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="#moduleasm">Module-Level Inline Assembly</a></li>
30 <li><a href="#datalayout">Data Layout</a></li>
33 <li><a href="#typesystem">Type System</a>
35 <li><a href="#t_primitive">Primitive Types</a>
37 <li><a href="#t_classifications">Type Classifications</a></li>
40 <li><a href="#t_derived">Derived Types</a>
42 <li><a href="#t_array">Array Type</a></li>
43 <li><a href="#t_function">Function Type</a></li>
44 <li><a href="#t_pointer">Pointer Type</a></li>
45 <li><a href="#t_struct">Structure Type</a></li>
46 <li><a href="#t_pstruct">Packed Structure Type</a></li>
47 <li><a href="#t_vector">Vector Type</a></li>
48 <li><a href="#t_opaque">Opaque Type</a></li>
53 <li><a href="#constants">Constants</a>
55 <li><a href="#simpleconstants">Simple Constants</a>
56 <li><a href="#aggregateconstants">Aggregate Constants</a>
57 <li><a href="#globalconstants">Global Variable and Function Addresses</a>
58 <li><a href="#undefvalues">Undefined Values</a>
59 <li><a href="#constantexprs">Constant Expressions</a>
62 <li><a href="#othervalues">Other Values</a>
64 <li><a href="#inlineasm">Inline Assembler Expressions</a>
67 <li><a href="#instref">Instruction Reference</a>
69 <li><a href="#terminators">Terminator Instructions</a>
71 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
72 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
73 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
74 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
75 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
76 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
79 <li><a href="#binaryops">Binary Operations</a>
81 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
82 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
83 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
84 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
85 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
86 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
87 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
88 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
89 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
92 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
94 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
95 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
96 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
97 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
98 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
99 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
102 <li><a href="#vectorops">Vector Operations</a>
104 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
105 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
106 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
109 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
111 <li><a href="#i_malloc">'<tt>malloc</tt>' Instruction</a></li>
112 <li><a href="#i_free">'<tt>free</tt>' Instruction</a></li>
113 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
114 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
115 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
116 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
119 <li><a href="#convertops">Conversion Operations</a>
121 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
122 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
123 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
124 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
125 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
126 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
127 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
128 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
129 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
130 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
131 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
132 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
134 <li><a href="#otherops">Other Operations</a>
136 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
137 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
138 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
139 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
140 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
141 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
146 <li><a href="#intrinsics">Intrinsic Functions</a>
148 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
150 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
151 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
152 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
155 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
157 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
158 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
159 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
162 <li><a href="#int_codegen">Code Generator Intrinsics</a>
164 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
165 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
166 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
167 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
168 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
169 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
170 <li><a href="#int_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
173 <li><a href="#int_libc">Standard C Library Intrinsics</a>
175 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
176 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
177 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
178 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
179 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
182 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
184 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
185 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
186 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
187 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
188 <li><a href="#int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic </a></li>
189 <li><a href="#int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic </a></li>
192 <li><a href="#int_debugger">Debugger intrinsics</a></li>
193 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
194 <li><a href="#int_atomics">Atomic Operations and Synchronization Intrinsics</a>
196 <li><a href="#int_lcs">'<tt>llvm.atomic.lcs.*</tt>' Intrinsic</a></li>
197 <li><a href="#int_ls">'<tt>llvm.atomic.ls.*</tt>' Intrinsic</a></li>
198 <li><a href="#int_las">'<tt>llvm.atomic.las.*</tt>' Intrinsic</a></li>
199 <li><a href="#int_lss">'<tt>llvm.atomic.lss.*</tt>' Intrinsic</a></li>
200 <li><a href="#int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a></li>
203 <li><a href="#int_general">General intrinsics</a>
205 <li><a href="#int_var_annotation">
206 <tt>llvm.var.annotation</tt>' Intrinsic</a></li>
213 <div class="doc_author">
214 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
215 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
218 <!-- *********************************************************************** -->
219 <div class="doc_section"> <a name="abstract">Abstract </a></div>
220 <!-- *********************************************************************** -->
222 <div class="doc_text">
223 <p>This document is a reference manual for the LLVM assembly language.
224 LLVM is an SSA based representation that provides type safety,
225 low-level operations, flexibility, and the capability of representing
226 'all' high-level languages cleanly. It is the common code
227 representation used throughout all phases of the LLVM compilation
231 <!-- *********************************************************************** -->
232 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
233 <!-- *********************************************************************** -->
235 <div class="doc_text">
237 <p>The LLVM code representation is designed to be used in three
238 different forms: as an in-memory compiler IR, as an on-disk bitcode
239 representation (suitable for fast loading by a Just-In-Time compiler),
240 and as a human readable assembly language representation. This allows
241 LLVM to provide a powerful intermediate representation for efficient
242 compiler transformations and analysis, while providing a natural means
243 to debug and visualize the transformations. The three different forms
244 of LLVM are all equivalent. This document describes the human readable
245 representation and notation.</p>
247 <p>The LLVM representation aims to be light-weight and low-level
248 while being expressive, typed, and extensible at the same time. It
249 aims to be a "universal IR" of sorts, by being at a low enough level
250 that high-level ideas may be cleanly mapped to it (similar to how
251 microprocessors are "universal IR's", allowing many source languages to
252 be mapped to them). By providing type information, LLVM can be used as
253 the target of optimizations: for example, through pointer analysis, it
254 can be proven that a C automatic variable is never accessed outside of
255 the current function... allowing it to be promoted to a simple SSA
256 value instead of a memory location.</p>
260 <!-- _______________________________________________________________________ -->
261 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
263 <div class="doc_text">
265 <p>It is important to note that this document describes 'well formed'
266 LLVM assembly language. There is a difference between what the parser
267 accepts and what is considered 'well formed'. For example, the
268 following instruction is syntactically okay, but not well formed:</p>
270 <div class="doc_code">
272 %x = <a href="#i_add">add</a> i32 1, %x
276 <p>...because the definition of <tt>%x</tt> does not dominate all of
277 its uses. The LLVM infrastructure provides a verification pass that may
278 be used to verify that an LLVM module is well formed. This pass is
279 automatically run by the parser after parsing input assembly and by
280 the optimizer before it outputs bitcode. The violations pointed out
281 by the verifier pass indicate bugs in transformation passes or input to
285 <!-- Describe the typesetting conventions here. -->
287 <!-- *********************************************************************** -->
288 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
289 <!-- *********************************************************************** -->
291 <div class="doc_text">
293 <p>LLVM uses three different forms of identifiers, for different
297 <li>Named values are represented as a string of characters with a '%' prefix.
298 For example, %foo, %DivisionByZero, %a.really.long.identifier. The actual
299 regular expression used is '<tt>%[a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
300 Identifiers which require other characters in their names can be surrounded
301 with quotes. In this way, anything except a <tt>"</tt> character can be used
304 <li>Unnamed values are represented as an unsigned numeric value with a '%'
305 prefix. For example, %12, %2, %44.</li>
307 <li>Constants, which are described in a <a href="#constants">section about
308 constants</a>, below.</li>
311 <p>LLVM requires that values start with a '%' sign for two reasons: Compilers
312 don't need to worry about name clashes with reserved words, and the set of
313 reserved words may be expanded in the future without penalty. Additionally,
314 unnamed identifiers allow a compiler to quickly come up with a temporary
315 variable without having to avoid symbol table conflicts.</p>
317 <p>Reserved words in LLVM are very similar to reserved words in other
318 languages. There are keywords for different opcodes
319 ('<tt><a href="#i_add">add</a></tt>',
320 '<tt><a href="#i_bitcast">bitcast</a></tt>',
321 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
322 href="#t_void">void</a></tt>', '<tt><a href="#t_primitive">i32</a></tt>', etc...),
323 and others. These reserved words cannot conflict with variable names, because
324 none of them start with a '%' character.</p>
326 <p>Here is an example of LLVM code to multiply the integer variable
327 '<tt>%X</tt>' by 8:</p>
331 <div class="doc_code">
333 %result = <a href="#i_mul">mul</a> i32 %X, 8
337 <p>After strength reduction:</p>
339 <div class="doc_code">
341 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
345 <p>And the hard way:</p>
347 <div class="doc_code">
349 <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
350 <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
351 %result = <a href="#i_add">add</a> i32 %1, %1
355 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
356 important lexical features of LLVM:</p>
360 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
363 <li>Unnamed temporaries are created when the result of a computation is not
364 assigned to a named value.</li>
366 <li>Unnamed temporaries are numbered sequentially</li>
370 <p>...and it also shows a convention that we follow in this document. When
371 demonstrating instructions, we will follow an instruction with a comment that
372 defines the type and name of value produced. Comments are shown in italic
377 <!-- *********************************************************************** -->
378 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
379 <!-- *********************************************************************** -->
381 <!-- ======================================================================= -->
382 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
385 <div class="doc_text">
387 <p>LLVM programs are composed of "Module"s, each of which is a
388 translation unit of the input programs. Each module consists of
389 functions, global variables, and symbol table entries. Modules may be
390 combined together with the LLVM linker, which merges function (and
391 global variable) definitions, resolves forward declarations, and merges
392 symbol table entries. Here is an example of the "hello world" module:</p>
394 <div class="doc_code">
395 <pre><i>; Declare the string constant as a global constant...</i>
396 <a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a
397 href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
399 <i>; External declaration of the puts function</i>
400 <a href="#functionstructure">declare</a> i32 @puts(i8 *) <i>; i32(i8 *)* </i>
402 <i>; Definition of main function</i>
403 define i32 @main() { <i>; i32()* </i>
404 <i>; Convert [13x i8 ]* to i8 *...</i>
406 href="#i_getelementptr">getelementptr</a> [13 x i8 ]* @.LC0, i64 0, i64 0 <i>; i8 *</i>
408 <i>; Call puts function to write out the string to stdout...</i>
410 href="#i_call">call</a> i32 @puts(i8 * %cast210) <i>; i32</i>
412 href="#i_ret">ret</a> i32 0<br>}<br>
416 <p>This example is made up of a <a href="#globalvars">global variable</a>
417 named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
418 function, and a <a href="#functionstructure">function definition</a>
419 for "<tt>main</tt>".</p>
421 <p>In general, a module is made up of a list of global values,
422 where both functions and global variables are global values. Global values are
423 represented by a pointer to a memory location (in this case, a pointer to an
424 array of char, and a pointer to a function), and have one of the following <a
425 href="#linkage">linkage types</a>.</p>
429 <!-- ======================================================================= -->
430 <div class="doc_subsection">
431 <a name="linkage">Linkage Types</a>
434 <div class="doc_text">
437 All Global Variables and Functions have one of the following types of linkage:
442 <dt><tt><b><a name="linkage_internal">internal</a></b></tt> </dt>
444 <dd>Global values with internal linkage are only directly accessible by
445 objects in the current module. In particular, linking code into a module with
446 an internal global value may cause the internal to be renamed as necessary to
447 avoid collisions. Because the symbol is internal to the module, all
448 references can be updated. This corresponds to the notion of the
449 '<tt>static</tt>' keyword in C.
452 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
454 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
455 the same name when linkage occurs. This is typically used to implement
456 inline functions, templates, or other code which must be generated in each
457 translation unit that uses it. Unreferenced <tt>linkonce</tt> globals are
458 allowed to be discarded.
461 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
463 <dd>"<tt>weak</tt>" linkage is exactly the same as <tt>linkonce</tt> linkage,
464 except that unreferenced <tt>weak</tt> globals may not be discarded. This is
465 used for globals that may be emitted in multiple translation units, but that
466 are not guaranteed to be emitted into every translation unit that uses them.
467 One example of this are common globals in C, such as "<tt>int X;</tt>" at
471 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
473 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
474 pointer to array type. When two global variables with appending linkage are
475 linked together, the two global arrays are appended together. This is the
476 LLVM, typesafe, equivalent of having the system linker append together
477 "sections" with identical names when .o files are linked.
480 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt>: </dt>
481 <dd>The semantics of this linkage follow the ELF model: the symbol is weak
482 until linked, if not linked, the symbol becomes null instead of being an
486 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
488 <dd>If none of the above identifiers are used, the global is externally
489 visible, meaning that it participates in linkage and can be used to resolve
490 external symbol references.
495 The next two types of linkage are targeted for Microsoft Windows platform
496 only. They are designed to support importing (exporting) symbols from (to)
501 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt>: </dt>
503 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
504 or variable via a global pointer to a pointer that is set up by the DLL
505 exporting the symbol. On Microsoft Windows targets, the pointer name is
506 formed by combining <code>_imp__</code> and the function or variable name.
509 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt>: </dt>
511 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
512 pointer to a pointer in a DLL, so that it can be referenced with the
513 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
514 name is formed by combining <code>_imp__</code> and the function or variable
520 <p><a name="linkage_external"></a>For example, since the "<tt>.LC0</tt>"
521 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
522 variable and was linked with this one, one of the two would be renamed,
523 preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
524 external (i.e., lacking any linkage declarations), they are accessible
525 outside of the current module.</p>
526 <p>It is illegal for a function <i>declaration</i>
527 to have any linkage type other than "externally visible", <tt>dllimport</tt>,
528 or <tt>extern_weak</tt>.</p>
529 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt> and <tt>weak</tt>
533 <!-- ======================================================================= -->
534 <div class="doc_subsection">
535 <a name="callingconv">Calling Conventions</a>
538 <div class="doc_text">
540 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
541 and <a href="#i_invoke">invokes</a> can all have an optional calling convention
542 specified for the call. The calling convention of any pair of dynamic
543 caller/callee must match, or the behavior of the program is undefined. The
544 following calling conventions are supported by LLVM, and more may be added in
548 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
550 <dd>This calling convention (the default if no other calling convention is
551 specified) matches the target C calling conventions. This calling convention
552 supports varargs function calls and tolerates some mismatch in the declared
553 prototype and implemented declaration of the function (as does normal C).
556 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
558 <dd>This calling convention attempts to make calls as fast as possible
559 (e.g. by passing things in registers). This calling convention allows the
560 target to use whatever tricks it wants to produce fast code for the target,
561 without having to conform to an externally specified ABI. Implementations of
562 this convention should allow arbitrary tail call optimization to be supported.
563 This calling convention does not support varargs and requires the prototype of
564 all callees to exactly match the prototype of the function definition.
567 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
569 <dd>This calling convention attempts to make code in the caller as efficient
570 as possible under the assumption that the call is not commonly executed. As
571 such, these calls often preserve all registers so that the call does not break
572 any live ranges in the caller side. This calling convention does not support
573 varargs and requires the prototype of all callees to exactly match the
574 prototype of the function definition.
577 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
579 <dd>Any calling convention may be specified by number, allowing
580 target-specific calling conventions to be used. Target specific calling
581 conventions start at 64.
585 <p>More calling conventions can be added/defined on an as-needed basis, to
586 support pascal conventions or any other well-known target-independent
591 <!-- ======================================================================= -->
592 <div class="doc_subsection">
593 <a name="visibility">Visibility Styles</a>
596 <div class="doc_text">
599 All Global Variables and Functions have one of the following visibility styles:
603 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
605 <dd>On ELF, default visibility means that the declaration is visible to other
606 modules and, in shared libraries, means that the declared entity may be
607 overridden. On Darwin, default visibility means that the declaration is
608 visible to other modules. Default visibility corresponds to "external
609 linkage" in the language.
612 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
614 <dd>Two declarations of an object with hidden visibility refer to the same
615 object if they are in the same shared object. Usually, hidden visibility
616 indicates that the symbol will not be placed into the dynamic symbol table,
617 so no other module (executable or shared library) can reference it
621 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
623 <dd>On ELF, protected visibility indicates that the symbol will be placed in
624 the dynamic symbol table, but that references within the defining module will
625 bind to the local symbol. That is, the symbol cannot be overridden by another
632 <!-- ======================================================================= -->
633 <div class="doc_subsection">
634 <a name="globalvars">Global Variables</a>
637 <div class="doc_text">
639 <p>Global variables define regions of memory allocated at compilation time
640 instead of run-time. Global variables may optionally be initialized, may have
641 an explicit section to be placed in, and may have an optional explicit alignment
642 specified. A variable may be defined as "thread_local", which means that it
643 will not be shared by threads (each thread will have a separated copy of the
644 variable). A variable may be defined as a global "constant," which indicates
645 that the contents of the variable will <b>never</b> be modified (enabling better
646 optimization, allowing the global data to be placed in the read-only section of
647 an executable, etc). Note that variables that need runtime initialization
648 cannot be marked "constant" as there is a store to the variable.</p>
651 LLVM explicitly allows <em>declarations</em> of global variables to be marked
652 constant, even if the final definition of the global is not. This capability
653 can be used to enable slightly better optimization of the program, but requires
654 the language definition to guarantee that optimizations based on the
655 'constantness' are valid for the translation units that do not include the
659 <p>As SSA values, global variables define pointer values that are in
660 scope (i.e. they dominate) all basic blocks in the program. Global
661 variables always define a pointer to their "content" type because they
662 describe a region of memory, and all memory objects in LLVM are
663 accessed through pointers.</p>
665 <p>LLVM allows an explicit section to be specified for globals. If the target
666 supports it, it will emit globals to the section specified.</p>
668 <p>An explicit alignment may be specified for a global. If not present, or if
669 the alignment is set to zero, the alignment of the global is set by the target
670 to whatever it feels convenient. If an explicit alignment is specified, the
671 global is forced to have at least that much alignment. All alignments must be
674 <p>For example, the following defines a global with an initializer, section,
677 <div class="doc_code">
679 @G = constant float 1.0, section "foo", align 4
686 <!-- ======================================================================= -->
687 <div class="doc_subsection">
688 <a name="functionstructure">Functions</a>
691 <div class="doc_text">
693 <p>LLVM function definitions consist of the "<tt>define</tt>" keyord,
694 an optional <a href="#linkage">linkage type</a>, an optional
695 <a href="#visibility">visibility style</a>, an optional
696 <a href="#callingconv">calling convention</a>, a return type, an optional
697 <a href="#paramattrs">parameter attribute</a> for the return type, a function
698 name, a (possibly empty) argument list (each with optional
699 <a href="#paramattrs">parameter attributes</a>), an optional section, an
700 optional alignment, an opening curly brace, a list of basic blocks, and a
703 LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
704 optional <a href="#linkage">linkage type</a>, an optional
705 <a href="#visibility">visibility style</a>, an optional
706 <a href="#callingconv">calling convention</a>, a return type, an optional
707 <a href="#paramattrs">parameter attribute</a> for the return type, a function
708 name, a possibly empty list of arguments, and an optional alignment.</p>
710 <p>A function definition contains a list of basic blocks, forming the CFG for
711 the function. Each basic block may optionally start with a label (giving the
712 basic block a symbol table entry), contains a list of instructions, and ends
713 with a <a href="#terminators">terminator</a> instruction (such as a branch or
714 function return).</p>
716 <p>The first basic block in a function is special in two ways: it is immediately
717 executed on entrance to the function, and it is not allowed to have predecessor
718 basic blocks (i.e. there can not be any branches to the entry block of a
719 function). Because the block can have no predecessors, it also cannot have any
720 <a href="#i_phi">PHI nodes</a>.</p>
722 <p>LLVM allows an explicit section to be specified for functions. If the target
723 supports it, it will emit functions to the section specified.</p>
725 <p>An explicit alignment may be specified for a function. If not present, or if
726 the alignment is set to zero, the alignment of the function is set by the target
727 to whatever it feels convenient. If an explicit alignment is specified, the
728 function is forced to have at least that much alignment. All alignments must be
734 <!-- ======================================================================= -->
735 <div class="doc_subsection">
736 <a name="aliasstructure">Aliases</a>
738 <div class="doc_text">
739 <p>Aliases act as "second name" for the aliasee value (which can be either
740 function or global variable or bitcast of global value). Aliases may have an
741 optional <a href="#linkage">linkage type</a>, and an
742 optional <a href="#visibility">visibility style</a>.</p>
746 <div class="doc_code">
748 @<Name> = [Linkage] [Visibility] alias <AliaseeTy> @<Aliasee>
756 <!-- ======================================================================= -->
757 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
758 <div class="doc_text">
759 <p>The return type and each parameter of a function type may have a set of
760 <i>parameter attributes</i> associated with them. Parameter attributes are
761 used to communicate additional information about the result or parameters of
762 a function. Parameter attributes are considered to be part of the function
763 type so two functions types that differ only by the parameter attributes
764 are different function types.</p>
766 <p>Parameter attributes are simple keywords that follow the type specified. If
767 multiple parameter attributes are needed, they are space separated. For
770 <div class="doc_code">
772 %someFunc = i16 (i8 signext %someParam) zeroext
773 %someFunc = i16 (i8 zeroext %someParam) zeroext
777 <p>Note that the two function types above are unique because the parameter has
778 a different attribute (<tt>signext</tt> in the first one, <tt>zeroext</tt> in
779 the second). Also note that the attribute for the function result
780 (<tt>zeroext</tt>) comes immediately after the argument list.</p>
782 <p>Currently, only the following parameter attributes are defined:</p>
784 <dt><tt>zeroext</tt></dt>
785 <dd>This indicates that the parameter should be zero extended just before
786 a call to this function.</dd>
787 <dt><tt>signext</tt></dt>
788 <dd>This indicates that the parameter should be sign extended just before
789 a call to this function.</dd>
790 <dt><tt>inreg</tt></dt>
791 <dd>This indicates that the parameter should be placed in register (if
792 possible) during assembling function call. Support for this attribute is
794 <dt><tt>sret</tt></dt>
795 <dd>This indicates that the parameter specifies the address of a structure
796 that is the return value of the function in the source program.</dd>
797 <dt><tt>noalias</tt></dt>
798 <dd>This indicates that the parameter not alias any other object or any
799 other "noalias" objects during the function call.
800 <dt><tt>noreturn</tt></dt>
801 <dd>This function attribute indicates that the function never returns. This
802 indicates to LLVM that every call to this function should be treated as if
803 an <tt>unreachable</tt> instruction immediately followed the call.</dd>
804 <dt><tt>nounwind</tt></dt>
805 <dd>This function attribute indicates that the function type does not use
806 the unwind instruction and does not allow stack unwinding to propagate
812 <!-- ======================================================================= -->
813 <div class="doc_subsection">
814 <a name="moduleasm">Module-Level Inline Assembly</a>
817 <div class="doc_text">
819 Modules may contain "module-level inline asm" blocks, which corresponds to the
820 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
821 LLVM and treated as a single unit, but may be separated in the .ll file if
822 desired. The syntax is very simple:
825 <div class="doc_code">
827 module asm "inline asm code goes here"
828 module asm "more can go here"
832 <p>The strings can contain any character by escaping non-printable characters.
833 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
838 The inline asm code is simply printed to the machine code .s file when
839 assembly code is generated.
843 <!-- ======================================================================= -->
844 <div class="doc_subsection">
845 <a name="datalayout">Data Layout</a>
848 <div class="doc_text">
849 <p>A module may specify a target specific data layout string that specifies how
850 data is to be laid out in memory. The syntax for the data layout is simply:</p>
851 <pre> target datalayout = "<i>layout specification</i>"</pre>
852 <p>The <i>layout specification</i> consists of a list of specifications
853 separated by the minus sign character ('-'). Each specification starts with a
854 letter and may include other information after the letter to define some
855 aspect of the data layout. The specifications accepted are as follows: </p>
858 <dd>Specifies that the target lays out data in big-endian form. That is, the
859 bits with the most significance have the lowest address location.</dd>
861 <dd>Specifies that hte target lays out data in little-endian form. That is,
862 the bits with the least significance have the lowest address location.</dd>
863 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
864 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
865 <i>preferred</i> alignments. All sizes are in bits. Specifying the <i>pref</i>
866 alignment is optional. If omitted, the preceding <tt>:</tt> should be omitted
868 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
869 <dd>This specifies the alignment for an integer type of a given bit
870 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
871 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
872 <dd>This specifies the alignment for a vector type of a given bit
874 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
875 <dd>This specifies the alignment for a floating point type of a given bit
876 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
878 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
879 <dd>This specifies the alignment for an aggregate type of a given bit
882 <p>When constructing the data layout for a given target, LLVM starts with a
883 default set of specifications which are then (possibly) overriden by the
884 specifications in the <tt>datalayout</tt> keyword. The default specifications
885 are given in this list:</p>
887 <li><tt>E</tt> - big endian</li>
888 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
889 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
890 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
891 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
892 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
893 <li><tt>i64:32:64</tt> - i64 has abi alignment of 32-bits but preferred
894 alignment of 64-bits</li>
895 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
896 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
897 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
898 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
899 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
901 <p>When llvm is determining the alignment for a given type, it uses the
904 <li>If the type sought is an exact match for one of the specifications, that
905 specification is used.</li>
906 <li>If no match is found, and the type sought is an integer type, then the
907 smallest integer type that is larger than the bitwidth of the sought type is
908 used. If none of the specifications are larger than the bitwidth then the the
909 largest integer type is used. For example, given the default specifications
910 above, the i7 type will use the alignment of i8 (next largest) while both
911 i65 and i256 will use the alignment of i64 (largest specified).</li>
912 <li>If no match is found, and the type sought is a vector type, then the
913 largest vector type that is smaller than the sought vector type will be used
914 as a fall back. This happens because <128 x double> can be implemented in
915 terms of 64 <2 x double>, for example.</li>
919 <!-- *********************************************************************** -->
920 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
921 <!-- *********************************************************************** -->
923 <div class="doc_text">
925 <p>The LLVM type system is one of the most important features of the
926 intermediate representation. Being typed enables a number of
927 optimizations to be performed on the IR directly, without having to do
928 extra analyses on the side before the transformation. A strong type
929 system makes it easier to read the generated code and enables novel
930 analyses and transformations that are not feasible to perform on normal
931 three address code representations.</p>
935 <!-- ======================================================================= -->
936 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
937 <div class="doc_text">
938 <p>The primitive types are the fundamental building blocks of the LLVM
939 system. The current set of primitive types is as follows:</p>
941 <table class="layout">
946 <tr><th>Type</th><th>Description</th></tr>
947 <tr><td><tt><a name="t_void">void</a></tt></td><td>No value</td></tr>
948 <tr><td><tt>label</tt></td><td>Branch destination</td></tr>
955 <tr><th>Type</th><th>Description</th></tr>
956 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
957 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
965 <!-- _______________________________________________________________________ -->
966 <div class="doc_subsubsection"> <a name="t_classifications">Type
967 Classifications</a> </div>
968 <div class="doc_text">
969 <p>These different primitive types fall into a few useful
972 <table border="1" cellspacing="0" cellpadding="4">
974 <tr><th>Classification</th><th>Types</th></tr>
976 <td><a name="t_integer">integer</a></td>
977 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
980 <td><a name="t_floating">floating point</a></td>
981 <td><tt>float, double</tt></td>
984 <td><a name="t_firstclass">first class</a></td>
985 <td><tt>i1, ..., float, double, <br/>
986 <a href="#t_pointer">pointer</a>,<a href="#t_vector">vector</a></tt>
992 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
993 most important. Values of these types are the only ones which can be
994 produced by instructions, passed as arguments, or used as operands to
995 instructions. This means that all structures and arrays must be
996 manipulated either by pointer or by component.</p>
999 <!-- ======================================================================= -->
1000 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1002 <div class="doc_text">
1004 <p>The real power in LLVM comes from the derived types in the system.
1005 This is what allows a programmer to represent arrays, functions,
1006 pointers, and other useful types. Note that these derived types may be
1007 recursive: For example, it is possible to have a two dimensional array.</p>
1011 <!-- _______________________________________________________________________ -->
1012 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1014 <div class="doc_text">
1017 <p>The integer type is a very simple derived type that simply specifies an
1018 arbitrary bit width for the integer type desired. Any bit width from 1 bit to
1019 2^23-1 (about 8 million) can be specified.</p>
1027 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1031 <table class="layout">
1041 <tt>i1942652</tt><br/>
1044 A boolean integer of 1 bit<br/>
1045 A nibble sized integer of 4 bits.<br/>
1046 A byte sized integer of 8 bits.<br/>
1047 A half word sized integer of 16 bits.<br/>
1048 A word sized integer of 32 bits.<br/>
1049 An integer whose bit width is the answer. <br/>
1050 A double word sized integer of 64 bits.<br/>
1051 A really big integer of over 1 million bits.<br/>
1057 <!-- _______________________________________________________________________ -->
1058 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1060 <div class="doc_text">
1064 <p>The array type is a very simple derived type that arranges elements
1065 sequentially in memory. The array type requires a size (number of
1066 elements) and an underlying data type.</p>
1071 [<# elements> x <elementtype>]
1074 <p>The number of elements is a constant integer value; elementtype may
1075 be any type with a size.</p>
1078 <table class="layout">
1081 <tt>[40 x i32 ]</tt><br/>
1082 <tt>[41 x i32 ]</tt><br/>
1083 <tt>[40 x i8]</tt><br/>
1086 Array of 40 32-bit integer values.<br/>
1087 Array of 41 32-bit integer values.<br/>
1088 Array of 40 8-bit integer values.<br/>
1092 <p>Here are some examples of multidimensional arrays:</p>
1093 <table class="layout">
1096 <tt>[3 x [4 x i32]]</tt><br/>
1097 <tt>[12 x [10 x float]]</tt><br/>
1098 <tt>[2 x [3 x [4 x i16]]]</tt><br/>
1101 3x4 array of 32-bit integer values.<br/>
1102 12x10 array of single precision floating point values.<br/>
1103 2x3x4 array of 16-bit integer values.<br/>
1108 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1109 length array. Normally, accesses past the end of an array are undefined in
1110 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
1111 As a special case, however, zero length arrays are recognized to be variable
1112 length. This allows implementation of 'pascal style arrays' with the LLVM
1113 type "{ i32, [0 x float]}", for example.</p>
1117 <!-- _______________________________________________________________________ -->
1118 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1119 <div class="doc_text">
1121 <p>The function type can be thought of as a function signature. It
1122 consists of a return type and a list of formal parameter types.
1123 Function types are usually used to build virtual function tables
1124 (which are structures of pointers to functions), for indirect function
1125 calls, and when defining a function.</p>
1127 The return type of a function type cannot be an aggregate type.
1130 <pre> <returntype> (<parameter list>)<br></pre>
1131 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1132 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1133 which indicates that the function takes a variable number of arguments.
1134 Variable argument functions can access their arguments with the <a
1135 href="#int_varargs">variable argument handling intrinsic</a> functions.</p>
1137 <table class="layout">
1139 <td class="left"><tt>i32 (i32)</tt></td>
1140 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1142 </tr><tr class="layout">
1143 <td class="left"><tt>float (i16 signext, i32 *) *
1145 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1146 an <tt>i16</tt> that should be sign extended and a
1147 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1150 </tr><tr class="layout">
1151 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1152 <td class="left">A vararg function that takes at least one
1153 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1154 which returns an integer. This is the signature for <tt>printf</tt> in
1161 <!-- _______________________________________________________________________ -->
1162 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1163 <div class="doc_text">
1165 <p>The structure type is used to represent a collection of data members
1166 together in memory. The packing of the field types is defined to match
1167 the ABI of the underlying processor. The elements of a structure may
1168 be any type that has a size.</p>
1169 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1170 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1171 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1174 <pre> { <type list> }<br></pre>
1176 <table class="layout">
1178 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1179 <td class="left">A triple of three <tt>i32</tt> values</td>
1180 </tr><tr class="layout">
1181 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1182 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1183 second element is a <a href="#t_pointer">pointer</a> to a
1184 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1185 an <tt>i32</tt>.</td>
1190 <!-- _______________________________________________________________________ -->
1191 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1193 <div class="doc_text">
1195 <p>The packed structure type is used to represent a collection of data members
1196 together in memory. There is no padding between fields. Further, the alignment
1197 of a packed structure is 1 byte. The elements of a packed structure may
1198 be any type that has a size.</p>
1199 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1200 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1201 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1204 <pre> < { <type list> } > <br></pre>
1206 <table class="layout">
1208 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1209 <td class="left">A triple of three <tt>i32</tt> values</td>
1210 </tr><tr class="layout">
1211 <td class="left"><tt>< { float, i32 (i32) * } ></tt></td>
1212 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1213 second element is a <a href="#t_pointer">pointer</a> to a
1214 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1215 an <tt>i32</tt>.</td>
1220 <!-- _______________________________________________________________________ -->
1221 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1222 <div class="doc_text">
1224 <p>As in many languages, the pointer type represents a pointer or
1225 reference to another object, which must live in memory.</p>
1227 <pre> <type> *<br></pre>
1229 <table class="layout">
1232 <tt>[4x i32]*</tt><br/>
1233 <tt>i32 (i32 *) *</tt><br/>
1236 A <a href="#t_pointer">pointer</a> to <a href="#t_array">array</a> of
1237 four <tt>i32</tt> values<br/>
1238 A <a href="#t_pointer">pointer</a> to a <a
1239 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1246 <!-- _______________________________________________________________________ -->
1247 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1248 <div class="doc_text">
1252 <p>A vector type is a simple derived type that represents a vector
1253 of elements. Vector types are used when multiple primitive data
1254 are operated in parallel using a single instruction (SIMD).
1255 A vector type requires a size (number of
1256 elements) and an underlying primitive data type. Vectors must have a power
1257 of two length (1, 2, 4, 8, 16 ...). Vector types are
1258 considered <a href="#t_firstclass">first class</a>.</p>
1263 < <# elements> x <elementtype> >
1266 <p>The number of elements is a constant integer value; elementtype may
1267 be any integer or floating point type.</p>
1271 <table class="layout">
1274 <tt><4 x i32></tt><br/>
1275 <tt><8 x float></tt><br/>
1276 <tt><2 x i64></tt><br/>
1279 Vector of 4 32-bit integer values.<br/>
1280 Vector of 8 floating-point values.<br/>
1281 Vector of 2 64-bit integer values.<br/>
1287 <!-- _______________________________________________________________________ -->
1288 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1289 <div class="doc_text">
1293 <p>Opaque types are used to represent unknown types in the system. This
1294 corresponds (for example) to the C notion of a foward declared structure type.
1295 In LLVM, opaque types can eventually be resolved to any type (not just a
1296 structure type).</p>
1306 <table class="layout">
1312 An opaque type.<br/>
1319 <!-- *********************************************************************** -->
1320 <div class="doc_section"> <a name="constants">Constants</a> </div>
1321 <!-- *********************************************************************** -->
1323 <div class="doc_text">
1325 <p>LLVM has several different basic types of constants. This section describes
1326 them all and their syntax.</p>
1330 <!-- ======================================================================= -->
1331 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1333 <div class="doc_text">
1336 <dt><b>Boolean constants</b></dt>
1338 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1339 constants of the <tt><a href="#t_primitive">i1</a></tt> type.
1342 <dt><b>Integer constants</b></dt>
1344 <dd>Standard integers (such as '4') are constants of the <a
1345 href="#t_integer">integer</a> type. Negative numbers may be used with
1349 <dt><b>Floating point constants</b></dt>
1351 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1352 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1353 notation (see below). Floating point constants must have a <a
1354 href="#t_floating">floating point</a> type. </dd>
1356 <dt><b>Null pointer constants</b></dt>
1358 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1359 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1363 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1364 of floating point constants. For example, the form '<tt>double
1365 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1366 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1367 (and the only time that they are generated by the disassembler) is when a
1368 floating point constant must be emitted but it cannot be represented as a
1369 decimal floating point number. For example, NaN's, infinities, and other
1370 special values are represented in their IEEE hexadecimal format so that
1371 assembly and disassembly do not cause any bits to change in the constants.</p>
1375 <!-- ======================================================================= -->
1376 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1379 <div class="doc_text">
1380 <p>Aggregate constants arise from aggregation of simple constants
1381 and smaller aggregate constants.</p>
1384 <dt><b>Structure constants</b></dt>
1386 <dd>Structure constants are represented with notation similar to structure
1387 type definitions (a comma separated list of elements, surrounded by braces
1388 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* %G }</tt>",
1389 where "<tt>%G</tt>" is declared as "<tt>@G = external global i32</tt>". Structure constants
1390 must have <a href="#t_struct">structure type</a>, and the number and
1391 types of elements must match those specified by the type.
1394 <dt><b>Array constants</b></dt>
1396 <dd>Array constants are represented with notation similar to array type
1397 definitions (a comma separated list of elements, surrounded by square brackets
1398 (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74 ]</tt>". Array
1399 constants must have <a href="#t_array">array type</a>, and the number and
1400 types of elements must match those specified by the type.
1403 <dt><b>Vector constants</b></dt>
1405 <dd>Vector constants are represented with notation similar to vector type
1406 definitions (a comma separated list of elements, surrounded by
1407 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32 42,
1408 i32 11, i32 74, i32 100 ></tt>". Vector constants must have <a
1409 href="#t_vector">vector type</a>, and the number and types of elements must
1410 match those specified by the type.
1413 <dt><b>Zero initialization</b></dt>
1415 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1416 value to zero of <em>any</em> type, including scalar and aggregate types.
1417 This is often used to avoid having to print large zero initializers (e.g. for
1418 large arrays) and is always exactly equivalent to using explicit zero
1425 <!-- ======================================================================= -->
1426 <div class="doc_subsection">
1427 <a name="globalconstants">Global Variable and Function Addresses</a>
1430 <div class="doc_text">
1432 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1433 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1434 constants. These constants are explicitly referenced when the <a
1435 href="#identifiers">identifier for the global</a> is used and always have <a
1436 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1439 <div class="doc_code">
1443 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1449 <!-- ======================================================================= -->
1450 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1451 <div class="doc_text">
1452 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1453 no specific value. Undefined values may be of any type and be used anywhere
1454 a constant is permitted.</p>
1456 <p>Undefined values indicate to the compiler that the program is well defined
1457 no matter what value is used, giving the compiler more freedom to optimize.
1461 <!-- ======================================================================= -->
1462 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1465 <div class="doc_text">
1467 <p>Constant expressions are used to allow expressions involving other constants
1468 to be used as constants. Constant expressions may be of any <a
1469 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1470 that does not have side effects (e.g. load and call are not supported). The
1471 following is the syntax for constant expressions:</p>
1474 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1475 <dd>Truncate a constant to another type. The bit size of CST must be larger
1476 than the bit size of TYPE. Both types must be integers.</dd>
1478 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1479 <dd>Zero extend a constant to another type. The bit size of CST must be
1480 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1482 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1483 <dd>Sign extend a constant to another type. The bit size of CST must be
1484 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1486 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1487 <dd>Truncate a floating point constant to another floating point type. The
1488 size of CST must be larger than the size of TYPE. Both types must be
1489 floating point.</dd>
1491 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1492 <dd>Floating point extend a constant to another type. The size of CST must be
1493 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1495 <dt><b><tt>fp2uint ( CST to TYPE )</tt></b></dt>
1496 <dd>Convert a floating point constant to the corresponding unsigned integer
1497 constant. TYPE must be an integer type. CST must be floating point. If the
1498 value won't fit in the integer type, the results are undefined.</dd>
1500 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
1501 <dd>Convert a floating point constant to the corresponding signed integer
1502 constant. TYPE must be an integer type. CST must be floating point. If the
1503 value won't fit in the integer type, the results are undefined.</dd>
1505 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
1506 <dd>Convert an unsigned integer constant to the corresponding floating point
1507 constant. TYPE must be floating point. CST must be of integer type. If the
1508 value won't fit in the floating point type, the results are undefined.</dd>
1510 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
1511 <dd>Convert a signed integer constant to the corresponding floating point
1512 constant. TYPE must be floating point. CST must be of integer type. If the
1513 value won't fit in the floating point type, the results are undefined.</dd>
1515 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
1516 <dd>Convert a pointer typed constant to the corresponding integer constant
1517 TYPE must be an integer type. CST must be of pointer type. The CST value is
1518 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1520 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
1521 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1522 pointer type. CST must be of integer type. The CST value is zero extended,
1523 truncated, or unchanged to make it fit in a pointer size. This one is
1524 <i>really</i> dangerous!</dd>
1526 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
1527 <dd>Convert a constant, CST, to another TYPE. The size of CST and TYPE must be
1528 identical (same number of bits). The conversion is done as if the CST value
1529 was stored to memory and read back as TYPE. In other words, no bits change
1530 with this operator, just the type. This can be used for conversion of
1531 vector types to any other type, as long as they have the same bit width. For
1532 pointers it is only valid to cast to another pointer type.
1535 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1537 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1538 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1539 instruction, the index list may have zero or more indexes, which are required
1540 to make sense for the type of "CSTPTR".</dd>
1542 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1544 <dd>Perform the <a href="#i_select">select operation</a> on
1547 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
1548 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
1550 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
1551 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
1553 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1555 <dd>Perform the <a href="#i_extractelement">extractelement
1556 operation</a> on constants.
1558 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1560 <dd>Perform the <a href="#i_insertelement">insertelement
1561 operation</a> on constants.</dd>
1564 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
1566 <dd>Perform the <a href="#i_shufflevector">shufflevector
1567 operation</a> on constants.</dd>
1569 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1571 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1572 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1573 binary</a> operations. The constraints on operands are the same as those for
1574 the corresponding instruction (e.g. no bitwise operations on floating point
1575 values are allowed).</dd>
1579 <!-- *********************************************************************** -->
1580 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1581 <!-- *********************************************************************** -->
1583 <!-- ======================================================================= -->
1584 <div class="doc_subsection">
1585 <a name="inlineasm">Inline Assembler Expressions</a>
1588 <div class="doc_text">
1591 LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
1592 Module-Level Inline Assembly</a>) through the use of a special value. This
1593 value represents the inline assembler as a string (containing the instructions
1594 to emit), a list of operand constraints (stored as a string), and a flag that
1595 indicates whether or not the inline asm expression has side effects. An example
1596 inline assembler expression is:
1599 <div class="doc_code">
1601 i32 (i32) asm "bswap $0", "=r,r"
1606 Inline assembler expressions may <b>only</b> be used as the callee operand of
1607 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
1610 <div class="doc_code">
1612 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
1617 Inline asms with side effects not visible in the constraint list must be marked
1618 as having side effects. This is done through the use of the
1619 '<tt>sideeffect</tt>' keyword, like so:
1622 <div class="doc_code">
1624 call void asm sideeffect "eieio", ""()
1628 <p>TODO: The format of the asm and constraints string still need to be
1629 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1630 need to be documented).
1635 <!-- *********************************************************************** -->
1636 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1637 <!-- *********************************************************************** -->
1639 <div class="doc_text">
1641 <p>The LLVM instruction set consists of several different
1642 classifications of instructions: <a href="#terminators">terminator
1643 instructions</a>, <a href="#binaryops">binary instructions</a>,
1644 <a href="#bitwiseops">bitwise binary instructions</a>, <a
1645 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1646 instructions</a>.</p>
1650 <!-- ======================================================================= -->
1651 <div class="doc_subsection"> <a name="terminators">Terminator
1652 Instructions</a> </div>
1654 <div class="doc_text">
1656 <p>As mentioned <a href="#functionstructure">previously</a>, every
1657 basic block in a program ends with a "Terminator" instruction, which
1658 indicates which block should be executed after the current block is
1659 finished. These terminator instructions typically yield a '<tt>void</tt>'
1660 value: they produce control flow, not values (the one exception being
1661 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1662 <p>There are six different terminator instructions: the '<a
1663 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1664 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1665 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1666 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1667 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1671 <!-- _______________________________________________________________________ -->
1672 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1673 Instruction</a> </div>
1674 <div class="doc_text">
1676 <pre> ret <type> <value> <i>; Return a value from a non-void function</i>
1677 ret void <i>; Return from void function</i>
1680 <p>The '<tt>ret</tt>' instruction is used to return control flow (and a
1681 value) from a function back to the caller.</p>
1682 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
1683 returns a value and then causes control flow, and one that just causes
1684 control flow to occur.</p>
1686 <p>The '<tt>ret</tt>' instruction may return any '<a
1687 href="#t_firstclass">first class</a>' type. Notice that a function is
1688 not <a href="#wellformed">well formed</a> if there exists a '<tt>ret</tt>'
1689 instruction inside of the function that returns a value that does not
1690 match the return type of the function.</p>
1692 <p>When the '<tt>ret</tt>' instruction is executed, control flow
1693 returns back to the calling function's context. If the caller is a "<a
1694 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1695 the instruction after the call. If the caller was an "<a
1696 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1697 at the beginning of the "normal" destination block. If the instruction
1698 returns a value, that value shall set the call or invoke instruction's
1701 <pre> ret i32 5 <i>; Return an integer value of 5</i>
1702 ret void <i>; Return from a void function</i>
1705 <!-- _______________________________________________________________________ -->
1706 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1707 <div class="doc_text">
1709 <pre> br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1712 <p>The '<tt>br</tt>' instruction is used to cause control flow to
1713 transfer to a different basic block in the current function. There are
1714 two forms of this instruction, corresponding to a conditional branch
1715 and an unconditional branch.</p>
1717 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1718 single '<tt>i1</tt>' value and two '<tt>label</tt>' values. The
1719 unconditional form of the '<tt>br</tt>' instruction takes a single
1720 '<tt>label</tt>' value as a target.</p>
1722 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
1723 argument is evaluated. If the value is <tt>true</tt>, control flows
1724 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1725 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1727 <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
1728 href="#i_ret">ret</a> i32 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> i32 0<br></pre>
1730 <!-- _______________________________________________________________________ -->
1731 <div class="doc_subsubsection">
1732 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
1735 <div class="doc_text">
1739 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
1744 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
1745 several different places. It is a generalization of the '<tt>br</tt>'
1746 instruction, allowing a branch to occur to one of many possible
1752 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
1753 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
1754 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
1755 table is not allowed to contain duplicate constant entries.</p>
1759 <p>The <tt>switch</tt> instruction specifies a table of values and
1760 destinations. When the '<tt>switch</tt>' instruction is executed, this
1761 table is searched for the given value. If the value is found, control flow is
1762 transfered to the corresponding destination; otherwise, control flow is
1763 transfered to the default destination.</p>
1765 <h5>Implementation:</h5>
1767 <p>Depending on properties of the target machine and the particular
1768 <tt>switch</tt> instruction, this instruction may be code generated in different
1769 ways. For example, it could be generated as a series of chained conditional
1770 branches or with a lookup table.</p>
1775 <i>; Emulate a conditional br instruction</i>
1776 %Val = <a href="#i_zext">zext</a> i1 %value to i32
1777 switch i32 %Val, label %truedest [i32 0, label %falsedest ]
1779 <i>; Emulate an unconditional br instruction</i>
1780 switch i32 0, label %dest [ ]
1782 <i>; Implement a jump table:</i>
1783 switch i32 %val, label %otherwise [ i32 0, label %onzero
1785 i32 2, label %ontwo ]
1789 <!-- _______________________________________________________________________ -->
1790 <div class="doc_subsubsection">
1791 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
1794 <div class="doc_text">
1799 <result> = invoke [<a href="#callingconv">cconv</a>] <ptr to function ty> %<function ptr val>(<function args>)
1800 to label <normal label> unwind label <exception label>
1805 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
1806 function, with the possibility of control flow transfer to either the
1807 '<tt>normal</tt>' label or the
1808 '<tt>exception</tt>' label. If the callee function returns with the
1809 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
1810 "normal" label. If the callee (or any indirect callees) returns with the "<a
1811 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
1812 continued at the dynamically nearest "exception" label.</p>
1816 <p>This instruction requires several arguments:</p>
1820 The optional "cconv" marker indicates which <a href="#callingconv">calling
1821 convention</a> the call should use. If none is specified, the call defaults
1822 to using C calling conventions.
1824 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
1825 function value being invoked. In most cases, this is a direct function
1826 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
1827 an arbitrary pointer to function value.
1830 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
1831 function to be invoked. </li>
1833 <li>'<tt>function args</tt>': argument list whose types match the function
1834 signature argument types. If the function signature indicates the function
1835 accepts a variable number of arguments, the extra arguments can be
1838 <li>'<tt>normal label</tt>': the label reached when the called function
1839 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
1841 <li>'<tt>exception label</tt>': the label reached when a callee returns with
1842 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
1848 <p>This instruction is designed to operate as a standard '<tt><a
1849 href="#i_call">call</a></tt>' instruction in most regards. The primary
1850 difference is that it establishes an association with a label, which is used by
1851 the runtime library to unwind the stack.</p>
1853 <p>This instruction is used in languages with destructors to ensure that proper
1854 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
1855 exception. Additionally, this is important for implementation of
1856 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
1860 %retval = invoke i32 %Test(i32 15) to label %Continue
1861 unwind label %TestCleanup <i>; {i32}:retval set</i>
1862 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Test(i32 15) to label %Continue
1863 unwind label %TestCleanup <i>; {i32}:retval set</i>
1868 <!-- _______________________________________________________________________ -->
1870 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
1871 Instruction</a> </div>
1873 <div class="doc_text">
1882 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
1883 at the first callee in the dynamic call stack which used an <a
1884 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
1885 primarily used to implement exception handling.</p>
1889 <p>The '<tt>unwind</tt>' intrinsic causes execution of the current function to
1890 immediately halt. The dynamic call stack is then searched for the first <a
1891 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
1892 execution continues at the "exceptional" destination block specified by the
1893 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
1894 dynamic call chain, undefined behavior results.</p>
1897 <!-- _______________________________________________________________________ -->
1899 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
1900 Instruction</a> </div>
1902 <div class="doc_text">
1911 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
1912 instruction is used to inform the optimizer that a particular portion of the
1913 code is not reachable. This can be used to indicate that the code after a
1914 no-return function cannot be reached, and other facts.</p>
1918 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
1923 <!-- ======================================================================= -->
1924 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
1925 <div class="doc_text">
1926 <p>Binary operators are used to do most of the computation in a
1927 program. They require two operands, execute an operation on them, and
1928 produce a single value. The operands might represent
1929 multiple data, as is the case with the <a href="#t_vector">vector</a> data type.
1930 The result value of a binary operator is not
1931 necessarily the same type as its operands.</p>
1932 <p>There are several different binary operators:</p>
1934 <!-- _______________________________________________________________________ -->
1935 <div class="doc_subsubsection"> <a name="i_add">'<tt>add</tt>'
1936 Instruction</a> </div>
1937 <div class="doc_text">
1939 <pre> <result> = add <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1942 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
1944 <p>The two arguments to the '<tt>add</tt>' instruction must be either <a
1945 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a> values.
1946 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
1947 Both arguments must have identical types.</p>
1949 <p>The value produced is the integer or floating point sum of the two
1952 <pre> <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
1955 <!-- _______________________________________________________________________ -->
1956 <div class="doc_subsubsection"> <a name="i_sub">'<tt>sub</tt>'
1957 Instruction</a> </div>
1958 <div class="doc_text">
1960 <pre> <result> = sub <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1963 <p>The '<tt>sub</tt>' instruction returns the difference of its two
1965 <p>Note that the '<tt>sub</tt>' instruction is used to represent the '<tt>neg</tt>'
1966 instruction present in most other intermediate representations.</p>
1968 <p>The two arguments to the '<tt>sub</tt>' instruction must be either <a
1969 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1971 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
1972 Both arguments must have identical types.</p>
1974 <p>The value produced is the integer or floating point difference of
1975 the two operands.</p>
1978 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
1979 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
1982 <!-- _______________________________________________________________________ -->
1983 <div class="doc_subsubsection"> <a name="i_mul">'<tt>mul</tt>'
1984 Instruction</a> </div>
1985 <div class="doc_text">
1987 <pre> <result> = mul <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1990 <p>The '<tt>mul</tt>' instruction returns the product of its two
1993 <p>The two arguments to the '<tt>mul</tt>' instruction must be either <a
1994 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1996 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
1997 Both arguments must have identical types.</p>
1999 <p>The value produced is the integer or floating point product of the
2001 <p>Because the operands are the same width, the result of an integer
2002 multiplication is the same whether the operands should be deemed unsigned or
2005 <pre> <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
2008 <!-- _______________________________________________________________________ -->
2009 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
2011 <div class="doc_text">
2013 <pre> <result> = udiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2016 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two
2019 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
2020 <a href="#t_integer">integer</a> values. Both arguments must have identical
2021 types. This instruction can also take <a href="#t_vector">vector</a> versions
2022 of the values in which case the elements must be integers.</p>
2024 <p>The value produced is the unsigned integer quotient of the two operands. This
2025 instruction always performs an unsigned division operation, regardless of
2026 whether the arguments are unsigned or not.</p>
2028 <pre> <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2031 <!-- _______________________________________________________________________ -->
2032 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
2034 <div class="doc_text">
2036 <pre> <result> = sdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2039 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
2042 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
2043 <a href="#t_integer">integer</a> values. Both arguments must have identical
2044 types. This instruction can also take <a href="#t_vector">vector</a> versions
2045 of the values in which case the elements must be integers.</p>
2047 <p>The value produced is the signed integer quotient of the two operands. This
2048 instruction always performs a signed division operation, regardless of whether
2049 the arguments are signed or not.</p>
2051 <pre> <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2054 <!-- _______________________________________________________________________ -->
2055 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
2056 Instruction</a> </div>
2057 <div class="doc_text">
2059 <pre> <result> = fdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2062 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
2065 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
2066 <a href="#t_floating">floating point</a> values. Both arguments must have
2067 identical types. This instruction can also take <a href="#t_vector">vector</a>
2068 versions of floating point values.</p>
2070 <p>The value produced is the floating point quotient of the two operands.</p>
2072 <pre> <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
2075 <!-- _______________________________________________________________________ -->
2076 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
2078 <div class="doc_text">
2080 <pre> <result> = urem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2083 <p>The '<tt>urem</tt>' instruction returns the remainder from the
2084 unsigned division of its two arguments.</p>
2086 <p>The two arguments to the '<tt>urem</tt>' instruction must be
2087 <a href="#t_integer">integer</a> values. Both arguments must have identical
2090 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
2091 This instruction always performs an unsigned division to get the remainder,
2092 regardless of whether the arguments are unsigned or not.</p>
2094 <pre> <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2098 <!-- _______________________________________________________________________ -->
2099 <div class="doc_subsubsection"> <a name="i_srem">'<tt>srem</tt>'
2100 Instruction</a> </div>
2101 <div class="doc_text">
2103 <pre> <result> = srem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2106 <p>The '<tt>srem</tt>' instruction returns the remainder from the
2107 signed division of its two operands.</p>
2109 <p>The two arguments to the '<tt>srem</tt>' instruction must be
2110 <a href="#t_integer">integer</a> values. Both arguments must have identical
2113 <p>This instruction returns the <i>remainder</i> of a division (where the result
2114 has the same sign as the dividend, <tt>var1</tt>), not the <i>modulo</i>
2115 operator (where the result has the same sign as the divisor, <tt>var2</tt>) of
2116 a value. For more information about the difference, see <a
2117 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
2118 Math Forum</a>. For a table of how this is implemented in various languages,
2119 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
2120 Wikipedia: modulo operation</a>.</p>
2122 <pre> <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2126 <!-- _______________________________________________________________________ -->
2127 <div class="doc_subsubsection"> <a name="i_frem">'<tt>frem</tt>'
2128 Instruction</a> </div>
2129 <div class="doc_text">
2131 <pre> <result> = frem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2134 <p>The '<tt>frem</tt>' instruction returns the remainder from the
2135 division of its two operands.</p>
2137 <p>The two arguments to the '<tt>frem</tt>' instruction must be
2138 <a href="#t_floating">floating point</a> values. Both arguments must have
2139 identical types.</p>
2141 <p>This instruction returns the <i>remainder</i> of a division.</p>
2143 <pre> <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
2147 <!-- ======================================================================= -->
2148 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
2149 Operations</a> </div>
2150 <div class="doc_text">
2151 <p>Bitwise binary operators are used to do various forms of
2152 bit-twiddling in a program. They are generally very efficient
2153 instructions and can commonly be strength reduced from other
2154 instructions. They require two operands, execute an operation on them,
2155 and produce a single value. The resulting value of the bitwise binary
2156 operators is always the same type as its first operand.</p>
2159 <!-- _______________________________________________________________________ -->
2160 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
2161 Instruction</a> </div>
2162 <div class="doc_text">
2164 <pre> <result> = shl <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2167 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2168 the left a specified number of bits.</p>
2170 <p>Both arguments to the '<tt>shl</tt>' instruction must be the same <a
2171 href="#t_integer">integer</a> type.</p>
2173 <p>The value produced is <tt>var1</tt> * 2<sup><tt>var2</tt></sup>.</p>
2174 <h5>Example:</h5><pre>
2175 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
2176 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
2177 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
2180 <!-- _______________________________________________________________________ -->
2181 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
2182 Instruction</a> </div>
2183 <div class="doc_text">
2185 <pre> <result> = lshr <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2189 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
2190 operand shifted to the right a specified number of bits with zero fill.</p>
2193 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
2194 <a href="#t_integer">integer</a> type.</p>
2197 <p>This instruction always performs a logical shift right operation. The most
2198 significant bits of the result will be filled with zero bits after the
2203 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
2204 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
2205 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
2206 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
2210 <!-- _______________________________________________________________________ -->
2211 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
2212 Instruction</a> </div>
2213 <div class="doc_text">
2216 <pre> <result> = ashr <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2220 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
2221 operand shifted to the right a specified number of bits with sign extension.</p>
2224 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
2225 <a href="#t_integer">integer</a> type.</p>
2228 <p>This instruction always performs an arithmetic shift right operation,
2229 The most significant bits of the result will be filled with the sign bit
2230 of <tt>var1</tt>.</p>
2234 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
2235 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
2236 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
2237 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
2241 <!-- _______________________________________________________________________ -->
2242 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
2243 Instruction</a> </div>
2244 <div class="doc_text">
2246 <pre> <result> = and <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2249 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
2250 its two operands.</p>
2252 <p>The two arguments to the '<tt>and</tt>' instruction must be <a
2253 href="#t_integer">integer</a> values. Both arguments must have
2254 identical types.</p>
2256 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
2258 <div style="align: center">
2259 <table border="1" cellspacing="0" cellpadding="4">
2290 <pre> <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
2291 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
2292 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
2295 <!-- _______________________________________________________________________ -->
2296 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
2297 <div class="doc_text">
2299 <pre> <result> = or <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2302 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
2303 or of its two operands.</p>
2305 <p>The two arguments to the '<tt>or</tt>' instruction must be <a
2306 href="#t_integer">integer</a> values. Both arguments must have
2307 identical types.</p>
2309 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
2311 <div style="align: center">
2312 <table border="1" cellspacing="0" cellpadding="4">
2343 <pre> <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
2344 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
2345 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
2348 <!-- _______________________________________________________________________ -->
2349 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
2350 Instruction</a> </div>
2351 <div class="doc_text">
2353 <pre> <result> = xor <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2356 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
2357 or of its two operands. The <tt>xor</tt> is used to implement the
2358 "one's complement" operation, which is the "~" operator in C.</p>
2360 <p>The two arguments to the '<tt>xor</tt>' instruction must be <a
2361 href="#t_integer">integer</a> values. Both arguments must have
2362 identical types.</p>
2364 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
2366 <div style="align: center">
2367 <table border="1" cellspacing="0" cellpadding="4">
2399 <pre> <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
2400 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
2401 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
2402 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
2406 <!-- ======================================================================= -->
2407 <div class="doc_subsection">
2408 <a name="vectorops">Vector Operations</a>
2411 <div class="doc_text">
2413 <p>LLVM supports several instructions to represent vector operations in a
2414 target-independent manner. These instructions cover the element-access and
2415 vector-specific operations needed to process vectors effectively. While LLVM
2416 does directly support these vector operations, many sophisticated algorithms
2417 will want to use target-specific intrinsics to take full advantage of a specific
2422 <!-- _______________________________________________________________________ -->
2423 <div class="doc_subsubsection">
2424 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2427 <div class="doc_text">
2432 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
2438 The '<tt>extractelement</tt>' instruction extracts a single scalar
2439 element from a vector at a specified index.
2446 The first operand of an '<tt>extractelement</tt>' instruction is a
2447 value of <a href="#t_vector">vector</a> type. The second operand is
2448 an index indicating the position from which to extract the element.
2449 The index may be a variable.</p>
2454 The result is a scalar of the same type as the element type of
2455 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2456 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2457 results are undefined.
2463 %result = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
2468 <!-- _______________________________________________________________________ -->
2469 <div class="doc_subsubsection">
2470 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
2473 <div class="doc_text">
2478 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
2484 The '<tt>insertelement</tt>' instruction inserts a scalar
2485 element into a vector at a specified index.
2492 The first operand of an '<tt>insertelement</tt>' instruction is a
2493 value of <a href="#t_vector">vector</a> type. The second operand is a
2494 scalar value whose type must equal the element type of the first
2495 operand. The third operand is an index indicating the position at
2496 which to insert the value. The index may be a variable.</p>
2501 The result is a vector of the same type as <tt>val</tt>. Its
2502 element values are those of <tt>val</tt> except at position
2503 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
2504 exceeds the length of <tt>val</tt>, the results are undefined.
2510 %result = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
2514 <!-- _______________________________________________________________________ -->
2515 <div class="doc_subsubsection">
2516 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
2519 <div class="doc_text">
2524 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <n x i32> <mask> <i>; yields <n x <ty>></i>
2530 The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
2531 from two input vectors, returning a vector of the same type.
2537 The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
2538 with types that match each other and types that match the result of the
2539 instruction. The third argument is a shuffle mask, which has the same number
2540 of elements as the other vector type, but whose element type is always 'i32'.
2544 The shuffle mask operand is required to be a constant vector with either
2545 constant integer or undef values.
2551 The elements of the two input vectors are numbered from left to right across
2552 both of the vectors. The shuffle mask operand specifies, for each element of
2553 the result vector, which element of the two input registers the result element
2554 gets. The element selector may be undef (meaning "don't care") and the second
2555 operand may be undef if performing a shuffle from only one vector.
2561 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
2562 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
2563 %result = shufflevector <4 x i32> %v1, <4 x i32> undef,
2564 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
2569 <!-- ======================================================================= -->
2570 <div class="doc_subsection">
2571 <a name="memoryops">Memory Access and Addressing Operations</a>
2574 <div class="doc_text">
2576 <p>A key design point of an SSA-based representation is how it
2577 represents memory. In LLVM, no memory locations are in SSA form, which
2578 makes things very simple. This section describes how to read, write,
2579 allocate, and free memory in LLVM.</p>
2583 <!-- _______________________________________________________________________ -->
2584 <div class="doc_subsubsection">
2585 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
2588 <div class="doc_text">
2593 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2598 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
2599 heap and returns a pointer to it.</p>
2603 <p>The '<tt>malloc</tt>' instruction allocates
2604 <tt>sizeof(<type>)*NumElements</tt>
2605 bytes of memory from the operating system and returns a pointer of the
2606 appropriate type to the program. If "NumElements" is specified, it is the
2607 number of elements allocated. If an alignment is specified, the value result
2608 of the allocation is guaranteed to be aligned to at least that boundary. If
2609 not specified, or if zero, the target can choose to align the allocation on any
2610 convenient boundary.</p>
2612 <p>'<tt>type</tt>' must be a sized type.</p>
2616 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
2617 a pointer is returned.</p>
2622 %array = malloc [4 x i8 ] <i>; yields {[%4 x i8]*}:array</i>
2624 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
2625 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
2626 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
2627 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
2628 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
2632 <!-- _______________________________________________________________________ -->
2633 <div class="doc_subsubsection">
2634 <a name="i_free">'<tt>free</tt>' Instruction</a>
2637 <div class="doc_text">
2642 free <type> <value> <i>; yields {void}</i>
2647 <p>The '<tt>free</tt>' instruction returns memory back to the unused
2648 memory heap to be reallocated in the future.</p>
2652 <p>'<tt>value</tt>' shall be a pointer value that points to a value
2653 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
2658 <p>Access to the memory pointed to by the pointer is no longer defined
2659 after this instruction executes.</p>
2664 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
2665 free [4 x i8]* %array
2669 <!-- _______________________________________________________________________ -->
2670 <div class="doc_subsubsection">
2671 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
2674 <div class="doc_text">
2679 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2684 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
2685 currently executing function, to be automatically released when this function
2686 returns to its caller.</p>
2690 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
2691 bytes of memory on the runtime stack, returning a pointer of the
2692 appropriate type to the program. If "NumElements" is specified, it is the
2693 number of elements allocated. If an alignment is specified, the value result
2694 of the allocation is guaranteed to be aligned to at least that boundary. If
2695 not specified, or if zero, the target can choose to align the allocation on any
2696 convenient boundary.</p>
2698 <p>'<tt>type</tt>' may be any sized type.</p>
2702 <p>Memory is allocated; a pointer is returned. '<tt>alloca</tt>'d
2703 memory is automatically released when the function returns. The '<tt>alloca</tt>'
2704 instruction is commonly used to represent automatic variables that must
2705 have an address available. When the function returns (either with the <tt><a
2706 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
2707 instructions), the memory is reclaimed.</p>
2712 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
2713 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
2714 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
2715 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
2719 <!-- _______________________________________________________________________ -->
2720 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
2721 Instruction</a> </div>
2722 <div class="doc_text">
2724 <pre> <result> = load <ty>* <pointer>[, align <alignment>]<br> <result> = volatile load <ty>* <pointer>[, align <alignment>]<br></pre>
2726 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
2728 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
2729 address from which to load. The pointer must point to a <a
2730 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
2731 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
2732 the number or order of execution of this <tt>load</tt> with other
2733 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
2736 <p>The location of memory pointed to is loaded.</p>
2738 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
2740 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
2741 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
2744 <!-- _______________________________________________________________________ -->
2745 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
2746 Instruction</a> </div>
2747 <div class="doc_text">
2749 <pre> store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
2750 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
2753 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
2755 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
2756 to store and an address at which to store it. The type of the '<tt><pointer></tt>'
2757 operand must be a pointer to the type of the '<tt><value></tt>'
2758 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
2759 optimizer is not allowed to modify the number or order of execution of
2760 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
2761 href="#i_store">store</a></tt> instructions.</p>
2763 <p>The contents of memory are updated to contain '<tt><value></tt>'
2764 at the location specified by the '<tt><pointer></tt>' operand.</p>
2766 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
2768 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
2769 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
2773 <!-- _______________________________________________________________________ -->
2774 <div class="doc_subsubsection">
2775 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
2778 <div class="doc_text">
2781 <result> = getelementptr <ty>* <ptrval>{, <ty> <idx>}*
2787 The '<tt>getelementptr</tt>' instruction is used to get the address of a
2788 subelement of an aggregate data structure.</p>
2792 <p>This instruction takes a list of integer operands that indicate what
2793 elements of the aggregate object to index to. The actual types of the arguments
2794 provided depend on the type of the first pointer argument. The
2795 '<tt>getelementptr</tt>' instruction is used to index down through the type
2796 levels of a structure or to a specific index in an array. When indexing into a
2797 structure, only <tt>i32</tt> integer constants are allowed. When indexing
2798 into an array or pointer, only integers of 32 or 64 bits are allowed, and will
2799 be sign extended to 64-bit values.</p>
2801 <p>For example, let's consider a C code fragment and how it gets
2802 compiled to LLVM:</p>
2804 <div class="doc_code">
2817 int *foo(struct ST *s) {
2818 return &s[1].Z.B[5][13];
2823 <p>The LLVM code generated by the GCC frontend is:</p>
2825 <div class="doc_code">
2827 %RT = type { i8 , [10 x [20 x i32]], i8 }
2828 %ST = type { i32, double, %RT }
2830 define i32* %foo(%ST* %s) {
2832 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
2840 <p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
2841 on the pointer type that is being indexed into. <a href="#t_pointer">Pointer</a>
2842 and <a href="#t_array">array</a> types can use a 32-bit or 64-bit
2843 <a href="#t_integer">integer</a> type but the value will always be sign extended
2844 to 64-bits. <a href="#t_struct">Structure</a> types require <tt>i32</tt>
2845 <b>constants</b>.</p>
2847 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
2848 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
2849 }</tt>' type, a structure. The second index indexes into the third element of
2850 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
2851 i8 }</tt>' type, another structure. The third index indexes into the second
2852 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
2853 array. The two dimensions of the array are subscripted into, yielding an
2854 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
2855 to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
2857 <p>Note that it is perfectly legal to index partially through a
2858 structure, returning a pointer to an inner element. Because of this,
2859 the LLVM code for the given testcase is equivalent to:</p>
2862 define i32* %foo(%ST* %s) {
2863 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
2864 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
2865 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
2866 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
2867 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
2872 <p>Note that it is undefined to access an array out of bounds: array and
2873 pointer indexes must always be within the defined bounds of the array type.
2874 The one exception for this rules is zero length arrays. These arrays are
2875 defined to be accessible as variable length arrays, which requires access
2876 beyond the zero'th element.</p>
2878 <p>The getelementptr instruction is often confusing. For some more insight
2879 into how it works, see <a href="GetElementPtr.html">the getelementptr
2885 <i>; yields [12 x i8]*:aptr</i>
2886 %aptr = getelementptr {i32, [12 x i8]}* %sptr, i64 0, i32 1
2890 <!-- ======================================================================= -->
2891 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
2893 <div class="doc_text">
2894 <p>The instructions in this category are the conversion instructions (casting)
2895 which all take a single operand and a type. They perform various bit conversions
2899 <!-- _______________________________________________________________________ -->
2900 <div class="doc_subsubsection">
2901 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
2903 <div class="doc_text">
2907 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
2912 The '<tt>trunc</tt>' instruction truncates its operand to the type <tt>ty2</tt>.
2917 The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
2918 be an <a href="#t_integer">integer</a> type, and a type that specifies the size
2919 and type of the result, which must be an <a href="#t_integer">integer</a>
2920 type. The bit size of <tt>value</tt> must be larger than the bit size of
2921 <tt>ty2</tt>. Equal sized types are not allowed.</p>
2925 The '<tt>trunc</tt>' instruction truncates the high order bits in <tt>value</tt>
2926 and converts the remaining bits to <tt>ty2</tt>. Since the source size must be
2927 larger than the destination size, <tt>trunc</tt> cannot be a <i>no-op cast</i>.
2928 It will always truncate bits.</p>
2932 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
2933 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
2934 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
2938 <!-- _______________________________________________________________________ -->
2939 <div class="doc_subsubsection">
2940 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
2942 <div class="doc_text">
2946 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
2950 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
2955 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
2956 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
2957 also be of <a href="#t_integer">integer</a> type. The bit size of the
2958 <tt>value</tt> must be smaller than the bit size of the destination type,
2962 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
2963 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
2965 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
2969 %X = zext i32 257 to i64 <i>; yields i64:257</i>
2970 %Y = zext i1 true to i32 <i>; yields i32:1</i>
2974 <!-- _______________________________________________________________________ -->
2975 <div class="doc_subsubsection">
2976 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
2978 <div class="doc_text">
2982 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
2986 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
2990 The '<tt>sext</tt>' instruction takes a value to cast, which must be of
2991 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
2992 also be of <a href="#t_integer">integer</a> type. The bit size of the
2993 <tt>value</tt> must be smaller than the bit size of the destination type,
2998 The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
2999 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size of
3000 the type <tt>ty2</tt>.</p>
3002 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
3006 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
3007 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
3011 <!-- _______________________________________________________________________ -->
3012 <div class="doc_subsubsection">
3013 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
3016 <div class="doc_text">
3021 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
3025 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
3030 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
3031 point</a> value to cast and a <a href="#t_floating">floating point</a> type to
3032 cast it to. The size of <tt>value</tt> must be larger than the size of
3033 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
3034 <i>no-op cast</i>.</p>
3037 <p> The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
3038 <a href="#t_floating">floating point</a> type to a smaller
3039 <a href="#t_floating">floating point</a> type. If the value cannot fit within
3040 the destination type, <tt>ty2</tt>, then the results are undefined.</p>
3044 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
3045 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
3049 <!-- _______________________________________________________________________ -->
3050 <div class="doc_subsubsection">
3051 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
3053 <div class="doc_text">
3057 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
3061 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
3062 floating point value.</p>
3065 <p>The '<tt>fpext</tt>' instruction takes a
3066 <a href="#t_floating">floating point</a> <tt>value</tt> to cast,
3067 and a <a href="#t_floating">floating point</a> type to cast it to. The source
3068 type must be smaller than the destination type.</p>
3071 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
3072 <a href="#t_floating">floating point</a> type to a larger
3073 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
3074 used to make a <i>no-op cast</i> because it always changes bits. Use
3075 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
3079 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
3080 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
3084 <!-- _______________________________________________________________________ -->
3085 <div class="doc_subsubsection">
3086 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
3088 <div class="doc_text">
3092 <result> = fp2uint <ty> <value> to <ty2> <i>; yields ty2</i>
3096 <p>The '<tt>fp2uint</tt>' converts a floating point <tt>value</tt> to its
3097 unsigned integer equivalent of type <tt>ty2</tt>.
3101 <p>The '<tt>fp2uint</tt>' instruction takes a value to cast, which must be a
3102 <a href="#t_floating">floating point</a> value, and a type to cast it to, which
3103 must be an <a href="#t_integer">integer</a> type.</p>
3106 <p> The '<tt>fp2uint</tt>' instruction converts its
3107 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3108 towards zero) unsigned integer value. If the value cannot fit in <tt>ty2</tt>,
3109 the results are undefined.</p>
3111 <p>When converting to i1, the conversion is done as a comparison against
3112 zero. If the <tt>value</tt> was zero, the i1 result will be <tt>false</tt>.
3113 If the <tt>value</tt> was non-zero, the i1 result will be <tt>true</tt>.</p>
3117 %X = fp2uint double 123.0 to i32 <i>; yields i32:123</i>
3118 %Y = fp2uint float 1.0E+300 to i1 <i>; yields i1:true</i>
3119 %X = fp2uint float 1.04E+17 to i8 <i>; yields undefined:1</i>
3123 <!-- _______________________________________________________________________ -->
3124 <div class="doc_subsubsection">
3125 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
3127 <div class="doc_text">
3131 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
3135 <p>The '<tt>fptosi</tt>' instruction converts
3136 <a href="#t_floating">floating point</a> <tt>value</tt> to type <tt>ty2</tt>.
3141 <p> The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
3142 <a href="#t_floating">floating point</a> value, and a type to cast it to, which
3143 must also be an <a href="#t_integer">integer</a> type.</p>
3146 <p>The '<tt>fptosi</tt>' instruction converts its
3147 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3148 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
3149 the results are undefined.</p>
3151 <p>When converting to i1, the conversion is done as a comparison against
3152 zero. If the <tt>value</tt> was zero, the i1 result will be <tt>false</tt>.
3153 If the <tt>value</tt> was non-zero, the i1 result will be <tt>true</tt>.</p>
3157 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
3158 %Y = fptosi float 1.0E-247 to i1 <i>; yields i1:true</i>
3159 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
3163 <!-- _______________________________________________________________________ -->
3164 <div class="doc_subsubsection">
3165 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
3167 <div class="doc_text">
3171 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3175 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
3176 integer and converts that value to the <tt>ty2</tt> type.</p>
3180 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be an
3181 <a href="#t_integer">integer</a> value, and a type to cast it to, which must
3182 be a <a href="#t_floating">floating point</a> type.</p>
3185 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
3186 integer quantity and converts it to the corresponding floating point value. If
3187 the value cannot fit in the floating point value, the results are undefined.</p>
3192 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
3193 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
3197 <!-- _______________________________________________________________________ -->
3198 <div class="doc_subsubsection">
3199 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
3201 <div class="doc_text">
3205 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3209 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed
3210 integer and converts that value to the <tt>ty2</tt> type.</p>
3213 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be an
3214 <a href="#t_integer">integer</a> value, and a type to cast it to, which must be
3215 a <a href="#t_floating">floating point</a> type.</p>
3218 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed
3219 integer quantity and converts it to the corresponding floating point value. If
3220 the value cannot fit in the floating point value, the results are undefined.</p>
3224 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
3225 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
3229 <!-- _______________________________________________________________________ -->
3230 <div class="doc_subsubsection">
3231 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
3233 <div class="doc_text">
3237 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
3241 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
3242 the integer type <tt>ty2</tt>.</p>
3245 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
3246 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
3247 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.
3250 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
3251 <tt>ty2</tt> by interpreting the pointer value as an integer and either
3252 truncating or zero extending that value to the size of the integer type. If
3253 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
3254 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
3255 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
3260 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
3261 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
3265 <!-- _______________________________________________________________________ -->
3266 <div class="doc_subsubsection">
3267 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
3269 <div class="doc_text">
3273 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
3277 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to
3278 a pointer type, <tt>ty2</tt>.</p>
3281 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
3282 value to cast, and a type to cast it to, which must be a
3283 <a href="#t_pointer">pointer</a> type.
3286 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
3287 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
3288 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
3289 size of a pointer then a truncation is done. If <tt>value</tt> is smaller than
3290 the size of a pointer then a zero extension is done. If they are the same size,
3291 nothing is done (<i>no-op cast</i>).</p>
3295 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
3296 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
3297 %Y = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
3301 <!-- _______________________________________________________________________ -->
3302 <div class="doc_subsubsection">
3303 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
3305 <div class="doc_text">
3309 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
3313 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3314 <tt>ty2</tt> without changing any bits.</p>
3317 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be
3318 a first class value, and a type to cast it to, which must also be a <a
3319 href="#t_firstclass">first class</a> type. The bit sizes of <tt>value</tt>
3320 and the destination type, <tt>ty2</tt>, must be identical. If the source
3321 type is a pointer, the destination type must also be a pointer.</p>
3324 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3325 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
3326 this conversion. The conversion is done as if the <tt>value</tt> had been
3327 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only be
3328 converted to other pointer types with this instruction. To convert pointers to
3329 other types, use the <a href="#i_inttoptr">inttoptr</a> or
3330 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
3334 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
3335 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
3336 %Z = bitcast <2xint> %V to i64; <i>; yields i64: %V</i>
3340 <!-- ======================================================================= -->
3341 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
3342 <div class="doc_text">
3343 <p>The instructions in this category are the "miscellaneous"
3344 instructions, which defy better classification.</p>
3347 <!-- _______________________________________________________________________ -->
3348 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
3350 <div class="doc_text">
3352 <pre> <result> = icmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3355 <p>The '<tt>icmp</tt>' instruction returns a boolean value based on comparison
3356 of its two integer operands.</p>
3358 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
3359 the condition code indicating the kind of comparison to perform. It is not
3360 a value, just a keyword. The possible condition code are:
3362 <li><tt>eq</tt>: equal</li>
3363 <li><tt>ne</tt>: not equal </li>
3364 <li><tt>ugt</tt>: unsigned greater than</li>
3365 <li><tt>uge</tt>: unsigned greater or equal</li>
3366 <li><tt>ult</tt>: unsigned less than</li>
3367 <li><tt>ule</tt>: unsigned less or equal</li>
3368 <li><tt>sgt</tt>: signed greater than</li>
3369 <li><tt>sge</tt>: signed greater or equal</li>
3370 <li><tt>slt</tt>: signed less than</li>
3371 <li><tt>sle</tt>: signed less or equal</li>
3373 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
3374 <a href="#t_pointer">pointer</a> typed. They must also be identical types.</p>
3376 <p>The '<tt>icmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3377 the condition code given as <tt>cond</tt>. The comparison performed always
3378 yields a <a href="#t_primitive">i1</a> result, as follows:
3380 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
3381 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3383 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
3384 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3385 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
3386 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3387 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
3388 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3389 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
3390 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3391 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
3392 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3393 <li><tt>sgt</tt>: interprets the operands as signed values and yields
3394 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3395 <li><tt>sge</tt>: interprets the operands as signed values and yields
3396 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3397 <li><tt>slt</tt>: interprets the operands as signed values and yields
3398 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3399 <li><tt>sle</tt>: interprets the operands as signed values and yields
3400 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3402 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
3403 values are compared as if they were integers.</p>
3406 <pre> <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
3407 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
3408 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
3409 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
3410 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
3411 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
3415 <!-- _______________________________________________________________________ -->
3416 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
3418 <div class="doc_text">
3420 <pre> <result> = fcmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3423 <p>The '<tt>fcmp</tt>' instruction returns a boolean value based on comparison
3424 of its floating point operands.</p>
3426 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
3427 the condition code indicating the kind of comparison to perform. It is not
3428 a value, just a keyword. The possible condition code are:
3430 <li><tt>false</tt>: no comparison, always returns false</li>
3431 <li><tt>oeq</tt>: ordered and equal</li>
3432 <li><tt>ogt</tt>: ordered and greater than </li>
3433 <li><tt>oge</tt>: ordered and greater than or equal</li>
3434 <li><tt>olt</tt>: ordered and less than </li>
3435 <li><tt>ole</tt>: ordered and less than or equal</li>
3436 <li><tt>one</tt>: ordered and not equal</li>
3437 <li><tt>ord</tt>: ordered (no nans)</li>
3438 <li><tt>ueq</tt>: unordered or equal</li>
3439 <li><tt>ugt</tt>: unordered or greater than </li>
3440 <li><tt>uge</tt>: unordered or greater than or equal</li>
3441 <li><tt>ult</tt>: unordered or less than </li>
3442 <li><tt>ule</tt>: unordered or less than or equal</li>
3443 <li><tt>une</tt>: unordered or not equal</li>
3444 <li><tt>uno</tt>: unordered (either nans)</li>
3445 <li><tt>true</tt>: no comparison, always returns true</li>
3447 <p><i>Ordered</i> means that neither operand is a QNAN while
3448 <i>unordered</i> means that either operand may be a QNAN.</p>
3449 <p>The <tt>val1</tt> and <tt>val2</tt> arguments must be
3450 <a href="#t_floating">floating point</a> typed. They must have identical
3453 <p>The '<tt>fcmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3454 the condition code given as <tt>cond</tt>. The comparison performed always
3455 yields a <a href="#t_primitive">i1</a> result, as follows:
3457 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
3458 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3459 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3460 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3461 <tt>var1</tt> is greather than <tt>var2</tt>.</li>
3462 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3463 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3464 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3465 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3466 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3467 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3468 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3469 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3470 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
3471 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
3472 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3473 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
3474 <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3475 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
3476 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3477 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
3478 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3479 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
3480 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3481 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
3482 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3483 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
3484 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
3488 <pre> <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
3489 <result> = icmp one float 4.0, 5.0 <i>; yields: result=true</i>
3490 <result> = icmp olt float 4.0, 5.0 <i>; yields: result=true</i>
3491 <result> = icmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
3495 <!-- _______________________________________________________________________ -->
3496 <div class="doc_subsubsection"> <a name="i_phi">'<tt>phi</tt>'
3497 Instruction</a> </div>
3498 <div class="doc_text">
3500 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
3502 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
3503 the SSA graph representing the function.</p>
3505 <p>The type of the incoming values is specified with the first type
3506 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
3507 as arguments, with one pair for each predecessor basic block of the
3508 current block. Only values of <a href="#t_firstclass">first class</a>
3509 type may be used as the value arguments to the PHI node. Only labels
3510 may be used as the label arguments.</p>
3511 <p>There must be no non-phi instructions between the start of a basic
3512 block and the PHI instructions: i.e. PHI instructions must be first in
3515 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
3516 specified by the pair corresponding to the predecessor basic block that executed
3517 just prior to the current block.</p>
3519 <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>
3522 <!-- _______________________________________________________________________ -->
3523 <div class="doc_subsubsection">
3524 <a name="i_select">'<tt>select</tt>' Instruction</a>
3527 <div class="doc_text">
3532 <result> = select i1 <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
3538 The '<tt>select</tt>' instruction is used to choose one value based on a
3539 condition, without branching.
3546 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.
3552 If the boolean condition evaluates to true, the instruction returns the first
3553 value argument; otherwise, it returns the second value argument.
3559 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
3564 <!-- _______________________________________________________________________ -->
3565 <div class="doc_subsubsection">
3566 <a name="i_call">'<tt>call</tt>' Instruction</a>
3569 <div class="doc_text">
3573 <result> = [tail] call [<a href="#callingconv">cconv</a>] <ty>* <fnptrval>(<param list>)
3578 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
3582 <p>This instruction requires several arguments:</p>
3586 <p>The optional "tail" marker indicates whether the callee function accesses
3587 any allocas or varargs in the caller. If the "tail" marker is present, the
3588 function call is eligible for tail call optimization. Note that calls may
3589 be marked "tail" even if they do not occur before a <a
3590 href="#i_ret"><tt>ret</tt></a> instruction.
3593 <p>The optional "cconv" marker indicates which <a href="#callingconv">calling
3594 convention</a> the call should use. If none is specified, the call defaults
3595 to using C calling conventions.
3598 <p>'<tt>ty</tt>': shall be the signature of the pointer to function value
3599 being invoked. The argument types must match the types implied by this
3600 signature. This type can be omitted if the function is not varargs and
3601 if the function type does not return a pointer to a function.</p>
3604 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
3605 be invoked. In most cases, this is a direct function invocation, but
3606 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
3607 to function value.</p>
3610 <p>'<tt>function args</tt>': argument list whose types match the
3611 function signature argument types. All arguments must be of
3612 <a href="#t_firstclass">first class</a> type. If the function signature
3613 indicates the function accepts a variable number of arguments, the extra
3614 arguments can be specified.</p>
3620 <p>The '<tt>call</tt>' instruction is used to cause control flow to
3621 transfer to a specified function, with its incoming arguments bound to
3622 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
3623 instruction in the called function, control flow continues with the
3624 instruction after the function call, and the return value of the
3625 function is bound to the result argument. This is a simpler case of
3626 the <a href="#i_invoke">invoke</a> instruction.</p>
3631 %retval = call i32 %test(i32 %argc)
3632 call i32(i8 *, ...) *%printf(i8 * %msg, i32 12, i8 42);
3633 %X = tail call i32 %foo()
3634 %Y = tail call <a href="#callingconv">fastcc</a> i32 %foo()
3639 <!-- _______________________________________________________________________ -->
3640 <div class="doc_subsubsection">
3641 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
3644 <div class="doc_text">
3649 <resultval> = va_arg <va_list*> <arglist>, <argty>
3654 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
3655 the "variable argument" area of a function call. It is used to implement the
3656 <tt>va_arg</tt> macro in C.</p>
3660 <p>This instruction takes a <tt>va_list*</tt> value and the type of
3661 the argument. It returns a value of the specified argument type and
3662 increments the <tt>va_list</tt> to point to the next argument. The
3663 actual type of <tt>va_list</tt> is target specific.</p>
3667 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
3668 type from the specified <tt>va_list</tt> and causes the
3669 <tt>va_list</tt> to point to the next argument. For more information,
3670 see the variable argument handling <a href="#int_varargs">Intrinsic
3673 <p>It is legal for this instruction to be called in a function which does not
3674 take a variable number of arguments, for example, the <tt>vfprintf</tt>
3677 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
3678 href="#intrinsics">intrinsic function</a> because it takes a type as an
3683 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
3687 <!-- *********************************************************************** -->
3688 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
3689 <!-- *********************************************************************** -->
3691 <div class="doc_text">
3693 <p>LLVM supports the notion of an "intrinsic function". These functions have
3694 well known names and semantics and are required to follow certain restrictions.
3695 Overall, these intrinsics represent an extension mechanism for the LLVM
3696 language that does not require changing all of the transformations in LLVM when
3697 adding to the language (or the bitcode reader/writer, the parser, etc...).</p>
3699 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
3700 prefix is reserved in LLVM for intrinsic names; thus, function names may not
3701 begin with this prefix. Intrinsic functions must always be external functions:
3702 you cannot define the body of intrinsic functions. Intrinsic functions may
3703 only be used in call or invoke instructions: it is illegal to take the address
3704 of an intrinsic function. Additionally, because intrinsic functions are part
3705 of the LLVM language, it is required if any are added that they be documented
3708 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
3709 a family of functions that perform the same operation but on different data
3710 types. This is most frequent with the integer types. Since LLVM can represent
3711 over 8 million different integer types, there is a way to declare an intrinsic
3712 that can be overloaded based on its arguments. Such an intrinsic will have the
3713 names of its argument types encoded into its function name, each
3714 preceded by a period. For example, the <tt>llvm.ctpop</tt> function can take an
3715 integer of any width. This leads to a family of functions such as
3716 <tt>i32 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i32 @llvm.ctpop.i29(i29 %val)</tt>.
3720 <p>To learn how to add an intrinsic function, please see the
3721 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.
3726 <!-- ======================================================================= -->
3727 <div class="doc_subsection">
3728 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
3731 <div class="doc_text">
3733 <p>Variable argument support is defined in LLVM with the <a
3734 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
3735 intrinsic functions. These functions are related to the similarly
3736 named macros defined in the <tt><stdarg.h></tt> header file.</p>
3738 <p>All of these functions operate on arguments that use a
3739 target-specific value type "<tt>va_list</tt>". The LLVM assembly
3740 language reference manual does not define what this type is, so all
3741 transformations should be prepared to handle these functions regardless of
3744 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
3745 instruction and the variable argument handling intrinsic functions are
3748 <div class="doc_code">
3750 define i32 @test(i32 %X, ...) {
3751 ; Initialize variable argument processing
3753 %ap2 = bitcast i8** %ap to i8*
3754 call void @llvm.va_start(i8* %ap2)
3756 ; Read a single integer argument
3757 %tmp = va_arg i8** %ap, i32
3759 ; Demonstrate usage of llvm.va_copy and llvm.va_end
3761 %aq2 = bitcast i8** %aq to i8*
3762 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
3763 call void @llvm.va_end(i8* %aq2)
3765 ; Stop processing of arguments.
3766 call void @llvm.va_end(i8* %ap2)
3770 declare void @llvm.va_start(i8*)
3771 declare void @llvm.va_copy(i8*, i8*)
3772 declare void @llvm.va_end(i8*)
3778 <!-- _______________________________________________________________________ -->
3779 <div class="doc_subsubsection">
3780 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
3784 <div class="doc_text">
3786 <pre> declare void %llvm.va_start(i8* <arglist>)<br></pre>
3788 <P>The '<tt>llvm.va_start</tt>' intrinsic initializes
3789 <tt>*<arglist></tt> for subsequent use by <tt><a
3790 href="#i_va_arg">va_arg</a></tt>.</p>
3794 <P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
3798 <P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
3799 macro available in C. In a target-dependent way, it initializes the
3800 <tt>va_list</tt> element to which the argument points, so that the next call to
3801 <tt>va_arg</tt> will produce the first variable argument passed to the function.
3802 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
3803 last argument of the function as the compiler can figure that out.</p>
3807 <!-- _______________________________________________________________________ -->
3808 <div class="doc_subsubsection">
3809 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
3812 <div class="doc_text">
3814 <pre> declare void @llvm.va_end(i8* <arglist>)<br></pre>
3817 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
3818 which has been initialized previously with <tt><a href="#int_va_start">llvm.va_start</a></tt>
3819 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
3823 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
3827 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
3828 macro available in C. In a target-dependent way, it destroys the
3829 <tt>va_list</tt> element to which the argument points. Calls to <a
3830 href="#int_va_start"><tt>llvm.va_start</tt></a> and <a href="#int_va_copy">
3831 <tt>llvm.va_copy</tt></a> must be matched exactly with calls to
3832 <tt>llvm.va_end</tt>.</p>
3836 <!-- _______________________________________________________________________ -->
3837 <div class="doc_subsubsection">
3838 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
3841 <div class="doc_text">
3846 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
3851 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
3852 from the source argument list to the destination argument list.</p>
3856 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
3857 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
3862 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
3863 macro available in C. In a target-dependent way, it copies the source
3864 <tt>va_list</tt> element into the destination <tt>va_list</tt> element. This
3865 intrinsic is necessary because the <tt><a href="#int_va_start">
3866 llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
3867 example, memory allocation.</p>
3871 <!-- ======================================================================= -->
3872 <div class="doc_subsection">
3873 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
3876 <div class="doc_text">
3879 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
3880 Collection</a> requires the implementation and generation of these intrinsics.
3881 These intrinsics allow identification of <a href="#int_gcroot">GC roots on the
3882 stack</a>, as well as garbage collector implementations that require <a
3883 href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a> barriers.
3884 Front-ends for type-safe garbage collected languages should generate these
3885 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
3886 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
3890 <!-- _______________________________________________________________________ -->
3891 <div class="doc_subsubsection">
3892 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
3895 <div class="doc_text">
3900 declare void @llvm.gcroot(<ty>** %ptrloc, <ty2>* %metadata)
3905 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
3906 the code generator, and allows some metadata to be associated with it.</p>
3910 <p>The first argument specifies the address of a stack object that contains the
3911 root pointer. The second pointer (which must be either a constant or a global
3912 value address) contains the meta-data to be associated with the root.</p>
3916 <p>At runtime, a call to this intrinsics stores a null pointer into the "ptrloc"
3917 location. At compile-time, the code generator generates information to allow
3918 the runtime to find the pointer at GC safe points.
3924 <!-- _______________________________________________________________________ -->
3925 <div class="doc_subsubsection">
3926 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
3929 <div class="doc_text">
3934 declare i8 * @llvm.gcread(i8 * %ObjPtr, i8 ** %Ptr)
3939 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
3940 locations, allowing garbage collector implementations that require read
3945 <p>The second argument is the address to read from, which should be an address
3946 allocated from the garbage collector. The first object is a pointer to the
3947 start of the referenced object, if needed by the language runtime (otherwise
3952 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
3953 instruction, but may be replaced with substantially more complex code by the
3954 garbage collector runtime, as needed.</p>
3959 <!-- _______________________________________________________________________ -->
3960 <div class="doc_subsubsection">
3961 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
3964 <div class="doc_text">
3969 declare void @llvm.gcwrite(i8 * %P1, i8 * %Obj, i8 ** %P2)
3974 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
3975 locations, allowing garbage collector implementations that require write
3976 barriers (such as generational or reference counting collectors).</p>
3980 <p>The first argument is the reference to store, the second is the start of the
3981 object to store it to, and the third is the address of the field of Obj to
3982 store to. If the runtime does not require a pointer to the object, Obj may be
3987 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
3988 instruction, but may be replaced with substantially more complex code by the
3989 garbage collector runtime, as needed.</p>
3995 <!-- ======================================================================= -->
3996 <div class="doc_subsection">
3997 <a name="int_codegen">Code Generator Intrinsics</a>
4000 <div class="doc_text">
4002 These intrinsics are provided by LLVM to expose special features that may only
4003 be implemented with code generator support.
4008 <!-- _______________________________________________________________________ -->
4009 <div class="doc_subsubsection">
4010 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
4013 <div class="doc_text">
4017 declare i8 *@llvm.returnaddress(i32 <level>)
4023 The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
4024 target-specific value indicating the return address of the current function
4025 or one of its callers.
4031 The argument to this intrinsic indicates which function to return the address
4032 for. Zero indicates the calling function, one indicates its caller, etc. The
4033 argument is <b>required</b> to be a constant integer value.
4039 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
4040 the return address of the specified call frame, or zero if it cannot be
4041 identified. The value returned by this intrinsic is likely to be incorrect or 0
4042 for arguments other than zero, so it should only be used for debugging purposes.
4046 Note that calling this intrinsic does not prevent function inlining or other
4047 aggressive transformations, so the value returned may not be that of the obvious
4048 source-language caller.
4053 <!-- _______________________________________________________________________ -->
4054 <div class="doc_subsubsection">
4055 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
4058 <div class="doc_text">
4062 declare i8 *@llvm.frameaddress(i32 <level>)
4068 The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
4069 target-specific frame pointer value for the specified stack frame.
4075 The argument to this intrinsic indicates which function to return the frame
4076 pointer for. Zero indicates the calling function, one indicates its caller,
4077 etc. The argument is <b>required</b> to be a constant integer value.
4083 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
4084 the frame address of the specified call frame, or zero if it cannot be
4085 identified. The value returned by this intrinsic is likely to be incorrect or 0
4086 for arguments other than zero, so it should only be used for debugging purposes.
4090 Note that calling this intrinsic does not prevent function inlining or other
4091 aggressive transformations, so the value returned may not be that of the obvious
4092 source-language caller.
4096 <!-- _______________________________________________________________________ -->
4097 <div class="doc_subsubsection">
4098 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
4101 <div class="doc_text">
4105 declare i8 *@llvm.stacksave()
4111 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
4112 the function stack, for use with <a href="#int_stackrestore">
4113 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
4114 features like scoped automatic variable sized arrays in C99.
4120 This intrinsic returns a opaque pointer value that can be passed to <a
4121 href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
4122 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
4123 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
4124 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
4125 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
4126 that were allocated after the <tt>llvm.stacksave</tt> was executed.
4131 <!-- _______________________________________________________________________ -->
4132 <div class="doc_subsubsection">
4133 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
4136 <div class="doc_text">
4140 declare void @llvm.stackrestore(i8 * %ptr)
4146 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
4147 the function stack to the state it was in when the corresponding <a
4148 href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
4149 useful for implementing language features like scoped automatic variable sized
4156 See the description for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.
4162 <!-- _______________________________________________________________________ -->
4163 <div class="doc_subsubsection">
4164 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
4167 <div class="doc_text">
4171 declare void @llvm.prefetch(i8 * <address>,
4172 i32 <rw>, i32 <locality>)
4179 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
4180 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
4182 effect on the behavior of the program but can change its performance
4189 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
4190 determining if the fetch should be for a read (0) or write (1), and
4191 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
4192 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
4193 <tt>locality</tt> arguments must be constant integers.
4199 This intrinsic does not modify the behavior of the program. In particular,
4200 prefetches cannot trap and do not produce a value. On targets that support this
4201 intrinsic, the prefetch can provide hints to the processor cache for better
4207 <!-- _______________________________________________________________________ -->
4208 <div class="doc_subsubsection">
4209 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
4212 <div class="doc_text">
4216 declare void @llvm.pcmarker( i32 <id> )
4223 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
4225 code to simulators and other tools. The method is target specific, but it is
4226 expected that the marker will use exported symbols to transmit the PC of the marker.
4227 The marker makes no guarantees that it will remain with any specific instruction
4228 after optimizations. It is possible that the presence of a marker will inhibit
4229 optimizations. The intended use is to be inserted after optimizations to allow
4230 correlations of simulation runs.
4236 <tt>id</tt> is a numerical id identifying the marker.
4242 This intrinsic does not modify the behavior of the program. Backends that do not
4243 support this intrinisic may ignore it.
4248 <!-- _______________________________________________________________________ -->
4249 <div class="doc_subsubsection">
4250 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
4253 <div class="doc_text">
4257 declare i64 @llvm.readcyclecounter( )
4264 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
4265 counter register (or similar low latency, high accuracy clocks) on those targets
4266 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
4267 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
4268 should only be used for small timings.
4274 When directly supported, reading the cycle counter should not modify any memory.
4275 Implementations are allowed to either return a application specific value or a
4276 system wide value. On backends without support, this is lowered to a constant 0.
4281 <!-- ======================================================================= -->
4282 <div class="doc_subsection">
4283 <a name="int_libc">Standard C Library Intrinsics</a>
4286 <div class="doc_text">
4288 LLVM provides intrinsics for a few important standard C library functions.
4289 These intrinsics allow source-language front-ends to pass information about the
4290 alignment of the pointer arguments to the code generator, providing opportunity
4291 for more efficient code generation.
4296 <!-- _______________________________________________________________________ -->
4297 <div class="doc_subsubsection">
4298 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
4301 <div class="doc_text">
4305 declare void @llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
4306 i32 <len>, i32 <align>)
4307 declare void @llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
4308 i64 <len>, i32 <align>)
4314 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4315 location to the destination location.
4319 Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
4320 intrinsics do not return a value, and takes an extra alignment argument.
4326 The first argument is a pointer to the destination, the second is a pointer to
4327 the source. The third argument is an integer argument
4328 specifying the number of bytes to copy, and the fourth argument is the alignment
4329 of the source and destination locations.
4333 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4334 the caller guarantees that both the source and destination pointers are aligned
4341 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4342 location to the destination location, which are not allowed to overlap. It
4343 copies "len" bytes of memory over. If the argument is known to be aligned to
4344 some boundary, this can be specified as the fourth argument, otherwise it should
4350 <!-- _______________________________________________________________________ -->
4351 <div class="doc_subsubsection">
4352 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
4355 <div class="doc_text">
4359 declare void @llvm.memmove.i32(i8 * <dest>, i8 * <src>,
4360 i32 <len>, i32 <align>)
4361 declare void @llvm.memmove.i64(i8 * <dest>, i8 * <src>,
4362 i64 <len>, i32 <align>)
4368 The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
4369 location to the destination location. It is similar to the
4370 '<tt>llvm.memcmp</tt>' intrinsic but allows the two memory locations to overlap.
4374 Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
4375 intrinsics do not return a value, and takes an extra alignment argument.
4381 The first argument is a pointer to the destination, the second is a pointer to
4382 the source. The third argument is an integer argument
4383 specifying the number of bytes to copy, and the fourth argument is the alignment
4384 of the source and destination locations.
4388 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4389 the caller guarantees that the source and destination pointers are aligned to
4396 The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
4397 location to the destination location, which may overlap. It
4398 copies "len" bytes of memory over. If the argument is known to be aligned to
4399 some boundary, this can be specified as the fourth argument, otherwise it should
4405 <!-- _______________________________________________________________________ -->
4406 <div class="doc_subsubsection">
4407 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
4410 <div class="doc_text">
4414 declare void @llvm.memset.i32(i8 * <dest>, i8 <val>,
4415 i32 <len>, i32 <align>)
4416 declare void @llvm.memset.i64(i8 * <dest>, i8 <val>,
4417 i64 <len>, i32 <align>)
4423 The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
4428 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
4429 does not return a value, and takes an extra alignment argument.
4435 The first argument is a pointer to the destination to fill, the second is the
4436 byte value to fill it with, the third argument is an integer
4437 argument specifying the number of bytes to fill, and the fourth argument is the
4438 known alignment of destination location.
4442 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4443 the caller guarantees that the destination pointer is aligned to that boundary.
4449 The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
4451 destination location. If the argument is known to be aligned to some boundary,
4452 this can be specified as the fourth argument, otherwise it should be set to 0 or
4458 <!-- _______________________________________________________________________ -->
4459 <div class="doc_subsubsection">
4460 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
4463 <div class="doc_text">
4467 declare float @llvm.sqrt.f32(float %Val)
4468 declare double @llvm.sqrt.f64(double %Val)
4474 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
4475 returning the same value as the libm '<tt>sqrt</tt>' function would. Unlike
4476 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
4477 negative numbers (which allows for better optimization).
4483 The argument and return value are floating point numbers of the same type.
4489 This function returns the sqrt of the specified operand if it is a nonnegative
4490 floating point number.
4494 <!-- _______________________________________________________________________ -->
4495 <div class="doc_subsubsection">
4496 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
4499 <div class="doc_text">
4503 declare float @llvm.powi.f32(float %Val, i32 %power)
4504 declare double @llvm.powi.f64(double %Val, i32 %power)
4510 The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
4511 specified (positive or negative) power. The order of evaluation of
4512 multiplications is not defined.
4518 The second argument is an integer power, and the first is a value to raise to
4525 This function returns the first value raised to the second power with an
4526 unspecified sequence of rounding operations.</p>
4530 <!-- ======================================================================= -->
4531 <div class="doc_subsection">
4532 <a name="int_manip">Bit Manipulation Intrinsics</a>
4535 <div class="doc_text">
4537 LLVM provides intrinsics for a few important bit manipulation operations.
4538 These allow efficient code generation for some algorithms.
4543 <!-- _______________________________________________________________________ -->
4544 <div class="doc_subsubsection">
4545 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
4548 <div class="doc_text">
4551 <p>This is an overloaded intrinsic function. You can use bswap on any integer
4552 type that is an even number of bytes (i.e. BitWidth % 16 == 0). Note the suffix
4553 that includes the type for the result and the operand.
4555 declare i16 @llvm.bswap.i16.i16(i16 <id>)
4556 declare i32 @llvm.bswap.i32.i32(i32 <id>)
4557 declare i64 @llvm.bswap.i64.i64(i64 <id>)
4563 The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
4564 values with an even number of bytes (positive multiple of 16 bits). These are
4565 useful for performing operations on data that is not in the target's native
4572 The <tt>llvm.bswap.16.i16</tt> intrinsic returns an i16 value that has the high
4573 and low byte of the input i16 swapped. Similarly, the <tt>llvm.bswap.i32</tt>
4574 intrinsic returns an i32 value that has the four bytes of the input i32
4575 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the returned
4576 i32 will have its bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i48.i48</tt>,
4577 <tt>llvm.bswap.i64.i64</tt> and other intrinsics extend this concept to
4578 additional even-byte lengths (6 bytes, 8 bytes and more, respectively).
4583 <!-- _______________________________________________________________________ -->
4584 <div class="doc_subsubsection">
4585 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
4588 <div class="doc_text">
4591 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
4592 width. Not all targets support all bit widths however.
4594 declare i32 @llvm.ctpop.i8 (i8 <src>)
4595 declare i32 @llvm.ctpop.i16(i16 <src>)
4596 declare i32 @llvm.ctpop.i32(i32 <src>)
4597 declare i32 @llvm.ctpop.i64(i64 <src>)
4598 declare i32 @llvm.ctpop.i256(i256 <src>)
4604 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
4611 The only argument is the value to be counted. The argument may be of any
4612 integer type. The return type must match the argument type.
4618 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
4622 <!-- _______________________________________________________________________ -->
4623 <div class="doc_subsubsection">
4624 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
4627 <div class="doc_text">
4630 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
4631 integer bit width. Not all targets support all bit widths however.
4633 declare i32 @llvm.ctlz.i8 (i8 <src>)
4634 declare i32 @llvm.ctlz.i16(i16 <src>)
4635 declare i32 @llvm.ctlz.i32(i32 <src>)
4636 declare i32 @llvm.ctlz.i64(i64 <src>)
4637 declare i32 @llvm.ctlz.i256(i256 <src>)
4643 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
4644 leading zeros in a variable.
4650 The only argument is the value to be counted. The argument may be of any
4651 integer type. The return type must match the argument type.
4657 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
4658 in a variable. If the src == 0 then the result is the size in bits of the type
4659 of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.
4665 <!-- _______________________________________________________________________ -->
4666 <div class="doc_subsubsection">
4667 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
4670 <div class="doc_text">
4673 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
4674 integer bit width. Not all targets support all bit widths however.
4676 declare i32 @llvm.cttz.i8 (i8 <src>)
4677 declare i32 @llvm.cttz.i16(i16 <src>)
4678 declare i32 @llvm.cttz.i32(i32 <src>)
4679 declare i32 @llvm.cttz.i64(i64 <src>)
4680 declare i32 @llvm.cttz.i256(i256 <src>)
4686 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
4693 The only argument is the value to be counted. The argument may be of any
4694 integer type. The return type must match the argument type.
4700 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
4701 in a variable. If the src == 0 then the result is the size in bits of the type
4702 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
4706 <!-- _______________________________________________________________________ -->
4707 <div class="doc_subsubsection">
4708 <a name="int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic</a>
4711 <div class="doc_text">
4714 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.select</tt>
4715 on any integer bit width.
4717 declare i17 @llvm.part.select.i17.i17 (i17 %val, i32 %loBit, i32 %hiBit)
4718 declare i29 @llvm.part.select.i29.i29 (i29 %val, i32 %loBit, i32 %hiBit)
4722 <p>The '<tt>llvm.part.select</tt>' family of intrinsic functions selects a
4723 range of bits from an integer value and returns them in the same bit width as
4724 the original value.</p>
4727 <p>The first argument, <tt>%val</tt> and the result may be integer types of
4728 any bit width but they must have the same bit width. The second and third
4729 arguments must be <tt>i32</tt> type since they specify only a bit index.</p>
4732 <p>The operation of the '<tt>llvm.part.select</tt>' intrinsic has two modes
4733 of operation: forwards and reverse. If <tt>%loBit</tt> is greater than
4734 <tt>%hiBits</tt> then the intrinsic operates in reverse mode. Otherwise it
4735 operates in forward mode.</p>
4736 <p>In forward mode, this intrinsic is the equivalent of shifting <tt>%val</tt>
4737 right by <tt>%loBit</tt> bits and then ANDing it with a mask with
4738 only the <tt>%hiBit - %loBit</tt> bits set, as follows:</p>
4740 <li>The <tt>%val</tt> is shifted right (LSHR) by the number of bits specified
4741 by <tt>%loBits</tt>. This normalizes the value to the low order bits.</li>
4742 <li>The <tt>%loBits</tt> value is subtracted from the <tt>%hiBits</tt> value
4743 to determine the number of bits to retain.</li>
4744 <li>A mask of the retained bits is created by shifting a -1 value.</li>
4745 <li>The mask is ANDed with <tt>%val</tt> to produce the result.
4747 <p>In reverse mode, a similar computation is made except that the bits are
4748 returned in the reverse order. So, for example, if <tt>X</tt> has the value
4749 <tt>i16 0x0ACF (101011001111)</tt> and we apply
4750 <tt>part.select(i16 X, 8, 3)</tt> to it, we get back the value
4751 <tt>i16 0x0026 (000000100110)</tt>.</p>
4754 <div class="doc_subsubsection">
4755 <a name="int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic</a>
4758 <div class="doc_text">
4761 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.set</tt>
4762 on any integer bit width.
4764 declare i17 @llvm.part.set.i17.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
4765 declare i29 @llvm.part.set.i29.i29.i9 (i29 %val, i9 %repl, i32 %lo, i32 %hi)
4769 <p>The '<tt>llvm.part.set</tt>' family of intrinsic functions replaces a range
4770 of bits in an integer value with another integer value. It returns the integer
4771 with the replaced bits.</p>
4774 <p>The first argument, <tt>%val</tt> and the result may be integer types of
4775 any bit width but they must have the same bit width. <tt>%val</tt> is the value
4776 whose bits will be replaced. The second argument, <tt>%repl</tt> may be an
4777 integer of any bit width. The third and fourth arguments must be <tt>i32</tt>
4778 type since they specify only a bit index.</p>
4781 <p>The operation of the '<tt>llvm.part.set</tt>' intrinsic has two modes
4782 of operation: forwards and reverse. If <tt>%lo</tt> is greater than
4783 <tt>%hi</tt> then the intrinsic operates in reverse mode. Otherwise it
4784 operates in forward mode.</p>
4785 <p>For both modes, the <tt>%repl</tt> value is prepared for use by either
4786 truncating it down to the size of the replacement area or zero extending it
4787 up to that size.</p>
4788 <p>In forward mode, the bits between <tt>%lo</tt> and <tt>%hi</tt> (inclusive)
4789 are replaced with corresponding bits from <tt>%repl</tt>. That is the 0th bit
4790 in <tt>%repl</tt> replaces the <tt>%lo</tt>th bit in <tt>%val</tt> and etc. up
4791 to the <tt>%hi</tt>th bit.
4792 <p>In reverse mode, a similar computation is made except that the bits are
4793 reversed. That is, the <tt>0</tt>th bit in <tt>%repl</tt> replaces the
4794 <tt>%hi</tt> bit in <tt>%val</tt> and etc. down to the <tt>%lo</tt>th bit.
4797 llvm.part.set(0xFFFF, 0, 4, 7) -> 0xFF0F
4798 llvm.part.set(0xFFFF, 0, 7, 4) -> 0xFF0F
4799 llvm.part.set(0xFFFF, 1, 7, 4) -> 0xFF8F
4800 llvm.part.set(0xFFFF, F, 8, 3) -> 0xFFE7
4801 llvm.part.set(0xFFFF, 0, 3, 8) -> 0xFE07
4805 <!-- ======================================================================= -->
4806 <div class="doc_subsection">
4807 <a name="int_debugger">Debugger Intrinsics</a>
4810 <div class="doc_text">
4812 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
4813 are described in the <a
4814 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
4815 Debugging</a> document.
4820 <!-- ======================================================================= -->
4821 <div class="doc_subsection">
4822 <a name="int_eh">Exception Handling Intrinsics</a>
4825 <div class="doc_text">
4826 <p> The LLVM exception handling intrinsics (which all start with
4827 <tt>llvm.eh.</tt> prefix), are described in the <a
4828 href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
4829 Handling</a> document. </p>
4832 <!-- ======================================================================= -->
4833 <div class="doc_subsection">
4834 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
4837 <div class="doc_text">
4839 These intrinsic functions expand the "universal IR" of LLVM to represent
4840 hardware constructs for atomic operations and memory synchronization. This
4841 provides an interface to the hardware, not an interface to the programmer. It
4842 is aimed at a low enough level to allow any programming models or APIs which
4843 need atomic behaviors to map cleanly onto it. It is also modeled primarily on
4844 hardware behavior. Just as hardware provides a "unviresal IR" for source
4845 languages, it also provides a starting point for developing a "universal"
4846 atomic operation and synchronization IR.
4849 These do <em>not</em> form an API such as high-level threading libraries,
4850 software transaction memory systems, atomic primitives, and intrinsic
4851 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
4852 application libraries. The hardware interface provided by LLVM should allow
4853 a clean implementation of all of these APIs and parallel programming models.
4854 No one model or paradigm should be selected above others unless the hardware
4855 itself ubiquitously does so.
4859 <!-- _______________________________________________________________________ -->
4860 <div class="doc_subsubsection">
4861 <a name="int_lcs">'<tt>llvm.atomic.lcs.*</tt>' Intrinsic</a>
4863 <div class="doc_text">
4866 This is an overloaded intrinsic. You can use <tt>llvm.atomic.lcs</tt> on any
4867 integer bit width. Not all targets support all bit widths however.</p>
4869 declare i8 @llvm.atomic.lcs.i8.i8p.i8.i8( i8* <ptr>, i8 <cmp>, i8 <val> )
4870 declare i16 @llvm.atomic.lcs.i16.i16p.i16.i16( i16* <ptr>, i16 <cmp>, i16 <val> )
4871 declare i32 @llvm.atomic.lcs.i32.i32p.i32.i32( i32* <ptr>, i32 <cmp>, i32 <val> )
4872 declare i64 @llvm.atomic.lcs.i64.i64p.i64.i64( i64* <ptr>, i64 <cmp>, i64 <val> )
4876 This loads a value in shared memory and compares it to a given value. If they
4877 are equal, it stores a new value into the shared memory.
4881 The <tt>llvm.atomic.lcs</tt> intrinsic takes three arguments. The result as
4882 well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
4883 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
4884 this integer type. While any bit width integer may be used, targets may only
4885 lower representations they support in hardware.
4889 This entire intrinsic must be executed atomically. It first loads the value
4890 in shared memory pointed to by <tt>ptr</tt> and compares it with the value
4891 <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the shared
4892 memory. The loaded value is yielded in all cases. This provides the
4893 equivalent of an atomic compare-and-swap operation within the SSA framework.
4900 %val1 = add i32 4, 4
4901 %result1 = call i32 @llvm.atomic.lcs( i32* %ptr, i32 4, %val1 )
4902 <i>; yields {i32}:result1 = 4</i>
4903 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
4904 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
4906 %val2 = add i32 1, 1
4907 %result2 = call i32 @llvm.atomic.lcs( i32* %ptr, i32 5, %val2 )
4908 <i>; yields {i32}:result2 = 8</i>
4909 %stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
4910 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
4914 <!-- _______________________________________________________________________ -->
4915 <div class="doc_subsubsection">
4916 <a name="int_ls">'<tt>llvm.atomic.ls.*</tt>' Intrinsic</a>
4918 <div class="doc_text">
4921 This is an overloaded intrinsic. You can use <tt>llvm.atomic.ls</tt> on any
4922 integer bit width. Not all targets support all bit widths however.</p>
4924 declare i8 @llvm.atomic.ls.i8.i8p.i8( i8* <ptr>, i8 <val> )
4925 declare i16 @llvm.atomic.ls.i16.i16p.i16( i16* <ptr>, i16 <val> )
4926 declare i32 @llvm.atomic.ls.i32.i32p.i32( i32* <ptr>, i32 <val> )
4927 declare i64 @llvm.atomic.ls.i64.i64p.i64( i64* <ptr>, i64 <val> )
4931 This intrinsic loads the value stored in shared memory at <tt>ptr</tt> and
4932 yields the value from memory. It then stores the value in <tt>val</tt> in the
4933 shared memory at <tt>ptr</tt>.
4937 The <tt>llvm.atomic.ls</tt> intrinsic takes two arguments. Both the
4938 <tt>val</tt> argument and the result must be integers of the same bit width.
4939 The first argument, <tt>ptr</tt>, must be a pointer to a value of this
4940 integer type. The targets may only lower integer representations they
4945 This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
4946 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
4947 equivalent of an atomic swap operation within the SSA framework.
4954 %val1 = add i32 4, 4
4955 %result1 = call i32 @llvm.atomic.ls( i32* %ptr, i32 %val1 )
4956 <i>; yields {i32}:result1 = 4</i>
4957 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
4958 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
4960 %val2 = add i32 1, 1
4961 %result2 = call i32 @llvm.atomic.ls( i32* %ptr, i32 %val2 )
4962 <i>; yields {i32}:result2 = 8</i>
4963 %stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
4964 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
4968 <!-- _______________________________________________________________________ -->
4969 <div class="doc_subsubsection">
4970 <a name="int_las">'<tt>llvm.atomic.las.*</tt>' Intrinsic</a>
4972 <div class="doc_text">
4975 This is an overloaded intrinsic. You can use <tt>llvm.atomic.las</tt> on any
4976 integer bit width. Not all targets support all bit widths however.</p>
4978 declare i8 @llvm.atomic.las.i8.i8p.i8( i8* <ptr>, i8 <delta> )
4979 declare i16 @llvm.atomic.las.i16.i16p.i16( i16* <ptr>, i16 <delta> )
4980 declare i32 @llvm.atomic.las.i32.i32p.i32( i32* <ptr>, i32 <delta> )
4981 declare i64 @llvm.atomic.las.i64.i64p.i64( i64* <ptr>, i64 <delta> )
4985 This intrinsic adds <tt>delta</tt> to the value stored in shared memory at
4986 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
4990 The intrinsic takes two arguments, the first a pointer to an integer value
4991 and the second an integer value. The result is also an integer value. These
4992 integer types can have any bit width, but they must all have the same bit
4993 width. The targets may only lower integer representations they support.
4997 This intrinsic does a series of operations atomically. It first loads the
4998 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
4999 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
5005 %result1 = call i32 @llvm.atomic.las( i32* %ptr, i32 4 )
5006 <i>; yields {i32}:result1 = 4</i>
5007 %result2 = call i32 @llvm.atomic.las( i32* %ptr, i32 2 )
5008 <i>; yields {i32}:result2 = 8</i>
5009 %result3 = call i32 @llvm.atomic.las( i32* %ptr, i32 5 )
5010 <i>; yields {i32}:result3 = 10</i>
5011 %memval = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
5015 <!-- _______________________________________________________________________ -->
5016 <div class="doc_subsubsection">
5017 <a name="int_lss">'<tt>llvm.atomic.lss.*</tt>' Intrinsic</a>
5019 <div class="doc_text">
5022 This is an overloaded intrinsic. You can use <tt>llvm.atomic.lss</tt> on any
5023 integer bit width. Not all targets support all bit widths however.</p>
5025 declare i8 @llvm.atomic.lss.i8.i8.i8( i8* <ptr>, i8 <delta> )
5026 declare i16 @llvm.atomic.lss.i16.i16.i16( i16* <ptr>, i16 <delta> )
5027 declare i32 @llvm.atomic.lss.i32.i32.i32( i32* <ptr>, i32 <delta> )
5028 declare i64 @llvm.atomic.lss.i64.i64.i64( i64* <ptr>, i64 <delta> )
5032 This intrinsic subtracts <tt>delta</tt> from the value stored in shared
5033 memory at <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
5037 The intrinsic takes two arguments, the first a pointer to an integer value
5038 and the second an integer value. The result is also an integer value. These
5039 integer types can have any bit width, but they must all have the same bit
5040 width. The targets may only lower integer representations they support.
5044 This intrinsic does a series of operations atomically. It first loads the
5045 value stored at <tt>ptr</tt>. It then subtracts <tt>delta</tt>,
5046 stores the result to <tt>ptr</tt>. It yields the original value stored
5053 %result1 = call i32 @llvm.atomic.lss( i32* %ptr, i32 4 )
5054 <i>; yields {i32}:result1 = 32</i>
5055 %result2 = call i32 @llvm.atomic.lss( i32* %ptr, i32 2 )
5056 <i>; yields {i32}:result2 = 28</i>
5057 %result3 = call i32 @llvm.atomic.lss( i32* %ptr, i32 5 )
5058 <i>; yields {i32}:result3 = 26</i>
5059 %memval = load i32* %ptr <i>; yields {i32}:memval1 = 21</i>
5063 <!-- _______________________________________________________________________ -->
5064 <div class="doc_subsubsection">
5065 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
5067 <div class="doc_text">
5070 declare void @llvm.memory.barrier( i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss> )
5074 The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
5075 specific pairs of memory access types.
5079 The <tt>llvm.memory.barrier</tt> intrinsic requires four boolean arguments.
5080 Each argument enables a specific barrier as listed below.
5082 <li><tt>ll</tt>: load-load barrier</li>
5083 <li><tt>ls</tt>: load-store barrier</li>
5084 <li><tt>sl</tt>: store-load barrier</li>
5085 <li><tt>ss</tt>: store-store barrier</li>
5090 This intrinsic causes the system to enforce some ordering constraints upon
5091 the loads and stores of the program. This barrier does not indicate
5092 <em>when</em> any events will occur, it only enforces an <em>order</em> in
5093 which they occur. For any of the specified pairs of load and store operations
5094 (f.ex. load-load, or store-load), all of the first operations preceding the
5095 barrier will complete before any of the second operations succeeding the
5096 barrier begin. Specifically the semantics for each pairing is as follows:
5098 <li><tt>ll</tt>: All loads before the barrier must complete before any load
5099 after the barrier begins.</li>
5100 <li><tt>ls</tt>: All loads before the barrier must complete before any
5101 store after the barrier begins.</li>
5102 <li><tt>ss</tt>: All stores before the barrier must complete before any
5103 store after the barrier begins.</li>
5104 <li><tt>sl</tt>: All stores before the barrier must complete before any
5105 load after the barrier begins.</li>
5107 These semantics are applied with a logical "and" behavior when more than one
5108 is enabled in a single memory barrier intrinsic.
5115 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
5116 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
5117 <i>; guarantee the above finishes</i>
5118 store i32 8, %ptr <i>; before this begins</i>
5122 <!-- ======================================================================= -->
5123 <div class="doc_subsection">
5124 <a name="int_general">General Intrinsics</a>
5127 <div class="doc_text">
5128 <p> This class of intrinsics is designed to be generic and has
5129 no specific purpose. </p>
5132 <!-- _______________________________________________________________________ -->
5133 <div class="doc_subsubsection">
5134 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
5137 <div class="doc_text">
5141 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
5147 The '<tt>llvm.var.annotation</tt>' intrinsic
5153 The first argument is a pointer to a value, the second is a pointer to a
5154 global string, the third is a pointer to a global string which is the source
5155 file name, and the last argument is the line number.
5161 This intrinsic allows annotation of local variables with arbitrary strings.
5162 This can be useful for special purpose optimizations that want to look for these
5163 annotations. These have no other defined use, they are ignored by code
5164 generation and optimization.
5168 <!-- *********************************************************************** -->
5171 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
5172 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
5173 <a href="http://validator.w3.org/check/referer"><img
5174 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a>
5176 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
5177 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
5178 Last modified: $Date$