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="#paramattrs">Parameter Attributes</a></li>
28 <li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
29 <li><a href="#datalayout">Data Layout</a></li>
32 <li><a href="#typesystem">Type System</a>
34 <li><a href="#t_primitive">Primitive Types</a>
36 <li><a href="#t_classifications">Type Classifications</a></li>
39 <li><a href="#t_derived">Derived Types</a>
41 <li><a href="#t_array">Array Type</a></li>
42 <li><a href="#t_function">Function Type</a></li>
43 <li><a href="#t_pointer">Pointer Type</a></li>
44 <li><a href="#t_struct">Structure Type</a></li>
45 <li><a href="#t_pstruct">Packed Structure Type</a></li>
46 <li><a href="#t_vector">Vector Type</a></li>
47 <li><a href="#t_opaque">Opaque Type</a></li>
52 <li><a href="#constants">Constants</a>
54 <li><a href="#simpleconstants">Simple Constants</a>
55 <li><a href="#aggregateconstants">Aggregate Constants</a>
56 <li><a href="#globalconstants">Global Variable and Function Addresses</a>
57 <li><a href="#undefvalues">Undefined Values</a>
58 <li><a href="#constantexprs">Constant Expressions</a>
61 <li><a href="#othervalues">Other Values</a>
63 <li><a href="#inlineasm">Inline Assembler Expressions</a>
66 <li><a href="#instref">Instruction Reference</a>
68 <li><a href="#terminators">Terminator Instructions</a>
70 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
71 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
72 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
73 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
74 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
75 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
78 <li><a href="#binaryops">Binary Operations</a>
80 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
81 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
82 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
83 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
84 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
85 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
86 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
87 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
88 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
91 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
93 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
94 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
95 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
96 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
97 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
98 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
101 <li><a href="#vectorops">Vector Operations</a>
103 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
104 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
105 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
108 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
110 <li><a href="#i_malloc">'<tt>malloc</tt>' Instruction</a></li>
111 <li><a href="#i_free">'<tt>free</tt>' Instruction</a></li>
112 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
113 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
114 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
115 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
118 <li><a href="#convertops">Conversion Operations</a>
120 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
121 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
122 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
123 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
124 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
125 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
126 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
127 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
128 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
129 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
130 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
131 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
133 <li><a href="#otherops">Other Operations</a>
135 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
136 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
137 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
138 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
139 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
140 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
145 <li><a href="#intrinsics">Intrinsic Functions</a>
147 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
149 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
150 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
151 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
154 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
156 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
157 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
158 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
161 <li><a href="#int_codegen">Code Generator Intrinsics</a>
163 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
164 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
165 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
166 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
167 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
168 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
169 <li><a href="#int_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
172 <li><a href="#int_libc">Standard C Library Intrinsics</a>
174 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
175 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
176 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
177 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
178 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
181 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
183 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
184 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
185 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
186 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
187 <li><a href="#int_part_select">'<tt>llvm.part_select.*</tt>' Intrinsic </a></li>
190 <li><a href="#int_debugger">Debugger intrinsics</a></li>
191 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
196 <div class="doc_author">
197 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
198 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
201 <!-- *********************************************************************** -->
202 <div class="doc_section"> <a name="abstract">Abstract </a></div>
203 <!-- *********************************************************************** -->
205 <div class="doc_text">
206 <p>This document is a reference manual for the LLVM assembly language.
207 LLVM is an SSA based representation that provides type safety,
208 low-level operations, flexibility, and the capability of representing
209 'all' high-level languages cleanly. It is the common code
210 representation used throughout all phases of the LLVM compilation
214 <!-- *********************************************************************** -->
215 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
216 <!-- *********************************************************************** -->
218 <div class="doc_text">
220 <p>The LLVM code representation is designed to be used in three
221 different forms: as an in-memory compiler IR, as an on-disk bytecode
222 representation (suitable for fast loading by a Just-In-Time compiler),
223 and as a human readable assembly language representation. This allows
224 LLVM to provide a powerful intermediate representation for efficient
225 compiler transformations and analysis, while providing a natural means
226 to debug and visualize the transformations. The three different forms
227 of LLVM are all equivalent. This document describes the human readable
228 representation and notation.</p>
230 <p>The LLVM representation aims to be light-weight and low-level
231 while being expressive, typed, and extensible at the same time. It
232 aims to be a "universal IR" of sorts, by being at a low enough level
233 that high-level ideas may be cleanly mapped to it (similar to how
234 microprocessors are "universal IR's", allowing many source languages to
235 be mapped to them). By providing type information, LLVM can be used as
236 the target of optimizations: for example, through pointer analysis, it
237 can be proven that a C automatic variable is never accessed outside of
238 the current function... allowing it to be promoted to a simple SSA
239 value instead of a memory location.</p>
243 <!-- _______________________________________________________________________ -->
244 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
246 <div class="doc_text">
248 <p>It is important to note that this document describes 'well formed'
249 LLVM assembly language. There is a difference between what the parser
250 accepts and what is considered 'well formed'. For example, the
251 following instruction is syntactically okay, but not well formed:</p>
254 %x = <a href="#i_add">add</a> i32 1, %x
257 <p>...because the definition of <tt>%x</tt> does not dominate all of
258 its uses. The LLVM infrastructure provides a verification pass that may
259 be used to verify that an LLVM module is well formed. This pass is
260 automatically run by the parser after parsing input assembly and by
261 the optimizer before it outputs bytecode. The violations pointed out
262 by the verifier pass indicate bugs in transformation passes or input to
265 <!-- Describe the typesetting conventions here. --> </div>
267 <!-- *********************************************************************** -->
268 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
269 <!-- *********************************************************************** -->
271 <div class="doc_text">
273 <p>LLVM uses three different forms of identifiers, for different
277 <li>Named values are represented as a string of characters with a '%' prefix.
278 For example, %foo, %DivisionByZero, %a.really.long.identifier. The actual
279 regular expression used is '<tt>%[a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
280 Identifiers which require other characters in their names can be surrounded
281 with quotes. In this way, anything except a <tt>"</tt> character can be used
284 <li>Unnamed values are represented as an unsigned numeric value with a '%'
285 prefix. For example, %12, %2, %44.</li>
287 <li>Constants, which are described in a <a href="#constants">section about
288 constants</a>, below.</li>
291 <p>LLVM requires that values start with a '%' sign for two reasons: Compilers
292 don't need to worry about name clashes with reserved words, and the set of
293 reserved words may be expanded in the future without penalty. Additionally,
294 unnamed identifiers allow a compiler to quickly come up with a temporary
295 variable without having to avoid symbol table conflicts.</p>
297 <p>Reserved words in LLVM are very similar to reserved words in other
298 languages. There are keywords for different opcodes
299 ('<tt><a href="#i_add">add</a></tt>',
300 '<tt><a href="#i_bitcast">bitcast</a></tt>',
301 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
302 href="#t_void">void</a></tt>', '<tt><a href="#t_primitive">i32</a></tt>', etc...),
303 and others. These reserved words cannot conflict with variable names, because
304 none of them start with a '%' character.</p>
306 <p>Here is an example of LLVM code to multiply the integer variable
307 '<tt>%X</tt>' by 8:</p>
312 %result = <a href="#i_mul">mul</a> i32 %X, 8
315 <p>After strength reduction:</p>
318 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
321 <p>And the hard way:</p>
324 <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
325 <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
326 %result = <a href="#i_add">add</a> i32 %1, %1
329 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
330 important lexical features of LLVM:</p>
334 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
337 <li>Unnamed temporaries are created when the result of a computation is not
338 assigned to a named value.</li>
340 <li>Unnamed temporaries are numbered sequentially</li>
344 <p>...and it also shows a convention that we follow in this document. When
345 demonstrating instructions, we will follow an instruction with a comment that
346 defines the type and name of value produced. Comments are shown in italic
351 <!-- *********************************************************************** -->
352 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
353 <!-- *********************************************************************** -->
355 <!-- ======================================================================= -->
356 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
359 <div class="doc_text">
361 <p>LLVM programs are composed of "Module"s, each of which is a
362 translation unit of the input programs. Each module consists of
363 functions, global variables, and symbol table entries. Modules may be
364 combined together with the LLVM linker, which merges function (and
365 global variable) definitions, resolves forward declarations, and merges
366 symbol table entries. Here is an example of the "hello world" module:</p>
368 <pre><i>; Declare the string constant as a global constant...</i>
369 <a href="#identifiers">%.LC0</a> = <a href="#linkage_internal">internal</a> <a
370 href="#globalvars">constant</a> <a href="#t_array">[13 x i8 ]</a> c"hello world\0A\00" <i>; [13 x i8 ]*</i>
372 <i>; External declaration of the puts function</i>
373 <a href="#functionstructure">declare</a> i32 %puts(i8 *) <i>; i32(i8 *)* </i>
375 <i>; Definition of main function</i>
376 define i32 %main() { <i>; i32()* </i>
377 <i>; Convert [13x i8 ]* to i8 *...</i>
379 href="#i_getelementptr">getelementptr</a> [13 x i8 ]* %.LC0, i64 0, i64 0 <i>; i8 *</i>
381 <i>; Call puts function to write out the string to stdout...</i>
383 href="#i_call">call</a> i32 %puts(i8 * %cast210) <i>; i32</i>
385 href="#i_ret">ret</a> i32 0<br>}<br></pre>
387 <p>This example is made up of a <a href="#globalvars">global variable</a>
388 named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
389 function, and a <a href="#functionstructure">function definition</a>
390 for "<tt>main</tt>".</p>
392 <p>In general, a module is made up of a list of global values,
393 where both functions and global variables are global values. Global values are
394 represented by a pointer to a memory location (in this case, a pointer to an
395 array of char, and a pointer to a function), and have one of the following <a
396 href="#linkage">linkage types</a>.</p>
400 <!-- ======================================================================= -->
401 <div class="doc_subsection">
402 <a name="linkage">Linkage Types</a>
405 <div class="doc_text">
408 All Global Variables and Functions have one of the following types of linkage:
413 <dt><tt><b><a name="linkage_internal">internal</a></b></tt> </dt>
415 <dd>Global values with internal linkage are only directly accessible by
416 objects in the current module. In particular, linking code into a module with
417 an internal global value may cause the internal to be renamed as necessary to
418 avoid collisions. Because the symbol is internal to the module, all
419 references can be updated. This corresponds to the notion of the
420 '<tt>static</tt>' keyword in C.
423 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
425 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
426 the same name when linkage occurs. This is typically used to implement
427 inline functions, templates, or other code which must be generated in each
428 translation unit that uses it. Unreferenced <tt>linkonce</tt> globals are
429 allowed to be discarded.
432 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
434 <dd>"<tt>weak</tt>" linkage is exactly the same as <tt>linkonce</tt> linkage,
435 except that unreferenced <tt>weak</tt> globals may not be discarded. This is
436 used for globals that may be emitted in multiple translation units, but that
437 are not guaranteed to be emitted into every translation unit that uses them.
438 One example of this are common globals in C, such as "<tt>int X;</tt>" at
442 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
444 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
445 pointer to array type. When two global variables with appending linkage are
446 linked together, the two global arrays are appended together. This is the
447 LLVM, typesafe, equivalent of having the system linker append together
448 "sections" with identical names when .o files are linked.
451 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt>: </dt>
452 <dd>The semantics of this linkage follow the ELF model: the symbol is weak
453 until linked, if not linked, the symbol becomes null instead of being an
458 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
460 <dd>If none of the above identifiers are used, the global is externally
461 visible, meaning that it participates in linkage and can be used to resolve
462 external symbol references.
466 The next two types of linkage are targeted for Microsoft Windows platform
467 only. They are designed to support importing (exporting) symbols from (to)
472 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt>: </dt>
474 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
475 or variable via a global pointer to a pointer that is set up by the DLL
476 exporting the symbol. On Microsoft Windows targets, the pointer name is
477 formed by combining <code>_imp__</code> and the function or variable name.
480 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt>: </dt>
482 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
483 pointer to a pointer in a DLL, so that it can be referenced with the
484 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
485 name is formed by combining <code>_imp__</code> and the function or variable
491 <p><a name="linkage_external"></a>For example, since the "<tt>.LC0</tt>"
492 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
493 variable and was linked with this one, one of the two would be renamed,
494 preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
495 external (i.e., lacking any linkage declarations), they are accessible
496 outside of the current module.</p>
497 <p>It is illegal for a function <i>declaration</i>
498 to have any linkage type other than "externally visible", <tt>dllimport</tt>,
499 or <tt>extern_weak</tt>.</p>
503 <!-- ======================================================================= -->
504 <div class="doc_subsection">
505 <a name="callingconv">Calling Conventions</a>
508 <div class="doc_text">
510 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
511 and <a href="#i_invoke">invokes</a> can all have an optional calling convention
512 specified for the call. The calling convention of any pair of dynamic
513 caller/callee must match, or the behavior of the program is undefined. The
514 following calling conventions are supported by LLVM, and more may be added in
518 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
520 <dd>This calling convention (the default if no other calling convention is
521 specified) matches the target C calling conventions. This calling convention
522 supports varargs function calls and tolerates some mismatch in the declared
523 prototype and implemented declaration of the function (as does normal C).
526 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
528 <dd>This calling convention attempts to make calls as fast as possible
529 (e.g. by passing things in registers). This calling convention allows the
530 target to use whatever tricks it wants to produce fast code for the target,
531 without having to conform to an externally specified ABI. Implementations of
532 this convention should allow arbitrary tail call optimization to be supported.
533 This calling convention does not support varargs and requires the prototype of
534 all callees to exactly match the prototype of the function definition.
537 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
539 <dd>This calling convention attempts to make code in the caller as efficient
540 as possible under the assumption that the call is not commonly executed. As
541 such, these calls often preserve all registers so that the call does not break
542 any live ranges in the caller side. This calling convention does not support
543 varargs and requires the prototype of all callees to exactly match the
544 prototype of the function definition.
547 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
549 <dd>Any calling convention may be specified by number, allowing
550 target-specific calling conventions to be used. Target specific calling
551 conventions start at 64.
555 <p>More calling conventions can be added/defined on an as-needed basis, to
556 support pascal conventions or any other well-known target-independent
561 <!-- ======================================================================= -->
562 <div class="doc_subsection">
563 <a name="visibility">Visibility Styles</a>
566 <div class="doc_text">
569 All Global Variables and Functions have one of the following visibility styles:
573 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
575 <dd>On ELF, default visibility means that the declaration is visible to other
576 modules and, in shared libraries, means that the declared entity may be
577 overridden. On Darwin, default visibility means that the declaration is
578 visible to other modules. Default visibility corresponds to "external
579 linkage" in the language.
582 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
584 <dd>Two declarations of an object with hidden visibility refer to the same
585 object if they are in the same shared object. Usually, hidden visibility
586 indicates that the symbol will not be placed into the dynamic symbol table,
587 so no other module (executable or shared library) can reference it
595 <!-- ======================================================================= -->
596 <div class="doc_subsection">
597 <a name="globalvars">Global Variables</a>
600 <div class="doc_text">
602 <p>Global variables define regions of memory allocated at compilation time
603 instead of run-time. Global variables may optionally be initialized, may have
604 an explicit section to be placed in, and may
605 have an optional explicit alignment specified. A
606 variable may be defined as a global "constant," which indicates that the
607 contents of the variable will <b>never</b> be modified (enabling better
608 optimization, allowing the global data to be placed in the read-only section of
609 an executable, etc). Note that variables that need runtime initialization
610 cannot be marked "constant" as there is a store to the variable.</p>
613 LLVM explicitly allows <em>declarations</em> of global variables to be marked
614 constant, even if the final definition of the global is not. This capability
615 can be used to enable slightly better optimization of the program, but requires
616 the language definition to guarantee that optimizations based on the
617 'constantness' are valid for the translation units that do not include the
621 <p>As SSA values, global variables define pointer values that are in
622 scope (i.e. they dominate) all basic blocks in the program. Global
623 variables always define a pointer to their "content" type because they
624 describe a region of memory, and all memory objects in LLVM are
625 accessed through pointers.</p>
627 <p>LLVM allows an explicit section to be specified for globals. If the target
628 supports it, it will emit globals to the section specified.</p>
630 <p>An explicit alignment may be specified for a global. If not present, or if
631 the alignment is set to zero, the alignment of the global is set by the target
632 to whatever it feels convenient. If an explicit alignment is specified, the
633 global is forced to have at least that much alignment. All alignments must be
636 <p>For example, the following defines a global with an initializer, section,
640 %G = constant float 1.0, section "foo", align 4
646 <!-- ======================================================================= -->
647 <div class="doc_subsection">
648 <a name="functionstructure">Functions</a>
651 <div class="doc_text">
653 <p>LLVM function definitions consist of the "<tt>define</tt>" keyord,
654 an optional <a href="#linkage">linkage type</a>, an optional
655 <a href="#visibility">visibility style</a>, an optional
656 <a href="#callingconv">calling convention</a>, a return type, an optional
657 <a href="#paramattrs">parameter attribute</a> for the return type, a function
658 name, a (possibly empty) argument list (each with optional
659 <a href="#paramattrs">parameter attributes</a>), an optional section, an
660 optional alignment, an opening curly brace, a list of basic blocks, and a
663 LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
664 optional <a href="#linkage">linkage type</a>, an optional
665 <a href="#visibility">visibility style</a>, an optional
666 <a href="#callingconv">calling convention</a>, a return type, an optional
667 <a href="#paramattrs">parameter attribute</a> for the return type, a function
668 name, a possibly empty list of arguments, and an optional alignment.</p>
670 <p>A function definition contains a list of basic blocks, forming the CFG for
671 the function. Each basic block may optionally start with a label (giving the
672 basic block a symbol table entry), contains a list of instructions, and ends
673 with a <a href="#terminators">terminator</a> instruction (such as a branch or
674 function return).</p>
676 <p>The first basic block in a program is special in two ways: it is immediately
677 executed on entrance to the function, and it is not allowed to have predecessor
678 basic blocks (i.e. there can not be any branches to the entry block of a
679 function). Because the block can have no predecessors, it also cannot have any
680 <a href="#i_phi">PHI nodes</a>.</p>
682 <p>LLVM functions are identified by their name and type signature. Hence, two
683 functions with the same name but different parameter lists or return values are
684 considered different functions, and LLVM will resolve references to each
687 <p>LLVM allows an explicit section to be specified for functions. If the target
688 supports it, it will emit functions to the section specified.</p>
690 <p>An explicit alignment may be specified for a function. If not present, or if
691 the alignment is set to zero, the alignment of the function is set by the target
692 to whatever it feels convenient. If an explicit alignment is specified, the
693 function is forced to have at least that much alignment. All alignments must be
698 <!-- ======================================================================= -->
699 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
700 <div class="doc_text">
701 <p>The return type and each parameter of a function type may have a set of
702 <i>parameter attributes</i> associated with them. Parameter attributes are
703 used to communicate additional information about the result or parameters of
704 a function. Parameter attributes are considered to be part of the function
705 type so two functions types that differ only by the parameter attributes
706 are different function types.</p>
708 <p>Parameter attributes are simple keywords that follow the type specified. If
709 multiple parameter attributes are needed, they are space separated. For
711 %someFunc = i16 (i8 sext %someParam) zext
712 %someFunc = i16 (i8 zext %someParam) zext</pre>
713 <p>Note that the two function types above are unique because the parameter has
714 a different attribute (sext in the first one, zext in the second). Also note
715 that the attribute for the function result (zext) comes immediately after the
718 <p>Currently, only the following parameter attributes are defined:</p>
720 <dt><tt>zext</tt></dt>
721 <dd>This indicates that the parameter should be zero extended just before
722 a call to this function.</dd>
723 <dt><tt>sext</tt></dt>
724 <dd>This indicates that the parameter should be sign extended just before
725 a call to this function.</dd>
726 <dt><tt>inreg</tt></dt>
727 <dd>This indicates that the parameter should be placed in register (if
728 possible) during assembling function call. Support for this attribute is
730 <dt><tt>sret</tt></dt>
731 <dd>This indicates that the parameter specifies the address of a structure
732 that is the return value of the function in the source program.</dd>
733 <dt><tt>noreturn</tt></dt>
734 <dd>This function attribute indicates that the function never returns. This
735 indicates to LLVM that every call to this function should be treated as if
736 an <tt>unreachable</tt> instruction immediately followed the call.</dd>
737 <dt><tt>nounwind</tt></dt>
738 <dd>This function attribute indicates that the function type does not use
739 the unwind instruction and does not allow stack unwinding to propagate
745 <!-- ======================================================================= -->
746 <div class="doc_subsection">
747 <a name="moduleasm">Module-Level Inline Assembly</a>
750 <div class="doc_text">
752 Modules may contain "module-level inline asm" blocks, which corresponds to the
753 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
754 LLVM and treated as a single unit, but may be separated in the .ll file if
755 desired. The syntax is very simple:
758 <div class="doc_code"><pre>
759 module asm "inline asm code goes here"
760 module asm "more can go here"
763 <p>The strings can contain any character by escaping non-printable characters.
764 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
769 The inline asm code is simply printed to the machine code .s file when
770 assembly code is generated.
774 <!-- ======================================================================= -->
775 <div class="doc_subsection">
776 <a name="datalayout">Data Layout</a>
779 <div class="doc_text">
780 <p>A module may specify a target specific data layout string that specifies how
781 data is to be laid out in memory. The syntax for the data layout is simply:<br/>
782 <pre> target datalayout = "<i>layout specification</i>"
784 The <i>layout specification</i> consists of a list of specifications separated
785 by the minus sign character ('-'). Each specification starts with a letter
786 and may include other information after the letter to define some aspect of the
787 data layout. The specifications accepted are as follows: </p>
790 <dd>Specifies that the target lays out data in big-endian form. That is, the
791 bits with the most significance have the lowest address location.</dd>
793 <dd>Specifies that hte target lays out data in little-endian form. That is,
794 the bits with the least significance have the lowest address location.</dd>
795 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
796 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
797 <i>preferred</i> alignments. All sizes are in bits. Specifying the <i>pref</i>
798 alignment is optional. If omitted, the preceding <tt>:</tt> should be omitted
800 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
801 <dd>This specifies the alignment for an integer type of a given bit
802 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
803 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
804 <dd>This specifies the alignment for a vector type of a given bit
806 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
807 <dd>This specifies the alignment for a floating point type of a given bit
808 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
810 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
811 <dd>This specifies the alignment for an aggregate type of a given bit
814 <p>When constructing the data layout for a given target, LLVM starts with a
815 default set of specifications which are then (possibly) overriden by the
816 specifications in the <tt>datalayout</tt> keyword. The default specifications
817 are given in this list:</p>
819 <li><tt>E</tt> - big endian</li>
820 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
821 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
822 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
823 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
824 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
825 <li><tt>i64:32:64</tt> - i64 has abi alignment of 32-bits but preferred
826 alignment of 64-bits</li>
827 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
828 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
829 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
830 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
831 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
833 <p>When llvm is determining the alignment for a given type, it uses the
836 <li>If the type sought is an exact match for one of the specifications, that
837 specification is used.</li>
838 <li>If no match is found, and the type sought is an integer type, then the
839 smallest integer type that is larger than the bitwidth of the sought type is
840 used. If none of the specifications are larger than the bitwidth then the the
841 largest integer type is used. For example, given the default specifications
842 above, the i7 type will use the alignment of i8 (next largest) while both
843 i65 and i256 will use the alignment of i64 (largest specified).</li>
844 <li>If no match is found, and the type sought is a vector type, then the
845 largest vector type that is smaller than the sought vector type will be used
846 as a fall back. This happens because <128 x double> can be implemented in
847 terms of 64 <2 x double>, for example.</li>
851 <!-- *********************************************************************** -->
852 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
853 <!-- *********************************************************************** -->
855 <div class="doc_text">
857 <p>The LLVM type system is one of the most important features of the
858 intermediate representation. Being typed enables a number of
859 optimizations to be performed on the IR directly, without having to do
860 extra analyses on the side before the transformation. A strong type
861 system makes it easier to read the generated code and enables novel
862 analyses and transformations that are not feasible to perform on normal
863 three address code representations.</p>
867 <!-- ======================================================================= -->
868 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
869 <div class="doc_text">
870 <p>The primitive types are the fundamental building blocks of the LLVM
871 system. The current set of primitive types is as follows:</p>
873 <table class="layout">
878 <tr><th>Type</th><th>Description</th></tr>
879 <tr><td><tt><a name="t_void">void</a></tt></td><td>No value</td></tr>
880 <tr><td><tt>i8</tt></td><td>8-bit value</td></tr>
881 <tr><td><tt>i32</tt></td><td>32-bit value</td></tr>
882 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
883 <tr><td><tt>label</tt></td><td>Branch destination</td></tr>
890 <tr><th>Type</th><th>Description</th></tr>
891 <tr><td><tt>i1</tt></td><td>True or False value</td></tr>
892 <tr><td><tt>i16</tt></td><td>16-bit value</td></tr>
893 <tr><td><tt>i64</tt></td><td>64-bit value</td></tr>
894 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
902 <!-- _______________________________________________________________________ -->
903 <div class="doc_subsubsection"> <a name="t_classifications">Type
904 Classifications</a> </div>
905 <div class="doc_text">
906 <p>These different primitive types fall into a few useful
909 <table border="1" cellspacing="0" cellpadding="4">
911 <tr><th>Classification</th><th>Types</th></tr>
913 <td><a name="t_integer">integer</a></td>
914 <td><tt>i1, i8, i16, i32, i64</tt></td>
917 <td><a name="t_floating">floating point</a></td>
918 <td><tt>float, double</tt></td>
921 <td><a name="t_firstclass">first class</a></td>
922 <td><tt>i1, i8, i16, i32, i64, float, double, <br/>
923 <a href="#t_pointer">pointer</a>,<a href="#t_vector">vector</a></tt>
929 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
930 most important. Values of these types are the only ones which can be
931 produced by instructions, passed as arguments, or used as operands to
932 instructions. This means that all structures and arrays must be
933 manipulated either by pointer or by component.</p>
936 <!-- ======================================================================= -->
937 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
939 <div class="doc_text">
941 <p>The real power in LLVM comes from the derived types in the system.
942 This is what allows a programmer to represent arrays, functions,
943 pointers, and other useful types. Note that these derived types may be
944 recursive: For example, it is possible to have a two dimensional array.</p>
948 <!-- _______________________________________________________________________ -->
949 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
951 <div class="doc_text">
955 <p>The array type is a very simple derived type that arranges elements
956 sequentially in memory. The array type requires a size (number of
957 elements) and an underlying data type.</p>
962 [<# elements> x <elementtype>]
965 <p>The number of elements is a constant integer value; elementtype may
966 be any type with a size.</p>
969 <table class="layout">
972 <tt>[40 x i32 ]</tt><br/>
973 <tt>[41 x i32 ]</tt><br/>
974 <tt>[40 x i8]</tt><br/>
977 Array of 40 32-bit integer values.<br/>
978 Array of 41 32-bit integer values.<br/>
979 Array of 40 8-bit integer values.<br/>
983 <p>Here are some examples of multidimensional arrays:</p>
984 <table class="layout">
987 <tt>[3 x [4 x i32]]</tt><br/>
988 <tt>[12 x [10 x float]]</tt><br/>
989 <tt>[2 x [3 x [4 x i16]]]</tt><br/>
992 3x4 array of 32-bit integer values.<br/>
993 12x10 array of single precision floating point values.<br/>
994 2x3x4 array of 16-bit integer values.<br/>
999 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1000 length array. Normally, accesses past the end of an array are undefined in
1001 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
1002 As a special case, however, zero length arrays are recognized to be variable
1003 length. This allows implementation of 'pascal style arrays' with the LLVM
1004 type "{ i32, [0 x float]}", for example.</p>
1008 <!-- _______________________________________________________________________ -->
1009 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1010 <div class="doc_text">
1012 <p>The function type can be thought of as a function signature. It
1013 consists of a return type and a list of formal parameter types.
1014 Function types are usually used to build virtual function tables
1015 (which are structures of pointers to functions), for indirect function
1016 calls, and when defining a function.</p>
1018 The return type of a function type cannot be an aggregate type.
1021 <pre> <returntype> (<parameter list>)<br></pre>
1022 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1023 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1024 which indicates that the function takes a variable number of arguments.
1025 Variable argument functions can access their arguments with the <a
1026 href="#int_varargs">variable argument handling intrinsic</a> functions.</p>
1028 <table class="layout">
1030 <td class="left"><tt>i32 (i32)</tt></td>
1031 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1033 </tr><tr class="layout">
1034 <td class="left"><tt>float (i16 sext, i32 *) *
1036 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1037 an <tt>i16</tt> that should be sign extended and a
1038 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1041 </tr><tr class="layout">
1042 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1043 <td class="left">A vararg function that takes at least one
1044 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1045 which returns an integer. This is the signature for <tt>printf</tt> in
1052 <!-- _______________________________________________________________________ -->
1053 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1054 <div class="doc_text">
1056 <p>The structure type is used to represent a collection of data members
1057 together in memory. The packing of the field types is defined to match
1058 the ABI of the underlying processor. The elements of a structure may
1059 be any type that has a size.</p>
1060 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1061 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1062 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1065 <pre> { <type list> }<br></pre>
1067 <table class="layout">
1070 <tt>{ i32, i32, i32 }</tt><br/>
1071 <tt>{ float, i32 (i32) * }</tt><br/>
1074 a triple of three <tt>i32</tt> values<br/>
1075 A pair, where the first element is a <tt>float</tt> and the second element
1076 is a <a href="#t_pointer">pointer</a> to a <a href="#t_function">function</a>
1077 that takes an <tt>i32</tt>, returning an <tt>i32</tt>.<br/>
1083 <!-- _______________________________________________________________________ -->
1084 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1086 <div class="doc_text">
1088 <p>The packed structure type is used to represent a collection of data members
1089 together in memory. There is no padding between fields. Further, the alignment
1090 of a packed structure is 1 byte. The elements of a packed structure may
1091 be any type that has a size.</p>
1092 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1093 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1094 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1097 <pre> < { <type list> } > <br></pre>
1099 <table class="layout">
1102 <tt> < { i32, i32, i32 } > </tt><br/>
1103 <tt> < { float, i32 (i32) * } > </tt><br/>
1106 a triple of three <tt>i32</tt> values<br/>
1107 A pair, where the first element is a <tt>float</tt> and the second element
1108 is a <a href="#t_pointer">pointer</a> to a <a href="#t_function">function</a>
1109 that takes an <tt>i32</tt>, returning an <tt>i32</tt>.<br/>
1115 <!-- _______________________________________________________________________ -->
1116 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1117 <div class="doc_text">
1119 <p>As in many languages, the pointer type represents a pointer or
1120 reference to another object, which must live in memory.</p>
1122 <pre> <type> *<br></pre>
1124 <table class="layout">
1127 <tt>[4x i32]*</tt><br/>
1128 <tt>i32 (i32 *) *</tt><br/>
1131 A <a href="#t_pointer">pointer</a> to <a href="#t_array">array</a> of
1132 four <tt>i32</tt> values<br/>
1133 A <a href="#t_pointer">pointer</a> to a <a
1134 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1141 <!-- _______________________________________________________________________ -->
1142 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1143 <div class="doc_text">
1147 <p>A vector type is a simple derived type that represents a vector
1148 of elements. Vector types are used when multiple primitive data
1149 are operated in parallel using a single instruction (SIMD).
1150 A vector type requires a size (number of
1151 elements) and an underlying primitive data type. Vectors must have a power
1152 of two length (1, 2, 4, 8, 16 ...). Vector types are
1153 considered <a href="#t_firstclass">first class</a>.</p>
1158 < <# elements> x <elementtype> >
1161 <p>The number of elements is a constant integer value; elementtype may
1162 be any integer or floating point type.</p>
1166 <table class="layout">
1169 <tt><4 x i32></tt><br/>
1170 <tt><8 x float></tt><br/>
1171 <tt><2 x i64></tt><br/>
1174 Vector of 4 32-bit integer values.<br/>
1175 Vector of 8 floating-point values.<br/>
1176 Vector of 2 64-bit integer values.<br/>
1182 <!-- _______________________________________________________________________ -->
1183 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1184 <div class="doc_text">
1188 <p>Opaque types are used to represent unknown types in the system. This
1189 corresponds (for example) to the C notion of a foward declared structure type.
1190 In LLVM, opaque types can eventually be resolved to any type (not just a
1191 structure type).</p>
1201 <table class="layout">
1207 An opaque type.<br/>
1214 <!-- *********************************************************************** -->
1215 <div class="doc_section"> <a name="constants">Constants</a> </div>
1216 <!-- *********************************************************************** -->
1218 <div class="doc_text">
1220 <p>LLVM has several different basic types of constants. This section describes
1221 them all and their syntax.</p>
1225 <!-- ======================================================================= -->
1226 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1228 <div class="doc_text">
1231 <dt><b>Boolean constants</b></dt>
1233 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1234 constants of the <tt><a href="#t_primitive">i1</a></tt> type.
1237 <dt><b>Integer constants</b></dt>
1239 <dd>Standard integers (such as '4') are constants of the <a
1240 href="#t_integer">integer</a> type. Negative numbers may be used with
1244 <dt><b>Floating point constants</b></dt>
1246 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1247 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1248 notation (see below). Floating point constants must have a <a
1249 href="#t_floating">floating point</a> type. </dd>
1251 <dt><b>Null pointer constants</b></dt>
1253 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1254 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1258 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1259 of floating point constants. For example, the form '<tt>double
1260 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1261 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1262 (and the only time that they are generated by the disassembler) is when a
1263 floating point constant must be emitted but it cannot be represented as a
1264 decimal floating point number. For example, NaN's, infinities, and other
1265 special values are represented in their IEEE hexadecimal format so that
1266 assembly and disassembly do not cause any bits to change in the constants.</p>
1270 <!-- ======================================================================= -->
1271 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1274 <div class="doc_text">
1275 <p>Aggregate constants arise from aggregation of simple constants
1276 and smaller aggregate constants.</p>
1279 <dt><b>Structure constants</b></dt>
1281 <dd>Structure constants are represented with notation similar to structure
1282 type definitions (a comma separated list of elements, surrounded by braces
1283 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* %G }</tt>",
1284 where "<tt>%G</tt>" is declared as "<tt>%G = external global i32</tt>". Structure constants
1285 must have <a href="#t_struct">structure type</a>, and the number and
1286 types of elements must match those specified by the type.
1289 <dt><b>Array constants</b></dt>
1291 <dd>Array constants are represented with notation similar to array type
1292 definitions (a comma separated list of elements, surrounded by square brackets
1293 (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74 ]</tt>". Array
1294 constants must have <a href="#t_array">array type</a>, and the number and
1295 types of elements must match those specified by the type.
1298 <dt><b>Vector constants</b></dt>
1300 <dd>Vector constants are represented with notation similar to vector type
1301 definitions (a comma separated list of elements, surrounded by
1302 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32 42,
1303 i32 11, i32 74, i32 100 ></tt>". VEctor constants must have <a
1304 href="#t_vector">vector type</a>, and the number and types of elements must
1305 match those specified by the type.
1308 <dt><b>Zero initialization</b></dt>
1310 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1311 value to zero of <em>any</em> type, including scalar and aggregate types.
1312 This is often used to avoid having to print large zero initializers (e.g. for
1313 large arrays) and is always exactly equivalent to using explicit zero
1320 <!-- ======================================================================= -->
1321 <div class="doc_subsection">
1322 <a name="globalconstants">Global Variable and Function Addresses</a>
1325 <div class="doc_text">
1327 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1328 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1329 constants. These constants are explicitly referenced when the <a
1330 href="#identifiers">identifier for the global</a> is used and always have <a
1331 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1337 %Z = global [2 x i32*] [ i32* %X, i32* %Y ]
1342 <!-- ======================================================================= -->
1343 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1344 <div class="doc_text">
1345 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1346 no specific value. Undefined values may be of any type and be used anywhere
1347 a constant is permitted.</p>
1349 <p>Undefined values indicate to the compiler that the program is well defined
1350 no matter what value is used, giving the compiler more freedom to optimize.
1354 <!-- ======================================================================= -->
1355 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1358 <div class="doc_text">
1360 <p>Constant expressions are used to allow expressions involving other constants
1361 to be used as constants. Constant expressions may be of any <a
1362 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1363 that does not have side effects (e.g. load and call are not supported). The
1364 following is the syntax for constant expressions:</p>
1367 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1368 <dd>Truncate a constant to another type. The bit size of CST must be larger
1369 than the bit size of TYPE. Both types must be integers.</dd>
1371 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1372 <dd>Zero extend a constant to another type. The bit size of CST must be
1373 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1375 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1376 <dd>Sign extend a constant to another type. The bit size of CST must be
1377 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1379 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1380 <dd>Truncate a floating point constant to another floating point type. The
1381 size of CST must be larger than the size of TYPE. Both types must be
1382 floating point.</dd>
1384 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1385 <dd>Floating point extend a constant to another type. The size of CST must be
1386 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1388 <dt><b><tt>fp2uint ( CST to TYPE )</tt></b></dt>
1389 <dd>Convert a floating point constant to the corresponding unsigned integer
1390 constant. TYPE must be an integer type. CST must be floating point. If the
1391 value won't fit in the integer type, the results are undefined.</dd>
1393 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
1394 <dd>Convert a floating point constant to the corresponding signed integer
1395 constant. TYPE must be an integer type. CST must be floating point. If the
1396 value won't fit in the integer type, the results are undefined.</dd>
1398 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
1399 <dd>Convert an unsigned integer constant to the corresponding floating point
1400 constant. TYPE must be floating point. CST must be of integer type. If the
1401 value won't fit in the floating point type, the results are undefined.</dd>
1403 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
1404 <dd>Convert a signed integer constant to the corresponding floating point
1405 constant. TYPE must be floating point. CST must be of integer type. If the
1406 value won't fit in the floating point type, the results are undefined.</dd>
1408 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
1409 <dd>Convert a pointer typed constant to the corresponding integer constant
1410 TYPE must be an integer type. CST must be of pointer type. The CST value is
1411 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1413 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
1414 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1415 pointer type. CST must be of integer type. The CST value is zero extended,
1416 truncated, or unchanged to make it fit in a pointer size. This one is
1417 <i>really</i> dangerous!</dd>
1419 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
1420 <dd>Convert a constant, CST, to another TYPE. The size of CST and TYPE must be
1421 identical (same number of bits). The conversion is done as if the CST value
1422 was stored to memory and read back as TYPE. In other words, no bits change
1423 with this operator, just the type. This can be used for conversion of
1424 vector types to any other type, as long as they have the same bit width. For
1425 pointers it is only valid to cast to another pointer type.
1428 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1430 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1431 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1432 instruction, the index list may have zero or more indexes, which are required
1433 to make sense for the type of "CSTPTR".</dd>
1435 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1437 <dd>Perform the <a href="#i_select">select operation</a> on
1440 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
1441 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
1443 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
1444 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
1446 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1448 <dd>Perform the <a href="#i_extractelement">extractelement
1449 operation</a> on constants.
1451 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1453 <dd>Perform the <a href="#i_insertelement">insertelement
1454 operation</a> on constants.</dd>
1457 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
1459 <dd>Perform the <a href="#i_shufflevector">shufflevector
1460 operation</a> on constants.</dd>
1462 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1464 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1465 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1466 binary</a> operations. The constraints on operands are the same as those for
1467 the corresponding instruction (e.g. no bitwise operations on floating point
1468 values are allowed).</dd>
1472 <!-- *********************************************************************** -->
1473 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1474 <!-- *********************************************************************** -->
1476 <!-- ======================================================================= -->
1477 <div class="doc_subsection">
1478 <a name="inlineasm">Inline Assembler Expressions</a>
1481 <div class="doc_text">
1484 LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
1485 Module-Level Inline Assembly</a>) through the use of a special value. This
1486 value represents the inline assembler as a string (containing the instructions
1487 to emit), a list of operand constraints (stored as a string), and a flag that
1488 indicates whether or not the inline asm expression has side effects. An example
1489 inline assembler expression is:
1493 i32 (i32) asm "bswap $0", "=r,r"
1497 Inline assembler expressions may <b>only</b> be used as the callee operand of
1498 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
1502 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
1506 Inline asms with side effects not visible in the constraint list must be marked
1507 as having side effects. This is done through the use of the
1508 '<tt>sideeffect</tt>' keyword, like so:
1512 call void asm sideeffect "eieio", ""()
1515 <p>TODO: The format of the asm and constraints string still need to be
1516 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1517 need to be documented).
1522 <!-- *********************************************************************** -->
1523 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1524 <!-- *********************************************************************** -->
1526 <div class="doc_text">
1528 <p>The LLVM instruction set consists of several different
1529 classifications of instructions: <a href="#terminators">terminator
1530 instructions</a>, <a href="#binaryops">binary instructions</a>,
1531 <a href="#bitwiseops">bitwise binary instructions</a>, <a
1532 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1533 instructions</a>.</p>
1537 <!-- ======================================================================= -->
1538 <div class="doc_subsection"> <a name="terminators">Terminator
1539 Instructions</a> </div>
1541 <div class="doc_text">
1543 <p>As mentioned <a href="#functionstructure">previously</a>, every
1544 basic block in a program ends with a "Terminator" instruction, which
1545 indicates which block should be executed after the current block is
1546 finished. These terminator instructions typically yield a '<tt>void</tt>'
1547 value: they produce control flow, not values (the one exception being
1548 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1549 <p>There are six different terminator instructions: the '<a
1550 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1551 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1552 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1553 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1554 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1558 <!-- _______________________________________________________________________ -->
1559 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1560 Instruction</a> </div>
1561 <div class="doc_text">
1563 <pre> ret <type> <value> <i>; Return a value from a non-void function</i>
1564 ret void <i>; Return from void function</i>
1567 <p>The '<tt>ret</tt>' instruction is used to return control flow (and a
1568 value) from a function back to the caller.</p>
1569 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
1570 returns a value and then causes control flow, and one that just causes
1571 control flow to occur.</p>
1573 <p>The '<tt>ret</tt>' instruction may return any '<a
1574 href="#t_firstclass">first class</a>' type. Notice that a function is
1575 not <a href="#wellformed">well formed</a> if there exists a '<tt>ret</tt>'
1576 instruction inside of the function that returns a value that does not
1577 match the return type of the function.</p>
1579 <p>When the '<tt>ret</tt>' instruction is executed, control flow
1580 returns back to the calling function's context. If the caller is a "<a
1581 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1582 the instruction after the call. If the caller was an "<a
1583 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1584 at the beginning of the "normal" destination block. If the instruction
1585 returns a value, that value shall set the call or invoke instruction's
1588 <pre> ret i32 5 <i>; Return an integer value of 5</i>
1589 ret void <i>; Return from a void function</i>
1592 <!-- _______________________________________________________________________ -->
1593 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1594 <div class="doc_text">
1596 <pre> br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1599 <p>The '<tt>br</tt>' instruction is used to cause control flow to
1600 transfer to a different basic block in the current function. There are
1601 two forms of this instruction, corresponding to a conditional branch
1602 and an unconditional branch.</p>
1604 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1605 single '<tt>i1</tt>' value and two '<tt>label</tt>' values. The
1606 unconditional form of the '<tt>br</tt>' instruction takes a single
1607 '<tt>label</tt>' value as a target.</p>
1609 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
1610 argument is evaluated. If the value is <tt>true</tt>, control flows
1611 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1612 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1614 <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
1615 href="#i_ret">ret</a> i32 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> i32 0<br></pre>
1617 <!-- _______________________________________________________________________ -->
1618 <div class="doc_subsubsection">
1619 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
1622 <div class="doc_text">
1626 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
1631 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
1632 several different places. It is a generalization of the '<tt>br</tt>'
1633 instruction, allowing a branch to occur to one of many possible
1639 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
1640 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
1641 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
1642 table is not allowed to contain duplicate constant entries.</p>
1646 <p>The <tt>switch</tt> instruction specifies a table of values and
1647 destinations. When the '<tt>switch</tt>' instruction is executed, this
1648 table is searched for the given value. If the value is found, control flow is
1649 transfered to the corresponding destination; otherwise, control flow is
1650 transfered to the default destination.</p>
1652 <h5>Implementation:</h5>
1654 <p>Depending on properties of the target machine and the particular
1655 <tt>switch</tt> instruction, this instruction may be code generated in different
1656 ways. For example, it could be generated as a series of chained conditional
1657 branches or with a lookup table.</p>
1662 <i>; Emulate a conditional br instruction</i>
1663 %Val = <a href="#i_zext">zext</a> i1 %value to i32
1664 switch i32 %Val, label %truedest [i32 0, label %falsedest ]
1666 <i>; Emulate an unconditional br instruction</i>
1667 switch i32 0, label %dest [ ]
1669 <i>; Implement a jump table:</i>
1670 switch i32 %val, label %otherwise [ i32 0, label %onzero
1672 i32 2, label %ontwo ]
1676 <!-- _______________________________________________________________________ -->
1677 <div class="doc_subsubsection">
1678 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
1681 <div class="doc_text">
1686 <result> = invoke [<a href="#callingconv">cconv</a>] <ptr to function ty> %<function ptr val>(<function args>)
1687 to label <normal label> unwind label <exception label>
1692 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
1693 function, with the possibility of control flow transfer to either the
1694 '<tt>normal</tt>' label or the
1695 '<tt>exception</tt>' label. If the callee function returns with the
1696 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
1697 "normal" label. If the callee (or any indirect callees) returns with the "<a
1698 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
1699 continued at the dynamically nearest "exception" label.</p>
1703 <p>This instruction requires several arguments:</p>
1707 The optional "cconv" marker indicates which <a href="#callingconv">calling
1708 convention</a> the call should use. If none is specified, the call defaults
1709 to using C calling conventions.
1711 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
1712 function value being invoked. In most cases, this is a direct function
1713 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
1714 an arbitrary pointer to function value.
1717 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
1718 function to be invoked. </li>
1720 <li>'<tt>function args</tt>': argument list whose types match the function
1721 signature argument types. If the function signature indicates the function
1722 accepts a variable number of arguments, the extra arguments can be
1725 <li>'<tt>normal label</tt>': the label reached when the called function
1726 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
1728 <li>'<tt>exception label</tt>': the label reached when a callee returns with
1729 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
1735 <p>This instruction is designed to operate as a standard '<tt><a
1736 href="#i_call">call</a></tt>' instruction in most regards. The primary
1737 difference is that it establishes an association with a label, which is used by
1738 the runtime library to unwind the stack.</p>
1740 <p>This instruction is used in languages with destructors to ensure that proper
1741 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
1742 exception. Additionally, this is important for implementation of
1743 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
1747 %retval = invoke i32 %Test(i32 15) to label %Continue
1748 unwind label %TestCleanup <i>; {i32}:retval set</i>
1749 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Test(i32 15) to label %Continue
1750 unwind label %TestCleanup <i>; {i32}:retval set</i>
1755 <!-- _______________________________________________________________________ -->
1757 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
1758 Instruction</a> </div>
1760 <div class="doc_text">
1769 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
1770 at the first callee in the dynamic call stack which used an <a
1771 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
1772 primarily used to implement exception handling.</p>
1776 <p>The '<tt>unwind</tt>' intrinsic causes execution of the current function to
1777 immediately halt. The dynamic call stack is then searched for the first <a
1778 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
1779 execution continues at the "exceptional" destination block specified by the
1780 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
1781 dynamic call chain, undefined behavior results.</p>
1784 <!-- _______________________________________________________________________ -->
1786 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
1787 Instruction</a> </div>
1789 <div class="doc_text">
1798 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
1799 instruction is used to inform the optimizer that a particular portion of the
1800 code is not reachable. This can be used to indicate that the code after a
1801 no-return function cannot be reached, and other facts.</p>
1805 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
1810 <!-- ======================================================================= -->
1811 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
1812 <div class="doc_text">
1813 <p>Binary operators are used to do most of the computation in a
1814 program. They require two operands, execute an operation on them, and
1815 produce a single value. The operands might represent
1816 multiple data, as is the case with the <a href="#t_vector">vector</a> data type.
1817 The result value of a binary operator is not
1818 necessarily the same type as its operands.</p>
1819 <p>There are several different binary operators:</p>
1821 <!-- _______________________________________________________________________ -->
1822 <div class="doc_subsubsection"> <a name="i_add">'<tt>add</tt>'
1823 Instruction</a> </div>
1824 <div class="doc_text">
1826 <pre> <result> = add <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1829 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
1831 <p>The two arguments to the '<tt>add</tt>' instruction must be either <a
1832 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a> values.
1833 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
1834 Both arguments must have identical types.</p>
1836 <p>The value produced is the integer or floating point sum of the two
1839 <pre> <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
1842 <!-- _______________________________________________________________________ -->
1843 <div class="doc_subsubsection"> <a name="i_sub">'<tt>sub</tt>'
1844 Instruction</a> </div>
1845 <div class="doc_text">
1847 <pre> <result> = sub <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1850 <p>The '<tt>sub</tt>' instruction returns the difference of its two
1852 <p>Note that the '<tt>sub</tt>' instruction is used to represent the '<tt>neg</tt>'
1853 instruction present in most other intermediate representations.</p>
1855 <p>The two arguments to the '<tt>sub</tt>' instruction must be either <a
1856 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1858 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
1859 Both arguments must have identical types.</p>
1861 <p>The value produced is the integer or floating point difference of
1862 the two operands.</p>
1864 <pre> <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
1865 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
1868 <!-- _______________________________________________________________________ -->
1869 <div class="doc_subsubsection"> <a name="i_mul">'<tt>mul</tt>'
1870 Instruction</a> </div>
1871 <div class="doc_text">
1873 <pre> <result> = mul <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1876 <p>The '<tt>mul</tt>' instruction returns the product of its two
1879 <p>The two arguments to the '<tt>mul</tt>' instruction must be either <a
1880 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1882 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
1883 Both arguments must have identical types.</p>
1885 <p>The value produced is the integer or floating point product of the
1887 <p>Because the operands are the same width, the result of an integer
1888 multiplication is the same whether the operands should be deemed unsigned or
1891 <pre> <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
1894 <!-- _______________________________________________________________________ -->
1895 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
1897 <div class="doc_text">
1899 <pre> <result> = udiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1902 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two
1905 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
1906 <a href="#t_integer">integer</a> values. Both arguments must have identical
1907 types. This instruction can also take <a href="#t_vector">vector</a> versions
1908 of the values in which case the elements must be integers.</p>
1910 <p>The value produced is the unsigned integer quotient of the two operands. This
1911 instruction always performs an unsigned division operation, regardless of
1912 whether the arguments are unsigned or not.</p>
1914 <pre> <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
1917 <!-- _______________________________________________________________________ -->
1918 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
1920 <div class="doc_text">
1922 <pre> <result> = sdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1925 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
1928 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
1929 <a href="#t_integer">integer</a> values. Both arguments must have identical
1930 types. This instruction can also take <a href="#t_vector">vector</a> versions
1931 of the values in which case the elements must be integers.</p>
1933 <p>The value produced is the signed integer quotient of the two operands. This
1934 instruction always performs a signed division operation, regardless of whether
1935 the arguments are signed or not.</p>
1937 <pre> <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
1940 <!-- _______________________________________________________________________ -->
1941 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
1942 Instruction</a> </div>
1943 <div class="doc_text">
1945 <pre> <result> = fdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1948 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
1951 <p>The two arguments to the '<tt>div</tt>' instruction must be
1952 <a href="#t_floating">floating point</a> values. Both arguments must have
1953 identical types. This instruction can also take <a href="#t_vector">vector</a>
1954 versions of the values in which case the elements must be floating point.</p>
1956 <p>The value produced is the floating point quotient of the two operands.</p>
1958 <pre> <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
1961 <!-- _______________________________________________________________________ -->
1962 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
1964 <div class="doc_text">
1966 <pre> <result> = urem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1969 <p>The '<tt>urem</tt>' instruction returns the remainder from the
1970 unsigned division of its two arguments.</p>
1972 <p>The two arguments to the '<tt>urem</tt>' instruction must be
1973 <a href="#t_integer">integer</a> values. Both arguments must have identical
1976 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
1977 This instruction always performs an unsigned division to get the remainder,
1978 regardless of whether the arguments are unsigned or not.</p>
1980 <pre> <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
1984 <!-- _______________________________________________________________________ -->
1985 <div class="doc_subsubsection"> <a name="i_srem">'<tt>srem</tt>'
1986 Instruction</a> </div>
1987 <div class="doc_text">
1989 <pre> <result> = srem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1992 <p>The '<tt>srem</tt>' instruction returns the remainder from the
1993 signed division of its two operands.</p>
1995 <p>The two arguments to the '<tt>srem</tt>' instruction must be
1996 <a href="#t_integer">integer</a> values. Both arguments must have identical
1999 <p>This instruction returns the <i>remainder</i> of a division (where the result
2000 has the same sign as the dividend, <tt>var1</tt>), not the <i>modulo</i>
2001 operator (where the result has the same sign as the divisor, <tt>var2</tt>) of
2002 a value. For more information about the difference, see <a
2003 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
2004 Math Forum</a>. For a table of how this is implemented in various languages,
2005 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
2006 Wikipedia: modulo operation</a>.</p>
2008 <pre> <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2012 <!-- _______________________________________________________________________ -->
2013 <div class="doc_subsubsection"> <a name="i_frem">'<tt>frem</tt>'
2014 Instruction</a> </div>
2015 <div class="doc_text">
2017 <pre> <result> = frem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2020 <p>The '<tt>frem</tt>' instruction returns the remainder from the
2021 division of its two operands.</p>
2023 <p>The two arguments to the '<tt>frem</tt>' instruction must be
2024 <a href="#t_floating">floating point</a> values. Both arguments must have
2025 identical types.</p>
2027 <p>This instruction returns the <i>remainder</i> of a division.</p>
2029 <pre> <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
2033 <!-- ======================================================================= -->
2034 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
2035 Operations</a> </div>
2036 <div class="doc_text">
2037 <p>Bitwise binary operators are used to do various forms of
2038 bit-twiddling in a program. They are generally very efficient
2039 instructions and can commonly be strength reduced from other
2040 instructions. They require two operands, execute an operation on them,
2041 and produce a single value. The resulting value of the bitwise binary
2042 operators is always the same type as its first operand.</p>
2045 <!-- _______________________________________________________________________ -->
2046 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
2047 Instruction</a> </div>
2048 <div class="doc_text">
2050 <pre> <result> = shl <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2053 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2054 the left a specified number of bits.</p>
2056 <p>Both arguments to the '<tt>shl</tt>' instruction must be the same <a
2057 href="#t_integer">integer</a> type.</p>
2059 <p>The value produced is <tt>var1</tt> * 2<sup><tt>var2</tt></sup>.</p>
2060 <h5>Example:</h5><pre>
2061 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
2062 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
2063 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
2066 <!-- _______________________________________________________________________ -->
2067 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
2068 Instruction</a> </div>
2069 <div class="doc_text">
2071 <pre> <result> = lshr <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2075 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
2076 operand shifted to the right a specified number of bits.</p>
2079 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
2080 <a href="#t_integer">integer</a> type.</p>
2083 <p>This instruction always performs a logical shift right operation. The most
2084 significant bits of the result will be filled with zero bits after the
2089 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
2090 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
2091 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
2092 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
2096 <!-- _______________________________________________________________________ -->
2097 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
2098 Instruction</a> </div>
2099 <div class="doc_text">
2102 <pre> <result> = ashr <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2106 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
2107 operand shifted to the right a specified number of bits.</p>
2110 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
2111 <a href="#t_integer">integer</a> type.</p>
2114 <p>This instruction always performs an arithmetic shift right operation,
2115 The most significant bits of the result will be filled with the sign bit
2116 of <tt>var1</tt>.</p>
2120 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
2121 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
2122 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
2123 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
2127 <!-- _______________________________________________________________________ -->
2128 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
2129 Instruction</a> </div>
2130 <div class="doc_text">
2132 <pre> <result> = and <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2135 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
2136 its two operands.</p>
2138 <p>The two arguments to the '<tt>and</tt>' instruction must be <a
2139 href="#t_integer">integer</a> values. Both arguments must have
2140 identical types.</p>
2142 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
2144 <div style="align: center">
2145 <table border="1" cellspacing="0" cellpadding="4">
2176 <pre> <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
2177 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
2178 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
2181 <!-- _______________________________________________________________________ -->
2182 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
2183 <div class="doc_text">
2185 <pre> <result> = or <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2188 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
2189 or of its two operands.</p>
2191 <p>The two arguments to the '<tt>or</tt>' instruction must be <a
2192 href="#t_integer">integer</a> values. Both arguments must have
2193 identical types.</p>
2195 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
2197 <div style="align: center">
2198 <table border="1" cellspacing="0" cellpadding="4">
2229 <pre> <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
2230 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
2231 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
2234 <!-- _______________________________________________________________________ -->
2235 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
2236 Instruction</a> </div>
2237 <div class="doc_text">
2239 <pre> <result> = xor <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2242 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
2243 or of its two operands. The <tt>xor</tt> is used to implement the
2244 "one's complement" operation, which is the "~" operator in C.</p>
2246 <p>The two arguments to the '<tt>xor</tt>' instruction must be <a
2247 href="#t_integer">integer</a> values. Both arguments must have
2248 identical types.</p>
2250 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
2252 <div style="align: center">
2253 <table border="1" cellspacing="0" cellpadding="4">
2285 <pre> <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
2286 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
2287 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
2288 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
2292 <!-- ======================================================================= -->
2293 <div class="doc_subsection">
2294 <a name="vectorops">Vector Operations</a>
2297 <div class="doc_text">
2299 <p>LLVM supports several instructions to represent vector operations in a
2300 target-independent manner. This instructions cover the element-access and
2301 vector-specific operations needed to process vectors effectively. While LLVM
2302 does directly support these vector operations, many sophisticated algorithms
2303 will want to use target-specific intrinsics to take full advantage of a specific
2308 <!-- _______________________________________________________________________ -->
2309 <div class="doc_subsubsection">
2310 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2313 <div class="doc_text">
2318 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
2324 The '<tt>extractelement</tt>' instruction extracts a single scalar
2325 element from a vector at a specified index.
2332 The first operand of an '<tt>extractelement</tt>' instruction is a
2333 value of <a href="#t_vector">vector</a> type. The second operand is
2334 an index indicating the position from which to extract the element.
2335 The index may be a variable.</p>
2340 The result is a scalar of the same type as the element type of
2341 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2342 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2343 results are undefined.
2349 %result = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
2354 <!-- _______________________________________________________________________ -->
2355 <div class="doc_subsubsection">
2356 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
2359 <div class="doc_text">
2364 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
2370 The '<tt>insertelement</tt>' instruction inserts a scalar
2371 element into a vector at a specified index.
2378 The first operand of an '<tt>insertelement</tt>' instruction is a
2379 value of <a href="#t_vector">vector</a> type. The second operand is a
2380 scalar value whose type must equal the element type of the first
2381 operand. The third operand is an index indicating the position at
2382 which to insert the value. The index may be a variable.</p>
2387 The result is a vector of the same type as <tt>val</tt>. Its
2388 element values are those of <tt>val</tt> except at position
2389 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
2390 exceeds the length of <tt>val</tt>, the results are undefined.
2396 %result = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
2400 <!-- _______________________________________________________________________ -->
2401 <div class="doc_subsubsection">
2402 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
2405 <div class="doc_text">
2410 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <n x i32> <mask> <i>; yields <n x <ty>></i>
2416 The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
2417 from two input vectors, returning a vector of the same type.
2423 The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
2424 with types that match each other and types that match the result of the
2425 instruction. The third argument is a shuffle mask, which has the same number
2426 of elements as the other vector type, but whose element type is always 'i32'.
2430 The shuffle mask operand is required to be a constant vector with either
2431 constant integer or undef values.
2437 The elements of the two input vectors are numbered from left to right across
2438 both of the vectors. The shuffle mask operand specifies, for each element of
2439 the result vector, which element of the two input registers the result element
2440 gets. The element selector may be undef (meaning "don't care") and the second
2441 operand may be undef if performing a shuffle from only one vector.
2447 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
2448 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
2449 %result = shufflevector <4 x i32> %v1, <4 x i32> undef,
2450 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
2455 <!-- ======================================================================= -->
2456 <div class="doc_subsection">
2457 <a name="memoryops">Memory Access and Addressing Operations</a>
2460 <div class="doc_text">
2462 <p>A key design point of an SSA-based representation is how it
2463 represents memory. In LLVM, no memory locations are in SSA form, which
2464 makes things very simple. This section describes how to read, write,
2465 allocate, and free memory in LLVM.</p>
2469 <!-- _______________________________________________________________________ -->
2470 <div class="doc_subsubsection">
2471 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
2474 <div class="doc_text">
2479 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2484 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
2485 heap and returns a pointer to it.</p>
2489 <p>The '<tt>malloc</tt>' instruction allocates
2490 <tt>sizeof(<type>)*NumElements</tt>
2491 bytes of memory from the operating system and returns a pointer of the
2492 appropriate type to the program. If "NumElements" is specified, it is the
2493 number of elements allocated. If an alignment is specified, the value result
2494 of the allocation is guaranteed to be aligned to at least that boundary. If
2495 not specified, or if zero, the target can choose to align the allocation on any
2496 convenient boundary.</p>
2498 <p>'<tt>type</tt>' must be a sized type.</p>
2502 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
2503 a pointer is returned.</p>
2508 %array = malloc [4 x i8 ] <i>; yields {[%4 x i8]*}:array</i>
2510 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
2511 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
2512 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
2513 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
2514 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
2518 <!-- _______________________________________________________________________ -->
2519 <div class="doc_subsubsection">
2520 <a name="i_free">'<tt>free</tt>' Instruction</a>
2523 <div class="doc_text">
2528 free <type> <value> <i>; yields {void}</i>
2533 <p>The '<tt>free</tt>' instruction returns memory back to the unused
2534 memory heap to be reallocated in the future.</p>
2538 <p>'<tt>value</tt>' shall be a pointer value that points to a value
2539 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
2544 <p>Access to the memory pointed to by the pointer is no longer defined
2545 after this instruction executes.</p>
2550 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
2551 free [4 x i8]* %array
2555 <!-- _______________________________________________________________________ -->
2556 <div class="doc_subsubsection">
2557 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
2560 <div class="doc_text">
2565 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2570 <p>The '<tt>alloca</tt>' instruction allocates memory on the current
2571 stack frame of the procedure that is live until the current function
2572 returns to its caller.</p>
2576 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
2577 bytes of memory on the runtime stack, returning a pointer of the
2578 appropriate type to the program. If "NumElements" is specified, it is the
2579 number of elements allocated. If an alignment is specified, the value result
2580 of the allocation is guaranteed to be aligned to at least that boundary. If
2581 not specified, or if zero, the target can choose to align the allocation on any
2582 convenient boundary.</p>
2584 <p>'<tt>type</tt>' may be any sized type.</p>
2588 <p>Memory is allocated; a pointer is returned. '<tt>alloca</tt>'d
2589 memory is automatically released when the function returns. The '<tt>alloca</tt>'
2590 instruction is commonly used to represent automatic variables that must
2591 have an address available. When the function returns (either with the <tt><a
2592 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
2593 instructions), the memory is reclaimed.</p>
2598 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
2599 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
2600 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
2601 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
2605 <!-- _______________________________________________________________________ -->
2606 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
2607 Instruction</a> </div>
2608 <div class="doc_text">
2610 <pre> <result> = load <ty>* <pointer><br> <result> = volatile load <ty>* <pointer><br></pre>
2612 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
2614 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
2615 address from which to load. The pointer must point to a <a
2616 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
2617 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
2618 the number or order of execution of this <tt>load</tt> with other
2619 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
2622 <p>The location of memory pointed to is loaded.</p>
2624 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
2626 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
2627 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
2630 <!-- _______________________________________________________________________ -->
2631 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
2632 Instruction</a> </div>
2633 <div class="doc_text">
2635 <pre> store <ty> <value>, <ty>* <pointer> <i>; yields {void}</i>
2636 volatile store <ty> <value>, <ty>* <pointer> <i>; yields {void}</i>
2639 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
2641 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
2642 to store and an address in which to store it. The type of the '<tt><pointer></tt>'
2643 operand must be a pointer to the type of the '<tt><value></tt>'
2644 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
2645 optimizer is not allowed to modify the number or order of execution of
2646 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
2647 href="#i_store">store</a></tt> instructions.</p>
2649 <p>The contents of memory are updated to contain '<tt><value></tt>'
2650 at the location specified by the '<tt><pointer></tt>' operand.</p>
2652 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
2654 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
2655 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
2659 <!-- _______________________________________________________________________ -->
2660 <div class="doc_subsubsection">
2661 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
2664 <div class="doc_text">
2667 <result> = getelementptr <ty>* <ptrval>{, <ty> <idx>}*
2673 The '<tt>getelementptr</tt>' instruction is used to get the address of a
2674 subelement of an aggregate data structure.</p>
2678 <p>This instruction takes a list of integer operands that indicate what
2679 elements of the aggregate object to index to. The actual types of the arguments
2680 provided depend on the type of the first pointer argument. The
2681 '<tt>getelementptr</tt>' instruction is used to index down through the type
2682 levels of a structure or to a specific index in an array. When indexing into a
2683 structure, only <tt>i32</tt> integer constants are allowed. When indexing
2684 into an array or pointer, only integers of 32 or 64 bits are allowed, and will
2685 be sign extended to 64-bit values.</p>
2687 <p>For example, let's consider a C code fragment and how it gets
2688 compiled to LLVM:</p>
2702 define i32 *foo(struct ST *s) {
2703 return &s[1].Z.B[5][13];
2707 <p>The LLVM code generated by the GCC frontend is:</p>
2710 %RT = type { i8 , [10 x [20 x i32]], i8 }
2711 %ST = type { i32, double, %RT }
2713 define i32* %foo(%ST* %s) {
2715 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
2722 <p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
2723 on the pointer type that is being indexed into. <a href="#t_pointer">Pointer</a>
2724 and <a href="#t_array">array</a> types can use a 32-bit or 64-bit
2725 <a href="#t_integer">integer</a> type but the value will always be sign extended
2726 to 64-bits. <a href="#t_struct">Structure</a> types, require <tt>i32</tt>
2727 <b>constants</b>.</p>
2729 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
2730 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
2731 }</tt>' type, a structure. The second index indexes into the third element of
2732 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
2733 i8 }</tt>' type, another structure. The third index indexes into the second
2734 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
2735 array. The two dimensions of the array are subscripted into, yielding an
2736 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
2737 to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
2739 <p>Note that it is perfectly legal to index partially through a
2740 structure, returning a pointer to an inner element. Because of this,
2741 the LLVM code for the given testcase is equivalent to:</p>
2744 define i32* %foo(%ST* %s) {
2745 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
2746 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
2747 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
2748 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
2749 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
2754 <p>Note that it is undefined to access an array out of bounds: array and
2755 pointer indexes must always be within the defined bounds of the array type.
2756 The one exception for this rules is zero length arrays. These arrays are
2757 defined to be accessible as variable length arrays, which requires access
2758 beyond the zero'th element.</p>
2760 <p>The getelementptr instruction is often confusing. For some more insight
2761 into how it works, see <a href="GetElementPtr.html">the getelementptr
2767 <i>; yields [12 x i8]*:aptr</i>
2768 %aptr = getelementptr {i32, [12 x i8]}* %sptr, i64 0, i32 1
2772 <!-- ======================================================================= -->
2773 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
2775 <div class="doc_text">
2776 <p>The instructions in this category are the conversion instructions (casting)
2777 which all take a single operand and a type. They perform various bit conversions
2781 <!-- _______________________________________________________________________ -->
2782 <div class="doc_subsubsection">
2783 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
2785 <div class="doc_text">
2789 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
2794 The '<tt>trunc</tt>' instruction truncates its operand to the type <tt>ty2</tt>.
2799 The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
2800 be an <a href="#t_integer">integer</a> type, and a type that specifies the size
2801 and type of the result, which must be an <a href="#t_integer">integer</a>
2802 type. The bit size of <tt>value</tt> must be larger than the bit size of
2803 <tt>ty2</tt>. Equal sized types are not allowed.</p>
2807 The '<tt>trunc</tt>' instruction truncates the high order bits in <tt>value</tt>
2808 and converts the remaining bits to <tt>ty2</tt>. Since the source size must be
2809 larger than the destination size, <tt>trunc</tt> cannot be a <i>no-op cast</i>.
2810 It will always truncate bits.</p>
2814 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
2815 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
2816 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
2820 <!-- _______________________________________________________________________ -->
2821 <div class="doc_subsubsection">
2822 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
2824 <div class="doc_text">
2828 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
2832 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
2837 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
2838 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
2839 also be of <a href="#t_integer">integer</a> type. The bit size of the
2840 <tt>value</tt> must be smaller than the bit size of the destination type,
2844 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
2845 bits until it reaches the size of the destination type, <tt>ty2</tt>. When the
2846 the operand and the type are the same size, no bit filling is done and the
2847 cast is considered a <i>no-op cast</i> because no bits change (only the type
2850 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
2854 %X = zext i32 257 to i64 <i>; yields i64:257</i>
2855 %Y = zext i1 true to i32 <i>; yields i32:1</i>
2859 <!-- _______________________________________________________________________ -->
2860 <div class="doc_subsubsection">
2861 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
2863 <div class="doc_text">
2867 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
2871 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
2875 The '<tt>sext</tt>' instruction takes a value to cast, which must be of
2876 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
2877 also be of <a href="#t_integer">integer</a> type. The bit size of the
2878 <tt>value</tt> must be smaller than the bit size of the destination type,
2883 The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
2884 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size of
2885 the type <tt>ty2</tt>. When the the operand and the type are the same size,
2886 no bit filling is done and the cast is considered a <i>no-op cast</i> because
2887 no bits change (only the type changes).</p>
2889 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
2893 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
2894 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
2898 <!-- _______________________________________________________________________ -->
2899 <div class="doc_subsubsection">
2900 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
2903 <div class="doc_text">
2908 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
2912 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
2917 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
2918 point</a> value to cast and a <a href="#t_floating">floating point</a> type to
2919 cast it to. The size of <tt>value</tt> must be larger than the size of
2920 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
2921 <i>no-op cast</i>.</p>
2924 <p> The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
2925 <a href="#t_floating">floating point</a> type to a smaller
2926 <a href="#t_floating">floating point</a> type. If the value cannot fit within
2927 the destination type, <tt>ty2</tt>, then the results are undefined.</p>
2931 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
2932 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
2936 <!-- _______________________________________________________________________ -->
2937 <div class="doc_subsubsection">
2938 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
2940 <div class="doc_text">
2944 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
2948 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
2949 floating point value.</p>
2952 <p>The '<tt>fpext</tt>' instruction takes a
2953 <a href="#t_floating">floating point</a> <tt>value</tt> to cast,
2954 and a <a href="#t_floating">floating point</a> type to cast it to. The source
2955 type must be smaller than the destination type.</p>
2958 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
2959 <a href="#t_floating">floating point</a> type to a larger
2960 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
2961 used to make a <i>no-op cast</i> because it always changes bits. Use
2962 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
2966 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
2967 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
2971 <!-- _______________________________________________________________________ -->
2972 <div class="doc_subsubsection">
2973 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
2975 <div class="doc_text">
2979 <result> = fp2uint <ty> <value> to <ty2> <i>; yields ty2</i>
2983 <p>The '<tt>fp2uint</tt>' converts a floating point <tt>value</tt> to its
2984 unsigned integer equivalent of type <tt>ty2</tt>.
2988 <p>The '<tt>fp2uint</tt>' instruction takes a value to cast, which must be a
2989 <a href="#t_floating">floating point</a> value, and a type to cast it to, which
2990 must be an <a href="#t_integer">integer</a> type.</p>
2993 <p> The '<tt>fp2uint</tt>' instruction converts its
2994 <a href="#t_floating">floating point</a> operand into the nearest (rounding
2995 towards zero) unsigned integer value. If the value cannot fit in <tt>ty2</tt>,
2996 the results are undefined.</p>
2998 <p>When converting to i1, the conversion is done as a comparison against
2999 zero. If the <tt>value</tt> was zero, the i1 result will be <tt>false</tt>.
3000 If the <tt>value</tt> was non-zero, the i1 result will be <tt>true</tt>.</p>
3004 %X = fp2uint double 123.0 to i32 <i>; yields i32:123</i>
3005 %Y = fp2uint float 1.0E+300 to i1 <i>; yields i1:true</i>
3006 %X = fp2uint float 1.04E+17 to i8 <i>; yields undefined:1</i>
3010 <!-- _______________________________________________________________________ -->
3011 <div class="doc_subsubsection">
3012 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
3014 <div class="doc_text">
3018 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
3022 <p>The '<tt>fptosi</tt>' instruction converts
3023 <a href="#t_floating">floating point</a> <tt>value</tt> to type <tt>ty2</tt>.
3028 <p> The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
3029 <a href="#t_floating">floating point</a> value, and a type to cast it to, which
3030 must also be an <a href="#t_integer">integer</a> type.</p>
3033 <p>The '<tt>fptosi</tt>' instruction converts its
3034 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3035 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
3036 the results are undefined.</p>
3038 <p>When converting to i1, the conversion is done as a comparison against
3039 zero. If the <tt>value</tt> was zero, the i1 result will be <tt>false</tt>.
3040 If the <tt>value</tt> was non-zero, the i1 result will be <tt>true</tt>.</p>
3044 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
3045 %Y = fptosi float 1.0E-247 to i1 <i>; yields i1:true</i>
3046 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
3050 <!-- _______________________________________________________________________ -->
3051 <div class="doc_subsubsection">
3052 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
3054 <div class="doc_text">
3058 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3062 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
3063 integer and converts that value to the <tt>ty2</tt> type.</p>
3067 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be an
3068 <a href="#t_integer">integer</a> value, and a type to cast it to, which must
3069 be a <a href="#t_floating">floating point</a> type.</p>
3072 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
3073 integer quantity and converts it to the corresponding floating point value. If
3074 the value cannot fit in the floating point value, the results are undefined.</p>
3079 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
3080 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
3084 <!-- _______________________________________________________________________ -->
3085 <div class="doc_subsubsection">
3086 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
3088 <div class="doc_text">
3092 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3096 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed
3097 integer and converts that value to the <tt>ty2</tt> type.</p>
3100 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be an
3101 <a href="#t_integer">integer</a> value, and a type to cast it to, which must be
3102 a <a href="#t_floating">floating point</a> type.</p>
3105 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed
3106 integer quantity and converts it to the corresponding floating point value. If
3107 the value cannot fit in the floating point value, the results are undefined.</p>
3111 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
3112 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
3116 <!-- _______________________________________________________________________ -->
3117 <div class="doc_subsubsection">
3118 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
3120 <div class="doc_text">
3124 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
3128 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
3129 the integer type <tt>ty2</tt>.</p>
3132 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
3133 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
3134 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.
3137 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
3138 <tt>ty2</tt> by interpreting the pointer value as an integer and either
3139 truncating or zero extending that value to the size of the integer type. If
3140 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
3141 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
3142 are the same size, then nothing is done (<i>no-op cast</i>).</p>
3146 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit</i>
3147 %Y = ptrtoint i32* %x to i64 <i>; yields zero extend on 32-bit</i>
3151 <!-- _______________________________________________________________________ -->
3152 <div class="doc_subsubsection">
3153 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
3155 <div class="doc_text">
3159 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
3163 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to
3164 a pointer type, <tt>ty2</tt>.</p>
3167 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
3168 value to cast, and a type to cast it to, which must be a
3169 <a href="#t_pointer">pointer</a> type.
3172 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
3173 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
3174 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
3175 size of a pointer then a truncation is done. If <tt>value</tt> is smaller than
3176 the size of a pointer then a zero extension is done. If they are the same size,
3177 nothing is done (<i>no-op cast</i>).</p>
3181 %X = inttoptr i32 255 to i32* <i>; yields zero extend on 64-bit</i>
3182 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit </i>
3183 %Y = inttoptr i16 0 to i32* <i>; yields zero extend on 32-bit</i>
3187 <!-- _______________________________________________________________________ -->
3188 <div class="doc_subsubsection">
3189 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
3191 <div class="doc_text">
3195 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
3199 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3200 <tt>ty2</tt> without changing any bits.</p>
3203 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be
3204 a first class value, and a type to cast it to, which must also be a <a
3205 href="#t_firstclass">first class</a> type. The bit sizes of <tt>value</tt>
3206 and the destination type, <tt>ty2</tt>, must be identical. If the source
3207 type is a pointer, the destination type must also be a pointer.</p>
3210 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3211 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
3212 this conversion. The conversion is done as if the <tt>value</tt> had been
3213 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only be
3214 converted to other pointer types with this instruction. To convert pointers to
3215 other types, use the <a href="#i_inttoptr">inttoptr</a> or
3216 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
3220 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
3221 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
3222 %Z = bitcast <2xint> %V to i64; <i>; yields i64: %V</i>
3226 <!-- ======================================================================= -->
3227 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
3228 <div class="doc_text">
3229 <p>The instructions in this category are the "miscellaneous"
3230 instructions, which defy better classification.</p>
3233 <!-- _______________________________________________________________________ -->
3234 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
3236 <div class="doc_text">
3238 <pre> <result> = icmp <cond> <ty> <var1>, <var2>
3239 <i>; yields {i1}:result</i>
3242 <p>The '<tt>icmp</tt>' instruction returns a boolean value based on comparison
3243 of its two integer operands.</p>
3245 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
3246 the condition code which indicates the kind of comparison to perform. It is not
3247 a value, just a keyword. The possibilities for the condition code are:
3249 <li><tt>eq</tt>: equal</li>
3250 <li><tt>ne</tt>: not equal </li>
3251 <li><tt>ugt</tt>: unsigned greater than</li>
3252 <li><tt>uge</tt>: unsigned greater or equal</li>
3253 <li><tt>ult</tt>: unsigned less than</li>
3254 <li><tt>ule</tt>: unsigned less or equal</li>
3255 <li><tt>sgt</tt>: signed greater than</li>
3256 <li><tt>sge</tt>: signed greater or equal</li>
3257 <li><tt>slt</tt>: signed less than</li>
3258 <li><tt>sle</tt>: signed less or equal</li>
3260 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
3261 <a href="#t_pointer">pointer</a> typed. They must also be identical types.</p>
3263 <p>The '<tt>icmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3264 the condition code given as <tt>cond</tt>. The comparison performed always
3265 yields a <a href="#t_primitive">i1</a> result, as follows:
3267 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
3268 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3270 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
3271 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3272 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
3273 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3274 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
3275 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3276 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
3277 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3278 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
3279 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3280 <li><tt>sgt</tt>: interprets the operands as signed values and yields
3281 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3282 <li><tt>sge</tt>: interprets the operands as signed values and yields
3283 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3284 <li><tt>slt</tt>: interprets the operands as signed values and yields
3285 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3286 <li><tt>sle</tt>: interprets the operands as signed values and yields
3287 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3289 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
3290 values are treated as integers and then compared.</p>
3293 <pre> <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
3294 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
3295 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
3296 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
3297 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
3298 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
3302 <!-- _______________________________________________________________________ -->
3303 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
3305 <div class="doc_text">
3307 <pre> <result> = fcmp <cond> <ty> <var1>, <var2>
3308 <i>; yields {i1}:result</i>
3311 <p>The '<tt>fcmp</tt>' instruction returns a boolean value based on comparison
3312 of its floating point operands.</p>
3314 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
3315 the condition code which indicates the kind of comparison to perform. It is not
3316 a value, just a keyword. The possibilities for the condition code are:
3318 <li><tt>false</tt>: no comparison, always returns false</li>
3319 <li><tt>oeq</tt>: ordered and equal</li>
3320 <li><tt>ogt</tt>: ordered and greater than </li>
3321 <li><tt>oge</tt>: ordered and greater than or equal</li>
3322 <li><tt>olt</tt>: ordered and less than </li>
3323 <li><tt>ole</tt>: ordered and less than or equal</li>
3324 <li><tt>one</tt>: ordered and not equal</li>
3325 <li><tt>ord</tt>: ordered (no nans)</li>
3326 <li><tt>ueq</tt>: unordered or equal</li>
3327 <li><tt>ugt</tt>: unordered or greater than </li>
3328 <li><tt>uge</tt>: unordered or greater than or equal</li>
3329 <li><tt>ult</tt>: unordered or less than </li>
3330 <li><tt>ule</tt>: unordered or less than or equal</li>
3331 <li><tt>une</tt>: unordered or not equal</li>
3332 <li><tt>uno</tt>: unordered (either nans)</li>
3333 <li><tt>true</tt>: no comparison, always returns true</li>
3335 <p>In the preceding, <i>ordered</i> means that neither operand is a QNAN while
3336 <i>unordered</i> means that either operand may be a QNAN.</p>
3337 <p>The <tt>val1</tt> and <tt>val2</tt> arguments must be
3338 <a href="#t_floating">floating point</a> typed. They must have identical
3340 <p>In the foregoing, <i>ordered</i> means that neither operand is a QNAN and
3341 <i>unordered</i> means that either operand is a QNAN.</p>
3343 <p>The '<tt>fcmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3344 the condition code given as <tt>cond</tt>. The comparison performed always
3345 yields a <a href="#t_primitive">i1</a> result, as follows:
3347 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
3348 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3349 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3350 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3351 <tt>var1</tt> is greather than <tt>var2</tt>.</li>
3352 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3353 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3354 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3355 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3356 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3357 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3358 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3359 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3360 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
3361 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
3362 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3363 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
3364 <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3365 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
3366 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3367 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
3368 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3369 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
3370 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3371 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
3372 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3373 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
3374 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
3378 <pre> <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
3379 <result> = icmp one float 4.0, 5.0 <i>; yields: result=true</i>
3380 <result> = icmp olt float 4.0, 5.0 <i>; yields: result=true</i>
3381 <result> = icmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
3385 <!-- _______________________________________________________________________ -->
3386 <div class="doc_subsubsection"> <a name="i_phi">'<tt>phi</tt>'
3387 Instruction</a> </div>
3388 <div class="doc_text">
3390 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
3392 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
3393 the SSA graph representing the function.</p>
3395 <p>The type of the incoming values are specified with the first type
3396 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
3397 as arguments, with one pair for each predecessor basic block of the
3398 current block. Only values of <a href="#t_firstclass">first class</a>
3399 type may be used as the value arguments to the PHI node. Only labels
3400 may be used as the label arguments.</p>
3401 <p>There must be no non-phi instructions between the start of a basic
3402 block and the PHI instructions: i.e. PHI instructions must be first in
3405 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the
3406 value specified by the parameter, depending on which basic block we
3407 came from in the last <a href="#terminators">terminator</a> instruction.</p>
3409 <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>
3412 <!-- _______________________________________________________________________ -->
3413 <div class="doc_subsubsection">
3414 <a name="i_select">'<tt>select</tt>' Instruction</a>
3417 <div class="doc_text">
3422 <result> = select i1 <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
3428 The '<tt>select</tt>' instruction is used to choose one value based on a
3429 condition, without branching.
3436 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.
3442 If the boolean condition evaluates to true, the instruction returns the first
3443 value argument; otherwise, it returns the second value argument.
3449 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
3454 <!-- _______________________________________________________________________ -->
3455 <div class="doc_subsubsection">
3456 <a name="i_call">'<tt>call</tt>' Instruction</a>
3459 <div class="doc_text">
3463 <result> = [tail] call [<a href="#callingconv">cconv</a>] <ty>* <fnptrval>(<param list>)
3468 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
3472 <p>This instruction requires several arguments:</p>
3476 <p>The optional "tail" marker indicates whether the callee function accesses
3477 any allocas or varargs in the caller. If the "tail" marker is present, the
3478 function call is eligible for tail call optimization. Note that calls may
3479 be marked "tail" even if they do not occur before a <a
3480 href="#i_ret"><tt>ret</tt></a> instruction.
3483 <p>The optional "cconv" marker indicates which <a href="#callingconv">calling
3484 convention</a> the call should use. If none is specified, the call defaults
3485 to using C calling conventions.
3488 <p>'<tt>ty</tt>': shall be the signature of the pointer to function value
3489 being invoked. The argument types must match the types implied by this
3490 signature. This type can be omitted if the function is not varargs and
3491 if the function type does not return a pointer to a function.</p>
3494 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
3495 be invoked. In most cases, this is a direct function invocation, but
3496 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
3497 to function value.</p>
3500 <p>'<tt>function args</tt>': argument list whose types match the
3501 function signature argument types. All arguments must be of
3502 <a href="#t_firstclass">first class</a> type. If the function signature
3503 indicates the function accepts a variable number of arguments, the extra
3504 arguments can be specified.</p>
3510 <p>The '<tt>call</tt>' instruction is used to cause control flow to
3511 transfer to a specified function, with its incoming arguments bound to
3512 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
3513 instruction in the called function, control flow continues with the
3514 instruction after the function call, and the return value of the
3515 function is bound to the result argument. This is a simpler case of
3516 the <a href="#i_invoke">invoke</a> instruction.</p>
3521 %retval = call i32 %test(i32 %argc)
3522 call i32(i8 *, ...) *%printf(i8 * %msg, i32 12, i8 42);
3523 %X = tail call i32 %foo()
3524 %Y = tail call <a href="#callingconv">fastcc</a> i32 %foo()
3529 <!-- _______________________________________________________________________ -->
3530 <div class="doc_subsubsection">
3531 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
3534 <div class="doc_text">
3539 <resultval> = va_arg <va_list*> <arglist>, <argty>
3544 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
3545 the "variable argument" area of a function call. It is used to implement the
3546 <tt>va_arg</tt> macro in C.</p>
3550 <p>This instruction takes a <tt>va_list*</tt> value and the type of
3551 the argument. It returns a value of the specified argument type and
3552 increments the <tt>va_list</tt> to point to the next argument. Again, the
3553 actual type of <tt>va_list</tt> is target specific.</p>
3557 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
3558 type from the specified <tt>va_list</tt> and causes the
3559 <tt>va_list</tt> to point to the next argument. For more information,
3560 see the variable argument handling <a href="#int_varargs">Intrinsic
3563 <p>It is legal for this instruction to be called in a function which does not
3564 take a variable number of arguments, for example, the <tt>vfprintf</tt>
3567 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
3568 href="#intrinsics">intrinsic function</a> because it takes a type as an
3573 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
3577 <!-- *********************************************************************** -->
3578 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
3579 <!-- *********************************************************************** -->
3581 <div class="doc_text">
3583 <p>LLVM supports the notion of an "intrinsic function". These functions have
3584 well known names and semantics and are required to follow certain restrictions.
3585 Overall, these intrinsics represent an extension mechanism for the LLVM
3586 language that does not require changing all of the transformations in LLVM to
3587 add to the language (or the bytecode reader/writer, the parser,
3590 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
3591 prefix is reserved in LLVM for intrinsic names; thus, functions may not be named
3592 this. Intrinsic functions must always be external functions: you cannot define
3593 the body of intrinsic functions. Intrinsic functions may only be used in call
3594 or invoke instructions: it is illegal to take the address of an intrinsic
3595 function. Additionally, because intrinsic functions are part of the LLVM
3596 language, it is required that they all be documented here if any are added.</p>
3598 <p>Some intrinsic functions can be overloaded. That is, the intrinsic represents
3599 a family of functions that perform the same operation but on different data
3600 types. This is most frequent with the integer types. Since LLVM can represent
3601 over 8 million different integer types, there is a way to declare an intrinsic
3602 that can be overloaded based on its arguments. Such intrinsics will have the
3603 names of the arbitrary types encoded into the intrinsic function name, each
3604 preceded by a period. For example, the <tt>llvm.ctpop</tt> function can take an
3605 integer of any width. This leads to a family of functions such as
3606 <tt>i32 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i32 @llvm.ctpop.i29(i29 %val)</tt>.
3610 <p>To learn how to add an intrinsic function, please see the
3611 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.
3616 <!-- ======================================================================= -->
3617 <div class="doc_subsection">
3618 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
3621 <div class="doc_text">
3623 <p>Variable argument support is defined in LLVM with the <a
3624 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
3625 intrinsic functions. These functions are related to the similarly
3626 named macros defined in the <tt><stdarg.h></tt> header file.</p>
3628 <p>All of these functions operate on arguments that use a
3629 target-specific value type "<tt>va_list</tt>". The LLVM assembly
3630 language reference manual does not define what this type is, so all
3631 transformations should be prepared to handle intrinsics with any type
3634 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
3635 instruction and the variable argument handling intrinsic functions are
3639 define i32 @test(i32 %X, ...) {
3640 ; Initialize variable argument processing
3642 %ap2 = bitcast i8** %ap to i8*
3643 call void @llvm.va_start(i8* %ap2)
3645 ; Read a single integer argument
3646 %tmp = va_arg i8 ** %ap, i32
3648 ; Demonstrate usage of llvm.va_copy and llvm.va_end
3650 %aq2 = bitcast i8** %aq to i8*
3651 call void @llvm.va_copy(i8 *%aq2, i8* %ap2)
3652 call void @llvm.va_end(i8* %aq2)
3654 ; Stop processing of arguments.
3655 call void @llvm.va_end(i8* %ap2)
3659 declare void @llvm.va_start(i8*)
3660 declare void @llvm.va_copy(i8*, i8*)
3661 declare void @llvm.va_end(i8*)
3665 <!-- _______________________________________________________________________ -->
3666 <div class="doc_subsubsection">
3667 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
3671 <div class="doc_text">
3673 <pre> declare void %llvm.va_start(i8* <arglist>)<br></pre>
3675 <P>The '<tt>llvm.va_start</tt>' intrinsic initializes
3676 <tt>*<arglist></tt> for subsequent use by <tt><a
3677 href="#i_va_arg">va_arg</a></tt>.</p>
3681 <P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
3685 <P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
3686 macro available in C. In a target-dependent way, it initializes the
3687 <tt>va_list</tt> element the argument points to, so that the next call to
3688 <tt>va_arg</tt> will produce the first variable argument passed to the function.
3689 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
3690 last argument of the function, the compiler can figure that out.</p>
3694 <!-- _______________________________________________________________________ -->
3695 <div class="doc_subsubsection">
3696 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
3699 <div class="doc_text">
3701 <pre> declare void @llvm.va_end(i8* <arglist>)<br></pre>
3704 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt><arglist></tt>
3705 which has been initialized previously with <tt><a href="#int_va_start">llvm.va_start</a></tt>
3706 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
3710 <p>The argument is a <tt>va_list</tt> to destroy.</p>
3714 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
3715 macro available in C. In a target-dependent way, it destroys the <tt>va_list</tt>.
3716 Calls to <a href="#int_va_start"><tt>llvm.va_start</tt></a> and <a
3717 href="#int_va_copy"><tt>llvm.va_copy</tt></a> must be matched exactly
3718 with calls to <tt>llvm.va_end</tt>.</p>
3722 <!-- _______________________________________________________________________ -->
3723 <div class="doc_subsubsection">
3724 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
3727 <div class="doc_text">
3732 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
3737 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position from
3738 the source argument list to the destination argument list.</p>
3742 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
3743 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
3748 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt> macro
3749 available in C. In a target-dependent way, it copies the source
3750 <tt>va_list</tt> element into the destination list. This intrinsic is necessary
3751 because the <tt><a href="#int_va_start">llvm.va_start</a></tt> intrinsic may be
3752 arbitrarily complex and require memory allocation, for example.</p>
3756 <!-- ======================================================================= -->
3757 <div class="doc_subsection">
3758 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
3761 <div class="doc_text">
3764 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
3765 Collection</a> requires the implementation and generation of these intrinsics.
3766 These intrinsics allow identification of <a href="#int_gcroot">GC roots on the
3767 stack</a>, as well as garbage collector implementations that require <a
3768 href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a> barriers.
3769 Front-ends for type-safe garbage collected languages should generate these
3770 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
3771 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
3775 <!-- _______________________________________________________________________ -->
3776 <div class="doc_subsubsection">
3777 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
3780 <div class="doc_text">
3785 declare void @llvm.gcroot(<ty>** %ptrloc, <ty2>* %metadata)
3790 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
3791 the code generator, and allows some metadata to be associated with it.</p>
3795 <p>The first argument specifies the address of a stack object that contains the
3796 root pointer. The second pointer (which must be either a constant or a global
3797 value address) contains the meta-data to be associated with the root.</p>
3801 <p>At runtime, a call to this intrinsics stores a null pointer into the "ptrloc"
3802 location. At compile-time, the code generator generates information to allow
3803 the runtime to find the pointer at GC safe points.
3809 <!-- _______________________________________________________________________ -->
3810 <div class="doc_subsubsection">
3811 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
3814 <div class="doc_text">
3819 declare i8 * @llvm.gcread(i8 * %ObjPtr, i8 ** %Ptr)
3824 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
3825 locations, allowing garbage collector implementations that require read
3830 <p>The second argument is the address to read from, which should be an address
3831 allocated from the garbage collector. The first object is a pointer to the
3832 start of the referenced object, if needed by the language runtime (otherwise
3837 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
3838 instruction, but may be replaced with substantially more complex code by the
3839 garbage collector runtime, as needed.</p>
3844 <!-- _______________________________________________________________________ -->
3845 <div class="doc_subsubsection">
3846 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
3849 <div class="doc_text">
3854 declare void @llvm.gcwrite(i8 * %P1, i8 * %Obj, i8 ** %P2)
3859 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
3860 locations, allowing garbage collector implementations that require write
3861 barriers (such as generational or reference counting collectors).</p>
3865 <p>The first argument is the reference to store, the second is the start of the
3866 object to store it to, and the third is the address of the field of Obj to
3867 store to. If the runtime does not require a pointer to the object, Obj may be
3872 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
3873 instruction, but may be replaced with substantially more complex code by the
3874 garbage collector runtime, as needed.</p>
3880 <!-- ======================================================================= -->
3881 <div class="doc_subsection">
3882 <a name="int_codegen">Code Generator Intrinsics</a>
3885 <div class="doc_text">
3887 These intrinsics are provided by LLVM to expose special features that may only
3888 be implemented with code generator support.
3893 <!-- _______________________________________________________________________ -->
3894 <div class="doc_subsubsection">
3895 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
3898 <div class="doc_text">
3902 declare i8 *@llvm.returnaddress(i32 <level>)
3908 The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
3909 target-specific value indicating the return address of the current function
3910 or one of its callers.
3916 The argument to this intrinsic indicates which function to return the address
3917 for. Zero indicates the calling function, one indicates its caller, etc. The
3918 argument is <b>required</b> to be a constant integer value.
3924 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
3925 the return address of the specified call frame, or zero if it cannot be
3926 identified. The value returned by this intrinsic is likely to be incorrect or 0
3927 for arguments other than zero, so it should only be used for debugging purposes.
3931 Note that calling this intrinsic does not prevent function inlining or other
3932 aggressive transformations, so the value returned may not be that of the obvious
3933 source-language caller.
3938 <!-- _______________________________________________________________________ -->
3939 <div class="doc_subsubsection">
3940 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
3943 <div class="doc_text">
3947 declare i8 *@llvm.frameaddress(i32 <level>)
3953 The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
3954 target-specific frame pointer value for the specified stack frame.
3960 The argument to this intrinsic indicates which function to return the frame
3961 pointer for. Zero indicates the calling function, one indicates its caller,
3962 etc. The argument is <b>required</b> to be a constant integer value.
3968 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
3969 the frame address of the specified call frame, or zero if it cannot be
3970 identified. The value returned by this intrinsic is likely to be incorrect or 0
3971 for arguments other than zero, so it should only be used for debugging purposes.
3975 Note that calling this intrinsic does not prevent function inlining or other
3976 aggressive transformations, so the value returned may not be that of the obvious
3977 source-language caller.
3981 <!-- _______________________________________________________________________ -->
3982 <div class="doc_subsubsection">
3983 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
3986 <div class="doc_text">
3990 declare i8 *@llvm.stacksave()
3996 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
3997 the function stack, for use with <a href="#int_stackrestore">
3998 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
3999 features like scoped automatic variable sized arrays in C99.
4005 This intrinsic returns a opaque pointer value that can be passed to <a
4006 href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
4007 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
4008 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
4009 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
4010 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
4011 that were allocated after the <tt>llvm.stacksave</tt> was executed.
4016 <!-- _______________________________________________________________________ -->
4017 <div class="doc_subsubsection">
4018 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
4021 <div class="doc_text">
4025 declare void @llvm.stackrestore(i8 * %ptr)
4031 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
4032 the function stack to the state it was in when the corresponding <a
4033 href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
4034 useful for implementing language features like scoped automatic variable sized
4041 See the description for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.
4047 <!-- _______________________________________________________________________ -->
4048 <div class="doc_subsubsection">
4049 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
4052 <div class="doc_text">
4056 declare void @llvm.prefetch(i8 * <address>,
4057 i32 <rw>, i32 <locality>)
4064 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
4065 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
4067 effect on the behavior of the program but can change its performance
4074 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
4075 determining if the fetch should be for a read (0) or write (1), and
4076 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
4077 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
4078 <tt>locality</tt> arguments must be constant integers.
4084 This intrinsic does not modify the behavior of the program. In particular,
4085 prefetches cannot trap and do not produce a value. On targets that support this
4086 intrinsic, the prefetch can provide hints to the processor cache for better
4092 <!-- _______________________________________________________________________ -->
4093 <div class="doc_subsubsection">
4094 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
4097 <div class="doc_text">
4101 declare void @llvm.pcmarker( i32 <id> )
4108 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
4110 code to simulators and other tools. The method is target specific, but it is
4111 expected that the marker will use exported symbols to transmit the PC of the marker.
4112 The marker makes no guarantees that it will remain with any specific instruction
4113 after optimizations. It is possible that the presence of a marker will inhibit
4114 optimizations. The intended use is to be inserted after optimizations to allow
4115 correlations of simulation runs.
4121 <tt>id</tt> is a numerical id identifying the marker.
4127 This intrinsic does not modify the behavior of the program. Backends that do not
4128 support this intrinisic may ignore it.
4133 <!-- _______________________________________________________________________ -->
4134 <div class="doc_subsubsection">
4135 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
4138 <div class="doc_text">
4142 declare i64 @llvm.readcyclecounter( )
4149 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
4150 counter register (or similar low latency, high accuracy clocks) on those targets
4151 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
4152 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
4153 should only be used for small timings.
4159 When directly supported, reading the cycle counter should not modify any memory.
4160 Implementations are allowed to either return a application specific value or a
4161 system wide value. On backends without support, this is lowered to a constant 0.
4166 <!-- ======================================================================= -->
4167 <div class="doc_subsection">
4168 <a name="int_libc">Standard C Library Intrinsics</a>
4171 <div class="doc_text">
4173 LLVM provides intrinsics for a few important standard C library functions.
4174 These intrinsics allow source-language front-ends to pass information about the
4175 alignment of the pointer arguments to the code generator, providing opportunity
4176 for more efficient code generation.
4181 <!-- _______________________________________________________________________ -->
4182 <div class="doc_subsubsection">
4183 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
4186 <div class="doc_text">
4190 declare void @llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
4191 i32 <len>, i32 <align>)
4192 declare void @llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
4193 i64 <len>, i32 <align>)
4199 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4200 location to the destination location.
4204 Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
4205 intrinsics do not return a value, and takes an extra alignment argument.
4211 The first argument is a pointer to the destination, the second is a pointer to
4212 the source. The third argument is an integer argument
4213 specifying the number of bytes to copy, and the fourth argument is the alignment
4214 of the source and destination locations.
4218 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4219 the caller guarantees that both the source and destination pointers are aligned
4226 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4227 location to the destination location, which are not allowed to overlap. It
4228 copies "len" bytes of memory over. If the argument is known to be aligned to
4229 some boundary, this can be specified as the fourth argument, otherwise it should
4235 <!-- _______________________________________________________________________ -->
4236 <div class="doc_subsubsection">
4237 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
4240 <div class="doc_text">
4244 declare void @llvm.memmove.i32(i8 * <dest>, i8 * <src>,
4245 i32 <len>, i32 <align>)
4246 declare void @llvm.memmove.i64(i8 * <dest>, i8 * <src>,
4247 i64 <len>, i32 <align>)
4253 The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
4254 location to the destination location. It is similar to the
4255 '<tt>llvm.memcmp</tt>' intrinsic but allows the two memory locations to overlap.
4259 Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
4260 intrinsics do not return a value, and takes an extra alignment argument.
4266 The first argument is a pointer to the destination, the second is a pointer to
4267 the source. The third argument is an integer argument
4268 specifying the number of bytes to copy, and the fourth argument is the alignment
4269 of the source and destination locations.
4273 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4274 the caller guarantees that the source and destination pointers are aligned to
4281 The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
4282 location to the destination location, which may overlap. It
4283 copies "len" bytes of memory over. If the argument is known to be aligned to
4284 some boundary, this can be specified as the fourth argument, otherwise it should
4290 <!-- _______________________________________________________________________ -->
4291 <div class="doc_subsubsection">
4292 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
4295 <div class="doc_text">
4299 declare void @llvm.memset.i32(i8 * <dest>, i8 <val>,
4300 i32 <len>, i32 <align>)
4301 declare void @llvm.memset.i64(i8 * <dest>, i8 <val>,
4302 i64 <len>, i32 <align>)
4308 The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
4313 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
4314 does not return a value, and takes an extra alignment argument.
4320 The first argument is a pointer to the destination to fill, the second is the
4321 byte value to fill it with, the third argument is an integer
4322 argument specifying the number of bytes to fill, and the fourth argument is the
4323 known alignment of destination location.
4327 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4328 the caller guarantees that the destination pointer is aligned to that boundary.
4334 The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
4336 destination location. If the argument is known to be aligned to some boundary,
4337 this can be specified as the fourth argument, otherwise it should be set to 0 or
4343 <!-- _______________________________________________________________________ -->
4344 <div class="doc_subsubsection">
4345 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
4348 <div class="doc_text">
4352 declare float @llvm.sqrt.f32(float %Val)
4353 declare double @llvm.sqrt.f64(double %Val)
4359 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
4360 returning the same value as the libm '<tt>sqrt</tt>' function would. Unlike
4361 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
4362 negative numbers (which allows for better optimization).
4368 The argument and return value are floating point numbers of the same type.
4374 This function returns the sqrt of the specified operand if it is a positive
4375 floating point number.
4379 <!-- _______________________________________________________________________ -->
4380 <div class="doc_subsubsection">
4381 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
4384 <div class="doc_text">
4388 declare float @llvm.powi.f32(float %Val, i32 %power)
4389 declare double @llvm.powi.f64(double %Val, i32 %power)
4395 The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
4396 specified (positive or negative) power. The order of evaluation of
4397 multiplications is not defined.
4403 The second argument is an integer power, and the first is a value to raise to
4410 This function returns the first value raised to the second power with an
4411 unspecified sequence of rounding operations.</p>
4415 <!-- ======================================================================= -->
4416 <div class="doc_subsection">
4417 <a name="int_manip">Bit Manipulation Intrinsics</a>
4420 <div class="doc_text">
4422 LLVM provides intrinsics for a few important bit manipulation operations.
4423 These allow efficient code generation for some algorithms.
4428 <!-- _______________________________________________________________________ -->
4429 <div class="doc_subsubsection">
4430 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
4433 <div class="doc_text">
4436 <p>This is an overloaded intrinsic function. You can use bswap on any integer
4437 type that is an even number of bytes (i.e. BitWidth % 16 == 0). Note the suffix
4438 that includes the type for the result and the operand.
4440 declare i16 @llvm.bswap.i16.i16(i16 <id>)
4441 declare i32 @llvm.bswap.i32.i32(i32 <id>)
4442 declare i64 @llvm.bswap.i64.i64(i64 <id>)
4448 The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
4449 values with an even number of bytes (positive multiple of 16 bits). These are
4450 useful for performing operations on data that is not in the target's native
4457 The <tt>llvm.bswap.16.i16</tt> intrinsic returns an i16 value that has the high
4458 and low byte of the input i16 swapped. Similarly, the <tt>llvm.bswap.i32</tt>
4459 intrinsic returns an i32 value that has the four bytes of the input i32
4460 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the returned
4461 i32 will have its bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i48.i48</tt>,
4462 <tt>llvm.bswap.i64.i64</tt> and other intrinsics extend this concept to
4463 additional even-byte lengths (6 bytes, 8 bytes and more, respectively).
4468 <!-- _______________________________________________________________________ -->
4469 <div class="doc_subsubsection">
4470 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
4473 <div class="doc_text">
4476 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
4477 width. Not all targets support all bit widths however.
4479 declare i32 @llvm.ctpop.i8 (i8 <src>)
4480 declare i32 @llvm.ctpop.i16(i16 <src>)
4481 declare i32 @llvm.ctpop.i32(i32 <src>)
4482 declare i32 @llvm.ctpop.i64(i64 <src>)
4483 declare i32 @llvm.ctpop.i256(i256 <src>)
4489 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
4496 The only argument is the value to be counted. The argument may be of any
4497 integer type. The return type must match the argument type.
4503 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
4507 <!-- _______________________________________________________________________ -->
4508 <div class="doc_subsubsection">
4509 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
4512 <div class="doc_text">
4515 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
4516 integer bit width. Not all targets support all bit widths however.
4518 declare i32 @llvm.ctlz.i8 (i8 <src>)
4519 declare i32 @llvm.ctlz.i16(i16 <src>)
4520 declare i32 @llvm.ctlz.i32(i32 <src>)
4521 declare i32 @llvm.ctlz.i64(i64 <src>)
4522 declare i32 @llvm.ctlz.i256(i256 <src>)
4528 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
4529 leading zeros in a variable.
4535 The only argument is the value to be counted. The argument may be of any
4536 integer type. The return type must match the argument type.
4542 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
4543 in a variable. If the src == 0 then the result is the size in bits of the type
4544 of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.
4550 <!-- _______________________________________________________________________ -->
4551 <div class="doc_subsubsection">
4552 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
4555 <div class="doc_text">
4558 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
4559 integer bit width. Not all targets support all bit widths however.
4561 declare i32 @llvm.cttz.i8 (i8 <src>)
4562 declare i32 @llvm.cttz.i16(i16 <src>)
4563 declare i32 @llvm.cttz.i32(i32 <src>)
4564 declare i32 @llvm.cttz.i64(i64 <src>)
4565 declare i32 @llvm.cttz.i256(i256 <src>)
4571 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
4578 The only argument is the value to be counted. The argument may be of any
4579 integer type. The return type must match the argument type.
4585 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
4586 in a variable. If the src == 0 then the result is the size in bits of the type
4587 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
4591 <!-- _______________________________________________________________________ -->
4592 <div class="doc_subsubsection">
4593 <a name="int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic</a>
4596 <div class="doc_text">
4599 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.select</tt>
4600 on any integer bit width.
4602 declare i17 @llvm.part.select.i17.i17 (i17 %val, i32 %loBit, i32 %hiBit)
4603 declare i29 @llvm.part.select.i29.i29 (i29 %val, i32 %loBit, i32 %hiBit)
4607 <p>The '<tt>llvm.part.select</tt>' family of intrinsic functions selects a
4608 range of bits from an integer value and returns them in the same bit width as
4609 the original value.</p>
4612 <p>The first argument, <tt>%val</tt> and the result may be integer types of
4613 any bit width but they must have the same bit width. The second and third
4614 arguments must be <tt>i32</tt> type since they specify only a bit index.</p>
4617 <p>The operation of the '<tt>llvm.part.select</tt>' intrinsic has two modes
4618 of operation: forwards and reverse. If <tt>%loBit</tt> is greater than
4619 <tt>%hiBits</tt> then the intrinsic operates in reverse mode. Otherwise it
4620 operates in forward mode.</p>
4621 <p>In forward mode, this intrinsic is the equivalent of shifting <tt>%val</tt>
4622 right by <tt>%loBit</tt> bits and then ANDing it with a mask with
4623 only the <tt>%hiBit - %loBit</tt> bits set, as follows:</p>
4625 <li>The <tt>%val</tt> is shifted right (LSHR) by the number of bits specified
4626 by <tt>%loBits</tt>. This normalizes the value to the low order bits.</li>
4627 <li>The <tt>%loBits</tt> value is subtracted from the <tt>%hiBits</tt> value
4628 to determine the number of bits to retain.</li>
4629 <li>A mask of the retained bits is created by shifting a -1 value.</li>
4630 <li>The mask is ANDed with <tt>%val</tt> to produce the result.
4632 <p>In reverse mode, a similar computation is made except that:</p>
4634 <li>The bits selected wrap around to include both the highest and lowest bits.
4635 For example, part.select(i16 X, 4, 7) selects bits from X with a mask of
4636 0x00F0 (forwards case) while part.select(i16 X, 8, 3) selects bits from X
4637 with a mask of 0xFF0F.</li>
4638 <li>The bits returned in the reverse case are reversed. So, if X has the value
4639 0x6ACF and we apply part.select(i16 X, 8, 3) to it, we get back the value
4644 <!-- ======================================================================= -->
4645 <div class="doc_subsection">
4646 <a name="int_debugger">Debugger Intrinsics</a>
4649 <div class="doc_text">
4651 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
4652 are described in the <a
4653 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
4654 Debugging</a> document.
4659 <!-- ======================================================================= -->
4660 <div class="doc_subsection">
4661 <a name="int_eh">Exception Handling Intrinsics</a>
4664 <div class="doc_text">
4665 <p> The LLVM exception handling intrinsics (which all start with
4666 <tt>llvm.eh.</tt> prefix), are described in the <a
4667 href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
4668 Handling</a> document. </p>
4672 <!-- *********************************************************************** -->
4675 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
4676 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
4677 <a href="http://validator.w3.org/check/referer"><img
4678 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a>
4680 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
4681 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
4682 Last modified: $Date$