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>
1035 <div class="doc_text">
1036 <p>The primitive types are the fundamental building blocks of the LLVM
1039 <!-- _______________________________________________________________________ -->
1040 <div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1042 <div class="doc_text">
1045 <tr><th>Type</th><th>Description</th></tr>
1046 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1047 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1048 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1049 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1050 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1055 <!-- _______________________________________________________________________ -->
1056 <div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1058 <div class="doc_text">
1060 <p>The void type does not represent any value and has no size.</p>
1069 <!-- _______________________________________________________________________ -->
1070 <div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1072 <div class="doc_text">
1074 <p>The label type represents code labels.</p>
1084 <!-- ======================================================================= -->
1085 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1087 <div class="doc_text">
1089 <p>The real power in LLVM comes from the derived types in the system.
1090 This is what allows a programmer to represent arrays, functions,
1091 pointers, and other useful types. Note that these derived types may be
1092 recursive: For example, it is possible to have a two dimensional array.</p>
1096 <!-- _______________________________________________________________________ -->
1097 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1099 <div class="doc_text">
1102 <p>The integer type is a very simple derived type that simply specifies an
1103 arbitrary bit width for the integer type desired. Any bit width from 1 bit to
1104 2^23-1 (about 8 million) can be specified.</p>
1112 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1116 <table class="layout">
1119 <td><tt>i1</tt></td>
1120 <td>a single-bit integer.</td>
1122 <td><tt>i32</tt></td>
1123 <td>a 32-bit integer.</td>
1125 <td><tt>i1942652</tt></td>
1126 <td>a really big integer of over 1 million bits.</td>
1132 <!-- _______________________________________________________________________ -->
1133 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1135 <div class="doc_text">
1139 <p>The array type is a very simple derived type that arranges elements
1140 sequentially in memory. The array type requires a size (number of
1141 elements) and an underlying data type.</p>
1146 [<# elements> x <elementtype>]
1149 <p>The number of elements is a constant integer value; elementtype may
1150 be any type with a size.</p>
1153 <table class="layout">
1155 <td class="left"><tt>[40 x i32]</tt></td>
1156 <td class="left">Array of 40 32-bit integer values.</td>
1159 <td class="left"><tt>[41 x i32]</tt></td>
1160 <td class="left">Array of 41 32-bit integer values.</td>
1163 <td class="left"><tt>[4 x i8]</tt></td>
1164 <td class="left">Array of 4 8-bit integer values.</td>
1167 <p>Here are some examples of multidimensional arrays:</p>
1168 <table class="layout">
1170 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1171 <td class="left">3x4 array of 32-bit integer values.</td>
1174 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1175 <td class="left">12x10 array of single precision floating point values.</td>
1178 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1179 <td class="left">2x3x4 array of 16-bit integer values.</td>
1183 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1184 length array. Normally, accesses past the end of an array are undefined in
1185 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
1186 As a special case, however, zero length arrays are recognized to be variable
1187 length. This allows implementation of 'pascal style arrays' with the LLVM
1188 type "{ i32, [0 x float]}", for example.</p>
1192 <!-- _______________________________________________________________________ -->
1193 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1194 <div class="doc_text">
1196 <p>The function type can be thought of as a function signature. It
1197 consists of a return type and a list of formal parameter types.
1198 Function types are usually used to build virtual function tables
1199 (which are structures of pointers to functions), for indirect function
1200 calls, and when defining a function.</p>
1202 The return type of a function type cannot be an aggregate type.
1205 <pre> <returntype> (<parameter list>)<br></pre>
1206 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1207 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1208 which indicates that the function takes a variable number of arguments.
1209 Variable argument functions can access their arguments with the <a
1210 href="#int_varargs">variable argument handling intrinsic</a> functions.</p>
1212 <table class="layout">
1214 <td class="left"><tt>i32 (i32)</tt></td>
1215 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1217 </tr><tr class="layout">
1218 <td class="left"><tt>float (i16 signext, i32 *) *
1220 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1221 an <tt>i16</tt> that should be sign extended and a
1222 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1225 </tr><tr class="layout">
1226 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1227 <td class="left">A vararg function that takes at least one
1228 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1229 which returns an integer. This is the signature for <tt>printf</tt> in
1236 <!-- _______________________________________________________________________ -->
1237 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1238 <div class="doc_text">
1240 <p>The structure type is used to represent a collection of data members
1241 together in memory. The packing of the field types is defined to match
1242 the ABI of the underlying processor. The elements of a structure may
1243 be any type that has a size.</p>
1244 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1245 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1246 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1249 <pre> { <type list> }<br></pre>
1251 <table class="layout">
1253 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1254 <td class="left">A triple of three <tt>i32</tt> values</td>
1255 </tr><tr class="layout">
1256 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1257 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1258 second element is a <a href="#t_pointer">pointer</a> to a
1259 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1260 an <tt>i32</tt>.</td>
1265 <!-- _______________________________________________________________________ -->
1266 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1268 <div class="doc_text">
1270 <p>The packed structure type is used to represent a collection of data members
1271 together in memory. There is no padding between fields. Further, the alignment
1272 of a packed structure is 1 byte. The elements of a packed structure may
1273 be any type that has a size.</p>
1274 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1275 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1276 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1279 <pre> < { <type list> } > <br></pre>
1281 <table class="layout">
1283 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1284 <td class="left">A triple of three <tt>i32</tt> values</td>
1285 </tr><tr class="layout">
1286 <td class="left"><tt>< { float, i32 (i32)* } ></tt></td>
1287 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1288 second element is a <a href="#t_pointer">pointer</a> to a
1289 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1290 an <tt>i32</tt>.</td>
1295 <!-- _______________________________________________________________________ -->
1296 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1297 <div class="doc_text">
1299 <p>As in many languages, the pointer type represents a pointer or
1300 reference to another object, which must live in memory. Pointer types may have
1301 an optional address space attribute defining the target-specific numbered
1302 address space where the pointed-to object resides. The default address space is
1305 <pre> <type> *<br></pre>
1307 <table class="layout">
1309 <td class="left"><tt>[4x i32]*</tt></td>
1310 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1311 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1314 <td class="left"><tt>i32 (i32 *) *</tt></td>
1315 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
1316 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1320 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1321 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1322 that resides in address space #5.</td>
1327 <!-- _______________________________________________________________________ -->
1328 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1329 <div class="doc_text">
1333 <p>A vector type is a simple derived type that represents a vector
1334 of elements. Vector types are used when multiple primitive data
1335 are operated in parallel using a single instruction (SIMD).
1336 A vector type requires a size (number of
1337 elements) and an underlying primitive data type. Vectors must have a power
1338 of two length (1, 2, 4, 8, 16 ...). Vector types are
1339 considered <a href="#t_firstclass">first class</a>.</p>
1344 < <# elements> x <elementtype> >
1347 <p>The number of elements is a constant integer value; elementtype may
1348 be any integer or floating point type.</p>
1352 <table class="layout">
1354 <td class="left"><tt><4 x i32></tt></td>
1355 <td class="left">Vector of 4 32-bit integer values.</td>
1358 <td class="left"><tt><8 x float></tt></td>
1359 <td class="left">Vector of 8 32-bit floating-point values.</td>
1362 <td class="left"><tt><2 x i64></tt></td>
1363 <td class="left">Vector of 2 64-bit integer values.</td>
1368 <!-- _______________________________________________________________________ -->
1369 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1370 <div class="doc_text">
1374 <p>Opaque types are used to represent unknown types in the system. This
1375 corresponds (for example) to the C notion of a forward declared structure type.
1376 In LLVM, opaque types can eventually be resolved to any type (not just a
1377 structure type).</p>
1387 <table class="layout">
1389 <td class="left"><tt>opaque</tt></td>
1390 <td class="left">An opaque type.</td>
1396 <!-- *********************************************************************** -->
1397 <div class="doc_section"> <a name="constants">Constants</a> </div>
1398 <!-- *********************************************************************** -->
1400 <div class="doc_text">
1402 <p>LLVM has several different basic types of constants. This section describes
1403 them all and their syntax.</p>
1407 <!-- ======================================================================= -->
1408 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1410 <div class="doc_text">
1413 <dt><b>Boolean constants</b></dt>
1415 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1416 constants of the <tt><a href="#t_primitive">i1</a></tt> type.
1419 <dt><b>Integer constants</b></dt>
1421 <dd>Standard integers (such as '4') are constants of the <a
1422 href="#t_integer">integer</a> type. Negative numbers may be used with
1426 <dt><b>Floating point constants</b></dt>
1428 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1429 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1430 notation (see below). Floating point constants must have a <a
1431 href="#t_floating">floating point</a> type. </dd>
1433 <dt><b>Null pointer constants</b></dt>
1435 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1436 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1440 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1441 of floating point constants. For example, the form '<tt>double
1442 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1443 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1444 (and the only time that they are generated by the disassembler) is when a
1445 floating point constant must be emitted but it cannot be represented as a
1446 decimal floating point number. For example, NaN's, infinities, and other
1447 special values are represented in their IEEE hexadecimal format so that
1448 assembly and disassembly do not cause any bits to change in the constants.</p>
1452 <!-- ======================================================================= -->
1453 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1456 <div class="doc_text">
1457 <p>Aggregate constants arise from aggregation of simple constants
1458 and smaller aggregate constants.</p>
1461 <dt><b>Structure constants</b></dt>
1463 <dd>Structure constants are represented with notation similar to structure
1464 type definitions (a comma separated list of elements, surrounded by braces
1465 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
1466 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>". Structure constants
1467 must have <a href="#t_struct">structure type</a>, and the number and
1468 types of elements must match those specified by the type.
1471 <dt><b>Array constants</b></dt>
1473 <dd>Array constants are represented with notation similar to array type
1474 definitions (a comma separated list of elements, surrounded by square brackets
1475 (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74 ]</tt>". Array
1476 constants must have <a href="#t_array">array type</a>, and the number and
1477 types of elements must match those specified by the type.
1480 <dt><b>Vector constants</b></dt>
1482 <dd>Vector constants are represented with notation similar to vector type
1483 definitions (a comma separated list of elements, surrounded by
1484 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32 42,
1485 i32 11, i32 74, i32 100 ></tt>". Vector constants must have <a
1486 href="#t_vector">vector type</a>, and the number and types of elements must
1487 match those specified by the type.
1490 <dt><b>Zero initialization</b></dt>
1492 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1493 value to zero of <em>any</em> type, including scalar and aggregate types.
1494 This is often used to avoid having to print large zero initializers (e.g. for
1495 large arrays) and is always exactly equivalent to using explicit zero
1502 <!-- ======================================================================= -->
1503 <div class="doc_subsection">
1504 <a name="globalconstants">Global Variable and Function Addresses</a>
1507 <div class="doc_text">
1509 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1510 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1511 constants. These constants are explicitly referenced when the <a
1512 href="#identifiers">identifier for the global</a> is used and always have <a
1513 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1516 <div class="doc_code">
1520 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1526 <!-- ======================================================================= -->
1527 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1528 <div class="doc_text">
1529 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1530 no specific value. Undefined values may be of any type and be used anywhere
1531 a constant is permitted.</p>
1533 <p>Undefined values indicate to the compiler that the program is well defined
1534 no matter what value is used, giving the compiler more freedom to optimize.
1538 <!-- ======================================================================= -->
1539 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1542 <div class="doc_text">
1544 <p>Constant expressions are used to allow expressions involving other constants
1545 to be used as constants. Constant expressions may be of any <a
1546 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1547 that does not have side effects (e.g. load and call are not supported). The
1548 following is the syntax for constant expressions:</p>
1551 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1552 <dd>Truncate a constant to another type. The bit size of CST must be larger
1553 than the bit size of TYPE. Both types must be integers.</dd>
1555 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1556 <dd>Zero extend a constant to another type. The bit size of CST must be
1557 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1559 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1560 <dd>Sign extend a constant to another type. The bit size of CST must be
1561 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1563 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1564 <dd>Truncate a floating point constant to another floating point type. The
1565 size of CST must be larger than the size of TYPE. Both types must be
1566 floating point.</dd>
1568 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1569 <dd>Floating point extend a constant to another type. The size of CST must be
1570 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1572 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
1573 <dd>Convert a floating point constant to the corresponding unsigned integer
1574 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1575 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1576 of the same number of elements. If the value won't fit in the integer type,
1577 the results are undefined.</dd>
1579 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
1580 <dd>Convert a floating point constant to the corresponding signed integer
1581 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1582 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1583 of the same number of elements. If the value won't fit in the integer type,
1584 the results are undefined.</dd>
1586 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
1587 <dd>Convert an unsigned integer constant to the corresponding floating point
1588 constant. TYPE must be a scalar or vector floating point type. CST must be of
1589 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1590 of the same number of elements. If the value won't fit in the floating point
1591 type, the results are undefined.</dd>
1593 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
1594 <dd>Convert a signed integer constant to the corresponding floating point
1595 constant. TYPE must be a scalar or vector floating point type. CST must be of
1596 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1597 of the same number of elements. If the value won't fit in the floating point
1598 type, the results are undefined.</dd>
1600 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
1601 <dd>Convert a pointer typed constant to the corresponding integer constant
1602 TYPE must be an integer type. CST must be of pointer type. The CST value is
1603 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1605 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
1606 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1607 pointer type. CST must be of integer type. The CST value is zero extended,
1608 truncated, or unchanged to make it fit in a pointer size. This one is
1609 <i>really</i> dangerous!</dd>
1611 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
1612 <dd>Convert a constant, CST, to another TYPE. The size of CST and TYPE must be
1613 identical (same number of bits). The conversion is done as if the CST value
1614 was stored to memory and read back as TYPE. In other words, no bits change
1615 with this operator, just the type. This can be used for conversion of
1616 vector types to any other type, as long as they have the same bit width. For
1617 pointers it is only valid to cast to another pointer type.
1620 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1622 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1623 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1624 instruction, the index list may have zero or more indexes, which are required
1625 to make sense for the type of "CSTPTR".</dd>
1627 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1629 <dd>Perform the <a href="#i_select">select operation</a> on
1632 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
1633 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
1635 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
1636 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
1638 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1640 <dd>Perform the <a href="#i_extractelement">extractelement
1641 operation</a> on constants.
1643 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1645 <dd>Perform the <a href="#i_insertelement">insertelement
1646 operation</a> on constants.</dd>
1649 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
1651 <dd>Perform the <a href="#i_shufflevector">shufflevector
1652 operation</a> on constants.</dd>
1654 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1656 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1657 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1658 binary</a> operations. The constraints on operands are the same as those for
1659 the corresponding instruction (e.g. no bitwise operations on floating point
1660 values are allowed).</dd>
1664 <!-- *********************************************************************** -->
1665 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1666 <!-- *********************************************************************** -->
1668 <!-- ======================================================================= -->
1669 <div class="doc_subsection">
1670 <a name="inlineasm">Inline Assembler Expressions</a>
1673 <div class="doc_text">
1676 LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
1677 Module-Level Inline Assembly</a>) through the use of a special value. This
1678 value represents the inline assembler as a string (containing the instructions
1679 to emit), a list of operand constraints (stored as a string), and a flag that
1680 indicates whether or not the inline asm expression has side effects. An example
1681 inline assembler expression is:
1684 <div class="doc_code">
1686 i32 (i32) asm "bswap $0", "=r,r"
1691 Inline assembler expressions may <b>only</b> be used as the callee operand of
1692 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
1695 <div class="doc_code">
1697 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
1702 Inline asms with side effects not visible in the constraint list must be marked
1703 as having side effects. This is done through the use of the
1704 '<tt>sideeffect</tt>' keyword, like so:
1707 <div class="doc_code">
1709 call void asm sideeffect "eieio", ""()
1713 <p>TODO: The format of the asm and constraints string still need to be
1714 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1715 need to be documented).
1720 <!-- *********************************************************************** -->
1721 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1722 <!-- *********************************************************************** -->
1724 <div class="doc_text">
1726 <p>The LLVM instruction set consists of several different
1727 classifications of instructions: <a href="#terminators">terminator
1728 instructions</a>, <a href="#binaryops">binary instructions</a>,
1729 <a href="#bitwiseops">bitwise binary instructions</a>, <a
1730 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1731 instructions</a>.</p>
1735 <!-- ======================================================================= -->
1736 <div class="doc_subsection"> <a name="terminators">Terminator
1737 Instructions</a> </div>
1739 <div class="doc_text">
1741 <p>As mentioned <a href="#functionstructure">previously</a>, every
1742 basic block in a program ends with a "Terminator" instruction, which
1743 indicates which block should be executed after the current block is
1744 finished. These terminator instructions typically yield a '<tt>void</tt>'
1745 value: they produce control flow, not values (the one exception being
1746 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1747 <p>There are six different terminator instructions: the '<a
1748 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1749 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1750 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1751 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1752 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1756 <!-- _______________________________________________________________________ -->
1757 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1758 Instruction</a> </div>
1759 <div class="doc_text">
1761 <pre> ret <type> <value> <i>; Return a value from a non-void function</i>
1762 ret void <i>; Return from void function</i>
1765 <p>The '<tt>ret</tt>' instruction is used to return control flow (and a
1766 value) from a function back to the caller.</p>
1767 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
1768 returns a value and then causes control flow, and one that just causes
1769 control flow to occur.</p>
1771 <p>The '<tt>ret</tt>' instruction may return any '<a
1772 href="#t_firstclass">first class</a>' type. Notice that a function is
1773 not <a href="#wellformed">well formed</a> if there exists a '<tt>ret</tt>'
1774 instruction inside of the function that returns a value that does not
1775 match the return type of the function.</p>
1777 <p>When the '<tt>ret</tt>' instruction is executed, control flow
1778 returns back to the calling function's context. If the caller is a "<a
1779 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1780 the instruction after the call. If the caller was an "<a
1781 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1782 at the beginning of the "normal" destination block. If the instruction
1783 returns a value, that value shall set the call or invoke instruction's
1786 <pre> ret i32 5 <i>; Return an integer value of 5</i>
1787 ret void <i>; Return from a void function</i>
1790 <!-- _______________________________________________________________________ -->
1791 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1792 <div class="doc_text">
1794 <pre> br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1797 <p>The '<tt>br</tt>' instruction is used to cause control flow to
1798 transfer to a different basic block in the current function. There are
1799 two forms of this instruction, corresponding to a conditional branch
1800 and an unconditional branch.</p>
1802 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1803 single '<tt>i1</tt>' value and two '<tt>label</tt>' values. The
1804 unconditional form of the '<tt>br</tt>' instruction takes a single
1805 '<tt>label</tt>' value as a target.</p>
1807 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
1808 argument is evaluated. If the value is <tt>true</tt>, control flows
1809 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1810 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1812 <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
1813 href="#i_ret">ret</a> i32 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> i32 0<br></pre>
1815 <!-- _______________________________________________________________________ -->
1816 <div class="doc_subsubsection">
1817 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
1820 <div class="doc_text">
1824 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
1829 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
1830 several different places. It is a generalization of the '<tt>br</tt>'
1831 instruction, allowing a branch to occur to one of many possible
1837 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
1838 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
1839 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
1840 table is not allowed to contain duplicate constant entries.</p>
1844 <p>The <tt>switch</tt> instruction specifies a table of values and
1845 destinations. When the '<tt>switch</tt>' instruction is executed, this
1846 table is searched for the given value. If the value is found, control flow is
1847 transfered to the corresponding destination; otherwise, control flow is
1848 transfered to the default destination.</p>
1850 <h5>Implementation:</h5>
1852 <p>Depending on properties of the target machine and the particular
1853 <tt>switch</tt> instruction, this instruction may be code generated in different
1854 ways. For example, it could be generated as a series of chained conditional
1855 branches or with a lookup table.</p>
1860 <i>; Emulate a conditional br instruction</i>
1861 %Val = <a href="#i_zext">zext</a> i1 %value to i32
1862 switch i32 %Val, label %truedest [i32 0, label %falsedest ]
1864 <i>; Emulate an unconditional br instruction</i>
1865 switch i32 0, label %dest [ ]
1867 <i>; Implement a jump table:</i>
1868 switch i32 %val, label %otherwise [ i32 0, label %onzero
1870 i32 2, label %ontwo ]
1874 <!-- _______________________________________________________________________ -->
1875 <div class="doc_subsubsection">
1876 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
1879 <div class="doc_text">
1884 <result> = invoke [<a href="#callingconv">cconv</a>] <ptr to function ty> %<function ptr val>(<function args>)
1885 to label <normal label> unwind label <exception label>
1890 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
1891 function, with the possibility of control flow transfer to either the
1892 '<tt>normal</tt>' label or the
1893 '<tt>exception</tt>' label. If the callee function returns with the
1894 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
1895 "normal" label. If the callee (or any indirect callees) returns with the "<a
1896 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
1897 continued at the dynamically nearest "exception" label.</p>
1901 <p>This instruction requires several arguments:</p>
1905 The optional "cconv" marker indicates which <a href="#callingconv">calling
1906 convention</a> the call should use. If none is specified, the call defaults
1907 to using C calling conventions.
1909 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
1910 function value being invoked. In most cases, this is a direct function
1911 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
1912 an arbitrary pointer to function value.
1915 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
1916 function to be invoked. </li>
1918 <li>'<tt>function args</tt>': argument list whose types match the function
1919 signature argument types. If the function signature indicates the function
1920 accepts a variable number of arguments, the extra arguments can be
1923 <li>'<tt>normal label</tt>': the label reached when the called function
1924 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
1926 <li>'<tt>exception label</tt>': the label reached when a callee returns with
1927 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
1933 <p>This instruction is designed to operate as a standard '<tt><a
1934 href="#i_call">call</a></tt>' instruction in most regards. The primary
1935 difference is that it establishes an association with a label, which is used by
1936 the runtime library to unwind the stack.</p>
1938 <p>This instruction is used in languages with destructors to ensure that proper
1939 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
1940 exception. Additionally, this is important for implementation of
1941 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
1945 %retval = invoke i32 %Test(i32 15) to label %Continue
1946 unwind label %TestCleanup <i>; {i32}:retval set</i>
1947 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Test(i32 15) to label %Continue
1948 unwind label %TestCleanup <i>; {i32}:retval set</i>
1953 <!-- _______________________________________________________________________ -->
1955 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
1956 Instruction</a> </div>
1958 <div class="doc_text">
1967 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
1968 at the first callee in the dynamic call stack which used an <a
1969 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
1970 primarily used to implement exception handling.</p>
1974 <p>The '<tt>unwind</tt>' intrinsic causes execution of the current function to
1975 immediately halt. The dynamic call stack is then searched for the first <a
1976 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
1977 execution continues at the "exceptional" destination block specified by the
1978 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
1979 dynamic call chain, undefined behavior results.</p>
1982 <!-- _______________________________________________________________________ -->
1984 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
1985 Instruction</a> </div>
1987 <div class="doc_text">
1996 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
1997 instruction is used to inform the optimizer that a particular portion of the
1998 code is not reachable. This can be used to indicate that the code after a
1999 no-return function cannot be reached, and other facts.</p>
2003 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
2008 <!-- ======================================================================= -->
2009 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
2010 <div class="doc_text">
2011 <p>Binary operators are used to do most of the computation in a
2012 program. They require two operands, execute an operation on them, and
2013 produce a single value. The operands might represent
2014 multiple data, as is the case with the <a href="#t_vector">vector</a> data type.
2015 The result value of a binary operator is not
2016 necessarily the same type as its operands.</p>
2017 <p>There are several different binary operators:</p>
2019 <!-- _______________________________________________________________________ -->
2020 <div class="doc_subsubsection"> <a name="i_add">'<tt>add</tt>'
2021 Instruction</a> </div>
2022 <div class="doc_text">
2024 <pre> <result> = add <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2027 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
2029 <p>The two arguments to the '<tt>add</tt>' instruction must be either <a
2030 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a> values.
2031 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2032 Both arguments must have identical types.</p>
2034 <p>The value produced is the integer or floating point sum of the two
2037 <pre> <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
2040 <!-- _______________________________________________________________________ -->
2041 <div class="doc_subsubsection"> <a name="i_sub">'<tt>sub</tt>'
2042 Instruction</a> </div>
2043 <div class="doc_text">
2045 <pre> <result> = sub <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2048 <p>The '<tt>sub</tt>' instruction returns the difference of its two
2050 <p>Note that the '<tt>sub</tt>' instruction is used to represent the '<tt>neg</tt>'
2051 instruction present in most other intermediate representations.</p>
2053 <p>The two arguments to the '<tt>sub</tt>' instruction must be either <a
2054 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
2056 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2057 Both arguments must have identical types.</p>
2059 <p>The value produced is the integer or floating point difference of
2060 the two operands.</p>
2063 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
2064 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
2067 <!-- _______________________________________________________________________ -->
2068 <div class="doc_subsubsection"> <a name="i_mul">'<tt>mul</tt>'
2069 Instruction</a> </div>
2070 <div class="doc_text">
2072 <pre> <result> = mul <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2075 <p>The '<tt>mul</tt>' instruction returns the product of its two
2078 <p>The two arguments to the '<tt>mul</tt>' instruction must be either <a
2079 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
2081 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2082 Both arguments must have identical types.</p>
2084 <p>The value produced is the integer or floating point product of the
2086 <p>Because the operands are the same width, the result of an integer
2087 multiplication is the same whether the operands should be deemed unsigned or
2090 <pre> <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
2093 <!-- _______________________________________________________________________ -->
2094 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
2096 <div class="doc_text">
2098 <pre> <result> = udiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2101 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two
2104 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
2105 <a href="#t_integer">integer</a> values. Both arguments must have identical
2106 types. This instruction can also take <a href="#t_vector">vector</a> versions
2107 of the values in which case the elements must be integers.</p>
2109 <p>The value produced is the unsigned integer quotient of the two operands. This
2110 instruction always performs an unsigned division operation, regardless of
2111 whether the arguments are unsigned or not.</p>
2113 <pre> <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2116 <!-- _______________________________________________________________________ -->
2117 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
2119 <div class="doc_text">
2121 <pre> <result> = sdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2124 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
2127 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
2128 <a href="#t_integer">integer</a> values. Both arguments must have identical
2129 types. This instruction can also take <a href="#t_vector">vector</a> versions
2130 of the values in which case the elements must be integers.</p>
2132 <p>The value produced is the signed integer quotient of the two operands. This
2133 instruction always performs a signed division operation, regardless of whether
2134 the arguments are signed or not.</p>
2136 <pre> <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2139 <!-- _______________________________________________________________________ -->
2140 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
2141 Instruction</a> </div>
2142 <div class="doc_text">
2144 <pre> <result> = fdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2147 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
2150 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
2151 <a href="#t_floating">floating point</a> values. Both arguments must have
2152 identical types. This instruction can also take <a href="#t_vector">vector</a>
2153 versions of floating point values.</p>
2155 <p>The value produced is the floating point quotient of the two operands.</p>
2157 <pre> <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
2160 <!-- _______________________________________________________________________ -->
2161 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
2163 <div class="doc_text">
2165 <pre> <result> = urem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2168 <p>The '<tt>urem</tt>' instruction returns the remainder from the
2169 unsigned division of its two arguments.</p>
2171 <p>The two arguments to the '<tt>urem</tt>' instruction must be
2172 <a href="#t_integer">integer</a> values. Both arguments must have identical
2173 types. This instruction can also take <a href="#t_vector">vector</a> versions
2174 of the values in which case the elements must be integers.</p>
2176 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
2177 This instruction always performs an unsigned division to get the remainder,
2178 regardless of whether the arguments are unsigned or not.</p>
2180 <pre> <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2184 <!-- _______________________________________________________________________ -->
2185 <div class="doc_subsubsection"> <a name="i_srem">'<tt>srem</tt>'
2186 Instruction</a> </div>
2187 <div class="doc_text">
2189 <pre> <result> = srem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2192 <p>The '<tt>srem</tt>' instruction returns the remainder from the
2193 signed division of its two operands. This instruction can also take
2194 <a href="#t_vector">vector</a> versions of the values in which case
2195 the elements must be integers.</p>
2198 <p>The two arguments to the '<tt>srem</tt>' instruction must be
2199 <a href="#t_integer">integer</a> values. Both arguments must have identical
2202 <p>This instruction returns the <i>remainder</i> of a division (where the result
2203 has the same sign as the dividend, <tt>var1</tt>), not the <i>modulo</i>
2204 operator (where the result has the same sign as the divisor, <tt>var2</tt>) of
2205 a value. For more information about the difference, see <a
2206 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
2207 Math Forum</a>. For a table of how this is implemented in various languages,
2208 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
2209 Wikipedia: modulo operation</a>.</p>
2211 <pre> <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2215 <!-- _______________________________________________________________________ -->
2216 <div class="doc_subsubsection"> <a name="i_frem">'<tt>frem</tt>'
2217 Instruction</a> </div>
2218 <div class="doc_text">
2220 <pre> <result> = frem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2223 <p>The '<tt>frem</tt>' instruction returns the remainder from the
2224 division of its two operands.</p>
2226 <p>The two arguments to the '<tt>frem</tt>' instruction must be
2227 <a href="#t_floating">floating point</a> values. Both arguments must have
2228 identical types. This instruction can also take <a href="#t_vector">vector</a>
2229 versions of floating point values.</p>
2231 <p>This instruction returns the <i>remainder</i> of a division.</p>
2233 <pre> <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
2237 <!-- ======================================================================= -->
2238 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
2239 Operations</a> </div>
2240 <div class="doc_text">
2241 <p>Bitwise binary operators are used to do various forms of
2242 bit-twiddling in a program. They are generally very efficient
2243 instructions and can commonly be strength reduced from other
2244 instructions. They require two operands, execute an operation on them,
2245 and produce a single value. The resulting value of the bitwise binary
2246 operators is always the same type as its first operand.</p>
2249 <!-- _______________________________________________________________________ -->
2250 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
2251 Instruction</a> </div>
2252 <div class="doc_text">
2254 <pre> <result> = shl <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2259 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2260 the left a specified number of bits.</p>
2264 <p>Both arguments to the '<tt>shl</tt>' instruction must be the same <a
2265 href="#t_integer">integer</a> type.</p>
2269 <p>The value produced is <tt>var1</tt> * 2<sup><tt>var2</tt></sup>. If
2270 <tt>var2</tt> is (statically or dynamically) equal to or larger than the number
2271 of bits in <tt>var1</tt>, the result is undefined.</p>
2273 <h5>Example:</h5><pre>
2274 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
2275 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
2276 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
2277 <result> = shl i32 1, 32 <i>; undefined</i>
2280 <!-- _______________________________________________________________________ -->
2281 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
2282 Instruction</a> </div>
2283 <div class="doc_text">
2285 <pre> <result> = lshr <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2289 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
2290 operand shifted to the right a specified number of bits with zero fill.</p>
2293 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
2294 <a href="#t_integer">integer</a> type.</p>
2298 <p>This instruction always performs a logical shift right operation. The most
2299 significant bits of the result will be filled with zero bits after the
2300 shift. If <tt>var2</tt> is (statically or dynamically) equal to or larger than
2301 the number of bits in <tt>var1</tt>, the result is undefined.</p>
2305 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
2306 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
2307 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
2308 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
2309 <result> = lshr i32 1, 32 <i>; undefined</i>
2313 <!-- _______________________________________________________________________ -->
2314 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
2315 Instruction</a> </div>
2316 <div class="doc_text">
2319 <pre> <result> = ashr <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2323 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
2324 operand shifted to the right a specified number of bits with sign extension.</p>
2327 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
2328 <a href="#t_integer">integer</a> type.</p>
2331 <p>This instruction always performs an arithmetic shift right operation,
2332 The most significant bits of the result will be filled with the sign bit
2333 of <tt>var1</tt>. If <tt>var2</tt> is (statically or dynamically) equal to or
2334 larger than the number of bits in <tt>var1</tt>, the result is undefined.
2339 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
2340 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
2341 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
2342 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
2343 <result> = ashr i32 1, 32 <i>; undefined</i>
2347 <!-- _______________________________________________________________________ -->
2348 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
2349 Instruction</a> </div>
2350 <div class="doc_text">
2352 <pre> <result> = and <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2355 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
2356 its two operands.</p>
2358 <p>The two arguments to the '<tt>and</tt>' instruction must be <a
2359 href="#t_integer">integer</a> values. Both arguments must have
2360 identical types.</p>
2362 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
2364 <div style="align: center">
2365 <table border="1" cellspacing="0" cellpadding="4">
2396 <pre> <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
2397 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
2398 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
2401 <!-- _______________________________________________________________________ -->
2402 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
2403 <div class="doc_text">
2405 <pre> <result> = or <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2408 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
2409 or of its two operands.</p>
2411 <p>The two arguments to the '<tt>or</tt>' instruction must be <a
2412 href="#t_integer">integer</a> values. Both arguments must have
2413 identical types.</p>
2415 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
2417 <div style="align: center">
2418 <table border="1" cellspacing="0" cellpadding="4">
2449 <pre> <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
2450 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
2451 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
2454 <!-- _______________________________________________________________________ -->
2455 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
2456 Instruction</a> </div>
2457 <div class="doc_text">
2459 <pre> <result> = xor <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2462 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
2463 or of its two operands. The <tt>xor</tt> is used to implement the
2464 "one's complement" operation, which is the "~" operator in C.</p>
2466 <p>The two arguments to the '<tt>xor</tt>' instruction must be <a
2467 href="#t_integer">integer</a> values. Both arguments must have
2468 identical types.</p>
2470 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
2472 <div style="align: center">
2473 <table border="1" cellspacing="0" cellpadding="4">
2505 <pre> <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
2506 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
2507 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
2508 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
2512 <!-- ======================================================================= -->
2513 <div class="doc_subsection">
2514 <a name="vectorops">Vector Operations</a>
2517 <div class="doc_text">
2519 <p>LLVM supports several instructions to represent vector operations in a
2520 target-independent manner. These instructions cover the element-access and
2521 vector-specific operations needed to process vectors effectively. While LLVM
2522 does directly support these vector operations, many sophisticated algorithms
2523 will want to use target-specific intrinsics to take full advantage of a specific
2528 <!-- _______________________________________________________________________ -->
2529 <div class="doc_subsubsection">
2530 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2533 <div class="doc_text">
2538 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
2544 The '<tt>extractelement</tt>' instruction extracts a single scalar
2545 element from a vector at a specified index.
2552 The first operand of an '<tt>extractelement</tt>' instruction is a
2553 value of <a href="#t_vector">vector</a> type. The second operand is
2554 an index indicating the position from which to extract the element.
2555 The index may be a variable.</p>
2560 The result is a scalar of the same type as the element type of
2561 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2562 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2563 results are undefined.
2569 %result = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
2574 <!-- _______________________________________________________________________ -->
2575 <div class="doc_subsubsection">
2576 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
2579 <div class="doc_text">
2584 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
2590 The '<tt>insertelement</tt>' instruction inserts a scalar
2591 element into a vector at a specified index.
2598 The first operand of an '<tt>insertelement</tt>' instruction is a
2599 value of <a href="#t_vector">vector</a> type. The second operand is a
2600 scalar value whose type must equal the element type of the first
2601 operand. The third operand is an index indicating the position at
2602 which to insert the value. The index may be a variable.</p>
2607 The result is a vector of the same type as <tt>val</tt>. Its
2608 element values are those of <tt>val</tt> except at position
2609 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
2610 exceeds the length of <tt>val</tt>, the results are undefined.
2616 %result = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
2620 <!-- _______________________________________________________________________ -->
2621 <div class="doc_subsubsection">
2622 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
2625 <div class="doc_text">
2630 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <n x i32> <mask> <i>; yields <n x <ty>></i>
2636 The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
2637 from two input vectors, returning a vector of the same type.
2643 The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
2644 with types that match each other and types that match the result of the
2645 instruction. The third argument is a shuffle mask, which has the same number
2646 of elements as the other vector type, but whose element type is always 'i32'.
2650 The shuffle mask operand is required to be a constant vector with either
2651 constant integer or undef values.
2657 The elements of the two input vectors are numbered from left to right across
2658 both of the vectors. The shuffle mask operand specifies, for each element of
2659 the result vector, which element of the two input registers the result element
2660 gets. The element selector may be undef (meaning "don't care") and the second
2661 operand may be undef if performing a shuffle from only one vector.
2667 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
2668 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
2669 %result = shufflevector <4 x i32> %v1, <4 x i32> undef,
2670 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
2675 <!-- ======================================================================= -->
2676 <div class="doc_subsection">
2677 <a name="memoryops">Memory Access and Addressing Operations</a>
2680 <div class="doc_text">
2682 <p>A key design point of an SSA-based representation is how it
2683 represents memory. In LLVM, no memory locations are in SSA form, which
2684 makes things very simple. This section describes how to read, write,
2685 allocate, and free memory in LLVM.</p>
2689 <!-- _______________________________________________________________________ -->
2690 <div class="doc_subsubsection">
2691 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
2694 <div class="doc_text">
2699 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2704 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
2705 heap and returns a pointer to it. The object is always allocated in the generic
2706 address space (address space zero).</p>
2710 <p>The '<tt>malloc</tt>' instruction allocates
2711 <tt>sizeof(<type>)*NumElements</tt>
2712 bytes of memory from the operating system and returns a pointer of the
2713 appropriate type to the program. If "NumElements" is specified, it is the
2714 number of elements allocated. If an alignment is specified, the value result
2715 of the allocation is guaranteed to be aligned to at least that boundary. If
2716 not specified, or if zero, the target can choose to align the allocation on any
2717 convenient boundary.</p>
2719 <p>'<tt>type</tt>' must be a sized type.</p>
2723 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
2724 a pointer is returned.</p>
2729 %array = malloc [4 x i8 ] <i>; yields {[%4 x i8]*}:array</i>
2731 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
2732 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
2733 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
2734 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
2735 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
2739 <!-- _______________________________________________________________________ -->
2740 <div class="doc_subsubsection">
2741 <a name="i_free">'<tt>free</tt>' Instruction</a>
2744 <div class="doc_text">
2749 free <type> <value> <i>; yields {void}</i>
2754 <p>The '<tt>free</tt>' instruction returns memory back to the unused
2755 memory heap to be reallocated in the future.</p>
2759 <p>'<tt>value</tt>' shall be a pointer value that points to a value
2760 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
2765 <p>Access to the memory pointed to by the pointer is no longer defined
2766 after this instruction executes.</p>
2771 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
2772 free [4 x i8]* %array
2776 <!-- _______________________________________________________________________ -->
2777 <div class="doc_subsubsection">
2778 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
2781 <div class="doc_text">
2786 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2791 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
2792 currently executing function, to be automatically released when this function
2793 returns to its caller. The object is always allocated in the generic address
2794 space (address space zero).</p>
2798 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
2799 bytes of memory on the runtime stack, returning a pointer of the
2800 appropriate type to the program. If "NumElements" is specified, it is the
2801 number of elements allocated. If an alignment is specified, the value result
2802 of the allocation is guaranteed to be aligned to at least that boundary. If
2803 not specified, or if zero, the target can choose to align the allocation on any
2804 convenient boundary.</p>
2806 <p>'<tt>type</tt>' may be any sized type.</p>
2810 <p>Memory is allocated; a pointer is returned. '<tt>alloca</tt>'d
2811 memory is automatically released when the function returns. The '<tt>alloca</tt>'
2812 instruction is commonly used to represent automatic variables that must
2813 have an address available. When the function returns (either with the <tt><a
2814 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
2815 instructions), the memory is reclaimed.</p>
2820 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
2821 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
2822 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
2823 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
2827 <!-- _______________________________________________________________________ -->
2828 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
2829 Instruction</a> </div>
2830 <div class="doc_text">
2832 <pre> <result> = load <ty>* <pointer>[, align <alignment>]<br> <result> = volatile load <ty>* <pointer>[, align <alignment>]<br></pre>
2834 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
2836 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
2837 address from which to load. The pointer must point to a <a
2838 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
2839 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
2840 the number or order of execution of this <tt>load</tt> with other
2841 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
2844 <p>The location of memory pointed to is loaded.</p>
2846 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
2848 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
2849 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
2852 <!-- _______________________________________________________________________ -->
2853 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
2854 Instruction</a> </div>
2855 <div class="doc_text">
2857 <pre> store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
2858 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
2861 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
2863 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
2864 to store and an address at which to store it. The type of the '<tt><pointer></tt>'
2865 operand must be a pointer to the type of the '<tt><value></tt>'
2866 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
2867 optimizer is not allowed to modify the number or order of execution of
2868 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
2869 href="#i_store">store</a></tt> instructions.</p>
2871 <p>The contents of memory are updated to contain '<tt><value></tt>'
2872 at the location specified by the '<tt><pointer></tt>' operand.</p>
2874 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
2875 store i32 3, i32* %ptr <i>; yields {void}</i>
2876 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
2880 <!-- _______________________________________________________________________ -->
2881 <div class="doc_subsubsection">
2882 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
2885 <div class="doc_text">
2888 <result> = getelementptr <ty>* <ptrval>{, <ty> <idx>}*
2894 The '<tt>getelementptr</tt>' instruction is used to get the address of a
2895 subelement of an aggregate data structure.</p>
2899 <p>This instruction takes a list of integer operands that indicate what
2900 elements of the aggregate object to index to. The actual types of the arguments
2901 provided depend on the type of the first pointer argument. The
2902 '<tt>getelementptr</tt>' instruction is used to index down through the type
2903 levels of a structure or to a specific index in an array. When indexing into a
2904 structure, only <tt>i32</tt> integer constants are allowed. When indexing
2905 into an array or pointer, only integers of 32 or 64 bits are allowed, and will
2906 be sign extended to 64-bit values.</p>
2908 <p>For example, let's consider a C code fragment and how it gets
2909 compiled to LLVM:</p>
2911 <div class="doc_code">
2924 int *foo(struct ST *s) {
2925 return &s[1].Z.B[5][13];
2930 <p>The LLVM code generated by the GCC frontend is:</p>
2932 <div class="doc_code">
2934 %RT = type { i8 , [10 x [20 x i32]], i8 }
2935 %ST = type { i32, double, %RT }
2937 define i32* %foo(%ST* %s) {
2939 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
2947 <p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
2948 on the pointer type that is being indexed into. <a href="#t_pointer">Pointer</a>
2949 and <a href="#t_array">array</a> types can use a 32-bit or 64-bit
2950 <a href="#t_integer">integer</a> type but the value will always be sign extended
2951 to 64-bits. <a href="#t_struct">Structure</a> types require <tt>i32</tt>
2952 <b>constants</b>.</p>
2954 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
2955 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
2956 }</tt>' type, a structure. The second index indexes into the third element of
2957 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
2958 i8 }</tt>' type, another structure. The third index indexes into the second
2959 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
2960 array. The two dimensions of the array are subscripted into, yielding an
2961 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
2962 to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
2964 <p>Note that it is perfectly legal to index partially through a
2965 structure, returning a pointer to an inner element. Because of this,
2966 the LLVM code for the given testcase is equivalent to:</p>
2969 define i32* %foo(%ST* %s) {
2970 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
2971 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
2972 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
2973 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
2974 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
2979 <p>Note that it is undefined to access an array out of bounds: array and
2980 pointer indexes must always be within the defined bounds of the array type.
2981 The one exception for this rules is zero length arrays. These arrays are
2982 defined to be accessible as variable length arrays, which requires access
2983 beyond the zero'th element.</p>
2985 <p>The getelementptr instruction is often confusing. For some more insight
2986 into how it works, see <a href="GetElementPtr.html">the getelementptr
2992 <i>; yields [12 x i8]*:aptr</i>
2993 %aptr = getelementptr {i32, [12 x i8]}* %sptr, i64 0, i32 1
2997 <!-- ======================================================================= -->
2998 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
3000 <div class="doc_text">
3001 <p>The instructions in this category are the conversion instructions (casting)
3002 which all take a single operand and a type. They perform various bit conversions
3006 <!-- _______________________________________________________________________ -->
3007 <div class="doc_subsubsection">
3008 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
3010 <div class="doc_text">
3014 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
3019 The '<tt>trunc</tt>' instruction truncates its operand to the type <tt>ty2</tt>.
3024 The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
3025 be an <a href="#t_integer">integer</a> type, and a type that specifies the size
3026 and type of the result, which must be an <a href="#t_integer">integer</a>
3027 type. The bit size of <tt>value</tt> must be larger than the bit size of
3028 <tt>ty2</tt>. Equal sized types are not allowed.</p>
3032 The '<tt>trunc</tt>' instruction truncates the high order bits in <tt>value</tt>
3033 and converts the remaining bits to <tt>ty2</tt>. Since the source size must be
3034 larger than the destination size, <tt>trunc</tt> cannot be a <i>no-op cast</i>.
3035 It will always truncate bits.</p>
3039 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
3040 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
3041 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
3045 <!-- _______________________________________________________________________ -->
3046 <div class="doc_subsubsection">
3047 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
3049 <div class="doc_text">
3053 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
3057 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
3062 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
3063 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3064 also be of <a href="#t_integer">integer</a> type. The bit size of the
3065 <tt>value</tt> must be smaller than the bit size of the destination type,
3069 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
3070 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
3072 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
3076 %X = zext i32 257 to i64 <i>; yields i64:257</i>
3077 %Y = zext i1 true to i32 <i>; yields i32:1</i>
3081 <!-- _______________________________________________________________________ -->
3082 <div class="doc_subsubsection">
3083 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
3085 <div class="doc_text">
3089 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
3093 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
3097 The '<tt>sext</tt>' instruction takes a value to cast, which must be of
3098 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3099 also be of <a href="#t_integer">integer</a> type. The bit size of the
3100 <tt>value</tt> must be smaller than the bit size of the destination type,
3105 The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
3106 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size of
3107 the type <tt>ty2</tt>.</p>
3109 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
3113 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
3114 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
3118 <!-- _______________________________________________________________________ -->
3119 <div class="doc_subsubsection">
3120 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
3123 <div class="doc_text">
3128 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
3132 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
3137 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
3138 point</a> value to cast and a <a href="#t_floating">floating point</a> type to
3139 cast it to. The size of <tt>value</tt> must be larger than the size of
3140 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
3141 <i>no-op cast</i>.</p>
3144 <p> The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
3145 <a href="#t_floating">floating point</a> type to a smaller
3146 <a href="#t_floating">floating point</a> type. If the value cannot fit within
3147 the destination type, <tt>ty2</tt>, then the results are undefined.</p>
3151 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
3152 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
3156 <!-- _______________________________________________________________________ -->
3157 <div class="doc_subsubsection">
3158 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
3160 <div class="doc_text">
3164 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
3168 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
3169 floating point value.</p>
3172 <p>The '<tt>fpext</tt>' instruction takes a
3173 <a href="#t_floating">floating point</a> <tt>value</tt> to cast,
3174 and a <a href="#t_floating">floating point</a> type to cast it to. The source
3175 type must be smaller than the destination type.</p>
3178 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
3179 <a href="#t_floating">floating point</a> type to a larger
3180 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
3181 used to make a <i>no-op cast</i> because it always changes bits. Use
3182 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
3186 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
3187 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
3191 <!-- _______________________________________________________________________ -->
3192 <div class="doc_subsubsection">
3193 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
3195 <div class="doc_text">
3199 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
3203 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
3204 unsigned integer equivalent of type <tt>ty2</tt>.
3208 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
3209 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3210 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3211 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3212 vector integer type with the same number of elements as <tt>ty</tt></p>
3215 <p> The '<tt>fptoui</tt>' instruction converts its
3216 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3217 towards zero) unsigned integer value. If the value cannot fit in <tt>ty2</tt>,
3218 the results are undefined.</p>
3222 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
3223 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
3224 %X = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
3228 <!-- _______________________________________________________________________ -->
3229 <div class="doc_subsubsection">
3230 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
3232 <div class="doc_text">
3236 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
3240 <p>The '<tt>fptosi</tt>' instruction converts
3241 <a href="#t_floating">floating point</a> <tt>value</tt> to type <tt>ty2</tt>.
3245 <p> The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
3246 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3247 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3248 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3249 vector integer type with the same number of elements as <tt>ty</tt></p>
3252 <p>The '<tt>fptosi</tt>' instruction converts its
3253 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3254 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
3255 the results are undefined.</p>
3259 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
3260 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
3261 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
3265 <!-- _______________________________________________________________________ -->
3266 <div class="doc_subsubsection">
3267 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
3269 <div class="doc_text">
3273 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3277 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
3278 integer and converts that value to the <tt>ty2</tt> type.</p>
3281 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
3282 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3283 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3284 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3285 floating point type with the same number of elements as <tt>ty</tt></p>
3288 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
3289 integer quantity and converts it to the corresponding floating point value. If
3290 the value cannot fit in the floating point value, the results are undefined.</p>
3294 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
3295 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
3299 <!-- _______________________________________________________________________ -->
3300 <div class="doc_subsubsection">
3301 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
3303 <div class="doc_text">
3307 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3311 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed
3312 integer and converts that value to the <tt>ty2</tt> type.</p>
3315 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
3316 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3317 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3318 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3319 floating point type with the same number of elements as <tt>ty</tt></p>
3322 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed
3323 integer quantity and converts it to the corresponding floating point value. If
3324 the value cannot fit in the floating point value, the results are undefined.</p>
3328 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
3329 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
3333 <!-- _______________________________________________________________________ -->
3334 <div class="doc_subsubsection">
3335 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
3337 <div class="doc_text">
3341 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
3345 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
3346 the integer type <tt>ty2</tt>.</p>
3349 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
3350 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
3351 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.
3354 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
3355 <tt>ty2</tt> by interpreting the pointer value as an integer and either
3356 truncating or zero extending that value to the size of the integer type. If
3357 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
3358 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
3359 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
3364 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
3365 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
3369 <!-- _______________________________________________________________________ -->
3370 <div class="doc_subsubsection">
3371 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
3373 <div class="doc_text">
3377 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
3381 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to
3382 a pointer type, <tt>ty2</tt>.</p>
3385 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
3386 value to cast, and a type to cast it to, which must be a
3387 <a href="#t_pointer">pointer</a> type.
3390 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
3391 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
3392 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
3393 size of a pointer then a truncation is done. If <tt>value</tt> is smaller than
3394 the size of a pointer then a zero extension is done. If they are the same size,
3395 nothing is done (<i>no-op cast</i>).</p>
3399 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
3400 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
3401 %Y = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
3405 <!-- _______________________________________________________________________ -->
3406 <div class="doc_subsubsection">
3407 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
3409 <div class="doc_text">
3413 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
3417 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3418 <tt>ty2</tt> without changing any bits.</p>
3421 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be
3422 a first class value, and a type to cast it to, which must also be a <a
3423 href="#t_firstclass">first class</a> type. The bit sizes of <tt>value</tt>
3424 and the destination type, <tt>ty2</tt>, must be identical. If the source
3425 type is a pointer, the destination type must also be a pointer.</p>
3428 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3429 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
3430 this conversion. The conversion is done as if the <tt>value</tt> had been
3431 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only be
3432 converted to other pointer types with this instruction. To convert pointers to
3433 other types, use the <a href="#i_inttoptr">inttoptr</a> or
3434 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
3438 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
3439 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
3440 %Z = bitcast <2xint> %V to i64; <i>; yields i64: %V</i>
3444 <!-- ======================================================================= -->
3445 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
3446 <div class="doc_text">
3447 <p>The instructions in this category are the "miscellaneous"
3448 instructions, which defy better classification.</p>
3451 <!-- _______________________________________________________________________ -->
3452 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
3454 <div class="doc_text">
3456 <pre> <result> = icmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3459 <p>The '<tt>icmp</tt>' instruction returns a boolean value based on comparison
3460 of its two integer operands.</p>
3462 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
3463 the condition code indicating the kind of comparison to perform. It is not
3464 a value, just a keyword. The possible condition code are:
3466 <li><tt>eq</tt>: equal</li>
3467 <li><tt>ne</tt>: not equal </li>
3468 <li><tt>ugt</tt>: unsigned greater than</li>
3469 <li><tt>uge</tt>: unsigned greater or equal</li>
3470 <li><tt>ult</tt>: unsigned less than</li>
3471 <li><tt>ule</tt>: unsigned less or equal</li>
3472 <li><tt>sgt</tt>: signed greater than</li>
3473 <li><tt>sge</tt>: signed greater or equal</li>
3474 <li><tt>slt</tt>: signed less than</li>
3475 <li><tt>sle</tt>: signed less or equal</li>
3477 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
3478 <a href="#t_pointer">pointer</a> typed. They must also be identical types.</p>
3480 <p>The '<tt>icmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3481 the condition code given as <tt>cond</tt>. The comparison performed always
3482 yields a <a href="#t_primitive">i1</a> result, as follows:
3484 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
3485 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3487 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
3488 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3489 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
3490 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3491 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
3492 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3493 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
3494 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3495 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
3496 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3497 <li><tt>sgt</tt>: interprets the operands as signed values and yields
3498 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3499 <li><tt>sge</tt>: interprets the operands as signed values and yields
3500 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3501 <li><tt>slt</tt>: interprets the operands as signed values and yields
3502 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3503 <li><tt>sle</tt>: interprets the operands as signed values and yields
3504 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3506 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
3507 values are compared as if they were integers.</p>
3510 <pre> <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
3511 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
3512 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
3513 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
3514 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
3515 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
3519 <!-- _______________________________________________________________________ -->
3520 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
3522 <div class="doc_text">
3524 <pre> <result> = fcmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3527 <p>The '<tt>fcmp</tt>' instruction returns a boolean value based on comparison
3528 of its floating point operands.</p>
3530 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
3531 the condition code indicating the kind of comparison to perform. It is not
3532 a value, just a keyword. The possible condition code are:
3534 <li><tt>false</tt>: no comparison, always returns false</li>
3535 <li><tt>oeq</tt>: ordered and equal</li>
3536 <li><tt>ogt</tt>: ordered and greater than </li>
3537 <li><tt>oge</tt>: ordered and greater than or equal</li>
3538 <li><tt>olt</tt>: ordered and less than </li>
3539 <li><tt>ole</tt>: ordered and less than or equal</li>
3540 <li><tt>one</tt>: ordered and not equal</li>
3541 <li><tt>ord</tt>: ordered (no nans)</li>
3542 <li><tt>ueq</tt>: unordered or equal</li>
3543 <li><tt>ugt</tt>: unordered or greater than </li>
3544 <li><tt>uge</tt>: unordered or greater than or equal</li>
3545 <li><tt>ult</tt>: unordered or less than </li>
3546 <li><tt>ule</tt>: unordered or less than or equal</li>
3547 <li><tt>une</tt>: unordered or not equal</li>
3548 <li><tt>uno</tt>: unordered (either nans)</li>
3549 <li><tt>true</tt>: no comparison, always returns true</li>
3551 <p><i>Ordered</i> means that neither operand is a QNAN while
3552 <i>unordered</i> means that either operand may be a QNAN.</p>
3553 <p>The <tt>val1</tt> and <tt>val2</tt> arguments must be
3554 <a href="#t_floating">floating point</a> typed. They must have identical
3557 <p>The '<tt>fcmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3558 the condition code given as <tt>cond</tt>. The comparison performed always
3559 yields a <a href="#t_primitive">i1</a> result, as follows:
3561 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
3562 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3563 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3564 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3565 <tt>var1</tt> is greather than <tt>var2</tt>.</li>
3566 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3567 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3568 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3569 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3570 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3571 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3572 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3573 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3574 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
3575 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
3576 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3577 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
3578 <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3579 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
3580 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3581 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
3582 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3583 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
3584 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3585 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
3586 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3587 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
3588 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
3592 <pre> <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
3593 <result> = icmp one float 4.0, 5.0 <i>; yields: result=true</i>
3594 <result> = icmp olt float 4.0, 5.0 <i>; yields: result=true</i>
3595 <result> = icmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
3599 <!-- _______________________________________________________________________ -->
3600 <div class="doc_subsubsection"> <a name="i_phi">'<tt>phi</tt>'
3601 Instruction</a> </div>
3602 <div class="doc_text">
3604 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
3606 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
3607 the SSA graph representing the function.</p>
3609 <p>The type of the incoming values is specified with the first type
3610 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
3611 as arguments, with one pair for each predecessor basic block of the
3612 current block. Only values of <a href="#t_firstclass">first class</a>
3613 type may be used as the value arguments to the PHI node. Only labels
3614 may be used as the label arguments.</p>
3615 <p>There must be no non-phi instructions between the start of a basic
3616 block and the PHI instructions: i.e. PHI instructions must be first in
3619 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
3620 specified by the pair corresponding to the predecessor basic block that executed
3621 just prior to the current block.</p>
3623 <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>
3626 <!-- _______________________________________________________________________ -->
3627 <div class="doc_subsubsection">
3628 <a name="i_select">'<tt>select</tt>' Instruction</a>
3631 <div class="doc_text">
3636 <result> = select i1 <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
3642 The '<tt>select</tt>' instruction is used to choose one value based on a
3643 condition, without branching.
3650 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.
3656 If the boolean condition evaluates to true, the instruction returns the first
3657 value argument; otherwise, it returns the second value argument.
3663 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
3668 <!-- _______________________________________________________________________ -->
3669 <div class="doc_subsubsection">
3670 <a name="i_call">'<tt>call</tt>' Instruction</a>
3673 <div class="doc_text">
3677 <result> = [tail] call [<a href="#callingconv">cconv</a>] <ty> [<fnty>*] <fnptrval>(<param list>)
3682 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
3686 <p>This instruction requires several arguments:</p>
3690 <p>The optional "tail" marker indicates whether the callee function accesses
3691 any allocas or varargs in the caller. If the "tail" marker is present, the
3692 function call is eligible for tail call optimization. Note that calls may
3693 be marked "tail" even if they do not occur before a <a
3694 href="#i_ret"><tt>ret</tt></a> instruction.
3697 <p>The optional "cconv" marker indicates which <a href="#callingconv">calling
3698 convention</a> the call should use. If none is specified, the call defaults
3699 to using C calling conventions.
3702 <p>'<tt>ty</tt>': the type of the call instruction itself which is also
3703 the type of the return value. Functions that return no value are marked
3704 <tt><a href="#t_void">void</a></tt>.</p>
3707 <p>'<tt>fnty</tt>': shall be the signature of the pointer to function
3708 value being invoked. The argument types must match the types implied by
3709 this signature. This type can be omitted if the function is not varargs
3710 and if the function type does not return a pointer to a function.</p>
3713 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
3714 be invoked. In most cases, this is a direct function invocation, but
3715 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
3716 to function value.</p>
3719 <p>'<tt>function args</tt>': argument list whose types match the
3720 function signature argument types. All arguments must be of
3721 <a href="#t_firstclass">first class</a> type. If the function signature
3722 indicates the function accepts a variable number of arguments, the extra
3723 arguments can be specified.</p>
3729 <p>The '<tt>call</tt>' instruction is used to cause control flow to
3730 transfer to a specified function, with its incoming arguments bound to
3731 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
3732 instruction in the called function, control flow continues with the
3733 instruction after the function call, and the return value of the
3734 function is bound to the result argument. This is a simpler case of
3735 the <a href="#i_invoke">invoke</a> instruction.</p>
3740 %retval = call i32 @test(i32 %argc)
3741 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42);
3742 %X = tail call i32 @foo()
3743 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo()
3744 %Z = call void %foo(i8 97 signext)
3749 <!-- _______________________________________________________________________ -->
3750 <div class="doc_subsubsection">
3751 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
3754 <div class="doc_text">
3759 <resultval> = va_arg <va_list*> <arglist>, <argty>
3764 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
3765 the "variable argument" area of a function call. It is used to implement the
3766 <tt>va_arg</tt> macro in C.</p>
3770 <p>This instruction takes a <tt>va_list*</tt> value and the type of
3771 the argument. It returns a value of the specified argument type and
3772 increments the <tt>va_list</tt> to point to the next argument. The
3773 actual type of <tt>va_list</tt> is target specific.</p>
3777 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
3778 type from the specified <tt>va_list</tt> and causes the
3779 <tt>va_list</tt> to point to the next argument. For more information,
3780 see the variable argument handling <a href="#int_varargs">Intrinsic
3783 <p>It is legal for this instruction to be called in a function which does not
3784 take a variable number of arguments, for example, the <tt>vfprintf</tt>
3787 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
3788 href="#intrinsics">intrinsic function</a> because it takes a type as an
3793 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
3797 <!-- *********************************************************************** -->
3798 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
3799 <!-- *********************************************************************** -->
3801 <div class="doc_text">
3803 <p>LLVM supports the notion of an "intrinsic function". These functions have
3804 well known names and semantics and are required to follow certain restrictions.
3805 Overall, these intrinsics represent an extension mechanism for the LLVM
3806 language that does not require changing all of the transformations in LLVM when
3807 adding to the language (or the bitcode reader/writer, the parser, etc...).</p>
3809 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
3810 prefix is reserved in LLVM for intrinsic names; thus, function names may not
3811 begin with this prefix. Intrinsic functions must always be external functions:
3812 you cannot define the body of intrinsic functions. Intrinsic functions may
3813 only be used in call or invoke instructions: it is illegal to take the address
3814 of an intrinsic function. Additionally, because intrinsic functions are part
3815 of the LLVM language, it is required if any are added that they be documented
3818 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
3819 a family of functions that perform the same operation but on different data
3820 types. Because LLVM can represent over 8 million different integer types,
3821 overloading is used commonly to allow an intrinsic function to operate on any
3822 integer type. One or more of the argument types or the result type can be
3823 overloaded to accept any integer type. Argument types may also be defined as
3824 exactly matching a previous argument's type or the result type. This allows an
3825 intrinsic function which accepts multiple arguments, but needs all of them to
3826 be of the same type, to only be overloaded with respect to a single argument or
3829 <p>Overloaded intrinsics will have the names of its overloaded argument types
3830 encoded into its function name, each preceded by a period. Only those types
3831 which are overloaded result in a name suffix. Arguments whose type is matched
3832 against another type do not. For example, the <tt>llvm.ctpop</tt> function can
3833 take an integer of any width and returns an integer of exactly the same integer
3834 width. This leads to a family of functions such as
3835 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29 %val)</tt>.
3836 Only one type, the return type, is overloaded, and only one type suffix is
3837 required. Because the argument's type is matched against the return type, it
3838 does not require its own name suffix.</p>
3840 <p>To learn how to add an intrinsic function, please see the
3841 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.
3846 <!-- ======================================================================= -->
3847 <div class="doc_subsection">
3848 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
3851 <div class="doc_text">
3853 <p>Variable argument support is defined in LLVM with the <a
3854 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
3855 intrinsic functions. These functions are related to the similarly
3856 named macros defined in the <tt><stdarg.h></tt> header file.</p>
3858 <p>All of these functions operate on arguments that use a
3859 target-specific value type "<tt>va_list</tt>". The LLVM assembly
3860 language reference manual does not define what this type is, so all
3861 transformations should be prepared to handle these functions regardless of
3864 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
3865 instruction and the variable argument handling intrinsic functions are
3868 <div class="doc_code">
3870 define i32 @test(i32 %X, ...) {
3871 ; Initialize variable argument processing
3873 %ap2 = bitcast i8** %ap to i8*
3874 call void @llvm.va_start(i8* %ap2)
3876 ; Read a single integer argument
3877 %tmp = va_arg i8** %ap, i32
3879 ; Demonstrate usage of llvm.va_copy and llvm.va_end
3881 %aq2 = bitcast i8** %aq to i8*
3882 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
3883 call void @llvm.va_end(i8* %aq2)
3885 ; Stop processing of arguments.
3886 call void @llvm.va_end(i8* %ap2)
3890 declare void @llvm.va_start(i8*)
3891 declare void @llvm.va_copy(i8*, i8*)
3892 declare void @llvm.va_end(i8*)
3898 <!-- _______________________________________________________________________ -->
3899 <div class="doc_subsubsection">
3900 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
3904 <div class="doc_text">
3906 <pre> declare void %llvm.va_start(i8* <arglist>)<br></pre>
3908 <P>The '<tt>llvm.va_start</tt>' intrinsic initializes
3909 <tt>*<arglist></tt> for subsequent use by <tt><a
3910 href="#i_va_arg">va_arg</a></tt>.</p>
3914 <P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
3918 <P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
3919 macro available in C. In a target-dependent way, it initializes the
3920 <tt>va_list</tt> element to which the argument points, so that the next call to
3921 <tt>va_arg</tt> will produce the first variable argument passed to the function.
3922 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
3923 last argument of the function as the compiler can figure that out.</p>
3927 <!-- _______________________________________________________________________ -->
3928 <div class="doc_subsubsection">
3929 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
3932 <div class="doc_text">
3934 <pre> declare void @llvm.va_end(i8* <arglist>)<br></pre>
3937 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
3938 which has been initialized previously with <tt><a href="#int_va_start">llvm.va_start</a></tt>
3939 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
3943 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
3947 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
3948 macro available in C. In a target-dependent way, it destroys the
3949 <tt>va_list</tt> element to which the argument points. Calls to <a
3950 href="#int_va_start"><tt>llvm.va_start</tt></a> and <a href="#int_va_copy">
3951 <tt>llvm.va_copy</tt></a> must be matched exactly with calls to
3952 <tt>llvm.va_end</tt>.</p>
3956 <!-- _______________________________________________________________________ -->
3957 <div class="doc_subsubsection">
3958 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
3961 <div class="doc_text">
3966 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
3971 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
3972 from the source argument list to the destination argument list.</p>
3976 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
3977 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
3982 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
3983 macro available in C. In a target-dependent way, it copies the source
3984 <tt>va_list</tt> element into the destination <tt>va_list</tt> element. This
3985 intrinsic is necessary because the <tt><a href="#int_va_start">
3986 llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
3987 example, memory allocation.</p>
3991 <!-- ======================================================================= -->
3992 <div class="doc_subsection">
3993 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
3996 <div class="doc_text">
3999 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
4000 Collection</a> requires the implementation and generation of these intrinsics.
4001 These intrinsics allow identification of <a href="#int_gcroot">GC roots on the
4002 stack</a>, as well as garbage collector implementations that require <a
4003 href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a> barriers.
4004 Front-ends for type-safe garbage collected languages should generate these
4005 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
4006 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
4009 <p>The garbage collection intrinsics only operate on objects in the generic
4010 address space (address space zero).</p>
4014 <!-- _______________________________________________________________________ -->
4015 <div class="doc_subsubsection">
4016 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
4019 <div class="doc_text">
4024 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
4029 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
4030 the code generator, and allows some metadata to be associated with it.</p>
4034 <p>The first argument specifies the address of a stack object that contains the
4035 root pointer. The second pointer (which must be either a constant or a global
4036 value address) contains the meta-data to be associated with the root.</p>
4040 <p>At runtime, a call to this intrinsics stores a null pointer into the "ptrloc"
4041 location. At compile-time, the code generator generates information to allow
4042 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
4043 intrinsic may only be used in a function which <a href="#gc">specifies a GC
4049 <!-- _______________________________________________________________________ -->
4050 <div class="doc_subsubsection">
4051 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
4054 <div class="doc_text">
4059 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
4064 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
4065 locations, allowing garbage collector implementations that require read
4070 <p>The second argument is the address to read from, which should be an address
4071 allocated from the garbage collector. The first object is a pointer to the
4072 start of the referenced object, if needed by the language runtime (otherwise
4077 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
4078 instruction, but may be replaced with substantially more complex code by the
4079 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
4080 may only be used in a function which <a href="#gc">specifies a GC
4086 <!-- _______________________________________________________________________ -->
4087 <div class="doc_subsubsection">
4088 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
4091 <div class="doc_text">
4096 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
4101 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
4102 locations, allowing garbage collector implementations that require write
4103 barriers (such as generational or reference counting collectors).</p>
4107 <p>The first argument is the reference to store, the second is the start of the
4108 object to store it to, and the third is the address of the field of Obj to
4109 store to. If the runtime does not require a pointer to the object, Obj may be
4114 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
4115 instruction, but may be replaced with substantially more complex code by the
4116 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
4117 may only be used in a function which <a href="#gc">specifies a GC
4124 <!-- ======================================================================= -->
4125 <div class="doc_subsection">
4126 <a name="int_codegen">Code Generator Intrinsics</a>
4129 <div class="doc_text">
4131 These intrinsics are provided by LLVM to expose special features that may only
4132 be implemented with code generator support.
4137 <!-- _______________________________________________________________________ -->
4138 <div class="doc_subsubsection">
4139 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
4142 <div class="doc_text">
4146 declare i8 *@llvm.returnaddress(i32 <level>)
4152 The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
4153 target-specific value indicating the return address of the current function
4154 or one of its callers.
4160 The argument to this intrinsic indicates which function to return the address
4161 for. Zero indicates the calling function, one indicates its caller, etc. The
4162 argument is <b>required</b> to be a constant integer value.
4168 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
4169 the return address of the specified call frame, or zero if it cannot be
4170 identified. The value returned by this intrinsic is likely to be incorrect or 0
4171 for arguments other than zero, so it should only be used for debugging purposes.
4175 Note that calling this intrinsic does not prevent function inlining or other
4176 aggressive transformations, so the value returned may not be that of the obvious
4177 source-language caller.
4182 <!-- _______________________________________________________________________ -->
4183 <div class="doc_subsubsection">
4184 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
4187 <div class="doc_text">
4191 declare i8 *@llvm.frameaddress(i32 <level>)
4197 The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
4198 target-specific frame pointer value for the specified stack frame.
4204 The argument to this intrinsic indicates which function to return the frame
4205 pointer for. Zero indicates the calling function, one indicates its caller,
4206 etc. The argument is <b>required</b> to be a constant integer value.
4212 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
4213 the frame address of the specified call frame, or zero if it cannot be
4214 identified. The value returned by this intrinsic is likely to be incorrect or 0
4215 for arguments other than zero, so it should only be used for debugging purposes.
4219 Note that calling this intrinsic does not prevent function inlining or other
4220 aggressive transformations, so the value returned may not be that of the obvious
4221 source-language caller.
4225 <!-- _______________________________________________________________________ -->
4226 <div class="doc_subsubsection">
4227 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
4230 <div class="doc_text">
4234 declare i8 *@llvm.stacksave()
4240 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
4241 the function stack, for use with <a href="#int_stackrestore">
4242 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
4243 features like scoped automatic variable sized arrays in C99.
4249 This intrinsic returns a opaque pointer value that can be passed to <a
4250 href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
4251 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
4252 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
4253 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
4254 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
4255 that were allocated after the <tt>llvm.stacksave</tt> was executed.
4260 <!-- _______________________________________________________________________ -->
4261 <div class="doc_subsubsection">
4262 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
4265 <div class="doc_text">
4269 declare void @llvm.stackrestore(i8 * %ptr)
4275 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
4276 the function stack to the state it was in when the corresponding <a
4277 href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
4278 useful for implementing language features like scoped automatic variable sized
4285 See the description for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.
4291 <!-- _______________________________________________________________________ -->
4292 <div class="doc_subsubsection">
4293 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
4296 <div class="doc_text">
4300 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>)
4307 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
4308 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
4310 effect on the behavior of the program but can change its performance
4317 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
4318 determining if the fetch should be for a read (0) or write (1), and
4319 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
4320 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
4321 <tt>locality</tt> arguments must be constant integers.
4327 This intrinsic does not modify the behavior of the program. In particular,
4328 prefetches cannot trap and do not produce a value. On targets that support this
4329 intrinsic, the prefetch can provide hints to the processor cache for better
4335 <!-- _______________________________________________________________________ -->
4336 <div class="doc_subsubsection">
4337 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
4340 <div class="doc_text">
4344 declare void @llvm.pcmarker(i32 <id>)
4351 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
4353 code to simulators and other tools. The method is target specific, but it is
4354 expected that the marker will use exported symbols to transmit the PC of the marker.
4355 The marker makes no guarantees that it will remain with any specific instruction
4356 after optimizations. It is possible that the presence of a marker will inhibit
4357 optimizations. The intended use is to be inserted after optimizations to allow
4358 correlations of simulation runs.
4364 <tt>id</tt> is a numerical id identifying the marker.
4370 This intrinsic does not modify the behavior of the program. Backends that do not
4371 support this intrinisic may ignore it.
4376 <!-- _______________________________________________________________________ -->
4377 <div class="doc_subsubsection">
4378 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
4381 <div class="doc_text">
4385 declare i64 @llvm.readcyclecounter( )
4392 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
4393 counter register (or similar low latency, high accuracy clocks) on those targets
4394 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
4395 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
4396 should only be used for small timings.
4402 When directly supported, reading the cycle counter should not modify any memory.
4403 Implementations are allowed to either return a application specific value or a
4404 system wide value. On backends without support, this is lowered to a constant 0.
4409 <!-- ======================================================================= -->
4410 <div class="doc_subsection">
4411 <a name="int_libc">Standard C Library Intrinsics</a>
4414 <div class="doc_text">
4416 LLVM provides intrinsics for a few important standard C library functions.
4417 These intrinsics allow source-language front-ends to pass information about the
4418 alignment of the pointer arguments to the code generator, providing opportunity
4419 for more efficient code generation.
4424 <!-- _______________________________________________________________________ -->
4425 <div class="doc_subsubsection">
4426 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
4429 <div class="doc_text">
4433 declare void @llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
4434 i32 <len>, i32 <align>)
4435 declare void @llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
4436 i64 <len>, i32 <align>)
4442 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4443 location to the destination location.
4447 Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
4448 intrinsics do not return a value, and takes an extra alignment argument.
4454 The first argument is a pointer to the destination, the second is a pointer to
4455 the source. The third argument is an integer argument
4456 specifying the number of bytes to copy, and the fourth argument is the alignment
4457 of the source and destination locations.
4461 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4462 the caller guarantees that both the source and destination pointers are aligned
4469 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4470 location to the destination location, which are not allowed to overlap. It
4471 copies "len" bytes of memory over. If the argument is known to be aligned to
4472 some boundary, this can be specified as the fourth argument, otherwise it should
4478 <!-- _______________________________________________________________________ -->
4479 <div class="doc_subsubsection">
4480 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
4483 <div class="doc_text">
4487 declare void @llvm.memmove.i32(i8 * <dest>, i8 * <src>,
4488 i32 <len>, i32 <align>)
4489 declare void @llvm.memmove.i64(i8 * <dest>, i8 * <src>,
4490 i64 <len>, i32 <align>)
4496 The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
4497 location to the destination location. It is similar to the
4498 '<tt>llvm.memcmp</tt>' intrinsic but allows the two memory locations to overlap.
4502 Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
4503 intrinsics do not return a value, and takes an extra alignment argument.
4509 The first argument is a pointer to the destination, the second is a pointer to
4510 the source. The third argument is an integer argument
4511 specifying the number of bytes to copy, and the fourth argument is the alignment
4512 of the source and destination locations.
4516 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4517 the caller guarantees that the source and destination pointers are aligned to
4524 The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
4525 location to the destination location, which may overlap. It
4526 copies "len" bytes of memory over. If the argument is known to be aligned to
4527 some boundary, this can be specified as the fourth argument, otherwise it should
4533 <!-- _______________________________________________________________________ -->
4534 <div class="doc_subsubsection">
4535 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
4538 <div class="doc_text">
4542 declare void @llvm.memset.i32(i8 * <dest>, i8 <val>,
4543 i32 <len>, i32 <align>)
4544 declare void @llvm.memset.i64(i8 * <dest>, i8 <val>,
4545 i64 <len>, i32 <align>)
4551 The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
4556 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
4557 does not return a value, and takes an extra alignment argument.
4563 The first argument is a pointer to the destination to fill, the second is the
4564 byte value to fill it with, the third argument is an integer
4565 argument specifying the number of bytes to fill, and the fourth argument is the
4566 known alignment of destination location.
4570 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4571 the caller guarantees that the destination pointer is aligned to that boundary.
4577 The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
4579 destination location. If the argument is known to be aligned to some boundary,
4580 this can be specified as the fourth argument, otherwise it should be set to 0 or
4586 <!-- _______________________________________________________________________ -->
4587 <div class="doc_subsubsection">
4588 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
4591 <div class="doc_text">
4594 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
4595 floating point or vector of floating point type. Not all targets support all
4598 declare float @llvm.sqrt.f32(float %Val)
4599 declare double @llvm.sqrt.f64(double %Val)
4600 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
4601 declare fp128 @llvm.sqrt.f128(fp128 %Val)
4602 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
4608 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
4609 returning the same value as the libm '<tt>sqrt</tt>' functions would. Unlike
4610 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
4611 negative numbers (which allows for better optimization).
4617 The argument and return value are floating point numbers of the same type.
4623 This function returns the sqrt of the specified operand if it is a nonnegative
4624 floating point number.
4628 <!-- _______________________________________________________________________ -->
4629 <div class="doc_subsubsection">
4630 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
4633 <div class="doc_text">
4636 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
4637 floating point or vector of floating point type. Not all targets support all
4640 declare float @llvm.powi.f32(float %Val, i32 %power)
4641 declare double @llvm.powi.f64(double %Val, i32 %power)
4642 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
4643 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
4644 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
4650 The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
4651 specified (positive or negative) power. The order of evaluation of
4652 multiplications is not defined. When a vector of floating point type is
4653 used, the second argument remains a scalar integer value.
4659 The second argument is an integer power, and the first is a value to raise to
4666 This function returns the first value raised to the second power with an
4667 unspecified sequence of rounding operations.</p>
4670 <!-- _______________________________________________________________________ -->
4671 <div class="doc_subsubsection">
4672 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
4675 <div class="doc_text">
4678 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
4679 floating point or vector of floating point type. Not all targets support all
4682 declare float @llvm.sin.f32(float %Val)
4683 declare double @llvm.sin.f64(double %Val)
4684 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
4685 declare fp128 @llvm.sin.f128(fp128 %Val)
4686 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
4692 The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.
4698 The argument and return value are floating point numbers of the same type.
4704 This function returns the sine of the specified operand, returning the
4705 same values as the libm <tt>sin</tt> functions would, and handles error
4706 conditions in the same way.</p>
4709 <!-- _______________________________________________________________________ -->
4710 <div class="doc_subsubsection">
4711 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
4714 <div class="doc_text">
4717 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
4718 floating point or vector of floating point type. Not all targets support all
4721 declare float @llvm.cos.f32(float %Val)
4722 declare double @llvm.cos.f64(double %Val)
4723 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
4724 declare fp128 @llvm.cos.f128(fp128 %Val)
4725 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
4731 The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.
4737 The argument and return value are floating point numbers of the same type.
4743 This function returns the cosine of the specified operand, returning the
4744 same values as the libm <tt>cos</tt> functions would, and handles error
4745 conditions in the same way.</p>
4748 <!-- _______________________________________________________________________ -->
4749 <div class="doc_subsubsection">
4750 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
4753 <div class="doc_text">
4756 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
4757 floating point or vector of floating point type. Not all targets support all
4760 declare float @llvm.pow.f32(float %Val, float %Power)
4761 declare double @llvm.pow.f64(double %Val, double %Power)
4762 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
4763 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
4764 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
4770 The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
4771 specified (positive or negative) power.
4777 The second argument is a floating point power, and the first is a value to
4778 raise to that power.
4784 This function returns the first value raised to the second power,
4786 same values as the libm <tt>pow</tt> functions would, and handles error
4787 conditions in the same way.</p>
4791 <!-- ======================================================================= -->
4792 <div class="doc_subsection">
4793 <a name="int_manip">Bit Manipulation Intrinsics</a>
4796 <div class="doc_text">
4798 LLVM provides intrinsics for a few important bit manipulation operations.
4799 These allow efficient code generation for some algorithms.
4804 <!-- _______________________________________________________________________ -->
4805 <div class="doc_subsubsection">
4806 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
4809 <div class="doc_text">
4812 <p>This is an overloaded intrinsic function. You can use bswap on any integer
4813 type that is an even number of bytes (i.e. BitWidth % 16 == 0).
4815 declare i16 @llvm.bswap.i16(i16 <id>)
4816 declare i32 @llvm.bswap.i32(i32 <id>)
4817 declare i64 @llvm.bswap.i64(i64 <id>)
4823 The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
4824 values with an even number of bytes (positive multiple of 16 bits). These are
4825 useful for performing operations on data that is not in the target's native
4832 The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
4833 and low byte of the input i16 swapped. Similarly, the <tt>llvm.bswap.i32</tt>
4834 intrinsic returns an i32 value that has the four bytes of the input i32
4835 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the returned
4836 i32 will have its bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i48</tt>,
4837 <tt>llvm.bswap.i64</tt> and other intrinsics extend this concept to
4838 additional even-byte lengths (6 bytes, 8 bytes and more, respectively).
4843 <!-- _______________________________________________________________________ -->
4844 <div class="doc_subsubsection">
4845 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
4848 <div class="doc_text">
4851 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
4852 width. Not all targets support all bit widths however.
4854 declare i8 @llvm.ctpop.i8 (i8 <src>)
4855 declare i16 @llvm.ctpop.i16(i16 <src>)
4856 declare i32 @llvm.ctpop.i32(i32 <src>)
4857 declare i64 @llvm.ctpop.i64(i64 <src>)
4858 declare i256 @llvm.ctpop.i256(i256 <src>)
4864 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
4871 The only argument is the value to be counted. The argument may be of any
4872 integer type. The return type must match the argument type.
4878 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
4882 <!-- _______________________________________________________________________ -->
4883 <div class="doc_subsubsection">
4884 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
4887 <div class="doc_text">
4890 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
4891 integer bit width. Not all targets support all bit widths however.
4893 declare i8 @llvm.ctlz.i8 (i8 <src>)
4894 declare i16 @llvm.ctlz.i16(i16 <src>)
4895 declare i32 @llvm.ctlz.i32(i32 <src>)
4896 declare i64 @llvm.ctlz.i64(i64 <src>)
4897 declare i256 @llvm.ctlz.i256(i256 <src>)
4903 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
4904 leading zeros in a variable.
4910 The only argument is the value to be counted. The argument may be of any
4911 integer type. The return type must match the argument type.
4917 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
4918 in a variable. If the src == 0 then the result is the size in bits of the type
4919 of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.
4925 <!-- _______________________________________________________________________ -->
4926 <div class="doc_subsubsection">
4927 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
4930 <div class="doc_text">
4933 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
4934 integer bit width. Not all targets support all bit widths however.
4936 declare i8 @llvm.cttz.i8 (i8 <src>)
4937 declare i16 @llvm.cttz.i16(i16 <src>)
4938 declare i32 @llvm.cttz.i32(i32 <src>)
4939 declare i64 @llvm.cttz.i64(i64 <src>)
4940 declare i256 @llvm.cttz.i256(i256 <src>)
4946 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
4953 The only argument is the value to be counted. The argument may be of any
4954 integer type. The return type must match the argument type.
4960 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
4961 in a variable. If the src == 0 then the result is the size in bits of the type
4962 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
4966 <!-- _______________________________________________________________________ -->
4967 <div class="doc_subsubsection">
4968 <a name="int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic</a>
4971 <div class="doc_text">
4974 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.select</tt>
4975 on any integer bit width.
4977 declare i17 @llvm.part.select.i17 (i17 %val, i32 %loBit, i32 %hiBit)
4978 declare i29 @llvm.part.select.i29 (i29 %val, i32 %loBit, i32 %hiBit)
4982 <p>The '<tt>llvm.part.select</tt>' family of intrinsic functions selects a
4983 range of bits from an integer value and returns them in the same bit width as
4984 the original value.</p>
4987 <p>The first argument, <tt>%val</tt> and the result may be integer types of
4988 any bit width but they must have the same bit width. The second and third
4989 arguments must be <tt>i32</tt> type since they specify only a bit index.</p>
4992 <p>The operation of the '<tt>llvm.part.select</tt>' intrinsic has two modes
4993 of operation: forwards and reverse. If <tt>%loBit</tt> is greater than
4994 <tt>%hiBits</tt> then the intrinsic operates in reverse mode. Otherwise it
4995 operates in forward mode.</p>
4996 <p>In forward mode, this intrinsic is the equivalent of shifting <tt>%val</tt>
4997 right by <tt>%loBit</tt> bits and then ANDing it with a mask with
4998 only the <tt>%hiBit - %loBit</tt> bits set, as follows:</p>
5000 <li>The <tt>%val</tt> is shifted right (LSHR) by the number of bits specified
5001 by <tt>%loBits</tt>. This normalizes the value to the low order bits.</li>
5002 <li>The <tt>%loBits</tt> value is subtracted from the <tt>%hiBits</tt> value
5003 to determine the number of bits to retain.</li>
5004 <li>A mask of the retained bits is created by shifting a -1 value.</li>
5005 <li>The mask is ANDed with <tt>%val</tt> to produce the result.
5007 <p>In reverse mode, a similar computation is made except that the bits are
5008 returned in the reverse order. So, for example, if <tt>X</tt> has the value
5009 <tt>i16 0x0ACF (101011001111)</tt> and we apply
5010 <tt>part.select(i16 X, 8, 3)</tt> to it, we get back the value
5011 <tt>i16 0x0026 (000000100110)</tt>.</p>
5014 <div class="doc_subsubsection">
5015 <a name="int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic</a>
5018 <div class="doc_text">
5021 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.set</tt>
5022 on any integer bit width.
5024 declare i17 @llvm.part.set.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
5025 declare i29 @llvm.part.set.i29.i9 (i29 %val, i9 %repl, i32 %lo, i32 %hi)
5029 <p>The '<tt>llvm.part.set</tt>' family of intrinsic functions replaces a range
5030 of bits in an integer value with another integer value. It returns the integer
5031 with the replaced bits.</p>
5034 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5035 any bit width but they must have the same bit width. <tt>%val</tt> is the value
5036 whose bits will be replaced. The second argument, <tt>%repl</tt> may be an
5037 integer of any bit width. The third and fourth arguments must be <tt>i32</tt>
5038 type since they specify only a bit index.</p>
5041 <p>The operation of the '<tt>llvm.part.set</tt>' intrinsic has two modes
5042 of operation: forwards and reverse. If <tt>%lo</tt> is greater than
5043 <tt>%hi</tt> then the intrinsic operates in reverse mode. Otherwise it
5044 operates in forward mode.</p>
5045 <p>For both modes, the <tt>%repl</tt> value is prepared for use by either
5046 truncating it down to the size of the replacement area or zero extending it
5047 up to that size.</p>
5048 <p>In forward mode, the bits between <tt>%lo</tt> and <tt>%hi</tt> (inclusive)
5049 are replaced with corresponding bits from <tt>%repl</tt>. That is the 0th bit
5050 in <tt>%repl</tt> replaces the <tt>%lo</tt>th bit in <tt>%val</tt> and etc. up
5051 to the <tt>%hi</tt>th bit.
5052 <p>In reverse mode, a similar computation is made except that the bits are
5053 reversed. That is, the <tt>0</tt>th bit in <tt>%repl</tt> replaces the
5054 <tt>%hi</tt> bit in <tt>%val</tt> and etc. down to the <tt>%lo</tt>th bit.
5057 llvm.part.set(0xFFFF, 0, 4, 7) -> 0xFF0F
5058 llvm.part.set(0xFFFF, 0, 7, 4) -> 0xFF0F
5059 llvm.part.set(0xFFFF, 1, 7, 4) -> 0xFF8F
5060 llvm.part.set(0xFFFF, F, 8, 3) -> 0xFFE7
5061 llvm.part.set(0xFFFF, 0, 3, 8) -> 0xFE07
5065 <!-- ======================================================================= -->
5066 <div class="doc_subsection">
5067 <a name="int_debugger">Debugger Intrinsics</a>
5070 <div class="doc_text">
5072 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
5073 are described in the <a
5074 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
5075 Debugging</a> document.
5080 <!-- ======================================================================= -->
5081 <div class="doc_subsection">
5082 <a name="int_eh">Exception Handling Intrinsics</a>
5085 <div class="doc_text">
5086 <p> The LLVM exception handling intrinsics (which all start with
5087 <tt>llvm.eh.</tt> prefix), are described in the <a
5088 href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
5089 Handling</a> document. </p>
5092 <!-- ======================================================================= -->
5093 <div class="doc_subsection">
5094 <a name="int_trampoline">Trampoline Intrinsic</a>
5097 <div class="doc_text">
5099 This intrinsic makes it possible to excise one parameter, marked with
5100 the <tt>nest</tt> attribute, from a function. The result is a callable
5101 function pointer lacking the nest parameter - the caller does not need
5102 to provide a value for it. Instead, the value to use is stored in
5103 advance in a "trampoline", a block of memory usually allocated
5104 on the stack, which also contains code to splice the nest value into the
5105 argument list. This is used to implement the GCC nested function address
5109 For example, if the function is
5110 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
5111 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as follows:</p>
5113 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
5114 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
5115 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
5116 %fp = bitcast i8* %p to i32 (i32, i32)*
5118 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
5119 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
5122 <!-- _______________________________________________________________________ -->
5123 <div class="doc_subsubsection">
5124 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
5126 <div class="doc_text">
5129 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
5133 This fills the memory pointed to by <tt>tramp</tt> with code
5134 and returns a function pointer suitable for executing it.
5138 The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
5139 pointers. The <tt>tramp</tt> argument must point to a sufficiently large
5140 and sufficiently aligned block of memory; this memory is written to by the
5141 intrinsic. Note that the size and the alignment are target-specific - LLVM
5142 currently provides no portable way of determining them, so a front-end that
5143 generates this intrinsic needs to have some target-specific knowledge.
5144 The <tt>func</tt> argument must hold a function bitcast to an <tt>i8*</tt>.
5148 The block of memory pointed to by <tt>tramp</tt> is filled with target
5149 dependent code, turning it into a function. A pointer to this function is
5150 returned, but needs to be bitcast to an
5151 <a href="#int_trampoline">appropriate function pointer type</a>
5152 before being called. The new function's signature is the same as that of
5153 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
5154 removed. At most one such <tt>nest</tt> argument is allowed, and it must be
5155 of pointer type. Calling the new function is equivalent to calling
5156 <tt>func</tt> with the same argument list, but with <tt>nval</tt> used for the
5157 missing <tt>nest</tt> argument. If, after calling
5158 <tt>llvm.init.trampoline</tt>, the memory pointed to by <tt>tramp</tt> is
5159 modified, then the effect of any later call to the returned function pointer is
5164 <!-- ======================================================================= -->
5165 <div class="doc_subsection">
5166 <a name="int_general">General Intrinsics</a>
5169 <div class="doc_text">
5170 <p> This class of intrinsics is designed to be generic and has
5171 no specific purpose. </p>
5174 <!-- _______________________________________________________________________ -->
5175 <div class="doc_subsubsection">
5176 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
5179 <div class="doc_text">
5183 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
5189 The '<tt>llvm.var.annotation</tt>' intrinsic
5195 The first argument is a pointer to a value, the second is a pointer to a
5196 global string, the third is a pointer to a global string which is the source
5197 file name, and the last argument is the line number.
5203 This intrinsic allows annotation of local variables with arbitrary strings.
5204 This can be useful for special purpose optimizations that want to look for these
5205 annotations. These have no other defined use, they are ignored by code
5206 generation and optimization.
5209 <!-- _______________________________________________________________________ -->
5210 <div class="doc_subsubsection">
5211 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
5214 <div class="doc_text">
5217 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
5218 any integer bit width.
5221 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
5222 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
5223 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
5224 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
5225 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
5231 The '<tt>llvm.annotation</tt>' intrinsic.
5237 The first argument is an integer value (result of some expression),
5238 the second is a pointer to a global string, the third is a pointer to a global
5239 string which is the source file name, and the last argument is the line number.
5240 It returns the value of the first argument.
5246 This intrinsic allows annotations to be put on arbitrary expressions
5247 with arbitrary strings. This can be useful for special purpose optimizations
5248 that want to look for these annotations. These have no other defined use, they
5249 are ignored by code generation and optimization.
5252 <!-- *********************************************************************** -->
5255 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
5256 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
5257 <a href="http://validator.w3.org/check/referer"><img
5258 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a>
5260 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
5261 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
5262 Last modified: $Date$