(i8 97 signext)
3828 <!-- _______________________________________________________________________ -->
3829 <div class="doc_subsubsection">
3830 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
3833 <div class="doc_text">
3838 <resultval> = va_arg <va_list*> <arglist>, <argty>
3843 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
3844 the "variable argument" area of a function call. It is used to implement the
3845 <tt>va_arg</tt> macro in C.</p>
3849 <p>This instruction takes a <tt>va_list*</tt> value and the type of
3850 the argument. It returns a value of the specified argument type and
3851 increments the <tt>va_list</tt> to point to the next argument. The
3852 actual type of <tt>va_list</tt> is target specific.</p>
3856 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
3857 type from the specified <tt>va_list</tt> and causes the
3858 <tt>va_list</tt> to point to the next argument. For more information,
3859 see the variable argument handling <a href="#int_varargs">Intrinsic
3862 <p>It is legal for this instruction to be called in a function which does not
3863 take a variable number of arguments, for example, the <tt>vfprintf</tt>
3866 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
3867 href="#intrinsics">intrinsic function</a> because it takes a type as an
3872 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
3876 <!-- *********************************************************************** -->
3877 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
3878 <!-- *********************************************************************** -->
3880 <div class="doc_text">
3882 <p>LLVM supports the notion of an "intrinsic function". These functions have
3883 well known names and semantics and are required to follow certain restrictions.
3884 Overall, these intrinsics represent an extension mechanism for the LLVM
3885 language that does not require changing all of the transformations in LLVM when
3886 adding to the language (or the bitcode reader/writer, the parser, etc...).</p>
3888 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
3889 prefix is reserved in LLVM for intrinsic names; thus, function names may not
3890 begin with this prefix. Intrinsic functions must always be external functions:
3891 you cannot define the body of intrinsic functions. Intrinsic functions may
3892 only be used in call or invoke instructions: it is illegal to take the address
3893 of an intrinsic function. Additionally, because intrinsic functions are part
3894 of the LLVM language, it is required if any are added that they be documented
3897 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
3898 a family of functions that perform the same operation but on different data
3899 types. Because LLVM can represent over 8 million different integer types,
3900 overloading is used commonly to allow an intrinsic function to operate on any
3901 integer type. One or more of the argument types or the result type can be
3902 overloaded to accept any integer type. Argument types may also be defined as
3903 exactly matching a previous argument's type or the result type. This allows an
3904 intrinsic function which accepts multiple arguments, but needs all of them to
3905 be of the same type, to only be overloaded with respect to a single argument or
3908 <p>Overloaded intrinsics will have the names of its overloaded argument types
3909 encoded into its function name, each preceded by a period. Only those types
3910 which are overloaded result in a name suffix. Arguments whose type is matched
3911 against another type do not. For example, the <tt>llvm.ctpop</tt> function can
3912 take an integer of any width and returns an integer of exactly the same integer
3913 width. This leads to a family of functions such as
3914 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29 %val)</tt>.
3915 Only one type, the return type, is overloaded, and only one type suffix is
3916 required. Because the argument's type is matched against the return type, it
3917 does not require its own name suffix.</p>
3919 <p>To learn how to add an intrinsic function, please see the
3920 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.
3925 <!-- ======================================================================= -->
3926 <div class="doc_subsection">
3927 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
3930 <div class="doc_text">
3932 <p>Variable argument support is defined in LLVM with the <a
3933 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
3934 intrinsic functions. These functions are related to the similarly
3935 named macros defined in the <tt><stdarg.h></tt> header file.</p>
3937 <p>All of these functions operate on arguments that use a
3938 target-specific value type "<tt>va_list</tt>". The LLVM assembly
3939 language reference manual does not define what this type is, so all
3940 transformations should be prepared to handle these functions regardless of
3943 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
3944 instruction and the variable argument handling intrinsic functions are
3947 <div class="doc_code">
3949 define i32 @test(i32 %X, ...) {
3950 ; Initialize variable argument processing
3952 %ap2 = bitcast i8** %ap to i8*
3953 call void @llvm.va_start(i8* %ap2)
3955 ; Read a single integer argument
3956 %tmp = va_arg i8** %ap, i32
3958 ; Demonstrate usage of llvm.va_copy and llvm.va_end
3960 %aq2 = bitcast i8** %aq to i8*
3961 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
3962 call void @llvm.va_end(i8* %aq2)
3964 ; Stop processing of arguments.
3965 call void @llvm.va_end(i8* %ap2)
3969 declare void @llvm.va_start(i8*)
3970 declare void @llvm.va_copy(i8*, i8*)
3971 declare void @llvm.va_end(i8*)
3977 <!-- _______________________________________________________________________ -->
3978 <div class="doc_subsubsection">
3979 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
3983 <div class="doc_text">
3985 <pre> declare void %llvm.va_start(i8* <arglist>)<br></pre>
3987 <P>The '<tt>llvm.va_start</tt>' intrinsic initializes
3988 <tt>*<arglist></tt> for subsequent use by <tt><a
3989 href="#i_va_arg">va_arg</a></tt>.</p>
3993 <P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
3997 <P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
3998 macro available in C. In a target-dependent way, it initializes the
3999 <tt>va_list</tt> element to which the argument points, so that the next call to
4000 <tt>va_arg</tt> will produce the first variable argument passed to the function.
4001 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
4002 last argument of the function as the compiler can figure that out.</p>
4006 <!-- _______________________________________________________________________ -->
4007 <div class="doc_subsubsection">
4008 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
4011 <div class="doc_text">
4013 <pre> declare void @llvm.va_end(i8* <arglist>)<br></pre>
4016 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
4017 which has been initialized previously with <tt><a href="#int_va_start">llvm.va_start</a></tt>
4018 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
4022 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
4026 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
4027 macro available in C. In a target-dependent way, it destroys the
4028 <tt>va_list</tt> element to which the argument points. Calls to <a
4029 href="#int_va_start"><tt>llvm.va_start</tt></a> and <a href="#int_va_copy">
4030 <tt>llvm.va_copy</tt></a> must be matched exactly with calls to
4031 <tt>llvm.va_end</tt>.</p>
4035 <!-- _______________________________________________________________________ -->
4036 <div class="doc_subsubsection">
4037 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
4040 <div class="doc_text">
4045 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
4050 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
4051 from the source argument list to the destination argument list.</p>
4055 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
4056 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
4061 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
4062 macro available in C. In a target-dependent way, it copies the source
4063 <tt>va_list</tt> element into the destination <tt>va_list</tt> element. This
4064 intrinsic is necessary because the <tt><a href="#int_va_start">
4065 llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
4066 example, memory allocation.</p>
4070 <!-- ======================================================================= -->
4071 <div class="doc_subsection">
4072 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
4075 <div class="doc_text">
4078 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
4079 Collection</a> requires the implementation and generation of these intrinsics.
4080 These intrinsics allow identification of <a href="#int_gcroot">GC roots on the
4081 stack</a>, as well as garbage collector implementations that require <a
4082 href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a> barriers.
4083 Front-ends for type-safe garbage collected languages should generate these
4084 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
4085 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
4088 <p>The garbage collection intrinsics only operate on objects in the generic
4089 address space (address space zero).</p>
4093 <!-- _______________________________________________________________________ -->
4094 <div class="doc_subsubsection">
4095 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
4098 <div class="doc_text">
4103 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
4108 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
4109 the code generator, and allows some metadata to be associated with it.</p>
4113 <p>The first argument specifies the address of a stack object that contains the
4114 root pointer. The second pointer (which must be either a constant or a global
4115 value address) contains the meta-data to be associated with the root.</p>
4119 <p>At runtime, a call to this intrinsics stores a null pointer into the "ptrloc"
4120 location. At compile-time, the code generator generates information to allow
4121 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
4122 intrinsic may only be used in a function which <a href="#gc">specifies a GC
4128 <!-- _______________________________________________________________________ -->
4129 <div class="doc_subsubsection">
4130 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
4133 <div class="doc_text">
4138 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
4143 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
4144 locations, allowing garbage collector implementations that require read
4149 <p>The second argument is the address to read from, which should be an address
4150 allocated from the garbage collector. The first object is a pointer to the
4151 start of the referenced object, if needed by the language runtime (otherwise
4156 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
4157 instruction, but may be replaced with substantially more complex code by the
4158 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
4159 may only be used in a function which <a href="#gc">specifies a GC
4165 <!-- _______________________________________________________________________ -->
4166 <div class="doc_subsubsection">
4167 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
4170 <div class="doc_text">
4175 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
4180 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
4181 locations, allowing garbage collector implementations that require write
4182 barriers (such as generational or reference counting collectors).</p>
4186 <p>The first argument is the reference to store, the second is the start of the
4187 object to store it to, and the third is the address of the field of Obj to
4188 store to. If the runtime does not require a pointer to the object, Obj may be
4193 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
4194 instruction, but may be replaced with substantially more complex code by the
4195 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
4196 may only be used in a function which <a href="#gc">specifies a GC
4203 <!-- ======================================================================= -->
4204 <div class="doc_subsection">
4205 <a name="int_codegen">Code Generator Intrinsics</a>
4208 <div class="doc_text">
4210 These intrinsics are provided by LLVM to expose special features that may only
4211 be implemented with code generator support.
4216 <!-- _______________________________________________________________________ -->
4217 <div class="doc_subsubsection">
4218 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
4221 <div class="doc_text">
4225 declare i8 *@llvm.returnaddress(i32 <level>)
4231 The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
4232 target-specific value indicating the return address of the current function
4233 or one of its callers.
4239 The argument to this intrinsic indicates which function to return the address
4240 for. Zero indicates the calling function, one indicates its caller, etc. The
4241 argument is <b>required</b> to be a constant integer value.
4247 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
4248 the return address of the specified call frame, or zero if it cannot be
4249 identified. The value returned by this intrinsic is likely to be incorrect or 0
4250 for arguments other than zero, so it should only be used for debugging purposes.
4254 Note that calling this intrinsic does not prevent function inlining or other
4255 aggressive transformations, so the value returned may not be that of the obvious
4256 source-language caller.
4261 <!-- _______________________________________________________________________ -->
4262 <div class="doc_subsubsection">
4263 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
4266 <div class="doc_text">
4270 declare i8 *@llvm.frameaddress(i32 <level>)
4276 The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
4277 target-specific frame pointer value for the specified stack frame.
4283 The argument to this intrinsic indicates which function to return the frame
4284 pointer for. Zero indicates the calling function, one indicates its caller,
4285 etc. The argument is <b>required</b> to be a constant integer value.
4291 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
4292 the frame address of the specified call frame, or zero if it cannot be
4293 identified. The value returned by this intrinsic is likely to be incorrect or 0
4294 for arguments other than zero, so it should only be used for debugging purposes.
4298 Note that calling this intrinsic does not prevent function inlining or other
4299 aggressive transformations, so the value returned may not be that of the obvious
4300 source-language caller.
4304 <!-- _______________________________________________________________________ -->
4305 <div class="doc_subsubsection">
4306 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
4309 <div class="doc_text">
4313 declare i8 *@llvm.stacksave()
4319 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
4320 the function stack, for use with <a href="#int_stackrestore">
4321 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
4322 features like scoped automatic variable sized arrays in C99.
4328 This intrinsic returns a opaque pointer value that can be passed to <a
4329 href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
4330 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
4331 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
4332 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
4333 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
4334 that were allocated after the <tt>llvm.stacksave</tt> was executed.
4339 <!-- _______________________________________________________________________ -->
4340 <div class="doc_subsubsection">
4341 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
4344 <div class="doc_text">
4348 declare void @llvm.stackrestore(i8 * %ptr)
4354 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
4355 the function stack to the state it was in when the corresponding <a
4356 href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
4357 useful for implementing language features like scoped automatic variable sized
4364 See the description for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.
4370 <!-- _______________________________________________________________________ -->
4371 <div class="doc_subsubsection">
4372 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
4375 <div class="doc_text">
4379 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>)
4386 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
4387 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
4389 effect on the behavior of the program but can change its performance
4396 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
4397 determining if the fetch should be for a read (0) or write (1), and
4398 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
4399 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
4400 <tt>locality</tt> arguments must be constant integers.
4406 This intrinsic does not modify the behavior of the program. In particular,
4407 prefetches cannot trap and do not produce a value. On targets that support this
4408 intrinsic, the prefetch can provide hints to the processor cache for better
4414 <!-- _______________________________________________________________________ -->
4415 <div class="doc_subsubsection">
4416 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
4419 <div class="doc_text">
4423 declare void @llvm.pcmarker(i32 <id>)
4430 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
4432 code to simulators and other tools. The method is target specific, but it is
4433 expected that the marker will use exported symbols to transmit the PC of the marker.
4434 The marker makes no guarantees that it will remain with any specific instruction
4435 after optimizations. It is possible that the presence of a marker will inhibit
4436 optimizations. The intended use is to be inserted after optimizations to allow
4437 correlations of simulation runs.
4443 <tt>id</tt> is a numerical id identifying the marker.
4449 This intrinsic does not modify the behavior of the program. Backends that do not
4450 support this intrinisic may ignore it.
4455 <!-- _______________________________________________________________________ -->
4456 <div class="doc_subsubsection">
4457 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
4460 <div class="doc_text">
4464 declare i64 @llvm.readcyclecounter( )
4471 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
4472 counter register (or similar low latency, high accuracy clocks) on those targets
4473 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
4474 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
4475 should only be used for small timings.
4481 When directly supported, reading the cycle counter should not modify any memory.
4482 Implementations are allowed to either return a application specific value or a
4483 system wide value. On backends without support, this is lowered to a constant 0.
4488 <!-- ======================================================================= -->
4489 <div class="doc_subsection">
4490 <a name="int_libc">Standard C Library Intrinsics</a>
4493 <div class="doc_text">
4495 LLVM provides intrinsics for a few important standard C library functions.
4496 These intrinsics allow source-language front-ends to pass information about the
4497 alignment of the pointer arguments to the code generator, providing opportunity
4498 for more efficient code generation.
4503 <!-- _______________________________________________________________________ -->
4504 <div class="doc_subsubsection">
4505 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
4508 <div class="doc_text">
4512 declare void @llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
4513 i32 <len>, i32 <align>)
4514 declare void @llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
4515 i64 <len>, i32 <align>)
4521 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4522 location to the destination location.
4526 Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
4527 intrinsics do not return a value, and takes an extra alignment argument.
4533 The first argument is a pointer to the destination, the second is a pointer to
4534 the source. The third argument is an integer argument
4535 specifying the number of bytes to copy, and the fourth argument is the alignment
4536 of the source and destination locations.
4540 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4541 the caller guarantees that both the source and destination pointers are aligned
4548 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4549 location to the destination location, which are not allowed to overlap. It
4550 copies "len" bytes of memory over. If the argument is known to be aligned to
4551 some boundary, this can be specified as the fourth argument, otherwise it should
4557 <!-- _______________________________________________________________________ -->
4558 <div class="doc_subsubsection">
4559 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
4562 <div class="doc_text">
4566 declare void @llvm.memmove.i32(i8 * <dest>, i8 * <src>,
4567 i32 <len>, i32 <align>)
4568 declare void @llvm.memmove.i64(i8 * <dest>, i8 * <src>,
4569 i64 <len>, i32 <align>)
4575 The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
4576 location to the destination location. It is similar to the
4577 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to overlap.
4581 Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
4582 intrinsics do not return a value, and takes an extra alignment argument.
4588 The first argument is a pointer to the destination, the second is a pointer to
4589 the source. The third argument is an integer argument
4590 specifying the number of bytes to copy, and the fourth argument is the alignment
4591 of the source and destination locations.
4595 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4596 the caller guarantees that the source and destination pointers are aligned to
4603 The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
4604 location to the destination location, which may overlap. It
4605 copies "len" bytes of memory over. If the argument is known to be aligned to
4606 some boundary, this can be specified as the fourth argument, otherwise it should
4612 <!-- _______________________________________________________________________ -->
4613 <div class="doc_subsubsection">
4614 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
4617 <div class="doc_text">
4621 declare void @llvm.memset.i32(i8 * <dest>, i8 <val>,
4622 i32 <len>, i32 <align>)
4623 declare void @llvm.memset.i64(i8 * <dest>, i8 <val>,
4624 i64 <len>, i32 <align>)
4630 The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
4635 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
4636 does not return a value, and takes an extra alignment argument.
4642 The first argument is a pointer to the destination to fill, the second is the
4643 byte value to fill it with, the third argument is an integer
4644 argument specifying the number of bytes to fill, and the fourth argument is the
4645 known alignment of destination location.
4649 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4650 the caller guarantees that the destination pointer is aligned to that boundary.
4656 The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
4658 destination location. If the argument is known to be aligned to some boundary,
4659 this can be specified as the fourth argument, otherwise it should be set to 0 or
4665 <!-- _______________________________________________________________________ -->
4666 <div class="doc_subsubsection">
4667 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
4670 <div class="doc_text">
4673 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
4674 floating point or vector of floating point type. Not all targets support all
4677 declare float @llvm.sqrt.f32(float %Val)
4678 declare double @llvm.sqrt.f64(double %Val)
4679 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
4680 declare fp128 @llvm.sqrt.f128(fp128 %Val)
4681 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
4687 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
4688 returning the same value as the libm '<tt>sqrt</tt>' functions would. Unlike
4689 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
4690 negative numbers other than -0.0 (which allows for better optimization, because
4691 there is no need to worry about errno being set). <tt>llvm.sqrt(-0.0)</tt> is
4692 defined to return -0.0 like IEEE sqrt.
4698 The argument and return value are floating point numbers of the same type.
4704 This function returns the sqrt of the specified operand if it is a nonnegative
4705 floating point number.
4709 <!-- _______________________________________________________________________ -->
4710 <div class="doc_subsubsection">
4711 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
4714 <div class="doc_text">
4717 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
4718 floating point or vector of floating point type. Not all targets support all
4721 declare float @llvm.powi.f32(float %Val, i32 %power)
4722 declare double @llvm.powi.f64(double %Val, i32 %power)
4723 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
4724 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
4725 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
4731 The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
4732 specified (positive or negative) power. The order of evaluation of
4733 multiplications is not defined. When a vector of floating point type is
4734 used, the second argument remains a scalar integer value.
4740 The second argument is an integer power, and the first is a value to raise to
4747 This function returns the first value raised to the second power with an
4748 unspecified sequence of rounding operations.</p>
4751 <!-- _______________________________________________________________________ -->
4752 <div class="doc_subsubsection">
4753 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
4756 <div class="doc_text">
4759 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
4760 floating point or vector of floating point type. Not all targets support all
4763 declare float @llvm.sin.f32(float %Val)
4764 declare double @llvm.sin.f64(double %Val)
4765 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
4766 declare fp128 @llvm.sin.f128(fp128 %Val)
4767 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
4773 The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.
4779 The argument and return value are floating point numbers of the same type.
4785 This function returns the sine of the specified operand, returning the
4786 same values as the libm <tt>sin</tt> functions would, and handles error
4787 conditions in the same way.</p>
4790 <!-- _______________________________________________________________________ -->
4791 <div class="doc_subsubsection">
4792 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
4795 <div class="doc_text">
4798 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
4799 floating point or vector of floating point type. Not all targets support all
4802 declare float @llvm.cos.f32(float %Val)
4803 declare double @llvm.cos.f64(double %Val)
4804 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
4805 declare fp128 @llvm.cos.f128(fp128 %Val)
4806 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
4812 The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.
4818 The argument and return value are floating point numbers of the same type.
4824 This function returns the cosine of the specified operand, returning the
4825 same values as the libm <tt>cos</tt> functions would, and handles error
4826 conditions in the same way.</p>
4829 <!-- _______________________________________________________________________ -->
4830 <div class="doc_subsubsection">
4831 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
4834 <div class="doc_text">
4837 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
4838 floating point or vector of floating point type. Not all targets support all
4841 declare float @llvm.pow.f32(float %Val, float %Power)
4842 declare double @llvm.pow.f64(double %Val, double %Power)
4843 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
4844 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
4845 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
4851 The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
4852 specified (positive or negative) power.
4858 The second argument is a floating point power, and the first is a value to
4859 raise to that power.
4865 This function returns the first value raised to the second power,
4867 same values as the libm <tt>pow</tt> functions would, and handles error
4868 conditions in the same way.</p>
4872 <!-- ======================================================================= -->
4873 <div class="doc_subsection">
4874 <a name="int_manip">Bit Manipulation Intrinsics</a>
4877 <div class="doc_text">
4879 LLVM provides intrinsics for a few important bit manipulation operations.
4880 These allow efficient code generation for some algorithms.
4885 <!-- _______________________________________________________________________ -->
4886 <div class="doc_subsubsection">
4887 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
4890 <div class="doc_text">
4893 <p>This is an overloaded intrinsic function. You can use bswap on any integer
4894 type that is an even number of bytes (i.e. BitWidth % 16 == 0).
4896 declare i16 @llvm.bswap.i16(i16 <id>)
4897 declare i32 @llvm.bswap.i32(i32 <id>)
4898 declare i64 @llvm.bswap.i64(i64 <id>)
4904 The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
4905 values with an even number of bytes (positive multiple of 16 bits). These are
4906 useful for performing operations on data that is not in the target's native
4913 The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
4914 and low byte of the input i16 swapped. Similarly, the <tt>llvm.bswap.i32</tt>
4915 intrinsic returns an i32 value that has the four bytes of the input i32
4916 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the returned
4917 i32 will have its bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i48</tt>,
4918 <tt>llvm.bswap.i64</tt> and other intrinsics extend this concept to
4919 additional even-byte lengths (6 bytes, 8 bytes and more, respectively).
4924 <!-- _______________________________________________________________________ -->
4925 <div class="doc_subsubsection">
4926 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
4929 <div class="doc_text">
4932 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
4933 width. Not all targets support all bit widths however.
4935 declare i8 @llvm.ctpop.i8 (i8 <src>)
4936 declare i16 @llvm.ctpop.i16(i16 <src>)
4937 declare i32 @llvm.ctpop.i32(i32 <src>)
4938 declare i64 @llvm.ctpop.i64(i64 <src>)
4939 declare i256 @llvm.ctpop.i256(i256 <src>)
4945 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
4952 The only argument is the value to be counted. The argument may be of any
4953 integer type. The return type must match the argument type.
4959 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
4963 <!-- _______________________________________________________________________ -->
4964 <div class="doc_subsubsection">
4965 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
4968 <div class="doc_text">
4971 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
4972 integer bit width. Not all targets support all bit widths however.
4974 declare i8 @llvm.ctlz.i8 (i8 <src>)
4975 declare i16 @llvm.ctlz.i16(i16 <src>)
4976 declare i32 @llvm.ctlz.i32(i32 <src>)
4977 declare i64 @llvm.ctlz.i64(i64 <src>)
4978 declare i256 @llvm.ctlz.i256(i256 <src>)
4984 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
4985 leading zeros in a variable.
4991 The only argument is the value to be counted. The argument may be of any
4992 integer type. The return type must match the argument type.
4998 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
4999 in a variable. If the src == 0 then the result is the size in bits of the type
5000 of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.
5006 <!-- _______________________________________________________________________ -->
5007 <div class="doc_subsubsection">
5008 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
5011 <div class="doc_text">
5014 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
5015 integer bit width. Not all targets support all bit widths however.
5017 declare i8 @llvm.cttz.i8 (i8 <src>)
5018 declare i16 @llvm.cttz.i16(i16 <src>)
5019 declare i32 @llvm.cttz.i32(i32 <src>)
5020 declare i64 @llvm.cttz.i64(i64 <src>)
5021 declare i256 @llvm.cttz.i256(i256 <src>)
5027 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
5034 The only argument is the value to be counted. The argument may be of any
5035 integer type. The return type must match the argument type.
5041 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
5042 in a variable. If the src == 0 then the result is the size in bits of the type
5043 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
5047 <!-- _______________________________________________________________________ -->
5048 <div class="doc_subsubsection">
5049 <a name="int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic</a>
5052 <div class="doc_text">
5055 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.select</tt>
5056 on any integer bit width.
5058 declare i17 @llvm.part.select.i17 (i17 %val, i32 %loBit, i32 %hiBit)
5059 declare i29 @llvm.part.select.i29 (i29 %val, i32 %loBit, i32 %hiBit)
5063 <p>The '<tt>llvm.part.select</tt>' family of intrinsic functions selects a
5064 range of bits from an integer value and returns them in the same bit width as
5065 the original value.</p>
5068 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5069 any bit width but they must have the same bit width. The second and third
5070 arguments must be <tt>i32</tt> type since they specify only a bit index.</p>
5073 <p>The operation of the '<tt>llvm.part.select</tt>' intrinsic has two modes
5074 of operation: forwards and reverse. If <tt>%loBit</tt> is greater than
5075 <tt>%hiBits</tt> then the intrinsic operates in reverse mode. Otherwise it
5076 operates in forward mode.</p>
5077 <p>In forward mode, this intrinsic is the equivalent of shifting <tt>%val</tt>
5078 right by <tt>%loBit</tt> bits and then ANDing it with a mask with
5079 only the <tt>%hiBit - %loBit</tt> bits set, as follows:</p>
5081 <li>The <tt>%val</tt> is shifted right (LSHR) by the number of bits specified
5082 by <tt>%loBits</tt>. This normalizes the value to the low order bits.</li>
5083 <li>The <tt>%loBits</tt> value is subtracted from the <tt>%hiBits</tt> value
5084 to determine the number of bits to retain.</li>
5085 <li>A mask of the retained bits is created by shifting a -1 value.</li>
5086 <li>The mask is ANDed with <tt>%val</tt> to produce the result.
5088 <p>In reverse mode, a similar computation is made except that the bits are
5089 returned in the reverse order. So, for example, if <tt>X</tt> has the value
5090 <tt>i16 0x0ACF (101011001111)</tt> and we apply
5091 <tt>part.select(i16 X, 8, 3)</tt> to it, we get back the value
5092 <tt>i16 0x0026 (000000100110)</tt>.</p>
5095 <div class="doc_subsubsection">
5096 <a name="int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic</a>
5099 <div class="doc_text">
5102 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.set</tt>
5103 on any integer bit width.
5105 declare i17 @llvm.part.set.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
5106 declare i29 @llvm.part.set.i29.i9 (i29 %val, i9 %repl, i32 %lo, i32 %hi)
5110 <p>The '<tt>llvm.part.set</tt>' family of intrinsic functions replaces a range
5111 of bits in an integer value with another integer value. It returns the integer
5112 with the replaced bits.</p>
5115 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5116 any bit width but they must have the same bit width. <tt>%val</tt> is the value
5117 whose bits will be replaced. The second argument, <tt>%repl</tt> may be an
5118 integer of any bit width. The third and fourth arguments must be <tt>i32</tt>
5119 type since they specify only a bit index.</p>
5122 <p>The operation of the '<tt>llvm.part.set</tt>' intrinsic has two modes
5123 of operation: forwards and reverse. If <tt>%lo</tt> is greater than
5124 <tt>%hi</tt> then the intrinsic operates in reverse mode. Otherwise it
5125 operates in forward mode.</p>
5126 <p>For both modes, the <tt>%repl</tt> value is prepared for use by either
5127 truncating it down to the size of the replacement area or zero extending it
5128 up to that size.</p>
5129 <p>In forward mode, the bits between <tt>%lo</tt> and <tt>%hi</tt> (inclusive)
5130 are replaced with corresponding bits from <tt>%repl</tt>. That is the 0th bit
5131 in <tt>%repl</tt> replaces the <tt>%lo</tt>th bit in <tt>%val</tt> and etc. up
5132 to the <tt>%hi</tt>th bit.
5133 <p>In reverse mode, a similar computation is made except that the bits are
5134 reversed. That is, the <tt>0</tt>th bit in <tt>%repl</tt> replaces the
5135 <tt>%hi</tt> bit in <tt>%val</tt> and etc. down to the <tt>%lo</tt>th bit.
5138 llvm.part.set(0xFFFF, 0, 4, 7) -> 0xFF0F
5139 llvm.part.set(0xFFFF, 0, 7, 4) -> 0xFF0F
5140 llvm.part.set(0xFFFF, 1, 7, 4) -> 0xFF8F
5141 llvm.part.set(0xFFFF, F, 8, 3) -> 0xFFE7
5142 llvm.part.set(0xFFFF, 0, 3, 8) -> 0xFE07
5146 <!-- ======================================================================= -->
5147 <div class="doc_subsection">
5148 <a name="int_debugger">Debugger Intrinsics</a>
5151 <div class="doc_text">
5153 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
5154 are described in the <a
5155 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
5156 Debugging</a> document.
5161 <!-- ======================================================================= -->
5162 <div class="doc_subsection">
5163 <a name="int_eh">Exception Handling Intrinsics</a>
5166 <div class="doc_text">
5167 <p> The LLVM exception handling intrinsics (which all start with
5168 <tt>llvm.eh.</tt> prefix), are described in the <a
5169 href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
5170 Handling</a> document. </p>
5173 <!-- ======================================================================= -->
5174 <div class="doc_subsection">
5175 <a name="int_trampoline">Trampoline Intrinsic</a>
5178 <div class="doc_text">
5180 This intrinsic makes it possible to excise one parameter, marked with
5181 the <tt>nest</tt> attribute, from a function. The result is a callable
5182 function pointer lacking the nest parameter - the caller does not need
5183 to provide a value for it. Instead, the value to use is stored in
5184 advance in a "trampoline", a block of memory usually allocated
5185 on the stack, which also contains code to splice the nest value into the
5186 argument list. This is used to implement the GCC nested function address
5190 For example, if the function is
5191 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
5192 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as follows:</p>
5194 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
5195 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
5196 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
5197 %fp = bitcast i8* %p to i32 (i32, i32)*
5199 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
5200 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
5203 <!-- _______________________________________________________________________ -->
5204 <div class="doc_subsubsection">
5205 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
5207 <div class="doc_text">
5210 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
5214 This fills the memory pointed to by <tt>tramp</tt> with code
5215 and returns a function pointer suitable for executing it.
5219 The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
5220 pointers. The <tt>tramp</tt> argument must point to a sufficiently large
5221 and sufficiently aligned block of memory; this memory is written to by the
5222 intrinsic. Note that the size and the alignment are target-specific - LLVM
5223 currently provides no portable way of determining them, so a front-end that
5224 generates this intrinsic needs to have some target-specific knowledge.
5225 The <tt>func</tt> argument must hold a function bitcast to an <tt>i8*</tt>.
5229 The block of memory pointed to by <tt>tramp</tt> is filled with target
5230 dependent code, turning it into a function. A pointer to this function is
5231 returned, but needs to be bitcast to an
5232 <a href="#int_trampoline">appropriate function pointer type</a>
5233 before being called. The new function's signature is the same as that of
5234 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
5235 removed. At most one such <tt>nest</tt> argument is allowed, and it must be
5236 of pointer type. Calling the new function is equivalent to calling
5237 <tt>func</tt> with the same argument list, but with <tt>nval</tt> used for the
5238 missing <tt>nest</tt> argument. If, after calling
5239 <tt>llvm.init.trampoline</tt>, the memory pointed to by <tt>tramp</tt> is
5240 modified, then the effect of any later call to the returned function pointer is
5245 <!-- ======================================================================= -->
5246 <div class="doc_subsection">
5247 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
5250 <div class="doc_text">
5252 These intrinsic functions expand the "universal IR" of LLVM to represent
5253 hardware constructs for atomic operations and memory synchronization. This
5254 provides an interface to the hardware, not an interface to the programmer. It
5255 is aimed at a low enough level to allow any programming models or APIs which
5256 need atomic behaviors to map cleanly onto it. It is also modeled primarily on
5257 hardware behavior. Just as hardware provides a "universal IR" for source
5258 languages, it also provides a starting point for developing a "universal"
5259 atomic operation and synchronization IR.
5262 These do <em>not</em> form an API such as high-level threading libraries,
5263 software transaction memory systems, atomic primitives, and intrinsic
5264 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
5265 application libraries. The hardware interface provided by LLVM should allow
5266 a clean implementation of all of these APIs and parallel programming models.
5267 No one model or paradigm should be selected above others unless the hardware
5268 itself ubiquitously does so.
5273 <!-- _______________________________________________________________________ -->
5274 <div class="doc_subsubsection">
5275 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
5277 <div class="doc_text">
5280 declare void @llvm.memory.barrier( i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>,
5286 The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
5287 specific pairs of memory access types.
5291 The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
5292 The first four arguments enables a specific barrier as listed below. The fith
5293 argument specifies that the barrier applies to io or device or uncached memory.
5297 <li><tt>ll</tt>: load-load barrier</li>
5298 <li><tt>ls</tt>: load-store barrier</li>
5299 <li><tt>sl</tt>: store-load barrier</li>
5300 <li><tt>ss</tt>: store-store barrier</li>
5301 <li><tt>device</tt>: barrier applies to device and uncached memory also.
5305 This intrinsic causes the system to enforce some ordering constraints upon
5306 the loads and stores of the program. This barrier does not indicate
5307 <em>when</em> any events will occur, it only enforces an <em>order</em> in
5308 which they occur. For any of the specified pairs of load and store operations
5309 (f.ex. load-load, or store-load), all of the first operations preceding the
5310 barrier will complete before any of the second operations succeeding the
5311 barrier begin. Specifically the semantics for each pairing is as follows:
5314 <li><tt>ll</tt>: All loads before the barrier must complete before any load
5315 after the barrier begins.</li>
5317 <li><tt>ls</tt>: All loads before the barrier must complete before any
5318 store after the barrier begins.</li>
5319 <li><tt>ss</tt>: All stores before the barrier must complete before any
5320 store after the barrier begins.</li>
5321 <li><tt>sl</tt>: All stores before the barrier must complete before any
5322 load after the barrier begins.</li>
5325 These semantics are applied with a logical "and" behavior when more than one
5326 is enabled in a single memory barrier intrinsic.
5329 Backends may implement stronger barriers than those requested when they do not
5330 support as fine grained a barrier as requested. Some architectures do not
5331 need all types of barriers and on such architectures, these become noops.
5338 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
5339 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
5340 <i>; guarantee the above finishes</i>
5341 store i32 8, %ptr <i>; before this begins</i>
5345 <!-- _______________________________________________________________________ -->
5346 <div class="doc_subsubsection">
5347 <a name="int_atomic_lcs">'<tt>llvm.atomic.lcs.*</tt>' Intrinsic</a>
5349 <div class="doc_text">
5352 This is an overloaded intrinsic. You can use <tt>llvm.atomic.lcs</tt> on any
5353 integer bit width. Not all targets support all bit widths however.</p>
5356 declare i8 @llvm.atomic.lcs.i8( i8* <ptr>, i8 <cmp>, i8 <val> )
5357 declare i16 @llvm.atomic.lcs.i16( i16* <ptr>, i16 <cmp>, i16 <val> )
5358 declare i32 @llvm.atomic.lcs.i32( i32* <ptr>, i32 <cmp>, i32 <val> )
5359 declare i64 @llvm.atomic.lcs.i64( i64* <ptr>, i64 <cmp>, i64 <val> )
5364 This loads a value in memory and compares it to a given value. If they are
5365 equal, it stores a new value into the memory.
5369 The <tt>llvm.atomic.lcs</tt> intrinsic takes three arguments. The result as
5370 well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
5371 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
5372 this integer type. While any bit width integer may be used, targets may only
5373 lower representations they support in hardware.
5378 This entire intrinsic must be executed atomically. It first loads the value
5379 in memory pointed to by <tt>ptr</tt> and compares it with the value
5380 <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the memory. The
5381 loaded value is yielded in all cases. This provides the equivalent of an
5382 atomic compare-and-swap operation within the SSA framework.
5390 %val1 = add i32 4, 4
5391 %result1 = call i32 @llvm.atomic.lcs.i32( i32* %ptr, i32 4, %val1 )
5392 <i>; yields {i32}:result1 = 4</i>
5393 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
5394 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
5396 %val2 = add i32 1, 1
5397 %result2 = call i32 @llvm.atomic.lcs.i32( i32* %ptr, i32 5, %val2 )
5398 <i>; yields {i32}:result2 = 8</i>
5399 %stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
5401 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
5405 <!-- _______________________________________________________________________ -->
5406 <div class="doc_subsubsection">
5407 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
5409 <div class="doc_text">
5413 This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
5414 integer bit width. Not all targets support all bit widths however.</p>
5416 declare i8 @llvm.atomic.swap.i8( i8* <ptr>, i8 <val> )
5417 declare i16 @llvm.atomic.swap.i16( i16* <ptr>, i16 <val> )
5418 declare i32 @llvm.atomic.swap.i32( i32* <ptr>, i32 <val> )
5419 declare i64 @llvm.atomic.swap.i64( i64* <ptr>, i64 <val> )
5424 This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
5425 the value from memory. It then stores the value in <tt>val</tt> in the memory
5431 The <tt>llvm.atomic.ls</tt> intrinsic takes two arguments. Both the
5432 <tt>val</tt> argument and the result must be integers of the same bit width.
5433 The first argument, <tt>ptr</tt>, must be a pointer to a value of this
5434 integer type. The targets may only lower integer representations they
5439 This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
5440 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
5441 equivalent of an atomic swap operation within the SSA framework.
5449 %val1 = add i32 4, 4
5450 %result1 = call i32 @llvm.atomic.swap.i32( i32* %ptr, i32 %val1 )
5451 <i>; yields {i32}:result1 = 4</i>
5452 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
5453 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
5455 %val2 = add i32 1, 1
5456 %result2 = call i32 @llvm.atomic.swap.i32( i32* %ptr, i32 %val2 )
5457 <i>; yields {i32}:result2 = 8</i>
5459 %stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
5460 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
5464 <!-- _______________________________________________________________________ -->
5465 <div class="doc_subsubsection">
5466 <a name="int_atomic_las">'<tt>llvm.atomic.las.*</tt>' Intrinsic</a>
5469 <div class="doc_text">
5472 This is an overloaded intrinsic. You can use <tt>llvm.atomic.las</tt> on any
5473 integer bit width. Not all targets support all bit widths however.</p>
5475 declare i8 @llvm.atomic.las.i8.( i8* <ptr>, i8 <delta> )
5476 declare i16 @llvm.atomic.las.i16.( i16* <ptr>, i16 <delta> )
5477 declare i32 @llvm.atomic.las.i32.( i32* <ptr>, i32 <delta> )
5478 declare i64 @llvm.atomic.las.i64.( i64* <ptr>, i64 <delta> )
5483 This intrinsic adds <tt>delta</tt> to the value stored in memory at
5484 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
5489 The intrinsic takes two arguments, the first a pointer to an integer value
5490 and the second an integer value. The result is also an integer value. These
5491 integer types can have any bit width, but they must all have the same bit
5492 width. The targets may only lower integer representations they support.
5496 This intrinsic does a series of operations atomically. It first loads the
5497 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
5498 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
5505 %result1 = call i32 @llvm.atomic.las.i32( i32* %ptr, i32 4 )
5506 <i>; yields {i32}:result1 = 4</i>
5507 %result2 = call i32 @llvm.atomic.las.i32( i32* %ptr, i32 2 )
5508 <i>; yields {i32}:result2 = 8</i>
5509 %result3 = call i32 @llvm.atomic.las.i32( i32* %ptr, i32 5 )
5510 <i>; yields {i32}:result3 = 10</i>
5511 %memval = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
5516 <!-- ======================================================================= -->
5517 <div class="doc_subsection">
5518 <a name="int_general">General Intrinsics</a>
5521 <div class="doc_text">
5522 <p> This class of intrinsics is designed to be generic and has
5523 no specific purpose. </p>
5526 <!-- _______________________________________________________________________ -->
5527 <div class="doc_subsubsection">
5528 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
5531 <div class="doc_text">
5535 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
5541 The '<tt>llvm.var.annotation</tt>' intrinsic
5547 The first argument is a pointer to a value, the second is a pointer to a
5548 global string, the third is a pointer to a global string which is the source
5549 file name, and the last argument is the line number.
5555 This intrinsic allows annotation of local variables with arbitrary strings.
5556 This can be useful for special purpose optimizations that want to look for these
5557 annotations. These have no other defined use, they are ignored by code
5558 generation and optimization.
5562 <!-- _______________________________________________________________________ -->
5563 <div class="doc_subsubsection">
5564 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
5567 <div class="doc_text">
5570 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
5571 any integer bit width.
5574 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
5575 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
5576 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
5577 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
5578 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
5584 The '<tt>llvm.annotation</tt>' intrinsic.
5590 The first argument is an integer value (result of some expression),
5591 the second is a pointer to a global string, the third is a pointer to a global
5592 string which is the source file name, and the last argument is the line number.
5593 It returns the value of the first argument.
5599 This intrinsic allows annotations to be put on arbitrary expressions
5600 with arbitrary strings. This can be useful for special purpose optimizations
5601 that want to look for these annotations. These have no other defined use, they
5602 are ignored by code generation and optimization.
5605 <!-- _______________________________________________________________________ -->
5606 <div class="doc_subsubsection">
5607 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
5610 <div class="doc_text">
5614 declare void @llvm.trap()
5620 The '<tt>llvm.trap</tt>' intrinsic
5632 This intrinsics is lowered to the target dependent trap instruction. If the
5633 target does not have a trap instruction, this intrinsic will be lowered to the
5634 call of the abort() function.
5638 <!-- *********************************************************************** -->
5641 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
5642 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
5643 <a href="http://validator.w3.org/check/referer"><img
5644 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
5646 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
5647 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
5648 Last modified: $Date$