1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <title>LLVM Assembly Language Reference Manual</title>
6 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
7 <meta name="author" content="Chris Lattner">
8 <meta name="description"
9 content="LLVM Assembly Language Reference Manual.">
10 <link rel="stylesheet" href="llvm.css" type="text/css">
15 <div class="doc_title"> LLVM Language Reference Manual </div>
17 <li><a href="#abstract">Abstract</a></li>
18 <li><a href="#introduction">Introduction</a></li>
19 <li><a href="#identifiers">Identifiers</a></li>
20 <li><a href="#highlevel">High Level Structure</a>
22 <li><a href="#modulestructure">Module Structure</a></li>
23 <li><a href="#linkage">Linkage Types</a></li>
24 <li><a href="#callingconv">Calling Conventions</a></li>
25 <li><a href="#globalvars">Global Variables</a></li>
26 <li><a href="#functionstructure">Functions</a></li>
27 <li><a href="#aliasstructure">Aliases</a>
28 <li><a href="#paramattrs">Parameter Attributes</a></li>
29 <li><a href="#gc">Garbage Collector Names</a></li>
30 <li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
31 <li><a href="#datalayout">Data Layout</a></li>
34 <li><a href="#typesystem">Type System</a>
36 <li><a href="#t_classifications">Type Classifications</a></li>
37 <li><a href="#t_primitive">Primitive Types</a>
39 <li><a href="#t_floating">Floating Point Types</a></li>
40 <li><a href="#t_void">Void Type</a></li>
41 <li><a href="#t_label">Label Type</a></li>
44 <li><a href="#t_derived">Derived Types</a>
46 <li><a href="#t_integer">Integer Type</a></li>
47 <li><a href="#t_array">Array Type</a></li>
48 <li><a href="#t_function">Function Type</a></li>
49 <li><a href="#t_pointer">Pointer Type</a></li>
50 <li><a href="#t_struct">Structure Type</a></li>
51 <li><a href="#t_pstruct">Packed Structure Type</a></li>
52 <li><a href="#t_vector">Vector Type</a></li>
53 <li><a href="#t_opaque">Opaque Type</a></li>
58 <li><a href="#constants">Constants</a>
60 <li><a href="#simpleconstants">Simple Constants</a>
61 <li><a href="#aggregateconstants">Aggregate Constants</a>
62 <li><a href="#globalconstants">Global Variable and Function Addresses</a>
63 <li><a href="#undefvalues">Undefined Values</a>
64 <li><a href="#constantexprs">Constant Expressions</a>
67 <li><a href="#othervalues">Other Values</a>
69 <li><a href="#inlineasm">Inline Assembler Expressions</a>
72 <li><a href="#instref">Instruction Reference</a>
74 <li><a href="#terminators">Terminator Instructions</a>
76 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
77 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
78 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
79 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
80 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
81 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
84 <li><a href="#binaryops">Binary Operations</a>
86 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
87 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
88 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
89 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
90 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
91 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
92 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
93 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
94 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
97 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
99 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
100 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
101 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
102 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
103 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
104 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
107 <li><a href="#vectorops">Vector Operations</a>
109 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
110 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
111 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
114 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
116 <li><a href="#i_malloc">'<tt>malloc</tt>' Instruction</a></li>
117 <li><a href="#i_free">'<tt>free</tt>' Instruction</a></li>
118 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
119 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
120 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
121 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
124 <li><a href="#convertops">Conversion Operations</a>
126 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
127 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
128 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
129 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
130 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
131 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
132 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
133 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
134 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
135 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
136 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
137 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
139 <li><a href="#otherops">Other Operations</a>
141 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
142 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
143 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
144 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
145 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
146 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
151 <li><a href="#intrinsics">Intrinsic Functions</a>
153 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
155 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
156 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
157 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
160 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
162 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
163 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
164 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
167 <li><a href="#int_codegen">Code Generator Intrinsics</a>
169 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
170 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
171 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
172 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
173 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
174 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
175 <li><a href="#int_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
178 <li><a href="#int_libc">Standard C Library Intrinsics</a>
180 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
181 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
182 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
183 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
184 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
185 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
186 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
187 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
190 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
192 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
193 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
194 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
195 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
196 <li><a href="#int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic </a></li>
197 <li><a href="#int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic </a></li>
200 <li><a href="#int_debugger">Debugger intrinsics</a></li>
201 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
202 <li><a href="#int_trampoline">Trampoline Intrinsic</a>
204 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
207 <li><a href="#int_general">General intrinsics</a>
209 <li><a href="#int_var_annotation">
210 <tt>llvm.var.annotation</tt>' Intrinsic</a></li>
213 <li><a href="#int_annotation">
214 <tt>llvm.annotation.*</tt>' Intrinsic</a></li>
221 <div class="doc_author">
222 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
223 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
226 <!-- *********************************************************************** -->
227 <div class="doc_section"> <a name="abstract">Abstract </a></div>
228 <!-- *********************************************************************** -->
230 <div class="doc_text">
231 <p>This document is a reference manual for the LLVM assembly language.
232 LLVM is an SSA based representation that provides type safety,
233 low-level operations, flexibility, and the capability of representing
234 'all' high-level languages cleanly. It is the common code
235 representation used throughout all phases of the LLVM compilation
239 <!-- *********************************************************************** -->
240 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
241 <!-- *********************************************************************** -->
243 <div class="doc_text">
245 <p>The LLVM code representation is designed to be used in three
246 different forms: as an in-memory compiler IR, as an on-disk bitcode
247 representation (suitable for fast loading by a Just-In-Time compiler),
248 and as a human readable assembly language representation. This allows
249 LLVM to provide a powerful intermediate representation for efficient
250 compiler transformations and analysis, while providing a natural means
251 to debug and visualize the transformations. The three different forms
252 of LLVM are all equivalent. This document describes the human readable
253 representation and notation.</p>
255 <p>The LLVM representation aims to be light-weight and low-level
256 while being expressive, typed, and extensible at the same time. It
257 aims to be a "universal IR" of sorts, by being at a low enough level
258 that high-level ideas may be cleanly mapped to it (similar to how
259 microprocessors are "universal IR's", allowing many source languages to
260 be mapped to them). By providing type information, LLVM can be used as
261 the target of optimizations: for example, through pointer analysis, it
262 can be proven that a C automatic variable is never accessed outside of
263 the current function... allowing it to be promoted to a simple SSA
264 value instead of a memory location.</p>
268 <!-- _______________________________________________________________________ -->
269 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
271 <div class="doc_text">
273 <p>It is important to note that this document describes 'well formed'
274 LLVM assembly language. There is a difference between what the parser
275 accepts and what is considered 'well formed'. For example, the
276 following instruction is syntactically okay, but not well formed:</p>
278 <div class="doc_code">
280 %x = <a href="#i_add">add</a> i32 1, %x
284 <p>...because the definition of <tt>%x</tt> does not dominate all of
285 its uses. The LLVM infrastructure provides a verification pass that may
286 be used to verify that an LLVM module is well formed. This pass is
287 automatically run by the parser after parsing input assembly and by
288 the optimizer before it outputs bitcode. The violations pointed out
289 by the verifier pass indicate bugs in transformation passes or input to
293 <!-- Describe the typesetting conventions here. -->
295 <!-- *********************************************************************** -->
296 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
297 <!-- *********************************************************************** -->
299 <div class="doc_text">
301 <p>LLVM identifiers come in two basic types: global and local. Global
302 identifiers (functions, global variables) begin with the @ character. Local
303 identifiers (register names, types) begin with the % character. Additionally,
304 there are three different formats for identifiers, for different purposes:
307 <li>Named values are represented as a string of characters with their prefix.
308 For example, %foo, @DivisionByZero, %a.really.long.identifier. The actual
309 regular expression used is '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
310 Identifiers which require other characters in their names can be surrounded
311 with quotes. In this way, anything except a <tt>"</tt> character can
312 be used in a named value.</li>
314 <li>Unnamed values are represented as an unsigned numeric value with their
315 prefix. For example, %12, @2, %44.</li>
317 <li>Constants, which are described in a <a href="#constants">section about
318 constants</a>, below.</li>
321 <p>LLVM requires that values start with a prefix for two reasons: Compilers
322 don't need to worry about name clashes with reserved words, and the set of
323 reserved words may be expanded in the future without penalty. Additionally,
324 unnamed identifiers allow a compiler to quickly come up with a temporary
325 variable without having to avoid symbol table conflicts.</p>
327 <p>Reserved words in LLVM are very similar to reserved words in other
328 languages. There are keywords for different opcodes
329 ('<tt><a href="#i_add">add</a></tt>',
330 '<tt><a href="#i_bitcast">bitcast</a></tt>',
331 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
332 href="#t_void">void</a></tt>', '<tt><a href="#t_primitive">i32</a></tt>', etc...),
333 and others. These reserved words cannot conflict with variable names, because
334 none of them start with a prefix character ('%' or '@').</p>
336 <p>Here is an example of LLVM code to multiply the integer variable
337 '<tt>%X</tt>' by 8:</p>
341 <div class="doc_code">
343 %result = <a href="#i_mul">mul</a> i32 %X, 8
347 <p>After strength reduction:</p>
349 <div class="doc_code">
351 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
355 <p>And the hard way:</p>
357 <div class="doc_code">
359 <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
360 <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
361 %result = <a href="#i_add">add</a> i32 %1, %1
365 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
366 important lexical features of LLVM:</p>
370 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
373 <li>Unnamed temporaries are created when the result of a computation is not
374 assigned to a named value.</li>
376 <li>Unnamed temporaries are numbered sequentially</li>
380 <p>...and it also shows a convention that we follow in this document. When
381 demonstrating instructions, we will follow an instruction with a comment that
382 defines the type and name of value produced. Comments are shown in italic
387 <!-- *********************************************************************** -->
388 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
389 <!-- *********************************************************************** -->
391 <!-- ======================================================================= -->
392 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
395 <div class="doc_text">
397 <p>LLVM programs are composed of "Module"s, each of which is a
398 translation unit of the input programs. Each module consists of
399 functions, global variables, and symbol table entries. Modules may be
400 combined together with the LLVM linker, which merges function (and
401 global variable) definitions, resolves forward declarations, and merges
402 symbol table entries. Here is an example of the "hello world" module:</p>
404 <div class="doc_code">
405 <pre><i>; Declare the string constant as a global constant...</i>
406 <a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a
407 href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
409 <i>; External declaration of the puts function</i>
410 <a href="#functionstructure">declare</a> i32 @puts(i8 *) <i>; i32(i8 *)* </i>
412 <i>; Definition of main function</i>
413 define i32 @main() { <i>; i32()* </i>
414 <i>; Convert [13x i8 ]* to i8 *...</i>
416 href="#i_getelementptr">getelementptr</a> [13 x i8 ]* @.LC0, i64 0, i64 0 <i>; i8 *</i>
418 <i>; Call puts function to write out the string to stdout...</i>
420 href="#i_call">call</a> i32 @puts(i8 * %cast210) <i>; i32</i>
422 href="#i_ret">ret</a> i32 0<br>}<br>
426 <p>This example is made up of a <a href="#globalvars">global variable</a>
427 named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
428 function, and a <a href="#functionstructure">function definition</a>
429 for "<tt>main</tt>".</p>
431 <p>In general, a module is made up of a list of global values,
432 where both functions and global variables are global values. Global values are
433 represented by a pointer to a memory location (in this case, a pointer to an
434 array of char, and a pointer to a function), and have one of the following <a
435 href="#linkage">linkage types</a>.</p>
439 <!-- ======================================================================= -->
440 <div class="doc_subsection">
441 <a name="linkage">Linkage Types</a>
444 <div class="doc_text">
447 All Global Variables and Functions have one of the following types of linkage:
452 <dt><tt><b><a name="linkage_internal">internal</a></b></tt> </dt>
454 <dd>Global values with internal linkage are only directly accessible by
455 objects in the current module. In particular, linking code into a module with
456 an internal global value may cause the internal to be renamed as necessary to
457 avoid collisions. Because the symbol is internal to the module, all
458 references can be updated. This corresponds to the notion of the
459 '<tt>static</tt>' keyword in C.
462 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
464 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
465 the same name when linkage occurs. This is typically used to implement
466 inline functions, templates, or other code which must be generated in each
467 translation unit that uses it. Unreferenced <tt>linkonce</tt> globals are
468 allowed to be discarded.
471 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
473 <dd>"<tt>weak</tt>" linkage is exactly the same as <tt>linkonce</tt> linkage,
474 except that unreferenced <tt>weak</tt> globals may not be discarded. This is
475 used for globals that may be emitted in multiple translation units, but that
476 are not guaranteed to be emitted into every translation unit that uses them.
477 One example of this are common globals in C, such as "<tt>int X;</tt>" at
481 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
483 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
484 pointer to array type. When two global variables with appending linkage are
485 linked together, the two global arrays are appended together. This is the
486 LLVM, typesafe, equivalent of having the system linker append together
487 "sections" with identical names when .o files are linked.
490 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt>: </dt>
491 <dd>The semantics of this linkage follow the ELF model: the symbol is weak
492 until linked, if not linked, the symbol becomes null instead of being an
496 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
498 <dd>If none of the above identifiers are used, the global is externally
499 visible, meaning that it participates in linkage and can be used to resolve
500 external symbol references.
505 The next two types of linkage are targeted for Microsoft Windows platform
506 only. They are designed to support importing (exporting) symbols from (to)
511 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt>: </dt>
513 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
514 or variable via a global pointer to a pointer that is set up by the DLL
515 exporting the symbol. On Microsoft Windows targets, the pointer name is
516 formed by combining <code>_imp__</code> and the function or variable name.
519 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt>: </dt>
521 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
522 pointer to a pointer in a DLL, so that it can be referenced with the
523 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
524 name is formed by combining <code>_imp__</code> and the function or variable
530 <p><a name="linkage_external"></a>For example, since the "<tt>.LC0</tt>"
531 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
532 variable and was linked with this one, one of the two would be renamed,
533 preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
534 external (i.e., lacking any linkage declarations), they are accessible
535 outside of the current module.</p>
536 <p>It is illegal for a function <i>declaration</i>
537 to have any linkage type other than "externally visible", <tt>dllimport</tt>,
538 or <tt>extern_weak</tt>.</p>
539 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt> and <tt>weak</tt>
543 <!-- ======================================================================= -->
544 <div class="doc_subsection">
545 <a name="callingconv">Calling Conventions</a>
548 <div class="doc_text">
550 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
551 and <a href="#i_invoke">invokes</a> can all have an optional calling convention
552 specified for the call. The calling convention of any pair of dynamic
553 caller/callee must match, or the behavior of the program is undefined. The
554 following calling conventions are supported by LLVM, and more may be added in
558 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
560 <dd>This calling convention (the default if no other calling convention is
561 specified) matches the target C calling conventions. This calling convention
562 supports varargs function calls and tolerates some mismatch in the declared
563 prototype and implemented declaration of the function (as does normal C).
566 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
568 <dd>This calling convention attempts to make calls as fast as possible
569 (e.g. by passing things in registers). This calling convention allows the
570 target to use whatever tricks it wants to produce fast code for the target,
571 without having to conform to an externally specified ABI. Implementations of
572 this convention should allow arbitrary tail call optimization to be supported.
573 This calling convention does not support varargs and requires the prototype of
574 all callees to exactly match the prototype of the function definition.
577 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
579 <dd>This calling convention attempts to make code in the caller as efficient
580 as possible under the assumption that the call is not commonly executed. As
581 such, these calls often preserve all registers so that the call does not break
582 any live ranges in the caller side. This calling convention does not support
583 varargs and requires the prototype of all callees to exactly match the
584 prototype of the function definition.
587 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
589 <dd>Any calling convention may be specified by number, allowing
590 target-specific calling conventions to be used. Target specific calling
591 conventions start at 64.
595 <p>More calling conventions can be added/defined on an as-needed basis, to
596 support pascal conventions or any other well-known target-independent
601 <!-- ======================================================================= -->
602 <div class="doc_subsection">
603 <a name="visibility">Visibility Styles</a>
606 <div class="doc_text">
609 All Global Variables and Functions have one of the following visibility styles:
613 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
615 <dd>On ELF, default visibility means that the declaration is visible to other
616 modules and, in shared libraries, means that the declared entity may be
617 overridden. On Darwin, default visibility means that the declaration is
618 visible to other modules. Default visibility corresponds to "external
619 linkage" in the language.
622 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
624 <dd>Two declarations of an object with hidden visibility refer to the same
625 object if they are in the same shared object. Usually, hidden visibility
626 indicates that the symbol will not be placed into the dynamic symbol table,
627 so no other module (executable or shared library) can reference it
631 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
633 <dd>On ELF, protected visibility indicates that the symbol will be placed in
634 the dynamic symbol table, but that references within the defining module will
635 bind to the local symbol. That is, the symbol cannot be overridden by another
642 <!-- ======================================================================= -->
643 <div class="doc_subsection">
644 <a name="globalvars">Global Variables</a>
647 <div class="doc_text">
649 <p>Global variables define regions of memory allocated at compilation time
650 instead of run-time. Global variables may optionally be initialized, may have
651 an explicit section to be placed in, and may have an optional explicit alignment
652 specified. A variable may be defined as "thread_local", which means that it
653 will not be shared by threads (each thread will have a separated copy of the
654 variable). A variable may be defined as a global "constant," which indicates
655 that the contents of the variable will <b>never</b> be modified (enabling better
656 optimization, allowing the global data to be placed in the read-only section of
657 an executable, etc). Note that variables that need runtime initialization
658 cannot be marked "constant" as there is a store to the variable.</p>
661 LLVM explicitly allows <em>declarations</em> of global variables to be marked
662 constant, even if the final definition of the global is not. This capability
663 can be used to enable slightly better optimization of the program, but requires
664 the language definition to guarantee that optimizations based on the
665 'constantness' are valid for the translation units that do not include the
669 <p>As SSA values, global variables define pointer values that are in
670 scope (i.e. they dominate) all basic blocks in the program. Global
671 variables always define a pointer to their "content" type because they
672 describe a region of memory, and all memory objects in LLVM are
673 accessed through pointers.</p>
675 <p>A global variable may be declared to reside in a target-specifc numbered
676 address space. For targets that support them, address spaces may affect how
677 optimizations are performed and/or what target instructions are used to access
678 the variable. The default address space is zero. The address space qualifier
679 must precede any other attributes.</p>
681 <p>LLVM allows an explicit section to be specified for globals. If the target
682 supports it, it will emit globals to the section specified.</p>
684 <p>An explicit alignment may be specified for a global. If not present, or if
685 the alignment is set to zero, the alignment of the global is set by the target
686 to whatever it feels convenient. If an explicit alignment is specified, the
687 global is forced to have at least that much alignment. All alignments must be
690 <p>For example, the following defines a global in a numbered address space with
691 an initializer, section, and alignment:</p>
693 <div class="doc_code">
695 @G = constant float 1.0 addrspace(5), section "foo", align 4
702 <!-- ======================================================================= -->
703 <div class="doc_subsection">
704 <a name="functionstructure">Functions</a>
707 <div class="doc_text">
709 <p>LLVM function definitions consist of the "<tt>define</tt>" keyord,
710 an optional <a href="#linkage">linkage type</a>, an optional
711 <a href="#visibility">visibility style</a>, an optional
712 <a href="#callingconv">calling convention</a>, a return type, an optional
713 <a href="#paramattrs">parameter attribute</a> for the return type, a function
714 name, a (possibly empty) argument list (each with optional
715 <a href="#paramattrs">parameter attributes</a>), an optional section, an
716 optional alignment, an optional <a href="#gc">garbage collector name</a>, an
717 opening curly brace, a list of basic blocks, and a closing curly brace.
719 LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
720 optional <a href="#linkage">linkage type</a>, an optional
721 <a href="#visibility">visibility style</a>, an optional
722 <a href="#callingconv">calling convention</a>, a return type, an optional
723 <a href="#paramattrs">parameter attribute</a> for the return type, a function
724 name, a possibly empty list of arguments, an optional alignment, and an optional
725 <a href="#gc">garbage collector name</a>.</p>
727 <p>A function definition contains a list of basic blocks, forming the CFG for
728 the function. Each basic block may optionally start with a label (giving the
729 basic block a symbol table entry), contains a list of instructions, and ends
730 with a <a href="#terminators">terminator</a> instruction (such as a branch or
731 function return).</p>
733 <p>The first basic block in a function is special in two ways: it is immediately
734 executed on entrance to the function, and it is not allowed to have predecessor
735 basic blocks (i.e. there can not be any branches to the entry block of a
736 function). Because the block can have no predecessors, it also cannot have any
737 <a href="#i_phi">PHI nodes</a>.</p>
739 <p>LLVM allows an explicit section to be specified for functions. If the target
740 supports it, it will emit functions to the section specified.</p>
742 <p>An explicit alignment may be specified for a function. If not present, or if
743 the alignment is set to zero, the alignment of the function is set by the target
744 to whatever it feels convenient. If an explicit alignment is specified, the
745 function is forced to have at least that much alignment. All alignments must be
751 <!-- ======================================================================= -->
752 <div class="doc_subsection">
753 <a name="aliasstructure">Aliases</a>
755 <div class="doc_text">
756 <p>Aliases act as "second name" for the aliasee value (which can be either
757 function or global variable or bitcast of global value). Aliases may have an
758 optional <a href="#linkage">linkage type</a>, and an
759 optional <a href="#visibility">visibility style</a>.</p>
763 <div class="doc_code">
765 @<Name> = [Linkage] [Visibility] alias <AliaseeTy> @<Aliasee>
773 <!-- ======================================================================= -->
774 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
775 <div class="doc_text">
776 <p>The return type and each parameter of a function type may have a set of
777 <i>parameter attributes</i> associated with them. Parameter attributes are
778 used to communicate additional information about the result or parameters of
779 a function. Parameter attributes are considered to be part of the function,
780 not of the function type, so functions with different parameter attributes
781 can have the same function type.</p>
783 <p>Parameter attributes are simple keywords that follow the type specified. If
784 multiple parameter attributes are needed, they are space separated. For
787 <div class="doc_code">
789 declare i32 @printf(i8* noalias , ...) nounwind
790 declare i32 @atoi(i8*) nounwind readonly
794 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
795 <tt>readonly</tt>) come immediately after the argument list.</p>
797 <p>Currently, only the following parameter attributes are defined:</p>
799 <dt><tt>zeroext</tt></dt>
800 <dd>This indicates that the parameter should be zero extended just before
801 a call to this function.</dd>
802 <dt><tt>signext</tt></dt>
803 <dd>This indicates that the parameter should be sign extended just before
804 a call to this function.</dd>
805 <dt><tt>inreg</tt></dt>
806 <dd>This indicates that the parameter should be placed in register (if
807 possible) during assembling function call. Support for this attribute is
809 <dt><tt>sret</tt></dt>
810 <dd>This indicates that the parameter specifies the address of a structure
811 that is the return value of the function in the source program.</dd>
812 <dt><tt>noalias</tt></dt>
813 <dd>This indicates that the parameter not alias any other object or any
814 other "noalias" objects during the function call.
815 <dt><tt>noreturn</tt></dt>
816 <dd>This function attribute indicates that the function never returns. This
817 indicates to LLVM that every call to this function should be treated as if
818 an <tt>unreachable</tt> instruction immediately followed the call.</dd>
819 <dt><tt>nounwind</tt></dt>
820 <dd>This function attribute indicates that the function type does not use
821 the unwind instruction and does not allow stack unwinding to propagate
823 <dt><tt>nest</tt></dt>
824 <dd>This indicates that the parameter can be excised using the
825 <a href="#int_trampoline">trampoline intrinsics</a>.</dd>
826 <dt><tt>readonly</tt></dt>
827 <dd>This function attribute indicates that the function has no side-effects
828 except for producing a return value or throwing an exception. The value
829 returned must only depend on the function arguments and/or global variables.
830 It may use values obtained by dereferencing pointers.</dd>
831 <dt><tt>readnone</tt></dt>
832 <dd>A <tt>readnone</tt> function has the same restrictions as a <tt>readonly</tt>
833 function, but in addition it is not allowed to dereference any pointer arguments
839 <!-- ======================================================================= -->
840 <div class="doc_subsection">
841 <a name="gc">Garbage Collector Names</a>
844 <div class="doc_text">
845 <p>Each function may specify a garbage collector name, which is simply a
848 <div class="doc_code"><pre
849 >define void @f() gc "name" { ...</pre></div>
851 <p>The compiler declares the supported values of <i>name</i>. Specifying a
852 collector which will cause the compiler to alter its output in order to support
853 the named garbage collection algorithm.</p>
856 <!-- ======================================================================= -->
857 <div class="doc_subsection">
858 <a name="moduleasm">Module-Level Inline Assembly</a>
861 <div class="doc_text">
863 Modules may contain "module-level inline asm" blocks, which corresponds to the
864 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
865 LLVM and treated as a single unit, but may be separated in the .ll file if
866 desired. The syntax is very simple:
869 <div class="doc_code">
871 module asm "inline asm code goes here"
872 module asm "more can go here"
876 <p>The strings can contain any character by escaping non-printable characters.
877 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
882 The inline asm code is simply printed to the machine code .s file when
883 assembly code is generated.
887 <!-- ======================================================================= -->
888 <div class="doc_subsection">
889 <a name="datalayout">Data Layout</a>
892 <div class="doc_text">
893 <p>A module may specify a target specific data layout string that specifies how
894 data is to be laid out in memory. The syntax for the data layout is simply:</p>
895 <pre> target datalayout = "<i>layout specification</i>"</pre>
896 <p>The <i>layout specification</i> consists of a list of specifications
897 separated by the minus sign character ('-'). Each specification starts with a
898 letter and may include other information after the letter to define some
899 aspect of the data layout. The specifications accepted are as follows: </p>
902 <dd>Specifies that the target lays out data in big-endian form. That is, the
903 bits with the most significance have the lowest address location.</dd>
905 <dd>Specifies that hte target lays out data in little-endian form. That is,
906 the bits with the least significance have the lowest address location.</dd>
907 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
908 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
909 <i>preferred</i> alignments. All sizes are in bits. Specifying the <i>pref</i>
910 alignment is optional. If omitted, the preceding <tt>:</tt> should be omitted
912 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
913 <dd>This specifies the alignment for an integer type of a given bit
914 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
915 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
916 <dd>This specifies the alignment for a vector type of a given bit
918 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
919 <dd>This specifies the alignment for a floating point type of a given bit
920 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
922 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
923 <dd>This specifies the alignment for an aggregate type of a given bit
926 <p>When constructing the data layout for a given target, LLVM starts with a
927 default set of specifications which are then (possibly) overriden by the
928 specifications in the <tt>datalayout</tt> keyword. The default specifications
929 are given in this list:</p>
931 <li><tt>E</tt> - big endian</li>
932 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
933 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
934 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
935 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
936 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
937 <li><tt>i64:32:64</tt> - i64 has abi alignment of 32-bits but preferred
938 alignment of 64-bits</li>
939 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
940 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
941 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
942 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
943 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
945 <p>When llvm is determining the alignment for a given type, it uses the
948 <li>If the type sought is an exact match for one of the specifications, that
949 specification is used.</li>
950 <li>If no match is found, and the type sought is an integer type, then the
951 smallest integer type that is larger than the bitwidth of the sought type is
952 used. If none of the specifications are larger than the bitwidth then the the
953 largest integer type is used. For example, given the default specifications
954 above, the i7 type will use the alignment of i8 (next largest) while both
955 i65 and i256 will use the alignment of i64 (largest specified).</li>
956 <li>If no match is found, and the type sought is a vector type, then the
957 largest vector type that is smaller than the sought vector type will be used
958 as a fall back. This happens because <128 x double> can be implemented in
959 terms of 64 <2 x double>, for example.</li>
963 <!-- *********************************************************************** -->
964 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
965 <!-- *********************************************************************** -->
967 <div class="doc_text">
969 <p>The LLVM type system is one of the most important features of the
970 intermediate representation. Being typed enables a number of
971 optimizations to be performed on the IR directly, without having to do
972 extra analyses on the side before the transformation. A strong type
973 system makes it easier to read the generated code and enables novel
974 analyses and transformations that are not feasible to perform on normal
975 three address code representations.</p>
979 <!-- ======================================================================= -->
980 <div class="doc_subsection"> <a name="t_classifications">Type
981 Classifications</a> </div>
982 <div class="doc_text">
983 <p>The types fall into a few useful
986 <table border="1" cellspacing="0" cellpadding="4">
988 <tr><th>Classification</th><th>Types</th></tr>
990 <td><a href="#t_integer">integer</a></td>
991 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
994 <td><a href="#t_floating">floating point</a></td>
995 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
998 <td><a name="t_firstclass">first class</a></td>
999 <td><a href="#t_integer">integer</a>,
1000 <a href="#t_floating">floating point</a>,
1001 <a href="#t_pointer">pointer</a>,
1002 <a href="#t_vector">vector</a>
1006 <td><a href="#t_primitive">primitive</a></td>
1007 <td><a href="#t_label">label</a>,
1008 <a href="#t_void">void</a>,
1009 <a href="#t_integer">integer</a>,
1010 <a href="#t_floating">floating point</a>.</td>
1013 <td><a href="#t_derived">derived</a></td>
1014 <td><a href="#t_integer">integer</a>,
1015 <a href="#t_array">array</a>,
1016 <a href="#t_function">function</a>,
1017 <a href="#t_pointer">pointer</a>,
1018 <a href="#t_struct">structure</a>,
1019 <a href="#t_pstruct">packed structure</a>,
1020 <a href="#t_vector">vector</a>,
1021 <a href="#t_opaque">opaque</a>.
1026 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
1027 most important. Values of these types are the only ones which can be
1028 produced by instructions, passed as arguments, or used as operands to
1029 instructions. This means that all structures and arrays must be
1030 manipulated either by pointer or by component.</p>
1033 <!-- ======================================================================= -->
1034 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
1036 <div class="doc_text">
1037 <p>The primitive types are the fundamental building blocks of the LLVM
1042 <!-- _______________________________________________________________________ -->
1043 <div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1045 <div class="doc_text">
1048 <tr><th>Type</th><th>Description</th></tr>
1049 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1050 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1051 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1052 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1053 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1058 <!-- _______________________________________________________________________ -->
1059 <div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1061 <div class="doc_text">
1063 <p>The void type does not represent any value and has no size.</p>
1072 <!-- _______________________________________________________________________ -->
1073 <div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1075 <div class="doc_text">
1077 <p>The label type represents code labels.</p>
1087 <!-- ======================================================================= -->
1088 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1090 <div class="doc_text">
1092 <p>The real power in LLVM comes from the derived types in the system.
1093 This is what allows a programmer to represent arrays, functions,
1094 pointers, and other useful types. Note that these derived types may be
1095 recursive: For example, it is possible to have a two dimensional array.</p>
1099 <!-- _______________________________________________________________________ -->
1100 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1102 <div class="doc_text">
1105 <p>The integer type is a very simple derived type that simply specifies an
1106 arbitrary bit width for the integer type desired. Any bit width from 1 bit to
1107 2^23-1 (about 8 million) can be specified.</p>
1115 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1119 <table class="layout">
1122 <td><tt>i1</tt></td>
1123 <td>a single-bit integer.</td>
1125 <td><tt>i32</tt></td>
1126 <td>a 32-bit integer.</td>
1128 <td><tt>i1942652</tt></td>
1129 <td>a really big integer of over 1 million bits.</td>
1135 <!-- _______________________________________________________________________ -->
1136 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1138 <div class="doc_text">
1142 <p>The array type is a very simple derived type that arranges elements
1143 sequentially in memory. The array type requires a size (number of
1144 elements) and an underlying data type.</p>
1149 [<# elements> x <elementtype>]
1152 <p>The number of elements is a constant integer value; elementtype may
1153 be any type with a size.</p>
1156 <table class="layout">
1158 <td class="left"><tt>[40 x i32]</tt></td>
1159 <td class="left">Array of 40 32-bit integer values.</td>
1162 <td class="left"><tt>[41 x i32]</tt></td>
1163 <td class="left">Array of 41 32-bit integer values.</td>
1166 <td class="left"><tt>[4 x i8]</tt></td>
1167 <td class="left">Array of 4 8-bit integer values.</td>
1170 <p>Here are some examples of multidimensional arrays:</p>
1171 <table class="layout">
1173 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1174 <td class="left">3x4 array of 32-bit integer values.</td>
1177 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1178 <td class="left">12x10 array of single precision floating point values.</td>
1181 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1182 <td class="left">2x3x4 array of 16-bit integer values.</td>
1186 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1187 length array. Normally, accesses past the end of an array are undefined in
1188 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
1189 As a special case, however, zero length arrays are recognized to be variable
1190 length. This allows implementation of 'pascal style arrays' with the LLVM
1191 type "{ i32, [0 x float]}", for example.</p>
1195 <!-- _______________________________________________________________________ -->
1196 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1197 <div class="doc_text">
1199 <p>The function type can be thought of as a function signature. It
1200 consists of a return type and a list of formal parameter types.
1201 Function types are usually used to build virtual function tables
1202 (which are structures of pointers to functions), for indirect function
1203 calls, and when defining a function.</p>
1205 The return type of a function type cannot be an aggregate type.
1208 <pre> <returntype> (<parameter list>)<br></pre>
1209 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1210 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1211 which indicates that the function takes a variable number of arguments.
1212 Variable argument functions can access their arguments with the <a
1213 href="#int_varargs">variable argument handling intrinsic</a> functions.</p>
1215 <table class="layout">
1217 <td class="left"><tt>i32 (i32)</tt></td>
1218 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1220 </tr><tr class="layout">
1221 <td class="left"><tt>float (i16 signext, i32 *) *
1223 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1224 an <tt>i16</tt> that should be sign extended and a
1225 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1228 </tr><tr class="layout">
1229 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1230 <td class="left">A vararg function that takes at least one
1231 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1232 which returns an integer. This is the signature for <tt>printf</tt> in
1239 <!-- _______________________________________________________________________ -->
1240 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1241 <div class="doc_text">
1243 <p>The structure type is used to represent a collection of data members
1244 together in memory. The packing of the field types is defined to match
1245 the ABI of the underlying processor. The elements of a structure may
1246 be any type that has a size.</p>
1247 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1248 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1249 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1252 <pre> { <type list> }<br></pre>
1254 <table class="layout">
1256 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1257 <td class="left">A triple of three <tt>i32</tt> values</td>
1258 </tr><tr class="layout">
1259 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1260 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1261 second element is a <a href="#t_pointer">pointer</a> to a
1262 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1263 an <tt>i32</tt>.</td>
1268 <!-- _______________________________________________________________________ -->
1269 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1271 <div class="doc_text">
1273 <p>The packed structure type is used to represent a collection of data members
1274 together in memory. There is no padding between fields. Further, the alignment
1275 of a packed structure is 1 byte. The elements of a packed structure may
1276 be any type that has a size.</p>
1277 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1278 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1279 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1282 <pre> < { <type list> } > <br></pre>
1284 <table class="layout">
1286 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1287 <td class="left">A triple of three <tt>i32</tt> values</td>
1288 </tr><tr class="layout">
1289 <td class="left"><tt>< { float, i32 (i32)* } ></tt></td>
1290 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1291 second element is a <a href="#t_pointer">pointer</a> to a
1292 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1293 an <tt>i32</tt>.</td>
1298 <!-- _______________________________________________________________________ -->
1299 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1300 <div class="doc_text">
1302 <p>As in many languages, the pointer type represents a pointer or
1303 reference to another object, which must live in memory. Pointer types may have
1304 an optional address space attribute defining the target-specific numbered
1305 address space where the pointed-to object resides. The default address space is
1308 <pre> <type> *<br></pre>
1310 <table class="layout">
1312 <td class="left"><tt>[4x i32]*</tt></td>
1313 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1314 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1317 <td class="left"><tt>i32 (i32 *) *</tt></td>
1318 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
1319 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1323 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1324 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1325 that resides in address space #5.</td>
1330 <!-- _______________________________________________________________________ -->
1331 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1332 <div class="doc_text">
1336 <p>A vector type is a simple derived type that represents a vector
1337 of elements. Vector types are used when multiple primitive data
1338 are operated in parallel using a single instruction (SIMD).
1339 A vector type requires a size (number of
1340 elements) and an underlying primitive data type. Vectors must have a power
1341 of two length (1, 2, 4, 8, 16 ...). Vector types are
1342 considered <a href="#t_firstclass">first class</a>.</p>
1347 < <# elements> x <elementtype> >
1350 <p>The number of elements is a constant integer value; elementtype may
1351 be any integer or floating point type.</p>
1355 <table class="layout">
1357 <td class="left"><tt><4 x i32></tt></td>
1358 <td class="left">Vector of 4 32-bit integer values.</td>
1361 <td class="left"><tt><8 x float></tt></td>
1362 <td class="left">Vector of 8 32-bit floating-point values.</td>
1365 <td class="left"><tt><2 x i64></tt></td>
1366 <td class="left">Vector of 2 64-bit integer values.</td>
1371 <!-- _______________________________________________________________________ -->
1372 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1373 <div class="doc_text">
1377 <p>Opaque types are used to represent unknown types in the system. This
1378 corresponds (for example) to the C notion of a forward declared structure type.
1379 In LLVM, opaque types can eventually be resolved to any type (not just a
1380 structure type).</p>
1390 <table class="layout">
1392 <td class="left"><tt>opaque</tt></td>
1393 <td class="left">An opaque type.</td>
1399 <!-- *********************************************************************** -->
1400 <div class="doc_section"> <a name="constants">Constants</a> </div>
1401 <!-- *********************************************************************** -->
1403 <div class="doc_text">
1405 <p>LLVM has several different basic types of constants. This section describes
1406 them all and their syntax.</p>
1410 <!-- ======================================================================= -->
1411 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1413 <div class="doc_text">
1416 <dt><b>Boolean constants</b></dt>
1418 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1419 constants of the <tt><a href="#t_primitive">i1</a></tt> type.
1422 <dt><b>Integer constants</b></dt>
1424 <dd>Standard integers (such as '4') are constants of the <a
1425 href="#t_integer">integer</a> type. Negative numbers may be used with
1429 <dt><b>Floating point constants</b></dt>
1431 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1432 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1433 notation (see below). Floating point constants must have a <a
1434 href="#t_floating">floating point</a> type. </dd>
1436 <dt><b>Null pointer constants</b></dt>
1438 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1439 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1443 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1444 of floating point constants. For example, the form '<tt>double
1445 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1446 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1447 (and the only time that they are generated by the disassembler) is when a
1448 floating point constant must be emitted but it cannot be represented as a
1449 decimal floating point number. For example, NaN's, infinities, and other
1450 special values are represented in their IEEE hexadecimal format so that
1451 assembly and disassembly do not cause any bits to change in the constants.</p>
1455 <!-- ======================================================================= -->
1456 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1459 <div class="doc_text">
1460 <p>Aggregate constants arise from aggregation of simple constants
1461 and smaller aggregate constants.</p>
1464 <dt><b>Structure constants</b></dt>
1466 <dd>Structure constants are represented with notation similar to structure
1467 type definitions (a comma separated list of elements, surrounded by braces
1468 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
1469 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>". Structure constants
1470 must have <a href="#t_struct">structure type</a>, and the number and
1471 types of elements must match those specified by the type.
1474 <dt><b>Array constants</b></dt>
1476 <dd>Array constants are represented with notation similar to array type
1477 definitions (a comma separated list of elements, surrounded by square brackets
1478 (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74 ]</tt>". Array
1479 constants must have <a href="#t_array">array type</a>, and the number and
1480 types of elements must match those specified by the type.
1483 <dt><b>Vector constants</b></dt>
1485 <dd>Vector constants are represented with notation similar to vector type
1486 definitions (a comma separated list of elements, surrounded by
1487 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32 42,
1488 i32 11, i32 74, i32 100 ></tt>". Vector constants must have <a
1489 href="#t_vector">vector type</a>, and the number and types of elements must
1490 match those specified by the type.
1493 <dt><b>Zero initialization</b></dt>
1495 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1496 value to zero of <em>any</em> type, including scalar and aggregate types.
1497 This is often used to avoid having to print large zero initializers (e.g. for
1498 large arrays) and is always exactly equivalent to using explicit zero
1505 <!-- ======================================================================= -->
1506 <div class="doc_subsection">
1507 <a name="globalconstants">Global Variable and Function Addresses</a>
1510 <div class="doc_text">
1512 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1513 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1514 constants. These constants are explicitly referenced when the <a
1515 href="#identifiers">identifier for the global</a> is used and always have <a
1516 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1519 <div class="doc_code">
1523 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1529 <!-- ======================================================================= -->
1530 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1531 <div class="doc_text">
1532 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1533 no specific value. Undefined values may be of any type and be used anywhere
1534 a constant is permitted.</p>
1536 <p>Undefined values indicate to the compiler that the program is well defined
1537 no matter what value is used, giving the compiler more freedom to optimize.
1541 <!-- ======================================================================= -->
1542 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1545 <div class="doc_text">
1547 <p>Constant expressions are used to allow expressions involving other constants
1548 to be used as constants. Constant expressions may be of any <a
1549 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1550 that does not have side effects (e.g. load and call are not supported). The
1551 following is the syntax for constant expressions:</p>
1554 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1555 <dd>Truncate a constant to another type. The bit size of CST must be larger
1556 than the bit size of TYPE. Both types must be integers.</dd>
1558 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1559 <dd>Zero extend a constant to another type. The bit size of CST must be
1560 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1562 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1563 <dd>Sign extend a constant to another type. The bit size of CST must be
1564 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1566 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1567 <dd>Truncate a floating point constant to another floating point type. The
1568 size of CST must be larger than the size of TYPE. Both types must be
1569 floating point.</dd>
1571 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1572 <dd>Floating point extend a constant to another type. The size of CST must be
1573 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1575 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
1576 <dd>Convert a floating point constant to the corresponding unsigned integer
1577 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1578 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1579 of the same number of elements. If the value won't fit in the integer type,
1580 the results are undefined.</dd>
1582 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
1583 <dd>Convert a floating point constant to the corresponding signed integer
1584 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1585 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1586 of the same number of elements. If the value won't fit in the integer type,
1587 the results are undefined.</dd>
1589 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
1590 <dd>Convert an unsigned integer constant to the corresponding floating point
1591 constant. TYPE must be a scalar or vector floating point type. CST must be of
1592 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1593 of the same number of elements. If the value won't fit in the floating point
1594 type, the results are undefined.</dd>
1596 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
1597 <dd>Convert a signed integer constant to the corresponding floating point
1598 constant. TYPE must be a scalar or vector floating point type. CST must be of
1599 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1600 of the same number of elements. If the value won't fit in the floating point
1601 type, the results are undefined.</dd>
1603 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
1604 <dd>Convert a pointer typed constant to the corresponding integer constant
1605 TYPE must be an integer type. CST must be of pointer type. The CST value is
1606 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1608 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
1609 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1610 pointer type. CST must be of integer type. The CST value is zero extended,
1611 truncated, or unchanged to make it fit in a pointer size. This one is
1612 <i>really</i> dangerous!</dd>
1614 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
1615 <dd>Convert a constant, CST, to another TYPE. The size of CST and TYPE must be
1616 identical (same number of bits). The conversion is done as if the CST value
1617 was stored to memory and read back as TYPE. In other words, no bits change
1618 with this operator, just the type. This can be used for conversion of
1619 vector types to any other type, as long as they have the same bit width. For
1620 pointers it is only valid to cast to another pointer type.
1623 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1625 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1626 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1627 instruction, the index list may have zero or more indexes, which are required
1628 to make sense for the type of "CSTPTR".</dd>
1630 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1632 <dd>Perform the <a href="#i_select">select operation</a> on
1635 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
1636 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
1638 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
1639 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
1641 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1643 <dd>Perform the <a href="#i_extractelement">extractelement
1644 operation</a> on constants.
1646 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1648 <dd>Perform the <a href="#i_insertelement">insertelement
1649 operation</a> on constants.</dd>
1652 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
1654 <dd>Perform the <a href="#i_shufflevector">shufflevector
1655 operation</a> on constants.</dd>
1657 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1659 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1660 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1661 binary</a> operations. The constraints on operands are the same as those for
1662 the corresponding instruction (e.g. no bitwise operations on floating point
1663 values are allowed).</dd>
1667 <!-- *********************************************************************** -->
1668 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1669 <!-- *********************************************************************** -->
1671 <!-- ======================================================================= -->
1672 <div class="doc_subsection">
1673 <a name="inlineasm">Inline Assembler Expressions</a>
1676 <div class="doc_text">
1679 LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
1680 Module-Level Inline Assembly</a>) through the use of a special value. This
1681 value represents the inline assembler as a string (containing the instructions
1682 to emit), a list of operand constraints (stored as a string), and a flag that
1683 indicates whether or not the inline asm expression has side effects. An example
1684 inline assembler expression is:
1687 <div class="doc_code">
1689 i32 (i32) asm "bswap $0", "=r,r"
1694 Inline assembler expressions may <b>only</b> be used as the callee operand of
1695 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
1698 <div class="doc_code">
1700 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
1705 Inline asms with side effects not visible in the constraint list must be marked
1706 as having side effects. This is done through the use of the
1707 '<tt>sideeffect</tt>' keyword, like so:
1710 <div class="doc_code">
1712 call void asm sideeffect "eieio", ""()
1716 <p>TODO: The format of the asm and constraints string still need to be
1717 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1718 need to be documented).
1723 <!-- *********************************************************************** -->
1724 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1725 <!-- *********************************************************************** -->
1727 <div class="doc_text">
1729 <p>The LLVM instruction set consists of several different
1730 classifications of instructions: <a href="#terminators">terminator
1731 instructions</a>, <a href="#binaryops">binary instructions</a>,
1732 <a href="#bitwiseops">bitwise binary instructions</a>, <a
1733 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1734 instructions</a>.</p>
1738 <!-- ======================================================================= -->
1739 <div class="doc_subsection"> <a name="terminators">Terminator
1740 Instructions</a> </div>
1742 <div class="doc_text">
1744 <p>As mentioned <a href="#functionstructure">previously</a>, every
1745 basic block in a program ends with a "Terminator" instruction, which
1746 indicates which block should be executed after the current block is
1747 finished. These terminator instructions typically yield a '<tt>void</tt>'
1748 value: they produce control flow, not values (the one exception being
1749 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1750 <p>There are six different terminator instructions: the '<a
1751 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1752 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1753 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1754 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1755 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1759 <!-- _______________________________________________________________________ -->
1760 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1761 Instruction</a> </div>
1762 <div class="doc_text">
1764 <pre> ret <type> <value> <i>; Return a value from a non-void function</i>
1765 ret void <i>; Return from void function</i>
1768 <p>The '<tt>ret</tt>' instruction is used to return control flow (and a
1769 value) from a function back to the caller.</p>
1770 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
1771 returns a value and then causes control flow, and one that just causes
1772 control flow to occur.</p>
1774 <p>The '<tt>ret</tt>' instruction may return any '<a
1775 href="#t_firstclass">first class</a>' type. Notice that a function is
1776 not <a href="#wellformed">well formed</a> if there exists a '<tt>ret</tt>'
1777 instruction inside of the function that returns a value that does not
1778 match the return type of the function.</p>
1780 <p>When the '<tt>ret</tt>' instruction is executed, control flow
1781 returns back to the calling function's context. If the caller is a "<a
1782 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1783 the instruction after the call. If the caller was an "<a
1784 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1785 at the beginning of the "normal" destination block. If the instruction
1786 returns a value, that value shall set the call or invoke instruction's
1789 <pre> ret i32 5 <i>; Return an integer value of 5</i>
1790 ret void <i>; Return from a void function</i>
1793 <!-- _______________________________________________________________________ -->
1794 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1795 <div class="doc_text">
1797 <pre> br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1800 <p>The '<tt>br</tt>' instruction is used to cause control flow to
1801 transfer to a different basic block in the current function. There are
1802 two forms of this instruction, corresponding to a conditional branch
1803 and an unconditional branch.</p>
1805 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1806 single '<tt>i1</tt>' value and two '<tt>label</tt>' values. The
1807 unconditional form of the '<tt>br</tt>' instruction takes a single
1808 '<tt>label</tt>' value as a target.</p>
1810 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
1811 argument is evaluated. If the value is <tt>true</tt>, control flows
1812 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1813 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1815 <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
1816 href="#i_ret">ret</a> i32 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> i32 0<br></pre>
1818 <!-- _______________________________________________________________________ -->
1819 <div class="doc_subsubsection">
1820 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
1823 <div class="doc_text">
1827 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
1832 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
1833 several different places. It is a generalization of the '<tt>br</tt>'
1834 instruction, allowing a branch to occur to one of many possible
1840 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
1841 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
1842 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
1843 table is not allowed to contain duplicate constant entries.</p>
1847 <p>The <tt>switch</tt> instruction specifies a table of values and
1848 destinations. When the '<tt>switch</tt>' instruction is executed, this
1849 table is searched for the given value. If the value is found, control flow is
1850 transfered to the corresponding destination; otherwise, control flow is
1851 transfered to the default destination.</p>
1853 <h5>Implementation:</h5>
1855 <p>Depending on properties of the target machine and the particular
1856 <tt>switch</tt> instruction, this instruction may be code generated in different
1857 ways. For example, it could be generated as a series of chained conditional
1858 branches or with a lookup table.</p>
1863 <i>; Emulate a conditional br instruction</i>
1864 %Val = <a href="#i_zext">zext</a> i1 %value to i32
1865 switch i32 %Val, label %truedest [i32 0, label %falsedest ]
1867 <i>; Emulate an unconditional br instruction</i>
1868 switch i32 0, label %dest [ ]
1870 <i>; Implement a jump table:</i>
1871 switch i32 %val, label %otherwise [ i32 0, label %onzero
1873 i32 2, label %ontwo ]
1877 <!-- _______________________________________________________________________ -->
1878 <div class="doc_subsubsection">
1879 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
1882 <div class="doc_text">
1887 <result> = invoke [<a href="#callingconv">cconv</a>] <ptr to function ty> %<function ptr val>(<function args>)
1888 to label <normal label> unwind label <exception label>
1893 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
1894 function, with the possibility of control flow transfer to either the
1895 '<tt>normal</tt>' label or the
1896 '<tt>exception</tt>' label. If the callee function returns with the
1897 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
1898 "normal" label. If the callee (or any indirect callees) returns with the "<a
1899 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
1900 continued at the dynamically nearest "exception" label.</p>
1904 <p>This instruction requires several arguments:</p>
1908 The optional "cconv" marker indicates which <a href="#callingconv">calling
1909 convention</a> the call should use. If none is specified, the call defaults
1910 to using C calling conventions.
1912 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
1913 function value being invoked. In most cases, this is a direct function
1914 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
1915 an arbitrary pointer to function value.
1918 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
1919 function to be invoked. </li>
1921 <li>'<tt>function args</tt>': argument list whose types match the function
1922 signature argument types. If the function signature indicates the function
1923 accepts a variable number of arguments, the extra arguments can be
1926 <li>'<tt>normal label</tt>': the label reached when the called function
1927 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
1929 <li>'<tt>exception label</tt>': the label reached when a callee returns with
1930 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
1936 <p>This instruction is designed to operate as a standard '<tt><a
1937 href="#i_call">call</a></tt>' instruction in most regards. The primary
1938 difference is that it establishes an association with a label, which is used by
1939 the runtime library to unwind the stack.</p>
1941 <p>This instruction is used in languages with destructors to ensure that proper
1942 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
1943 exception. Additionally, this is important for implementation of
1944 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
1948 %retval = invoke i32 %Test(i32 15) to label %Continue
1949 unwind label %TestCleanup <i>; {i32}:retval set</i>
1950 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Test(i32 15) to label %Continue
1951 unwind label %TestCleanup <i>; {i32}:retval set</i>
1956 <!-- _______________________________________________________________________ -->
1958 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
1959 Instruction</a> </div>
1961 <div class="doc_text">
1970 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
1971 at the first callee in the dynamic call stack which used an <a
1972 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
1973 primarily used to implement exception handling.</p>
1977 <p>The '<tt>unwind</tt>' intrinsic causes execution of the current function to
1978 immediately halt. The dynamic call stack is then searched for the first <a
1979 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
1980 execution continues at the "exceptional" destination block specified by the
1981 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
1982 dynamic call chain, undefined behavior results.</p>
1985 <!-- _______________________________________________________________________ -->
1987 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
1988 Instruction</a> </div>
1990 <div class="doc_text">
1999 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
2000 instruction is used to inform the optimizer that a particular portion of the
2001 code is not reachable. This can be used to indicate that the code after a
2002 no-return function cannot be reached, and other facts.</p>
2006 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
2011 <!-- ======================================================================= -->
2012 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
2013 <div class="doc_text">
2014 <p>Binary operators are used to do most of the computation in a
2015 program. They require two operands, execute an operation on them, and
2016 produce a single value. The operands might represent
2017 multiple data, as is the case with the <a href="#t_vector">vector</a> data type.
2018 The result value of a binary operator is not
2019 necessarily the same type as its operands.</p>
2020 <p>There are several different binary operators:</p>
2022 <!-- _______________________________________________________________________ -->
2023 <div class="doc_subsubsection"> <a name="i_add">'<tt>add</tt>'
2024 Instruction</a> </div>
2025 <div class="doc_text">
2027 <pre> <result> = add <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2030 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
2032 <p>The two arguments to the '<tt>add</tt>' instruction must be either <a
2033 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a> values.
2034 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2035 Both arguments must have identical types.</p>
2037 <p>The value produced is the integer or floating point sum of the two
2040 <pre> <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
2043 <!-- _______________________________________________________________________ -->
2044 <div class="doc_subsubsection"> <a name="i_sub">'<tt>sub</tt>'
2045 Instruction</a> </div>
2046 <div class="doc_text">
2048 <pre> <result> = sub <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2051 <p>The '<tt>sub</tt>' instruction returns the difference of its two
2053 <p>Note that the '<tt>sub</tt>' instruction is used to represent the '<tt>neg</tt>'
2054 instruction present in most other intermediate representations.</p>
2056 <p>The two arguments to the '<tt>sub</tt>' instruction must be either <a
2057 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
2059 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2060 Both arguments must have identical types.</p>
2062 <p>The value produced is the integer or floating point difference of
2063 the two operands.</p>
2066 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
2067 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
2070 <!-- _______________________________________________________________________ -->
2071 <div class="doc_subsubsection"> <a name="i_mul">'<tt>mul</tt>'
2072 Instruction</a> </div>
2073 <div class="doc_text">
2075 <pre> <result> = mul <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2078 <p>The '<tt>mul</tt>' instruction returns the product of its two
2081 <p>The two arguments to the '<tt>mul</tt>' instruction must be either <a
2082 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
2084 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2085 Both arguments must have identical types.</p>
2087 <p>The value produced is the integer or floating point product of the
2089 <p>Because the operands are the same width, the result of an integer
2090 multiplication is the same whether the operands should be deemed unsigned or
2093 <pre> <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
2096 <!-- _______________________________________________________________________ -->
2097 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
2099 <div class="doc_text">
2101 <pre> <result> = udiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2104 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two
2107 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
2108 <a href="#t_integer">integer</a> values. Both arguments must have identical
2109 types. This instruction can also take <a href="#t_vector">vector</a> versions
2110 of the values in which case the elements must be integers.</p>
2112 <p>The value produced is the unsigned integer quotient of the two operands. This
2113 instruction always performs an unsigned division operation, regardless of
2114 whether the arguments are unsigned or not.</p>
2116 <pre> <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2119 <!-- _______________________________________________________________________ -->
2120 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
2122 <div class="doc_text">
2124 <pre> <result> = sdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2127 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
2130 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
2131 <a href="#t_integer">integer</a> values. Both arguments must have identical
2132 types. This instruction can also take <a href="#t_vector">vector</a> versions
2133 of the values in which case the elements must be integers.</p>
2135 <p>The value produced is the signed integer quotient of the two operands. This
2136 instruction always performs a signed division operation, regardless of whether
2137 the arguments are signed or not.</p>
2139 <pre> <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2142 <!-- _______________________________________________________________________ -->
2143 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
2144 Instruction</a> </div>
2145 <div class="doc_text">
2147 <pre> <result> = fdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2150 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
2153 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
2154 <a href="#t_floating">floating point</a> values. Both arguments must have
2155 identical types. This instruction can also take <a href="#t_vector">vector</a>
2156 versions of floating point values.</p>
2158 <p>The value produced is the floating point quotient of the two operands.</p>
2160 <pre> <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
2163 <!-- _______________________________________________________________________ -->
2164 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
2166 <div class="doc_text">
2168 <pre> <result> = urem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2171 <p>The '<tt>urem</tt>' instruction returns the remainder from the
2172 unsigned division of its two arguments.</p>
2174 <p>The two arguments to the '<tt>urem</tt>' instruction must be
2175 <a href="#t_integer">integer</a> values. Both arguments must have identical
2176 types. This instruction can also take <a href="#t_vector">vector</a> versions
2177 of the values in which case the elements must be integers.</p>
2179 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
2180 This instruction always performs an unsigned division to get the remainder,
2181 regardless of whether the arguments are unsigned or not.</p>
2183 <pre> <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2187 <!-- _______________________________________________________________________ -->
2188 <div class="doc_subsubsection"> <a name="i_srem">'<tt>srem</tt>'
2189 Instruction</a> </div>
2190 <div class="doc_text">
2192 <pre> <result> = srem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2195 <p>The '<tt>srem</tt>' instruction returns the remainder from the
2196 signed division of its two operands. This instruction can also take
2197 <a href="#t_vector">vector</a> versions of the values in which case
2198 the elements must be integers.</p>
2201 <p>The two arguments to the '<tt>srem</tt>' instruction must be
2202 <a href="#t_integer">integer</a> values. Both arguments must have identical
2205 <p>This instruction returns the <i>remainder</i> of a division (where the result
2206 has the same sign as the dividend, <tt>var1</tt>), not the <i>modulo</i>
2207 operator (where the result has the same sign as the divisor, <tt>var2</tt>) of
2208 a value. For more information about the difference, see <a
2209 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
2210 Math Forum</a>. For a table of how this is implemented in various languages,
2211 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
2212 Wikipedia: modulo operation</a>.</p>
2214 <pre> <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2218 <!-- _______________________________________________________________________ -->
2219 <div class="doc_subsubsection"> <a name="i_frem">'<tt>frem</tt>'
2220 Instruction</a> </div>
2221 <div class="doc_text">
2223 <pre> <result> = frem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2226 <p>The '<tt>frem</tt>' instruction returns the remainder from the
2227 division of its two operands.</p>
2229 <p>The two arguments to the '<tt>frem</tt>' instruction must be
2230 <a href="#t_floating">floating point</a> values. Both arguments must have
2231 identical types. This instruction can also take <a href="#t_vector">vector</a>
2232 versions of floating point values.</p>
2234 <p>This instruction returns the <i>remainder</i> of a division.</p>
2236 <pre> <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
2240 <!-- ======================================================================= -->
2241 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
2242 Operations</a> </div>
2243 <div class="doc_text">
2244 <p>Bitwise binary operators are used to do various forms of
2245 bit-twiddling in a program. They are generally very efficient
2246 instructions and can commonly be strength reduced from other
2247 instructions. They require two operands, execute an operation on them,
2248 and produce a single value. The resulting value of the bitwise binary
2249 operators is always the same type as its first operand.</p>
2252 <!-- _______________________________________________________________________ -->
2253 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
2254 Instruction</a> </div>
2255 <div class="doc_text">
2257 <pre> <result> = shl <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2262 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2263 the left a specified number of bits.</p>
2267 <p>Both arguments to the '<tt>shl</tt>' instruction must be the same <a
2268 href="#t_integer">integer</a> type.</p>
2272 <p>The value produced is <tt>var1</tt> * 2<sup><tt>var2</tt></sup>. If
2273 <tt>var2</tt> is (statically or dynamically) equal to or larger than the number
2274 of bits in <tt>var1</tt>, the result is undefined.</p>
2276 <h5>Example:</h5><pre>
2277 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
2278 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
2279 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
2280 <result> = shl i32 1, 32 <i>; undefined</i>
2283 <!-- _______________________________________________________________________ -->
2284 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
2285 Instruction</a> </div>
2286 <div class="doc_text">
2288 <pre> <result> = lshr <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2292 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
2293 operand shifted to the right a specified number of bits with zero fill.</p>
2296 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
2297 <a href="#t_integer">integer</a> type.</p>
2301 <p>This instruction always performs a logical shift right operation. The most
2302 significant bits of the result will be filled with zero bits after the
2303 shift. If <tt>var2</tt> is (statically or dynamically) equal to or larger than
2304 the number of bits in <tt>var1</tt>, the result is undefined.</p>
2308 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
2309 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
2310 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
2311 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
2312 <result> = lshr i32 1, 32 <i>; undefined</i>
2316 <!-- _______________________________________________________________________ -->
2317 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
2318 Instruction</a> </div>
2319 <div class="doc_text">
2322 <pre> <result> = ashr <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2326 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
2327 operand shifted to the right a specified number of bits with sign extension.</p>
2330 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
2331 <a href="#t_integer">integer</a> type.</p>
2334 <p>This instruction always performs an arithmetic shift right operation,
2335 The most significant bits of the result will be filled with the sign bit
2336 of <tt>var1</tt>. If <tt>var2</tt> is (statically or dynamically) equal to or
2337 larger than the number of bits in <tt>var1</tt>, the result is undefined.
2342 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
2343 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
2344 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
2345 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
2346 <result> = ashr i32 1, 32 <i>; undefined</i>
2350 <!-- _______________________________________________________________________ -->
2351 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
2352 Instruction</a> </div>
2353 <div class="doc_text">
2355 <pre> <result> = and <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2358 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
2359 its two operands.</p>
2361 <p>The two arguments to the '<tt>and</tt>' instruction must be <a
2362 href="#t_integer">integer</a> values. Both arguments must have
2363 identical types.</p>
2365 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
2367 <div style="align: center">
2368 <table border="1" cellspacing="0" cellpadding="4">
2399 <pre> <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
2400 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
2401 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
2404 <!-- _______________________________________________________________________ -->
2405 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
2406 <div class="doc_text">
2408 <pre> <result> = or <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2411 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
2412 or of its two operands.</p>
2414 <p>The two arguments to the '<tt>or</tt>' instruction must be <a
2415 href="#t_integer">integer</a> values. Both arguments must have
2416 identical types.</p>
2418 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
2420 <div style="align: center">
2421 <table border="1" cellspacing="0" cellpadding="4">
2452 <pre> <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
2453 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
2454 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
2457 <!-- _______________________________________________________________________ -->
2458 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
2459 Instruction</a> </div>
2460 <div class="doc_text">
2462 <pre> <result> = xor <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2465 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
2466 or of its two operands. The <tt>xor</tt> is used to implement the
2467 "one's complement" operation, which is the "~" operator in C.</p>
2469 <p>The two arguments to the '<tt>xor</tt>' instruction must be <a
2470 href="#t_integer">integer</a> values. Both arguments must have
2471 identical types.</p>
2473 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
2475 <div style="align: center">
2476 <table border="1" cellspacing="0" cellpadding="4">
2508 <pre> <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
2509 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
2510 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
2511 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
2515 <!-- ======================================================================= -->
2516 <div class="doc_subsection">
2517 <a name="vectorops">Vector Operations</a>
2520 <div class="doc_text">
2522 <p>LLVM supports several instructions to represent vector operations in a
2523 target-independent manner. These instructions cover the element-access and
2524 vector-specific operations needed to process vectors effectively. While LLVM
2525 does directly support these vector operations, many sophisticated algorithms
2526 will want to use target-specific intrinsics to take full advantage of a specific
2531 <!-- _______________________________________________________________________ -->
2532 <div class="doc_subsubsection">
2533 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2536 <div class="doc_text">
2541 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
2547 The '<tt>extractelement</tt>' instruction extracts a single scalar
2548 element from a vector at a specified index.
2555 The first operand of an '<tt>extractelement</tt>' instruction is a
2556 value of <a href="#t_vector">vector</a> type. The second operand is
2557 an index indicating the position from which to extract the element.
2558 The index may be a variable.</p>
2563 The result is a scalar of the same type as the element type of
2564 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2565 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2566 results are undefined.
2572 %result = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
2577 <!-- _______________________________________________________________________ -->
2578 <div class="doc_subsubsection">
2579 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
2582 <div class="doc_text">
2587 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
2593 The '<tt>insertelement</tt>' instruction inserts a scalar
2594 element into a vector at a specified index.
2601 The first operand of an '<tt>insertelement</tt>' instruction is a
2602 value of <a href="#t_vector">vector</a> type. The second operand is a
2603 scalar value whose type must equal the element type of the first
2604 operand. The third operand is an index indicating the position at
2605 which to insert the value. The index may be a variable.</p>
2610 The result is a vector of the same type as <tt>val</tt>. Its
2611 element values are those of <tt>val</tt> except at position
2612 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
2613 exceeds the length of <tt>val</tt>, the results are undefined.
2619 %result = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
2623 <!-- _______________________________________________________________________ -->
2624 <div class="doc_subsubsection">
2625 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
2628 <div class="doc_text">
2633 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <n x i32> <mask> <i>; yields <n x <ty>></i>
2639 The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
2640 from two input vectors, returning a vector of the same type.
2646 The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
2647 with types that match each other and types that match the result of the
2648 instruction. The third argument is a shuffle mask, which has the same number
2649 of elements as the other vector type, but whose element type is always 'i32'.
2653 The shuffle mask operand is required to be a constant vector with either
2654 constant integer or undef values.
2660 The elements of the two input vectors are numbered from left to right across
2661 both of the vectors. The shuffle mask operand specifies, for each element of
2662 the result vector, which element of the two input registers the result element
2663 gets. The element selector may be undef (meaning "don't care") and the second
2664 operand may be undef if performing a shuffle from only one vector.
2670 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
2671 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
2672 %result = shufflevector <4 x i32> %v1, <4 x i32> undef,
2673 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
2678 <!-- ======================================================================= -->
2679 <div class="doc_subsection">
2680 <a name="memoryops">Memory Access and Addressing Operations</a>
2683 <div class="doc_text">
2685 <p>A key design point of an SSA-based representation is how it
2686 represents memory. In LLVM, no memory locations are in SSA form, which
2687 makes things very simple. This section describes how to read, write,
2688 allocate, and free memory in LLVM.</p>
2692 <!-- _______________________________________________________________________ -->
2693 <div class="doc_subsubsection">
2694 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
2697 <div class="doc_text">
2702 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2707 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
2708 heap and returns a pointer to it. The object is always allocated in the generic
2709 address space (address space zero).</p>
2713 <p>The '<tt>malloc</tt>' instruction allocates
2714 <tt>sizeof(<type>)*NumElements</tt>
2715 bytes of memory from the operating system and returns a pointer of the
2716 appropriate type to the program. If "NumElements" is specified, it is the
2717 number of elements allocated. If an alignment is specified, the value result
2718 of the allocation is guaranteed to be aligned to at least that boundary. If
2719 not specified, or if zero, the target can choose to align the allocation on any
2720 convenient boundary.</p>
2722 <p>'<tt>type</tt>' must be a sized type.</p>
2726 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
2727 a pointer is returned.</p>
2732 %array = malloc [4 x i8 ] <i>; yields {[%4 x i8]*}:array</i>
2734 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
2735 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
2736 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
2737 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
2738 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
2742 <!-- _______________________________________________________________________ -->
2743 <div class="doc_subsubsection">
2744 <a name="i_free">'<tt>free</tt>' Instruction</a>
2747 <div class="doc_text">
2752 free <type> <value> <i>; yields {void}</i>
2757 <p>The '<tt>free</tt>' instruction returns memory back to the unused
2758 memory heap to be reallocated in the future.</p>
2762 <p>'<tt>value</tt>' shall be a pointer value that points to a value
2763 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
2768 <p>Access to the memory pointed to by the pointer is no longer defined
2769 after this instruction executes.</p>
2774 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
2775 free [4 x i8]* %array
2779 <!-- _______________________________________________________________________ -->
2780 <div class="doc_subsubsection">
2781 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
2784 <div class="doc_text">
2789 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2794 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
2795 currently executing function, to be automatically released when this function
2796 returns to its caller. The object is always allocated in the generic address
2797 space (address space zero).</p>
2801 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
2802 bytes of memory on the runtime stack, returning a pointer of the
2803 appropriate type to the program. If "NumElements" is specified, it is the
2804 number of elements allocated. If an alignment is specified, the value result
2805 of the allocation is guaranteed to be aligned to at least that boundary. If
2806 not specified, or if zero, the target can choose to align the allocation on any
2807 convenient boundary.</p>
2809 <p>'<tt>type</tt>' may be any sized type.</p>
2813 <p>Memory is allocated; a pointer is returned. '<tt>alloca</tt>'d
2814 memory is automatically released when the function returns. The '<tt>alloca</tt>'
2815 instruction is commonly used to represent automatic variables that must
2816 have an address available. When the function returns (either with the <tt><a
2817 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
2818 instructions), the memory is reclaimed.</p>
2823 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
2824 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
2825 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
2826 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
2830 <!-- _______________________________________________________________________ -->
2831 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
2832 Instruction</a> </div>
2833 <div class="doc_text">
2835 <pre> <result> = load <ty>* <pointer>[, align <alignment>]<br> <result> = volatile load <ty>* <pointer>[, align <alignment>]<br></pre>
2837 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
2839 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
2840 address from which to load. The pointer must point to a <a
2841 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
2842 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
2843 the number or order of execution of this <tt>load</tt> with other
2844 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
2847 <p>The location of memory pointed to is loaded.</p>
2849 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
2851 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
2852 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
2855 <!-- _______________________________________________________________________ -->
2856 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
2857 Instruction</a> </div>
2858 <div class="doc_text">
2860 <pre> store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
2861 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
2864 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
2866 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
2867 to store and an address at which to store it. The type of the '<tt><pointer></tt>'
2868 operand must be a pointer to the type of the '<tt><value></tt>'
2869 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
2870 optimizer is not allowed to modify the number or order of execution of
2871 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
2872 href="#i_store">store</a></tt> instructions.</p>
2874 <p>The contents of memory are updated to contain '<tt><value></tt>'
2875 at the location specified by the '<tt><pointer></tt>' operand.</p>
2877 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
2878 store i32 3, i32* %ptr <i>; yields {void}</i>
2879 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
2883 <!-- _______________________________________________________________________ -->
2884 <div class="doc_subsubsection">
2885 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
2888 <div class="doc_text">
2891 <result> = getelementptr <ty>* <ptrval>{, <ty> <idx>}*
2897 The '<tt>getelementptr</tt>' instruction is used to get the address of a
2898 subelement of an aggregate data structure.</p>
2902 <p>This instruction takes a list of integer operands that indicate what
2903 elements of the aggregate object to index to. The actual types of the arguments
2904 provided depend on the type of the first pointer argument. The
2905 '<tt>getelementptr</tt>' instruction is used to index down through the type
2906 levels of a structure or to a specific index in an array. When indexing into a
2907 structure, only <tt>i32</tt> integer constants are allowed. When indexing
2908 into an array or pointer, only integers of 32 or 64 bits are allowed, and will
2909 be sign extended to 64-bit values.</p>
2911 <p>For example, let's consider a C code fragment and how it gets
2912 compiled to LLVM:</p>
2914 <div class="doc_code">
2927 int *foo(struct ST *s) {
2928 return &s[1].Z.B[5][13];
2933 <p>The LLVM code generated by the GCC frontend is:</p>
2935 <div class="doc_code">
2937 %RT = type { i8 , [10 x [20 x i32]], i8 }
2938 %ST = type { i32, double, %RT }
2940 define i32* %foo(%ST* %s) {
2942 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
2950 <p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
2951 on the pointer type that is being indexed into. <a href="#t_pointer">Pointer</a>
2952 and <a href="#t_array">array</a> types can use a 32-bit or 64-bit
2953 <a href="#t_integer">integer</a> type but the value will always be sign extended
2954 to 64-bits. <a href="#t_struct">Structure</a> types require <tt>i32</tt>
2955 <b>constants</b>.</p>
2957 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
2958 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
2959 }</tt>' type, a structure. The second index indexes into the third element of
2960 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
2961 i8 }</tt>' type, another structure. The third index indexes into the second
2962 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
2963 array. The two dimensions of the array are subscripted into, yielding an
2964 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
2965 to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
2967 <p>Note that it is perfectly legal to index partially through a
2968 structure, returning a pointer to an inner element. Because of this,
2969 the LLVM code for the given testcase is equivalent to:</p>
2972 define i32* %foo(%ST* %s) {
2973 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
2974 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
2975 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
2976 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
2977 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
2982 <p>Note that it is undefined to access an array out of bounds: array and
2983 pointer indexes must always be within the defined bounds of the array type.
2984 The one exception for this rules is zero length arrays. These arrays are
2985 defined to be accessible as variable length arrays, which requires access
2986 beyond the zero'th element.</p>
2988 <p>The getelementptr instruction is often confusing. For some more insight
2989 into how it works, see <a href="GetElementPtr.html">the getelementptr
2995 <i>; yields [12 x i8]*:aptr</i>
2996 %aptr = getelementptr {i32, [12 x i8]}* %sptr, i64 0, i32 1
3000 <!-- ======================================================================= -->
3001 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
3003 <div class="doc_text">
3004 <p>The instructions in this category are the conversion instructions (casting)
3005 which all take a single operand and a type. They perform various bit conversions
3009 <!-- _______________________________________________________________________ -->
3010 <div class="doc_subsubsection">
3011 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
3013 <div class="doc_text">
3017 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
3022 The '<tt>trunc</tt>' instruction truncates its operand to the type <tt>ty2</tt>.
3027 The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
3028 be an <a href="#t_integer">integer</a> type, and a type that specifies the size
3029 and type of the result, which must be an <a href="#t_integer">integer</a>
3030 type. The bit size of <tt>value</tt> must be larger than the bit size of
3031 <tt>ty2</tt>. Equal sized types are not allowed.</p>
3035 The '<tt>trunc</tt>' instruction truncates the high order bits in <tt>value</tt>
3036 and converts the remaining bits to <tt>ty2</tt>. Since the source size must be
3037 larger than the destination size, <tt>trunc</tt> cannot be a <i>no-op cast</i>.
3038 It will always truncate bits.</p>
3042 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
3043 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
3044 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
3048 <!-- _______________________________________________________________________ -->
3049 <div class="doc_subsubsection">
3050 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
3052 <div class="doc_text">
3056 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
3060 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
3065 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
3066 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3067 also be of <a href="#t_integer">integer</a> type. The bit size of the
3068 <tt>value</tt> must be smaller than the bit size of the destination type,
3072 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
3073 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
3075 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
3079 %X = zext i32 257 to i64 <i>; yields i64:257</i>
3080 %Y = zext i1 true to i32 <i>; yields i32:1</i>
3084 <!-- _______________________________________________________________________ -->
3085 <div class="doc_subsubsection">
3086 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
3088 <div class="doc_text">
3092 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
3096 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
3100 The '<tt>sext</tt>' instruction takes a value to cast, which must be of
3101 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3102 also be of <a href="#t_integer">integer</a> type. The bit size of the
3103 <tt>value</tt> must be smaller than the bit size of the destination type,
3108 The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
3109 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size of
3110 the type <tt>ty2</tt>.</p>
3112 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
3116 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
3117 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
3121 <!-- _______________________________________________________________________ -->
3122 <div class="doc_subsubsection">
3123 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
3126 <div class="doc_text">
3131 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
3135 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
3140 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
3141 point</a> value to cast and a <a href="#t_floating">floating point</a> type to
3142 cast it to. The size of <tt>value</tt> must be larger than the size of
3143 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
3144 <i>no-op cast</i>.</p>
3147 <p> The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
3148 <a href="#t_floating">floating point</a> type to a smaller
3149 <a href="#t_floating">floating point</a> type. If the value cannot fit within
3150 the destination type, <tt>ty2</tt>, then the results are undefined.</p>
3154 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
3155 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
3159 <!-- _______________________________________________________________________ -->
3160 <div class="doc_subsubsection">
3161 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
3163 <div class="doc_text">
3167 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
3171 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
3172 floating point value.</p>
3175 <p>The '<tt>fpext</tt>' instruction takes a
3176 <a href="#t_floating">floating point</a> <tt>value</tt> to cast,
3177 and a <a href="#t_floating">floating point</a> type to cast it to. The source
3178 type must be smaller than the destination type.</p>
3181 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
3182 <a href="#t_floating">floating point</a> type to a larger
3183 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
3184 used to make a <i>no-op cast</i> because it always changes bits. Use
3185 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
3189 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
3190 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
3194 <!-- _______________________________________________________________________ -->
3195 <div class="doc_subsubsection">
3196 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
3198 <div class="doc_text">
3202 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
3206 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
3207 unsigned integer equivalent of type <tt>ty2</tt>.
3211 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
3212 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3213 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3214 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3215 vector integer type with the same number of elements as <tt>ty</tt></p>
3218 <p> The '<tt>fptoui</tt>' instruction converts its
3219 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3220 towards zero) unsigned integer value. If the value cannot fit in <tt>ty2</tt>,
3221 the results are undefined.</p>
3225 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
3226 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
3227 %X = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
3231 <!-- _______________________________________________________________________ -->
3232 <div class="doc_subsubsection">
3233 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
3235 <div class="doc_text">
3239 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
3243 <p>The '<tt>fptosi</tt>' instruction converts
3244 <a href="#t_floating">floating point</a> <tt>value</tt> to type <tt>ty2</tt>.
3248 <p> The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
3249 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3250 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3251 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3252 vector integer type with the same number of elements as <tt>ty</tt></p>
3255 <p>The '<tt>fptosi</tt>' instruction converts its
3256 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3257 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
3258 the results are undefined.</p>
3262 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
3263 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
3264 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
3268 <!-- _______________________________________________________________________ -->
3269 <div class="doc_subsubsection">
3270 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
3272 <div class="doc_text">
3276 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3280 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
3281 integer and converts that value to the <tt>ty2</tt> type.</p>
3284 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
3285 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3286 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3287 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3288 floating point type with the same number of elements as <tt>ty</tt></p>
3291 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
3292 integer quantity and converts it to the corresponding floating point value. If
3293 the value cannot fit in the floating point value, the results are undefined.</p>
3297 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
3298 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
3302 <!-- _______________________________________________________________________ -->
3303 <div class="doc_subsubsection">
3304 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
3306 <div class="doc_text">
3310 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3314 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed
3315 integer and converts that value to the <tt>ty2</tt> type.</p>
3318 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
3319 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3320 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3321 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3322 floating point type with the same number of elements as <tt>ty</tt></p>
3325 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed
3326 integer quantity and converts it to the corresponding floating point value. If
3327 the value cannot fit in the floating point value, the results are undefined.</p>
3331 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
3332 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
3336 <!-- _______________________________________________________________________ -->
3337 <div class="doc_subsubsection">
3338 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
3340 <div class="doc_text">
3344 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
3348 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
3349 the integer type <tt>ty2</tt>.</p>
3352 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
3353 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
3354 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.
3357 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
3358 <tt>ty2</tt> by interpreting the pointer value as an integer and either
3359 truncating or zero extending that value to the size of the integer type. If
3360 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
3361 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
3362 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
3367 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
3368 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
3372 <!-- _______________________________________________________________________ -->
3373 <div class="doc_subsubsection">
3374 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
3376 <div class="doc_text">
3380 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
3384 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to
3385 a pointer type, <tt>ty2</tt>.</p>
3388 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
3389 value to cast, and a type to cast it to, which must be a
3390 <a href="#t_pointer">pointer</a> type.
3393 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
3394 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
3395 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
3396 size of a pointer then a truncation is done. If <tt>value</tt> is smaller than
3397 the size of a pointer then a zero extension is done. If they are the same size,
3398 nothing is done (<i>no-op cast</i>).</p>
3402 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
3403 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
3404 %Y = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
3408 <!-- _______________________________________________________________________ -->
3409 <div class="doc_subsubsection">
3410 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
3412 <div class="doc_text">
3416 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
3420 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3421 <tt>ty2</tt> without changing any bits.</p>
3424 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be
3425 a first class value, and a type to cast it to, which must also be a <a
3426 href="#t_firstclass">first class</a> type. The bit sizes of <tt>value</tt>
3427 and the destination type, <tt>ty2</tt>, must be identical. If the source
3428 type is a pointer, the destination type must also be a pointer.</p>
3431 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3432 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
3433 this conversion. The conversion is done as if the <tt>value</tt> had been
3434 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only be
3435 converted to other pointer types with this instruction. To convert pointers to
3436 other types, use the <a href="#i_inttoptr">inttoptr</a> or
3437 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
3441 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
3442 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
3443 %Z = bitcast <2xint> %V to i64; <i>; yields i64: %V</i>
3447 <!-- ======================================================================= -->
3448 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
3449 <div class="doc_text">
3450 <p>The instructions in this category are the "miscellaneous"
3451 instructions, which defy better classification.</p>
3454 <!-- _______________________________________________________________________ -->
3455 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
3457 <div class="doc_text">
3459 <pre> <result> = icmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3462 <p>The '<tt>icmp</tt>' instruction returns a boolean value based on comparison
3463 of its two integer operands.</p>
3465 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
3466 the condition code indicating the kind of comparison to perform. It is not
3467 a value, just a keyword. The possible condition code are:
3469 <li><tt>eq</tt>: equal</li>
3470 <li><tt>ne</tt>: not equal </li>
3471 <li><tt>ugt</tt>: unsigned greater than</li>
3472 <li><tt>uge</tt>: unsigned greater or equal</li>
3473 <li><tt>ult</tt>: unsigned less than</li>
3474 <li><tt>ule</tt>: unsigned less or equal</li>
3475 <li><tt>sgt</tt>: signed greater than</li>
3476 <li><tt>sge</tt>: signed greater or equal</li>
3477 <li><tt>slt</tt>: signed less than</li>
3478 <li><tt>sle</tt>: signed less or equal</li>
3480 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
3481 <a href="#t_pointer">pointer</a> typed. They must also be identical types.</p>
3483 <p>The '<tt>icmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3484 the condition code given as <tt>cond</tt>. The comparison performed always
3485 yields a <a href="#t_primitive">i1</a> result, as follows:
3487 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
3488 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3490 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
3491 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3492 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
3493 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3494 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
3495 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3496 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
3497 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3498 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
3499 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3500 <li><tt>sgt</tt>: interprets the operands as signed values and yields
3501 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3502 <li><tt>sge</tt>: interprets the operands as signed values and yields
3503 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3504 <li><tt>slt</tt>: interprets the operands as signed values and yields
3505 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3506 <li><tt>sle</tt>: interprets the operands as signed values and yields
3507 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3509 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
3510 values are compared as if they were integers.</p>
3513 <pre> <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
3514 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
3515 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
3516 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
3517 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
3518 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
3522 <!-- _______________________________________________________________________ -->
3523 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
3525 <div class="doc_text">
3527 <pre> <result> = fcmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3530 <p>The '<tt>fcmp</tt>' instruction returns a boolean value based on comparison
3531 of its floating point operands.</p>
3533 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
3534 the condition code indicating the kind of comparison to perform. It is not
3535 a value, just a keyword. The possible condition code are:
3537 <li><tt>false</tt>: no comparison, always returns false</li>
3538 <li><tt>oeq</tt>: ordered and equal</li>
3539 <li><tt>ogt</tt>: ordered and greater than </li>
3540 <li><tt>oge</tt>: ordered and greater than or equal</li>
3541 <li><tt>olt</tt>: ordered and less than </li>
3542 <li><tt>ole</tt>: ordered and less than or equal</li>
3543 <li><tt>one</tt>: ordered and not equal</li>
3544 <li><tt>ord</tt>: ordered (no nans)</li>
3545 <li><tt>ueq</tt>: unordered or equal</li>
3546 <li><tt>ugt</tt>: unordered or greater than </li>
3547 <li><tt>uge</tt>: unordered or greater than or equal</li>
3548 <li><tt>ult</tt>: unordered or less than </li>
3549 <li><tt>ule</tt>: unordered or less than or equal</li>
3550 <li><tt>une</tt>: unordered or not equal</li>
3551 <li><tt>uno</tt>: unordered (either nans)</li>
3552 <li><tt>true</tt>: no comparison, always returns true</li>
3554 <p><i>Ordered</i> means that neither operand is a QNAN while
3555 <i>unordered</i> means that either operand may be a QNAN.</p>
3556 <p>The <tt>val1</tt> and <tt>val2</tt> arguments must be
3557 <a href="#t_floating">floating point</a> typed. They must have identical
3560 <p>The '<tt>fcmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3561 the condition code given as <tt>cond</tt>. The comparison performed always
3562 yields a <a href="#t_primitive">i1</a> result, as follows:
3564 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
3565 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3566 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3567 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3568 <tt>var1</tt> is greather than <tt>var2</tt>.</li>
3569 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3570 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3571 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3572 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3573 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3574 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3575 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3576 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3577 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
3578 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
3579 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3580 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
3581 <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3582 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
3583 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3584 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
3585 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3586 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
3587 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3588 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
3589 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3590 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
3591 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
3595 <pre> <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
3596 <result> = icmp one float 4.0, 5.0 <i>; yields: result=true</i>
3597 <result> = icmp olt float 4.0, 5.0 <i>; yields: result=true</i>
3598 <result> = icmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
3602 <!-- _______________________________________________________________________ -->
3603 <div class="doc_subsubsection"> <a name="i_phi">'<tt>phi</tt>'
3604 Instruction</a> </div>
3605 <div class="doc_text">
3607 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
3609 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
3610 the SSA graph representing the function.</p>
3612 <p>The type of the incoming values is specified with the first type
3613 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
3614 as arguments, with one pair for each predecessor basic block of the
3615 current block. Only values of <a href="#t_firstclass">first class</a>
3616 type may be used as the value arguments to the PHI node. Only labels
3617 may be used as the label arguments.</p>
3618 <p>There must be no non-phi instructions between the start of a basic
3619 block and the PHI instructions: i.e. PHI instructions must be first in
3622 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
3623 specified by the pair corresponding to the predecessor basic block that executed
3624 just prior to the current block.</p>
3626 <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>
3629 <!-- _______________________________________________________________________ -->
3630 <div class="doc_subsubsection">
3631 <a name="i_select">'<tt>select</tt>' Instruction</a>
3634 <div class="doc_text">
3639 <result> = select i1 <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
3645 The '<tt>select</tt>' instruction is used to choose one value based on a
3646 condition, without branching.
3653 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.
3659 If the boolean condition evaluates to true, the instruction returns the first
3660 value argument; otherwise, it returns the second value argument.
3666 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
3671 <!-- _______________________________________________________________________ -->
3672 <div class="doc_subsubsection">
3673 <a name="i_call">'<tt>call</tt>' Instruction</a>
3676 <div class="doc_text">
3680 <result> = [tail] call [<a href="#callingconv">cconv</a>] <ty> [<fnty>*] <fnptrval>(<param list>)
3685 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
3689 <p>This instruction requires several arguments:</p>
3693 <p>The optional "tail" marker indicates whether the callee function accesses
3694 any allocas or varargs in the caller. If the "tail" marker is present, the
3695 function call is eligible for tail call optimization. Note that calls may
3696 be marked "tail" even if they do not occur before a <a
3697 href="#i_ret"><tt>ret</tt></a> instruction.
3700 <p>The optional "cconv" marker indicates which <a href="#callingconv">calling
3701 convention</a> the call should use. If none is specified, the call defaults
3702 to using C calling conventions.
3705 <p>'<tt>ty</tt>': the type of the call instruction itself which is also
3706 the type of the return value. Functions that return no value are marked
3707 <tt><a href="#t_void">void</a></tt>.</p>
3710 <p>'<tt>fnty</tt>': shall be the signature of the pointer to function
3711 value being invoked. The argument types must match the types implied by
3712 this signature. This type can be omitted if the function is not varargs
3713 and if the function type does not return a pointer to a function.</p>
3716 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
3717 be invoked. In most cases, this is a direct function invocation, but
3718 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
3719 to function value.</p>
3722 <p>'<tt>function args</tt>': argument list whose types match the
3723 function signature argument types. All arguments must be of
3724 <a href="#t_firstclass">first class</a> type. If the function signature
3725 indicates the function accepts a variable number of arguments, the extra
3726 arguments can be specified.</p>
3732 <p>The '<tt>call</tt>' instruction is used to cause control flow to
3733 transfer to a specified function, with its incoming arguments bound to
3734 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
3735 instruction in the called function, control flow continues with the
3736 instruction after the function call, and the return value of the
3737 function is bound to the result argument. This is a simpler case of
3738 the <a href="#i_invoke">invoke</a> instruction.</p>
3743 %retval = call i32 @test(i32 %argc)
3744 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42);
3745 %X = tail call i32 @foo()
3746 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo()
3747 %Z = call void %foo(i8 97 signext)
3752 <!-- _______________________________________________________________________ -->
3753 <div class="doc_subsubsection">
3754 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
3757 <div class="doc_text">
3762 <resultval> = va_arg <va_list*> <arglist>, <argty>
3767 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
3768 the "variable argument" area of a function call. It is used to implement the
3769 <tt>va_arg</tt> macro in C.</p>
3773 <p>This instruction takes a <tt>va_list*</tt> value and the type of
3774 the argument. It returns a value of the specified argument type and
3775 increments the <tt>va_list</tt> to point to the next argument. The
3776 actual type of <tt>va_list</tt> is target specific.</p>
3780 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
3781 type from the specified <tt>va_list</tt> and causes the
3782 <tt>va_list</tt> to point to the next argument. For more information,
3783 see the variable argument handling <a href="#int_varargs">Intrinsic
3786 <p>It is legal for this instruction to be called in a function which does not
3787 take a variable number of arguments, for example, the <tt>vfprintf</tt>
3790 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
3791 href="#intrinsics">intrinsic function</a> because it takes a type as an
3796 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
3800 <!-- *********************************************************************** -->
3801 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
3802 <!-- *********************************************************************** -->
3804 <div class="doc_text">
3806 <p>LLVM supports the notion of an "intrinsic function". These functions have
3807 well known names and semantics and are required to follow certain restrictions.
3808 Overall, these intrinsics represent an extension mechanism for the LLVM
3809 language that does not require changing all of the transformations in LLVM when
3810 adding to the language (or the bitcode reader/writer, the parser, etc...).</p>
3812 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
3813 prefix is reserved in LLVM for intrinsic names; thus, function names may not
3814 begin with this prefix. Intrinsic functions must always be external functions:
3815 you cannot define the body of intrinsic functions. Intrinsic functions may
3816 only be used in call or invoke instructions: it is illegal to take the address
3817 of an intrinsic function. Additionally, because intrinsic functions are part
3818 of the LLVM language, it is required if any are added that they be documented
3821 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
3822 a family of functions that perform the same operation but on different data
3823 types. Because LLVM can represent over 8 million different integer types,
3824 overloading is used commonly to allow an intrinsic function to operate on any
3825 integer type. One or more of the argument types or the result type can be
3826 overloaded to accept any integer type. Argument types may also be defined as
3827 exactly matching a previous argument's type or the result type. This allows an
3828 intrinsic function which accepts multiple arguments, but needs all of them to
3829 be of the same type, to only be overloaded with respect to a single argument or
3832 <p>Overloaded intrinsics will have the names of its overloaded argument types
3833 encoded into its function name, each preceded by a period. Only those types
3834 which are overloaded result in a name suffix. Arguments whose type is matched
3835 against another type do not. For example, the <tt>llvm.ctpop</tt> function can
3836 take an integer of any width and returns an integer of exactly the same integer
3837 width. This leads to a family of functions such as
3838 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29 %val)</tt>.
3839 Only one type, the return type, is overloaded, and only one type suffix is
3840 required. Because the argument's type is matched against the return type, it
3841 does not require its own name suffix.</p>
3843 <p>To learn how to add an intrinsic function, please see the
3844 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.
3849 <!-- ======================================================================= -->
3850 <div class="doc_subsection">
3851 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
3854 <div class="doc_text">
3856 <p>Variable argument support is defined in LLVM with the <a
3857 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
3858 intrinsic functions. These functions are related to the similarly
3859 named macros defined in the <tt><stdarg.h></tt> header file.</p>
3861 <p>All of these functions operate on arguments that use a
3862 target-specific value type "<tt>va_list</tt>". The LLVM assembly
3863 language reference manual does not define what this type is, so all
3864 transformations should be prepared to handle these functions regardless of
3867 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
3868 instruction and the variable argument handling intrinsic functions are
3871 <div class="doc_code">
3873 define i32 @test(i32 %X, ...) {
3874 ; Initialize variable argument processing
3876 %ap2 = bitcast i8** %ap to i8*
3877 call void @llvm.va_start(i8* %ap2)
3879 ; Read a single integer argument
3880 %tmp = va_arg i8** %ap, i32
3882 ; Demonstrate usage of llvm.va_copy and llvm.va_end
3884 %aq2 = bitcast i8** %aq to i8*
3885 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
3886 call void @llvm.va_end(i8* %aq2)
3888 ; Stop processing of arguments.
3889 call void @llvm.va_end(i8* %ap2)
3893 declare void @llvm.va_start(i8*)
3894 declare void @llvm.va_copy(i8*, i8*)
3895 declare void @llvm.va_end(i8*)
3901 <!-- _______________________________________________________________________ -->
3902 <div class="doc_subsubsection">
3903 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
3907 <div class="doc_text">
3909 <pre> declare void %llvm.va_start(i8* <arglist>)<br></pre>
3911 <P>The '<tt>llvm.va_start</tt>' intrinsic initializes
3912 <tt>*<arglist></tt> for subsequent use by <tt><a
3913 href="#i_va_arg">va_arg</a></tt>.</p>
3917 <P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
3921 <P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
3922 macro available in C. In a target-dependent way, it initializes the
3923 <tt>va_list</tt> element to which the argument points, so that the next call to
3924 <tt>va_arg</tt> will produce the first variable argument passed to the function.
3925 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
3926 last argument of the function as the compiler can figure that out.</p>
3930 <!-- _______________________________________________________________________ -->
3931 <div class="doc_subsubsection">
3932 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
3935 <div class="doc_text">
3937 <pre> declare void @llvm.va_end(i8* <arglist>)<br></pre>
3940 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
3941 which has been initialized previously with <tt><a href="#int_va_start">llvm.va_start</a></tt>
3942 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
3946 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
3950 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
3951 macro available in C. In a target-dependent way, it destroys the
3952 <tt>va_list</tt> element to which the argument points. Calls to <a
3953 href="#int_va_start"><tt>llvm.va_start</tt></a> and <a href="#int_va_copy">
3954 <tt>llvm.va_copy</tt></a> must be matched exactly with calls to
3955 <tt>llvm.va_end</tt>.</p>
3959 <!-- _______________________________________________________________________ -->
3960 <div class="doc_subsubsection">
3961 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
3964 <div class="doc_text">
3969 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
3974 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
3975 from the source argument list to the destination argument list.</p>
3979 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
3980 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
3985 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
3986 macro available in C. In a target-dependent way, it copies the source
3987 <tt>va_list</tt> element into the destination <tt>va_list</tt> element. This
3988 intrinsic is necessary because the <tt><a href="#int_va_start">
3989 llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
3990 example, memory allocation.</p>
3994 <!-- ======================================================================= -->
3995 <div class="doc_subsection">
3996 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
3999 <div class="doc_text">
4002 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
4003 Collection</a> requires the implementation and generation of these intrinsics.
4004 These intrinsics allow identification of <a href="#int_gcroot">GC roots on the
4005 stack</a>, as well as garbage collector implementations that require <a
4006 href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a> barriers.
4007 Front-ends for type-safe garbage collected languages should generate these
4008 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
4009 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
4012 <p>The garbage collection intrinsics only operate on objects in the generic
4013 address space (address space zero).</p>
4017 <!-- _______________________________________________________________________ -->
4018 <div class="doc_subsubsection">
4019 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
4022 <div class="doc_text">
4027 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
4032 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
4033 the code generator, and allows some metadata to be associated with it.</p>
4037 <p>The first argument specifies the address of a stack object that contains the
4038 root pointer. The second pointer (which must be either a constant or a global
4039 value address) contains the meta-data to be associated with the root.</p>
4043 <p>At runtime, a call to this intrinsics stores a null pointer into the "ptrloc"
4044 location. At compile-time, the code generator generates information to allow
4045 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
4046 intrinsic may only be used in a function which <a href="#gc">specifies a GC
4052 <!-- _______________________________________________________________________ -->
4053 <div class="doc_subsubsection">
4054 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
4057 <div class="doc_text">
4062 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
4067 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
4068 locations, allowing garbage collector implementations that require read
4073 <p>The second argument is the address to read from, which should be an address
4074 allocated from the garbage collector. The first object is a pointer to the
4075 start of the referenced object, if needed by the language runtime (otherwise
4080 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
4081 instruction, but may be replaced with substantially more complex code by the
4082 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
4083 may only be used in a function which <a href="#gc">specifies a GC
4089 <!-- _______________________________________________________________________ -->
4090 <div class="doc_subsubsection">
4091 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
4094 <div class="doc_text">
4099 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
4104 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
4105 locations, allowing garbage collector implementations that require write
4106 barriers (such as generational or reference counting collectors).</p>
4110 <p>The first argument is the reference to store, the second is the start of the
4111 object to store it to, and the third is the address of the field of Obj to
4112 store to. If the runtime does not require a pointer to the object, Obj may be
4117 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
4118 instruction, but may be replaced with substantially more complex code by the
4119 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
4120 may only be used in a function which <a href="#gc">specifies a GC
4127 <!-- ======================================================================= -->
4128 <div class="doc_subsection">
4129 <a name="int_codegen">Code Generator Intrinsics</a>
4132 <div class="doc_text">
4134 These intrinsics are provided by LLVM to expose special features that may only
4135 be implemented with code generator support.
4140 <!-- _______________________________________________________________________ -->
4141 <div class="doc_subsubsection">
4142 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
4145 <div class="doc_text">
4149 declare i8 *@llvm.returnaddress(i32 <level>)
4155 The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
4156 target-specific value indicating the return address of the current function
4157 or one of its callers.
4163 The argument to this intrinsic indicates which function to return the address
4164 for. Zero indicates the calling function, one indicates its caller, etc. The
4165 argument is <b>required</b> to be a constant integer value.
4171 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
4172 the return address of the specified call frame, or zero if it cannot be
4173 identified. The value returned by this intrinsic is likely to be incorrect or 0
4174 for arguments other than zero, so it should only be used for debugging purposes.
4178 Note that calling this intrinsic does not prevent function inlining or other
4179 aggressive transformations, so the value returned may not be that of the obvious
4180 source-language caller.
4185 <!-- _______________________________________________________________________ -->
4186 <div class="doc_subsubsection">
4187 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
4190 <div class="doc_text">
4194 declare i8 *@llvm.frameaddress(i32 <level>)
4200 The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
4201 target-specific frame pointer value for the specified stack frame.
4207 The argument to this intrinsic indicates which function to return the frame
4208 pointer for. Zero indicates the calling function, one indicates its caller,
4209 etc. The argument is <b>required</b> to be a constant integer value.
4215 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
4216 the frame address of the specified call frame, or zero if it cannot be
4217 identified. The value returned by this intrinsic is likely to be incorrect or 0
4218 for arguments other than zero, so it should only be used for debugging purposes.
4222 Note that calling this intrinsic does not prevent function inlining or other
4223 aggressive transformations, so the value returned may not be that of the obvious
4224 source-language caller.
4228 <!-- _______________________________________________________________________ -->
4229 <div class="doc_subsubsection">
4230 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
4233 <div class="doc_text">
4237 declare i8 *@llvm.stacksave()
4243 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
4244 the function stack, for use with <a href="#int_stackrestore">
4245 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
4246 features like scoped automatic variable sized arrays in C99.
4252 This intrinsic returns a opaque pointer value that can be passed to <a
4253 href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
4254 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
4255 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
4256 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
4257 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
4258 that were allocated after the <tt>llvm.stacksave</tt> was executed.
4263 <!-- _______________________________________________________________________ -->
4264 <div class="doc_subsubsection">
4265 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
4268 <div class="doc_text">
4272 declare void @llvm.stackrestore(i8 * %ptr)
4278 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
4279 the function stack to the state it was in when the corresponding <a
4280 href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
4281 useful for implementing language features like scoped automatic variable sized
4288 See the description for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.
4294 <!-- _______________________________________________________________________ -->
4295 <div class="doc_subsubsection">
4296 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
4299 <div class="doc_text">
4303 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>)
4310 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
4311 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
4313 effect on the behavior of the program but can change its performance
4320 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
4321 determining if the fetch should be for a read (0) or write (1), and
4322 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
4323 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
4324 <tt>locality</tt> arguments must be constant integers.
4330 This intrinsic does not modify the behavior of the program. In particular,
4331 prefetches cannot trap and do not produce a value. On targets that support this
4332 intrinsic, the prefetch can provide hints to the processor cache for better
4338 <!-- _______________________________________________________________________ -->
4339 <div class="doc_subsubsection">
4340 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
4343 <div class="doc_text">
4347 declare void @llvm.pcmarker(i32 <id>)
4354 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
4356 code to simulators and other tools. The method is target specific, but it is
4357 expected that the marker will use exported symbols to transmit the PC of the marker.
4358 The marker makes no guarantees that it will remain with any specific instruction
4359 after optimizations. It is possible that the presence of a marker will inhibit
4360 optimizations. The intended use is to be inserted after optimizations to allow
4361 correlations of simulation runs.
4367 <tt>id</tt> is a numerical id identifying the marker.
4373 This intrinsic does not modify the behavior of the program. Backends that do not
4374 support this intrinisic may ignore it.
4379 <!-- _______________________________________________________________________ -->
4380 <div class="doc_subsubsection">
4381 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
4384 <div class="doc_text">
4388 declare i64 @llvm.readcyclecounter( )
4395 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
4396 counter register (or similar low latency, high accuracy clocks) on those targets
4397 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
4398 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
4399 should only be used for small timings.
4405 When directly supported, reading the cycle counter should not modify any memory.
4406 Implementations are allowed to either return a application specific value or a
4407 system wide value. On backends without support, this is lowered to a constant 0.
4412 <!-- ======================================================================= -->
4413 <div class="doc_subsection">
4414 <a name="int_libc">Standard C Library Intrinsics</a>
4417 <div class="doc_text">
4419 LLVM provides intrinsics for a few important standard C library functions.
4420 These intrinsics allow source-language front-ends to pass information about the
4421 alignment of the pointer arguments to the code generator, providing opportunity
4422 for more efficient code generation.
4427 <!-- _______________________________________________________________________ -->
4428 <div class="doc_subsubsection">
4429 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
4432 <div class="doc_text">
4436 declare void @llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
4437 i32 <len>, i32 <align>)
4438 declare void @llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
4439 i64 <len>, i32 <align>)
4445 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4446 location to the destination location.
4450 Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
4451 intrinsics do not return a value, and takes an extra alignment argument.
4457 The first argument is a pointer to the destination, the second is a pointer to
4458 the source. The third argument is an integer argument
4459 specifying the number of bytes to copy, and the fourth argument is the alignment
4460 of the source and destination locations.
4464 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4465 the caller guarantees that both the source and destination pointers are aligned
4472 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4473 location to the destination location, which are not allowed to overlap. It
4474 copies "len" bytes of memory over. If the argument is known to be aligned to
4475 some boundary, this can be specified as the fourth argument, otherwise it should
4481 <!-- _______________________________________________________________________ -->
4482 <div class="doc_subsubsection">
4483 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
4486 <div class="doc_text">
4490 declare void @llvm.memmove.i32(i8 * <dest>, i8 * <src>,
4491 i32 <len>, i32 <align>)
4492 declare void @llvm.memmove.i64(i8 * <dest>, i8 * <src>,
4493 i64 <len>, i32 <align>)
4499 The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
4500 location to the destination location. It is similar to the
4501 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to overlap.
4505 Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
4506 intrinsics do not return a value, and takes an extra alignment argument.
4512 The first argument is a pointer to the destination, the second is a pointer to
4513 the source. The third argument is an integer argument
4514 specifying the number of bytes to copy, and the fourth argument is the alignment
4515 of the source and destination locations.
4519 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4520 the caller guarantees that the source and destination pointers are aligned to
4527 The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
4528 location to the destination location, which may overlap. It
4529 copies "len" bytes of memory over. If the argument is known to be aligned to
4530 some boundary, this can be specified as the fourth argument, otherwise it should
4536 <!-- _______________________________________________________________________ -->
4537 <div class="doc_subsubsection">
4538 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
4541 <div class="doc_text">
4545 declare void @llvm.memset.i32(i8 * <dest>, i8 <val>,
4546 i32 <len>, i32 <align>)
4547 declare void @llvm.memset.i64(i8 * <dest>, i8 <val>,
4548 i64 <len>, i32 <align>)
4554 The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
4559 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
4560 does not return a value, and takes an extra alignment argument.
4566 The first argument is a pointer to the destination to fill, the second is the
4567 byte value to fill it with, the third argument is an integer
4568 argument specifying the number of bytes to fill, and the fourth argument is the
4569 known alignment of destination location.
4573 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4574 the caller guarantees that the destination pointer is aligned to that boundary.
4580 The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
4582 destination location. If the argument is known to be aligned to some boundary,
4583 this can be specified as the fourth argument, otherwise it should be set to 0 or
4589 <!-- _______________________________________________________________________ -->
4590 <div class="doc_subsubsection">
4591 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
4594 <div class="doc_text">
4597 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
4598 floating point or vector of floating point type. Not all targets support all
4601 declare float @llvm.sqrt.f32(float %Val)
4602 declare double @llvm.sqrt.f64(double %Val)
4603 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
4604 declare fp128 @llvm.sqrt.f128(fp128 %Val)
4605 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
4611 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
4612 returning the same value as the libm '<tt>sqrt</tt>' functions would. Unlike
4613 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
4614 negative numbers (which allows for better optimization).
4620 The argument and return value are floating point numbers of the same type.
4626 This function returns the sqrt of the specified operand if it is a nonnegative
4627 floating point number.
4631 <!-- _______________________________________________________________________ -->
4632 <div class="doc_subsubsection">
4633 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
4636 <div class="doc_text">
4639 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
4640 floating point or vector of floating point type. Not all targets support all
4643 declare float @llvm.powi.f32(float %Val, i32 %power)
4644 declare double @llvm.powi.f64(double %Val, i32 %power)
4645 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
4646 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
4647 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
4653 The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
4654 specified (positive or negative) power. The order of evaluation of
4655 multiplications is not defined. When a vector of floating point type is
4656 used, the second argument remains a scalar integer value.
4662 The second argument is an integer power, and the first is a value to raise to
4669 This function returns the first value raised to the second power with an
4670 unspecified sequence of rounding operations.</p>
4673 <!-- _______________________________________________________________________ -->
4674 <div class="doc_subsubsection">
4675 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
4678 <div class="doc_text">
4681 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
4682 floating point or vector of floating point type. Not all targets support all
4685 declare float @llvm.sin.f32(float %Val)
4686 declare double @llvm.sin.f64(double %Val)
4687 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
4688 declare fp128 @llvm.sin.f128(fp128 %Val)
4689 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
4695 The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.
4701 The argument and return value are floating point numbers of the same type.
4707 This function returns the sine of the specified operand, returning the
4708 same values as the libm <tt>sin</tt> functions would, and handles error
4709 conditions in the same way.</p>
4712 <!-- _______________________________________________________________________ -->
4713 <div class="doc_subsubsection">
4714 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
4717 <div class="doc_text">
4720 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
4721 floating point or vector of floating point type. Not all targets support all
4724 declare float @llvm.cos.f32(float %Val)
4725 declare double @llvm.cos.f64(double %Val)
4726 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
4727 declare fp128 @llvm.cos.f128(fp128 %Val)
4728 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
4734 The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.
4740 The argument and return value are floating point numbers of the same type.
4746 This function returns the cosine of the specified operand, returning the
4747 same values as the libm <tt>cos</tt> functions would, and handles error
4748 conditions in the same way.</p>
4751 <!-- _______________________________________________________________________ -->
4752 <div class="doc_subsubsection">
4753 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
4756 <div class="doc_text">
4759 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
4760 floating point or vector of floating point type. Not all targets support all
4763 declare float @llvm.pow.f32(float %Val, float %Power)
4764 declare double @llvm.pow.f64(double %Val, double %Power)
4765 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
4766 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
4767 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
4773 The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
4774 specified (positive or negative) power.
4780 The second argument is a floating point power, and the first is a value to
4781 raise to that power.
4787 This function returns the first value raised to the second power,
4789 same values as the libm <tt>pow</tt> functions would, and handles error
4790 conditions in the same way.</p>
4794 <!-- ======================================================================= -->
4795 <div class="doc_subsection">
4796 <a name="int_manip">Bit Manipulation Intrinsics</a>
4799 <div class="doc_text">
4801 LLVM provides intrinsics for a few important bit manipulation operations.
4802 These allow efficient code generation for some algorithms.
4807 <!-- _______________________________________________________________________ -->
4808 <div class="doc_subsubsection">
4809 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
4812 <div class="doc_text">
4815 <p>This is an overloaded intrinsic function. You can use bswap on any integer
4816 type that is an even number of bytes (i.e. BitWidth % 16 == 0).
4818 declare i16 @llvm.bswap.i16(i16 <id>)
4819 declare i32 @llvm.bswap.i32(i32 <id>)
4820 declare i64 @llvm.bswap.i64(i64 <id>)
4826 The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
4827 values with an even number of bytes (positive multiple of 16 bits). These are
4828 useful for performing operations on data that is not in the target's native
4835 The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
4836 and low byte of the input i16 swapped. Similarly, the <tt>llvm.bswap.i32</tt>
4837 intrinsic returns an i32 value that has the four bytes of the input i32
4838 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the returned
4839 i32 will have its bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i48</tt>,
4840 <tt>llvm.bswap.i64</tt> and other intrinsics extend this concept to
4841 additional even-byte lengths (6 bytes, 8 bytes and more, respectively).
4846 <!-- _______________________________________________________________________ -->
4847 <div class="doc_subsubsection">
4848 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
4851 <div class="doc_text">
4854 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
4855 width. Not all targets support all bit widths however.
4857 declare i8 @llvm.ctpop.i8 (i8 <src>)
4858 declare i16 @llvm.ctpop.i16(i16 <src>)
4859 declare i32 @llvm.ctpop.i32(i32 <src>)
4860 declare i64 @llvm.ctpop.i64(i64 <src>)
4861 declare i256 @llvm.ctpop.i256(i256 <src>)
4867 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
4874 The only argument is the value to be counted. The argument may be of any
4875 integer type. The return type must match the argument type.
4881 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
4885 <!-- _______________________________________________________________________ -->
4886 <div class="doc_subsubsection">
4887 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
4890 <div class="doc_text">
4893 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
4894 integer bit width. Not all targets support all bit widths however.
4896 declare i8 @llvm.ctlz.i8 (i8 <src>)
4897 declare i16 @llvm.ctlz.i16(i16 <src>)
4898 declare i32 @llvm.ctlz.i32(i32 <src>)
4899 declare i64 @llvm.ctlz.i64(i64 <src>)
4900 declare i256 @llvm.ctlz.i256(i256 <src>)
4906 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
4907 leading zeros in a variable.
4913 The only argument is the value to be counted. The argument may be of any
4914 integer type. The return type must match the argument type.
4920 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
4921 in a variable. If the src == 0 then the result is the size in bits of the type
4922 of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.
4928 <!-- _______________________________________________________________________ -->
4929 <div class="doc_subsubsection">
4930 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
4933 <div class="doc_text">
4936 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
4937 integer bit width. Not all targets support all bit widths however.
4939 declare i8 @llvm.cttz.i8 (i8 <src>)
4940 declare i16 @llvm.cttz.i16(i16 <src>)
4941 declare i32 @llvm.cttz.i32(i32 <src>)
4942 declare i64 @llvm.cttz.i64(i64 <src>)
4943 declare i256 @llvm.cttz.i256(i256 <src>)
4949 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
4956 The only argument is the value to be counted. The argument may be of any
4957 integer type. The return type must match the argument type.
4963 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
4964 in a variable. If the src == 0 then the result is the size in bits of the type
4965 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
4969 <!-- _______________________________________________________________________ -->
4970 <div class="doc_subsubsection">
4971 <a name="int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic</a>
4974 <div class="doc_text">
4977 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.select</tt>
4978 on any integer bit width.
4980 declare i17 @llvm.part.select.i17 (i17 %val, i32 %loBit, i32 %hiBit)
4981 declare i29 @llvm.part.select.i29 (i29 %val, i32 %loBit, i32 %hiBit)
4985 <p>The '<tt>llvm.part.select</tt>' family of intrinsic functions selects a
4986 range of bits from an integer value and returns them in the same bit width as
4987 the original value.</p>
4990 <p>The first argument, <tt>%val</tt> and the result may be integer types of
4991 any bit width but they must have the same bit width. The second and third
4992 arguments must be <tt>i32</tt> type since they specify only a bit index.</p>
4995 <p>The operation of the '<tt>llvm.part.select</tt>' intrinsic has two modes
4996 of operation: forwards and reverse. If <tt>%loBit</tt> is greater than
4997 <tt>%hiBits</tt> then the intrinsic operates in reverse mode. Otherwise it
4998 operates in forward mode.</p>
4999 <p>In forward mode, this intrinsic is the equivalent of shifting <tt>%val</tt>
5000 right by <tt>%loBit</tt> bits and then ANDing it with a mask with
5001 only the <tt>%hiBit - %loBit</tt> bits set, as follows:</p>
5003 <li>The <tt>%val</tt> is shifted right (LSHR) by the number of bits specified
5004 by <tt>%loBits</tt>. This normalizes the value to the low order bits.</li>
5005 <li>The <tt>%loBits</tt> value is subtracted from the <tt>%hiBits</tt> value
5006 to determine the number of bits to retain.</li>
5007 <li>A mask of the retained bits is created by shifting a -1 value.</li>
5008 <li>The mask is ANDed with <tt>%val</tt> to produce the result.
5010 <p>In reverse mode, a similar computation is made except that the bits are
5011 returned in the reverse order. So, for example, if <tt>X</tt> has the value
5012 <tt>i16 0x0ACF (101011001111)</tt> and we apply
5013 <tt>part.select(i16 X, 8, 3)</tt> to it, we get back the value
5014 <tt>i16 0x0026 (000000100110)</tt>.</p>
5017 <div class="doc_subsubsection">
5018 <a name="int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic</a>
5021 <div class="doc_text">
5024 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.set</tt>
5025 on any integer bit width.
5027 declare i17 @llvm.part.set.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
5028 declare i29 @llvm.part.set.i29.i9 (i29 %val, i9 %repl, i32 %lo, i32 %hi)
5032 <p>The '<tt>llvm.part.set</tt>' family of intrinsic functions replaces a range
5033 of bits in an integer value with another integer value. It returns the integer
5034 with the replaced bits.</p>
5037 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5038 any bit width but they must have the same bit width. <tt>%val</tt> is the value
5039 whose bits will be replaced. The second argument, <tt>%repl</tt> may be an
5040 integer of any bit width. The third and fourth arguments must be <tt>i32</tt>
5041 type since they specify only a bit index.</p>
5044 <p>The operation of the '<tt>llvm.part.set</tt>' intrinsic has two modes
5045 of operation: forwards and reverse. If <tt>%lo</tt> is greater than
5046 <tt>%hi</tt> then the intrinsic operates in reverse mode. Otherwise it
5047 operates in forward mode.</p>
5048 <p>For both modes, the <tt>%repl</tt> value is prepared for use by either
5049 truncating it down to the size of the replacement area or zero extending it
5050 up to that size.</p>
5051 <p>In forward mode, the bits between <tt>%lo</tt> and <tt>%hi</tt> (inclusive)
5052 are replaced with corresponding bits from <tt>%repl</tt>. That is the 0th bit
5053 in <tt>%repl</tt> replaces the <tt>%lo</tt>th bit in <tt>%val</tt> and etc. up
5054 to the <tt>%hi</tt>th bit.
5055 <p>In reverse mode, a similar computation is made except that the bits are
5056 reversed. That is, the <tt>0</tt>th bit in <tt>%repl</tt> replaces the
5057 <tt>%hi</tt> bit in <tt>%val</tt> and etc. down to the <tt>%lo</tt>th bit.
5060 llvm.part.set(0xFFFF, 0, 4, 7) -> 0xFF0F
5061 llvm.part.set(0xFFFF, 0, 7, 4) -> 0xFF0F
5062 llvm.part.set(0xFFFF, 1, 7, 4) -> 0xFF8F
5063 llvm.part.set(0xFFFF, F, 8, 3) -> 0xFFE7
5064 llvm.part.set(0xFFFF, 0, 3, 8) -> 0xFE07
5068 <!-- ======================================================================= -->
5069 <div class="doc_subsection">
5070 <a name="int_debugger">Debugger Intrinsics</a>
5073 <div class="doc_text">
5075 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
5076 are described in the <a
5077 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
5078 Debugging</a> document.
5083 <!-- ======================================================================= -->
5084 <div class="doc_subsection">
5085 <a name="int_eh">Exception Handling Intrinsics</a>
5088 <div class="doc_text">
5089 <p> The LLVM exception handling intrinsics (which all start with
5090 <tt>llvm.eh.</tt> prefix), are described in the <a
5091 href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
5092 Handling</a> document. </p>
5095 <!-- ======================================================================= -->
5096 <div class="doc_subsection">
5097 <a name="int_trampoline">Trampoline Intrinsic</a>
5100 <div class="doc_text">
5102 This intrinsic makes it possible to excise one parameter, marked with
5103 the <tt>nest</tt> attribute, from a function. The result is a callable
5104 function pointer lacking the nest parameter - the caller does not need
5105 to provide a value for it. Instead, the value to use is stored in
5106 advance in a "trampoline", a block of memory usually allocated
5107 on the stack, which also contains code to splice the nest value into the
5108 argument list. This is used to implement the GCC nested function address
5112 For example, if the function is
5113 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
5114 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as follows:</p>
5116 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
5117 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
5118 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
5119 %fp = bitcast i8* %p to i32 (i32, i32)*
5121 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
5122 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
5125 <!-- _______________________________________________________________________ -->
5126 <div class="doc_subsubsection">
5127 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
5129 <div class="doc_text">
5132 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
5136 This fills the memory pointed to by <tt>tramp</tt> with code
5137 and returns a function pointer suitable for executing it.
5141 The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
5142 pointers. The <tt>tramp</tt> argument must point to a sufficiently large
5143 and sufficiently aligned block of memory; this memory is written to by the
5144 intrinsic. Note that the size and the alignment are target-specific - LLVM
5145 currently provides no portable way of determining them, so a front-end that
5146 generates this intrinsic needs to have some target-specific knowledge.
5147 The <tt>func</tt> argument must hold a function bitcast to an <tt>i8*</tt>.
5151 The block of memory pointed to by <tt>tramp</tt> is filled with target
5152 dependent code, turning it into a function. A pointer to this function is
5153 returned, but needs to be bitcast to an
5154 <a href="#int_trampoline">appropriate function pointer type</a>
5155 before being called. The new function's signature is the same as that of
5156 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
5157 removed. At most one such <tt>nest</tt> argument is allowed, and it must be
5158 of pointer type. Calling the new function is equivalent to calling
5159 <tt>func</tt> with the same argument list, but with <tt>nval</tt> used for the
5160 missing <tt>nest</tt> argument. If, after calling
5161 <tt>llvm.init.trampoline</tt>, the memory pointed to by <tt>tramp</tt> is
5162 modified, then the effect of any later call to the returned function pointer is
5167 <!-- ======================================================================= -->
5168 <div class="doc_subsection">
5169 <a name="int_general">General Intrinsics</a>
5172 <div class="doc_text">
5173 <p> This class of intrinsics is designed to be generic and has
5174 no specific purpose. </p>
5177 <!-- _______________________________________________________________________ -->
5178 <div class="doc_subsubsection">
5179 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
5182 <div class="doc_text">
5186 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
5192 The '<tt>llvm.var.annotation</tt>' intrinsic
5198 The first argument is a pointer to a value, the second is a pointer to a
5199 global string, the third is a pointer to a global string which is the source
5200 file name, and the last argument is the line number.
5206 This intrinsic allows annotation of local variables with arbitrary strings.
5207 This can be useful for special purpose optimizations that want to look for these
5208 annotations. These have no other defined use, they are ignored by code
5209 generation and optimization.
5212 <!-- _______________________________________________________________________ -->
5213 <div class="doc_subsubsection">
5214 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
5217 <div class="doc_text">
5220 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
5221 any integer bit width.
5224 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
5225 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
5226 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
5227 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
5228 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
5234 The '<tt>llvm.annotation</tt>' intrinsic.
5240 The first argument is an integer value (result of some expression),
5241 the second is a pointer to a global string, the third is a pointer to a global
5242 string which is the source file name, and the last argument is the line number.
5243 It returns the value of the first argument.
5249 This intrinsic allows annotations to be put on arbitrary expressions
5250 with arbitrary strings. This can be useful for special purpose optimizations
5251 that want to look for these annotations. These have no other defined use, they
5252 are ignored by code generation and optimization.
5255 <!-- *********************************************************************** -->
5258 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
5259 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
5260 <a href="http://validator.w3.org/check/referer"><img
5261 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
5263 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
5264 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
5265 Last modified: $Date$