1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <title>LLVM Assembly Language Reference Manual</title>
6 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
7 <meta name="author" content="Chris Lattner">
8 <meta name="description"
9 content="LLVM Assembly Language Reference Manual.">
10 <link rel="stylesheet" href="llvm.css" type="text/css">
15 <div class="doc_title"> LLVM Language Reference Manual </div>
17 <li><a href="#abstract">Abstract</a></li>
18 <li><a href="#introduction">Introduction</a></li>
19 <li><a href="#identifiers">Identifiers</a></li>
20 <li><a href="#highlevel">High Level Structure</a>
22 <li><a href="#modulestructure">Module Structure</a></li>
23 <li><a href="#linkage">Linkage Types</a></li>
24 <li><a href="#callingconv">Calling Conventions</a></li>
25 <li><a href="#globalvars">Global Variables</a></li>
26 <li><a href="#functionstructure">Functions</a></li>
27 <li><a href="#aliasstructure">Aliases</a>
28 <li><a href="#paramattrs">Parameter Attributes</a></li>
29 <li><a href="#gc">Garbage Collector Names</a></li>
30 <li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
31 <li><a href="#datalayout">Data Layout</a></li>
34 <li><a href="#typesystem">Type System</a>
36 <li><a href="#t_classifications">Type Classifications</a></li>
37 <li><a href="#t_primitive">Primitive Types</a>
39 <li><a href="#t_floating">Floating Point Types</a></li>
40 <li><a href="#t_void">Void Type</a></li>
41 <li><a href="#t_label">Label Type</a></li>
44 <li><a href="#t_derived">Derived Types</a>
46 <li><a href="#t_integer">Integer Type</a></li>
47 <li><a href="#t_array">Array Type</a></li>
48 <li><a href="#t_function">Function Type</a></li>
49 <li><a href="#t_pointer">Pointer Type</a></li>
50 <li><a href="#t_struct">Structure Type</a></li>
51 <li><a href="#t_pstruct">Packed Structure Type</a></li>
52 <li><a href="#t_vector">Vector Type</a></li>
53 <li><a href="#t_opaque">Opaque Type</a></li>
58 <li><a href="#constants">Constants</a>
60 <li><a href="#simpleconstants">Simple Constants</a>
61 <li><a href="#aggregateconstants">Aggregate Constants</a>
62 <li><a href="#globalconstants">Global Variable and Function Addresses</a>
63 <li><a href="#undefvalues">Undefined Values</a>
64 <li><a href="#constantexprs">Constant Expressions</a>
67 <li><a href="#othervalues">Other Values</a>
69 <li><a href="#inlineasm">Inline Assembler Expressions</a>
72 <li><a href="#instref">Instruction Reference</a>
74 <li><a href="#terminators">Terminator Instructions</a>
76 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
77 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
78 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
79 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
80 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
81 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
84 <li><a href="#binaryops">Binary Operations</a>
86 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
87 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
88 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
89 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
90 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
91 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
92 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
93 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
94 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
97 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
99 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
100 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
101 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
102 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
103 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
104 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
107 <li><a href="#vectorops">Vector Operations</a>
109 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
110 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
111 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
114 <li><a href="#aggregateops">Aggregate Operations</a>
116 <li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li>
117 <li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li>
120 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
122 <li><a href="#i_malloc">'<tt>malloc</tt>' Instruction</a></li>
123 <li><a href="#i_free">'<tt>free</tt>' Instruction</a></li>
124 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
125 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
126 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
127 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
130 <li><a href="#convertops">Conversion Operations</a>
132 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
133 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
134 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
135 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
136 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
137 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
138 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
139 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
140 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
141 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
142 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
143 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
145 <li><a href="#otherops">Other Operations</a>
147 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
148 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
149 <li><a href="#i_vicmp">'<tt>vicmp</tt>' Instruction</a></li>
150 <li><a href="#i_vfcmp">'<tt>vfcmp</tt>' Instruction</a></li>
151 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
152 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
153 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
154 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
155 <li><a href="#i_getresult">'<tt>getresult</tt>' Instruction</a></li>
160 <li><a href="#intrinsics">Intrinsic Functions</a>
162 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
164 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
165 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
166 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
169 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
171 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
172 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
173 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
176 <li><a href="#int_codegen">Code Generator Intrinsics</a>
178 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
179 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
180 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
181 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
182 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
183 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
184 <li><a href="#int_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
187 <li><a href="#int_libc">Standard C Library Intrinsics</a>
189 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
190 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
191 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
192 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
193 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
194 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
195 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
196 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
199 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
201 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
202 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
203 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
204 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
205 <li><a href="#int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic </a></li>
206 <li><a href="#int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic </a></li>
209 <li><a href="#int_debugger">Debugger intrinsics</a></li>
210 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
211 <li><a href="#int_trampoline">Trampoline Intrinsic</a>
213 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
216 <li><a href="#int_atomics">Atomic intrinsics</a>
218 <li><a href="#int_memory_barrier"><tt>llvm.memory_barrier</tt></a></li>
219 <li><a href="#int_atomic_lcs"><tt>llvm.atomic.lcs</tt></a></li>
220 <li><a href="#int_atomic_las"><tt>llvm.atomic.las</tt></a></li>
221 <li><a href="#int_atomic_swap"><tt>llvm.atomic.swap</tt></a></li>
224 <li><a href="#int_general">General intrinsics</a>
226 <li><a href="#int_var_annotation">
227 <tt>llvm.var.annotation</tt>' Intrinsic</a></li>
228 <li><a href="#int_annotation">
229 <tt>llvm.annotation.*</tt>' Intrinsic</a></li>
230 <li><a href="#int_trap">
231 <tt>llvm.trap</tt>' Intrinsic</a></li>
238 <div class="doc_author">
239 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
240 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
243 <!-- *********************************************************************** -->
244 <div class="doc_section"> <a name="abstract">Abstract </a></div>
245 <!-- *********************************************************************** -->
247 <div class="doc_text">
248 <p>This document is a reference manual for the LLVM assembly language.
249 LLVM is an SSA based representation that provides type safety,
250 low-level operations, flexibility, and the capability of representing
251 'all' high-level languages cleanly. It is the common code
252 representation used throughout all phases of the LLVM compilation
256 <!-- *********************************************************************** -->
257 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
258 <!-- *********************************************************************** -->
260 <div class="doc_text">
262 <p>The LLVM code representation is designed to be used in three
263 different forms: as an in-memory compiler IR, as an on-disk bitcode
264 representation (suitable for fast loading by a Just-In-Time compiler),
265 and as a human readable assembly language representation. This allows
266 LLVM to provide a powerful intermediate representation for efficient
267 compiler transformations and analysis, while providing a natural means
268 to debug and visualize the transformations. The three different forms
269 of LLVM are all equivalent. This document describes the human readable
270 representation and notation.</p>
272 <p>The LLVM representation aims to be light-weight and low-level
273 while being expressive, typed, and extensible at the same time. It
274 aims to be a "universal IR" of sorts, by being at a low enough level
275 that high-level ideas may be cleanly mapped to it (similar to how
276 microprocessors are "universal IR's", allowing many source languages to
277 be mapped to them). By providing type information, LLVM can be used as
278 the target of optimizations: for example, through pointer analysis, it
279 can be proven that a C automatic variable is never accessed outside of
280 the current function... allowing it to be promoted to a simple SSA
281 value instead of a memory location.</p>
285 <!-- _______________________________________________________________________ -->
286 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
288 <div class="doc_text">
290 <p>It is important to note that this document describes 'well formed'
291 LLVM assembly language. There is a difference between what the parser
292 accepts and what is considered 'well formed'. For example, the
293 following instruction is syntactically okay, but not well formed:</p>
295 <div class="doc_code">
297 %x = <a href="#i_add">add</a> i32 1, %x
301 <p>...because the definition of <tt>%x</tt> does not dominate all of
302 its uses. The LLVM infrastructure provides a verification pass that may
303 be used to verify that an LLVM module is well formed. This pass is
304 automatically run by the parser after parsing input assembly and by
305 the optimizer before it outputs bitcode. The violations pointed out
306 by the verifier pass indicate bugs in transformation passes or input to
310 <!-- Describe the typesetting conventions here. -->
312 <!-- *********************************************************************** -->
313 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
314 <!-- *********************************************************************** -->
316 <div class="doc_text">
318 <p>LLVM identifiers come in two basic types: global and local. Global
319 identifiers (functions, global variables) begin with the @ character. Local
320 identifiers (register names, types) begin with the % character. Additionally,
321 there are three different formats for identifiers, for different purposes:
324 <li>Named values are represented as a string of characters with their prefix.
325 For example, %foo, @DivisionByZero, %a.really.long.identifier. The actual
326 regular expression used is '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
327 Identifiers which require other characters in their names can be surrounded
328 with quotes. In this way, anything except a <tt>"</tt> character can
329 be used in a named value.</li>
331 <li>Unnamed values are represented as an unsigned numeric value with their
332 prefix. For example, %12, @2, %44.</li>
334 <li>Constants, which are described in a <a href="#constants">section about
335 constants</a>, below.</li>
338 <p>LLVM requires that values start with a prefix for two reasons: Compilers
339 don't need to worry about name clashes with reserved words, and the set of
340 reserved words may be expanded in the future without penalty. Additionally,
341 unnamed identifiers allow a compiler to quickly come up with a temporary
342 variable without having to avoid symbol table conflicts.</p>
344 <p>Reserved words in LLVM are very similar to reserved words in other
345 languages. There are keywords for different opcodes
346 ('<tt><a href="#i_add">add</a></tt>',
347 '<tt><a href="#i_bitcast">bitcast</a></tt>',
348 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
349 href="#t_void">void</a></tt>', '<tt><a href="#t_primitive">i32</a></tt>', etc...),
350 and others. These reserved words cannot conflict with variable names, because
351 none of them start with a prefix character ('%' or '@').</p>
353 <p>Here is an example of LLVM code to multiply the integer variable
354 '<tt>%X</tt>' by 8:</p>
358 <div class="doc_code">
360 %result = <a href="#i_mul">mul</a> i32 %X, 8
364 <p>After strength reduction:</p>
366 <div class="doc_code">
368 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
372 <p>And the hard way:</p>
374 <div class="doc_code">
376 <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
377 <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
378 %result = <a href="#i_add">add</a> i32 %1, %1
382 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
383 important lexical features of LLVM:</p>
387 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
390 <li>Unnamed temporaries are created when the result of a computation is not
391 assigned to a named value.</li>
393 <li>Unnamed temporaries are numbered sequentially</li>
397 <p>...and it also shows a convention that we follow in this document. When
398 demonstrating instructions, we will follow an instruction with a comment that
399 defines the type and name of value produced. Comments are shown in italic
404 <!-- *********************************************************************** -->
405 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
406 <!-- *********************************************************************** -->
408 <!-- ======================================================================= -->
409 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
412 <div class="doc_text">
414 <p>LLVM programs are composed of "Module"s, each of which is a
415 translation unit of the input programs. Each module consists of
416 functions, global variables, and symbol table entries. Modules may be
417 combined together with the LLVM linker, which merges function (and
418 global variable) definitions, resolves forward declarations, and merges
419 symbol table entries. Here is an example of the "hello world" module:</p>
421 <div class="doc_code">
422 <pre><i>; Declare the string constant as a global constant...</i>
423 <a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a
424 href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
426 <i>; External declaration of the puts function</i>
427 <a href="#functionstructure">declare</a> i32 @puts(i8 *) <i>; i32(i8 *)* </i>
429 <i>; Definition of main function</i>
430 define i32 @main() { <i>; i32()* </i>
431 <i>; Convert [13x i8 ]* to i8 *...</i>
433 href="#i_getelementptr">getelementptr</a> [13 x i8 ]* @.LC0, i64 0, i64 0 <i>; i8 *</i>
435 <i>; Call puts function to write out the string to stdout...</i>
437 href="#i_call">call</a> i32 @puts(i8 * %cast210) <i>; i32</i>
439 href="#i_ret">ret</a> i32 0<br>}<br>
443 <p>This example is made up of a <a href="#globalvars">global variable</a>
444 named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
445 function, and a <a href="#functionstructure">function definition</a>
446 for "<tt>main</tt>".</p>
448 <p>In general, a module is made up of a list of global values,
449 where both functions and global variables are global values. Global values are
450 represented by a pointer to a memory location (in this case, a pointer to an
451 array of char, and a pointer to a function), and have one of the following <a
452 href="#linkage">linkage types</a>.</p>
456 <!-- ======================================================================= -->
457 <div class="doc_subsection">
458 <a name="linkage">Linkage Types</a>
461 <div class="doc_text">
464 All Global Variables and Functions have one of the following types of linkage:
469 <dt><tt><b><a name="linkage_internal">internal</a></b></tt>: </dt>
471 <dd>Global values with internal linkage are only directly accessible by
472 objects in the current module. In particular, linking code into a module with
473 an internal global value may cause the internal to be renamed as necessary to
474 avoid collisions. Because the symbol is internal to the module, all
475 references can be updated. This corresponds to the notion of the
476 '<tt>static</tt>' keyword in C.
479 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
481 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
482 the same name when linkage occurs. This is typically used to implement
483 inline functions, templates, or other code which must be generated in each
484 translation unit that uses it. Unreferenced <tt>linkonce</tt> globals are
485 allowed to be discarded.
488 <dt><tt><b><a name="linkage_common">common</a></b></tt>: </dt>
490 <dd>"<tt>common</tt>" linkage is exactly the same as <tt>linkonce</tt>
491 linkage, except that unreferenced <tt>common</tt> globals may not be
492 discarded. This is used for globals that may be emitted in multiple
493 translation units, but that are not guaranteed to be emitted into every
494 translation unit that uses them. One example of this is tentative
495 definitions in C, such as "<tt>int X;</tt>" at global scope.
498 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
500 <dd>"<tt>weak</tt>" linkage is the same as <tt>common</tt> linkage, except
501 that some targets may choose to emit different assembly sequences for them
502 for target-dependent reasons. This is used for globals that are declared
503 "weak" in C source code.
506 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
508 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
509 pointer to array type. When two global variables with appending linkage are
510 linked together, the two global arrays are appended together. This is the
511 LLVM, typesafe, equivalent of having the system linker append together
512 "sections" with identical names when .o files are linked.
515 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt>: </dt>
516 <dd>The semantics of this linkage follow the ELF model: the symbol is weak
517 until linked, if not linked, the symbol becomes null instead of being an
521 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
523 <dd>If none of the above identifiers are used, the global is externally
524 visible, meaning that it participates in linkage and can be used to resolve
525 external symbol references.
530 The next two types of linkage are targeted for Microsoft Windows platform
531 only. They are designed to support importing (exporting) symbols from (to)
536 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt>: </dt>
538 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
539 or variable via a global pointer to a pointer that is set up by the DLL
540 exporting the symbol. On Microsoft Windows targets, the pointer name is
541 formed by combining <code>_imp__</code> and the function or variable name.
544 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt>: </dt>
546 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
547 pointer to a pointer in a DLL, so that it can be referenced with the
548 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
549 name is formed by combining <code>_imp__</code> and the function or variable
555 <p><a name="linkage_external"></a>For example, since the "<tt>.LC0</tt>"
556 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
557 variable and was linked with this one, one of the two would be renamed,
558 preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
559 external (i.e., lacking any linkage declarations), they are accessible
560 outside of the current module.</p>
561 <p>It is illegal for a function <i>declaration</i>
562 to have any linkage type other than "externally visible", <tt>dllimport</tt>,
563 or <tt>extern_weak</tt>.</p>
564 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt> and <tt>weak</tt>
568 <!-- ======================================================================= -->
569 <div class="doc_subsection">
570 <a name="callingconv">Calling Conventions</a>
573 <div class="doc_text">
575 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
576 and <a href="#i_invoke">invokes</a> can all have an optional calling convention
577 specified for the call. The calling convention of any pair of dynamic
578 caller/callee must match, or the behavior of the program is undefined. The
579 following calling conventions are supported by LLVM, and more may be added in
583 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
585 <dd>This calling convention (the default if no other calling convention is
586 specified) matches the target C calling conventions. This calling convention
587 supports varargs function calls and tolerates some mismatch in the declared
588 prototype and implemented declaration of the function (as does normal C).
591 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
593 <dd>This calling convention attempts to make calls as fast as possible
594 (e.g. by passing things in registers). This calling convention allows the
595 target to use whatever tricks it wants to produce fast code for the target,
596 without having to conform to an externally specified ABI. Implementations of
597 this convention should allow arbitrary
598 <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> to be
599 supported. This calling convention does not support varargs and requires the
600 prototype of all callees to exactly match the prototype of the function
604 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
606 <dd>This calling convention attempts to make code in the caller as efficient
607 as possible under the assumption that the call is not commonly executed. As
608 such, these calls often preserve all registers so that the call does not break
609 any live ranges in the caller side. This calling convention does not support
610 varargs and requires the prototype of all callees to exactly match the
611 prototype of the function definition.
614 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
616 <dd>Any calling convention may be specified by number, allowing
617 target-specific calling conventions to be used. Target specific calling
618 conventions start at 64.
622 <p>More calling conventions can be added/defined on an as-needed basis, to
623 support pascal conventions or any other well-known target-independent
628 <!-- ======================================================================= -->
629 <div class="doc_subsection">
630 <a name="visibility">Visibility Styles</a>
633 <div class="doc_text">
636 All Global Variables and Functions have one of the following visibility styles:
640 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
642 <dd>On ELF, default visibility means that the declaration is visible to other
643 modules and, in shared libraries, means that the declared entity may be
644 overridden. On Darwin, default visibility means that the declaration is
645 visible to other modules. Default visibility corresponds to "external
646 linkage" in the language.
649 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
651 <dd>Two declarations of an object with hidden visibility refer to the same
652 object if they are in the same shared object. Usually, hidden visibility
653 indicates that the symbol will not be placed into the dynamic symbol table,
654 so no other module (executable or shared library) can reference it
658 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
660 <dd>On ELF, protected visibility indicates that the symbol will be placed in
661 the dynamic symbol table, but that references within the defining module will
662 bind to the local symbol. That is, the symbol cannot be overridden by another
669 <!-- ======================================================================= -->
670 <div class="doc_subsection">
671 <a name="globalvars">Global Variables</a>
674 <div class="doc_text">
676 <p>Global variables define regions of memory allocated at compilation time
677 instead of run-time. Global variables may optionally be initialized, may have
678 an explicit section to be placed in, and may have an optional explicit alignment
679 specified. A variable may be defined as "thread_local", which means that it
680 will not be shared by threads (each thread will have a separated copy of the
681 variable). A variable may be defined as a global "constant," which indicates
682 that the contents of the variable will <b>never</b> be modified (enabling better
683 optimization, allowing the global data to be placed in the read-only section of
684 an executable, etc). Note that variables that need runtime initialization
685 cannot be marked "constant" as there is a store to the variable.</p>
688 LLVM explicitly allows <em>declarations</em> of global variables to be marked
689 constant, even if the final definition of the global is not. This capability
690 can be used to enable slightly better optimization of the program, but requires
691 the language definition to guarantee that optimizations based on the
692 'constantness' are valid for the translation units that do not include the
696 <p>As SSA values, global variables define pointer values that are in
697 scope (i.e. they dominate) all basic blocks in the program. Global
698 variables always define a pointer to their "content" type because they
699 describe a region of memory, and all memory objects in LLVM are
700 accessed through pointers.</p>
702 <p>A global variable may be declared to reside in a target-specifc numbered
703 address space. For targets that support them, address spaces may affect how
704 optimizations are performed and/or what target instructions are used to access
705 the variable. The default address space is zero. The address space qualifier
706 must precede any other attributes.</p>
708 <p>LLVM allows an explicit section to be specified for globals. If the target
709 supports it, it will emit globals to the section specified.</p>
711 <p>An explicit alignment may be specified for a global. If not present, or if
712 the alignment is set to zero, the alignment of the global is set by the target
713 to whatever it feels convenient. If an explicit alignment is specified, the
714 global is forced to have at least that much alignment. All alignments must be
717 <p>For example, the following defines a global in a numbered address space with
718 an initializer, section, and alignment:</p>
720 <div class="doc_code">
722 @G = constant float 1.0 addrspace(5), section "foo", align 4
729 <!-- ======================================================================= -->
730 <div class="doc_subsection">
731 <a name="functionstructure">Functions</a>
734 <div class="doc_text">
736 <p>LLVM function definitions consist of the "<tt>define</tt>" keyord,
737 an optional <a href="#linkage">linkage type</a>, an optional
738 <a href="#visibility">visibility style</a>, an optional
739 <a href="#callingconv">calling convention</a>, a return type, an optional
740 <a href="#paramattrs">parameter attribute</a> for the return type, a function
741 name, a (possibly empty) argument list (each with optional
742 <a href="#paramattrs">parameter attributes</a>), an optional section, an
743 optional alignment, an optional <a href="#gc">garbage collector name</a>, an
744 opening curly brace, a list of basic blocks, and a closing curly brace.
746 LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
747 optional <a href="#linkage">linkage type</a>, an optional
748 <a href="#visibility">visibility style</a>, an optional
749 <a href="#callingconv">calling convention</a>, a return type, an optional
750 <a href="#paramattrs">parameter attribute</a> for the return type, a function
751 name, a possibly empty list of arguments, an optional alignment, and an optional
752 <a href="#gc">garbage collector name</a>.</p>
754 <p>A function definition contains a list of basic blocks, forming the CFG for
755 the function. Each basic block may optionally start with a label (giving the
756 basic block a symbol table entry), contains a list of instructions, and ends
757 with a <a href="#terminators">terminator</a> instruction (such as a branch or
758 function return).</p>
760 <p>The first basic block in a function is special in two ways: it is immediately
761 executed on entrance to the function, and it is not allowed to have predecessor
762 basic blocks (i.e. there can not be any branches to the entry block of a
763 function). Because the block can have no predecessors, it also cannot have any
764 <a href="#i_phi">PHI nodes</a>.</p>
766 <p>LLVM allows an explicit section to be specified for functions. If the target
767 supports it, it will emit functions to the section specified.</p>
769 <p>An explicit alignment may be specified for a function. If not present, or if
770 the alignment is set to zero, the alignment of the function is set by the target
771 to whatever it feels convenient. If an explicit alignment is specified, the
772 function is forced to have at least that much alignment. All alignments must be
778 <!-- ======================================================================= -->
779 <div class="doc_subsection">
780 <a name="aliasstructure">Aliases</a>
782 <div class="doc_text">
783 <p>Aliases act as "second name" for the aliasee value (which can be either
784 function, global variable, another alias or bitcast of global value). Aliases
785 may have an optional <a href="#linkage">linkage type</a>, and an
786 optional <a href="#visibility">visibility style</a>.</p>
790 <div class="doc_code">
792 @<Name> = [Linkage] [Visibility] alias <AliaseeTy> @<Aliasee>
800 <!-- ======================================================================= -->
801 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
802 <div class="doc_text">
803 <p>The return type and each parameter of a function type may have a set of
804 <i>parameter attributes</i> associated with them. Parameter attributes are
805 used to communicate additional information about the result or parameters of
806 a function. Parameter attributes are considered to be part of the function,
807 not of the function type, so functions with different parameter attributes
808 can have the same function type.</p>
810 <p>Parameter attributes are simple keywords that follow the type specified. If
811 multiple parameter attributes are needed, they are space separated. For
814 <div class="doc_code">
816 declare i32 @printf(i8* noalias , ...) nounwind
817 declare i32 @atoi(i8*) nounwind readonly
821 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
822 <tt>readonly</tt>) come immediately after the argument list.</p>
824 <p>Currently, only the following parameter attributes are defined:</p>
826 <dt><tt>zeroext</tt></dt>
827 <dd>This indicates that the parameter should be zero extended just before
828 a call to this function.</dd>
830 <dt><tt>signext</tt></dt>
831 <dd>This indicates that the parameter should be sign extended just before
832 a call to this function.</dd>
834 <dt><tt>inreg</tt></dt>
835 <dd>This indicates that the parameter should be placed in register (if
836 possible) during assembling function call. Support for this attribute is
839 <dt><tt>byval</tt></dt>
840 <dd>This indicates that the pointer parameter should really be passed by
841 value to the function. The attribute implies that a hidden copy of the
842 pointee is made between the caller and the callee, so the callee is unable
843 to modify the value in the callee. This attribute is only valid on llvm
844 pointer arguments. It is generally used to pass structs and arrays by
845 value, but is also valid on scalars (even though this is silly).</dd>
847 <dt><tt>sret</tt></dt>
848 <dd>This indicates that the pointer parameter specifies the address of a
849 structure that is the return value of the function in the source program.
850 Loads and stores to the structure are assumed not to trap.
851 May only be applied to the first parameter.</dd>
853 <dt><tt>noalias</tt></dt>
854 <dd>This indicates that the parameter does not alias any global or any other
855 parameter. The caller is responsible for ensuring that this is the case,
856 usually by placing the value in a stack allocation.</dd>
858 <dt><tt>noreturn</tt></dt>
859 <dd>This function attribute indicates that the function never returns. This
860 indicates to LLVM that every call to this function should be treated as if
861 an <tt>unreachable</tt> instruction immediately followed the call.</dd>
863 <dt><tt>nounwind</tt></dt>
864 <dd>This function attribute indicates that no exceptions unwind out of the
865 function. Usually this is because the function makes no use of exceptions,
866 but it may also be that the function catches any exceptions thrown when
869 <dt><tt>nest</tt></dt>
870 <dd>This indicates that the parameter can be excised using the
871 <a href="#int_trampoline">trampoline intrinsics</a>.</dd>
872 <dt><tt>readonly</tt></dt>
873 <dd>This function attribute indicates that the function has no side-effects
874 except for producing a return value or throwing an exception. The value
875 returned must only depend on the function arguments and/or global variables.
876 It may use values obtained by dereferencing pointers.</dd>
877 <dt><tt>readnone</tt></dt>
878 <dd>A <tt>readnone</tt> function has the same restrictions as a <tt>readonly</tt>
879 function, but in addition it is not allowed to dereference any pointer arguments
885 <!-- ======================================================================= -->
886 <div class="doc_subsection">
887 <a name="gc">Garbage Collector Names</a>
890 <div class="doc_text">
891 <p>Each function may specify a garbage collector name, which is simply a
894 <div class="doc_code"><pre
895 >define void @f() gc "name" { ...</pre></div>
897 <p>The compiler declares the supported values of <i>name</i>. Specifying a
898 collector which will cause the compiler to alter its output in order to support
899 the named garbage collection algorithm.</p>
902 <!-- ======================================================================= -->
903 <div class="doc_subsection">
904 <a name="moduleasm">Module-Level Inline Assembly</a>
907 <div class="doc_text">
909 Modules may contain "module-level inline asm" blocks, which corresponds to the
910 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
911 LLVM and treated as a single unit, but may be separated in the .ll file if
912 desired. The syntax is very simple:
915 <div class="doc_code">
917 module asm "inline asm code goes here"
918 module asm "more can go here"
922 <p>The strings can contain any character by escaping non-printable characters.
923 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
928 The inline asm code is simply printed to the machine code .s file when
929 assembly code is generated.
933 <!-- ======================================================================= -->
934 <div class="doc_subsection">
935 <a name="datalayout">Data Layout</a>
938 <div class="doc_text">
939 <p>A module may specify a target specific data layout string that specifies how
940 data is to be laid out in memory. The syntax for the data layout is simply:</p>
941 <pre> target datalayout = "<i>layout specification</i>"</pre>
942 <p>The <i>layout specification</i> consists of a list of specifications
943 separated by the minus sign character ('-'). Each specification starts with a
944 letter and may include other information after the letter to define some
945 aspect of the data layout. The specifications accepted are as follows: </p>
948 <dd>Specifies that the target lays out data in big-endian form. That is, the
949 bits with the most significance have the lowest address location.</dd>
951 <dd>Specifies that hte target lays out data in little-endian form. That is,
952 the bits with the least significance have the lowest address location.</dd>
953 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
954 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
955 <i>preferred</i> alignments. All sizes are in bits. Specifying the <i>pref</i>
956 alignment is optional. If omitted, the preceding <tt>:</tt> should be omitted
958 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
959 <dd>This specifies the alignment for an integer type of a given bit
960 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
961 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
962 <dd>This specifies the alignment for a vector type of a given bit
964 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
965 <dd>This specifies the alignment for a floating point type of a given bit
966 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
968 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
969 <dd>This specifies the alignment for an aggregate type of a given bit
972 <p>When constructing the data layout for a given target, LLVM starts with a
973 default set of specifications which are then (possibly) overriden by the
974 specifications in the <tt>datalayout</tt> keyword. The default specifications
975 are given in this list:</p>
977 <li><tt>E</tt> - big endian</li>
978 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
979 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
980 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
981 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
982 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
983 <li><tt>i64:32:64</tt> - i64 has abi alignment of 32-bits but preferred
984 alignment of 64-bits</li>
985 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
986 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
987 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
988 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
989 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
991 <p>When llvm is determining the alignment for a given type, it uses the
994 <li>If the type sought is an exact match for one of the specifications, that
995 specification is used.</li>
996 <li>If no match is found, and the type sought is an integer type, then the
997 smallest integer type that is larger than the bitwidth of the sought type is
998 used. If none of the specifications are larger than the bitwidth then the the
999 largest integer type is used. For example, given the default specifications
1000 above, the i7 type will use the alignment of i8 (next largest) while both
1001 i65 and i256 will use the alignment of i64 (largest specified).</li>
1002 <li>If no match is found, and the type sought is a vector type, then the
1003 largest vector type that is smaller than the sought vector type will be used
1004 as a fall back. This happens because <128 x double> can be implemented in
1005 terms of 64 <2 x double>, for example.</li>
1009 <!-- *********************************************************************** -->
1010 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
1011 <!-- *********************************************************************** -->
1013 <div class="doc_text">
1015 <p>The LLVM type system is one of the most important features of the
1016 intermediate representation. Being typed enables a number of
1017 optimizations to be performed on the IR directly, without having to do
1018 extra analyses on the side before the transformation. A strong type
1019 system makes it easier to read the generated code and enables novel
1020 analyses and transformations that are not feasible to perform on normal
1021 three address code representations.</p>
1025 <!-- ======================================================================= -->
1026 <div class="doc_subsection"> <a name="t_classifications">Type
1027 Classifications</a> </div>
1028 <div class="doc_text">
1029 <p>The types fall into a few useful
1030 classifications:</p>
1032 <table border="1" cellspacing="0" cellpadding="4">
1034 <tr><th>Classification</th><th>Types</th></tr>
1036 <td><a href="#t_integer">integer</a></td>
1037 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1040 <td><a href="#t_floating">floating point</a></td>
1041 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1044 <td><a name="t_firstclass">first class</a></td>
1045 <td><a href="#t_integer">integer</a>,
1046 <a href="#t_floating">floating point</a>,
1047 <a href="#t_pointer">pointer</a>,
1048 <a href="#t_vector">vector</a>
1049 <a href="#t_struct">structure</a>,
1050 <a href="#t_array">array</a>,
1051 <a href="#t_label">label</a>.
1055 <td><a href="#t_primitive">primitive</a></td>
1056 <td><a href="#t_label">label</a>,
1057 <a href="#t_void">void</a>,
1058 <a href="#t_integer">integer</a>,
1059 <a href="#t_floating">floating point</a>.</td>
1062 <td><a href="#t_derived">derived</a></td>
1063 <td><a href="#t_integer">integer</a>,
1064 <a href="#t_array">array</a>,
1065 <a href="#t_function">function</a>,
1066 <a href="#t_pointer">pointer</a>,
1067 <a href="#t_struct">structure</a>,
1068 <a href="#t_pstruct">packed structure</a>,
1069 <a href="#t_vector">vector</a>,
1070 <a href="#t_opaque">opaque</a>.
1075 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
1076 most important. Values of these types are the only ones which can be
1077 produced by instructions, passed as arguments, or used as operands to
1081 <!-- ======================================================================= -->
1082 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
1084 <div class="doc_text">
1085 <p>The primitive types are the fundamental building blocks of the LLVM
1090 <!-- _______________________________________________________________________ -->
1091 <div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1093 <div class="doc_text">
1096 <tr><th>Type</th><th>Description</th></tr>
1097 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1098 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1099 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1100 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1101 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1106 <!-- _______________________________________________________________________ -->
1107 <div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1109 <div class="doc_text">
1111 <p>The void type does not represent any value and has no size.</p>
1120 <!-- _______________________________________________________________________ -->
1121 <div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1123 <div class="doc_text">
1125 <p>The label type represents code labels.</p>
1135 <!-- ======================================================================= -->
1136 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1138 <div class="doc_text">
1140 <p>The real power in LLVM comes from the derived types in the system.
1141 This is what allows a programmer to represent arrays, functions,
1142 pointers, and other useful types. Note that these derived types may be
1143 recursive: For example, it is possible to have a two dimensional array.</p>
1147 <!-- _______________________________________________________________________ -->
1148 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1150 <div class="doc_text">
1153 <p>The integer type is a very simple derived type that simply specifies an
1154 arbitrary bit width for the integer type desired. Any bit width from 1 bit to
1155 2^23-1 (about 8 million) can be specified.</p>
1163 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1167 <table class="layout">
1170 <td><tt>i1</tt></td>
1171 <td>a single-bit integer.</td>
1173 <td><tt>i32</tt></td>
1174 <td>a 32-bit integer.</td>
1176 <td><tt>i1942652</tt></td>
1177 <td>a really big integer of over 1 million bits.</td>
1183 <!-- _______________________________________________________________________ -->
1184 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1186 <div class="doc_text">
1190 <p>The array type is a very simple derived type that arranges elements
1191 sequentially in memory. The array type requires a size (number of
1192 elements) and an underlying data type.</p>
1197 [<# elements> x <elementtype>]
1200 <p>The number of elements is a constant integer value; elementtype may
1201 be any type with a size.</p>
1204 <table class="layout">
1206 <td class="left"><tt>[40 x i32]</tt></td>
1207 <td class="left">Array of 40 32-bit integer values.</td>
1210 <td class="left"><tt>[41 x i32]</tt></td>
1211 <td class="left">Array of 41 32-bit integer values.</td>
1214 <td class="left"><tt>[4 x i8]</tt></td>
1215 <td class="left">Array of 4 8-bit integer values.</td>
1218 <p>Here are some examples of multidimensional arrays:</p>
1219 <table class="layout">
1221 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1222 <td class="left">3x4 array of 32-bit integer values.</td>
1225 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1226 <td class="left">12x10 array of single precision floating point values.</td>
1229 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1230 <td class="left">2x3x4 array of 16-bit integer values.</td>
1234 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1235 length array. Normally, accesses past the end of an array are undefined in
1236 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
1237 As a special case, however, zero length arrays are recognized to be variable
1238 length. This allows implementation of 'pascal style arrays' with the LLVM
1239 type "{ i32, [0 x float]}", for example.</p>
1243 <!-- _______________________________________________________________________ -->
1244 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1245 <div class="doc_text">
1249 <p>The function type can be thought of as a function signature. It
1250 consists of a return type and a list of formal parameter types. The
1251 return type of a function type is a scalar type, a void type, or a struct type.
1252 If the return type is a struct type then all struct elements must be of first
1253 class types, and the struct must have at least one element.</p>
1258 <returntype list> (<parameter list>)
1261 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1262 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1263 which indicates that the function takes a variable number of arguments.
1264 Variable argument functions can access their arguments with the <a
1265 href="#int_varargs">variable argument handling intrinsic</a> functions.
1266 '<tt><returntype list></tt>' is a comma-separated list of
1267 <a href="#t_firstclass">first class</a> type specifiers.</p>
1270 <table class="layout">
1272 <td class="left"><tt>i32 (i32)</tt></td>
1273 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1275 </tr><tr class="layout">
1276 <td class="left"><tt>float (i16 signext, i32 *) *
1278 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1279 an <tt>i16</tt> that should be sign extended and a
1280 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1283 </tr><tr class="layout">
1284 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1285 <td class="left">A vararg function that takes at least one
1286 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1287 which returns an integer. This is the signature for <tt>printf</tt> in
1290 </tr><tr class="layout">
1291 <td class="left"><tt>{i32, i32} (i32)</tt></td>
1292 <td class="left">A function taking an <tt>i32></tt>, returning two
1293 <tt> i32 </tt> values as an aggregate of type <tt>{ i32, i32 }</tt>
1299 <!-- _______________________________________________________________________ -->
1300 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1301 <div class="doc_text">
1303 <p>The structure type is used to represent a collection of data members
1304 together in memory. The packing of the field types is defined to match
1305 the ABI of the underlying processor. The elements of a structure may
1306 be any type that has a size.</p>
1307 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1308 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1309 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1312 <pre> { <type list> }<br></pre>
1314 <table class="layout">
1316 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1317 <td class="left">A triple of three <tt>i32</tt> values</td>
1318 </tr><tr class="layout">
1319 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1320 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1321 second element is a <a href="#t_pointer">pointer</a> to a
1322 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1323 an <tt>i32</tt>.</td>
1328 <!-- _______________________________________________________________________ -->
1329 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1331 <div class="doc_text">
1333 <p>The packed structure type is used to represent a collection of data members
1334 together in memory. There is no padding between fields. Further, the alignment
1335 of a packed structure is 1 byte. The elements of a packed structure may
1336 be any type that has a size.</p>
1337 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1338 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1339 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1342 <pre> < { <type list> } > <br></pre>
1344 <table class="layout">
1346 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1347 <td class="left">A triple of three <tt>i32</tt> values</td>
1348 </tr><tr class="layout">
1349 <td class="left"><tt>< { float, i32 (i32)* } ></tt></td>
1350 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1351 second element is a <a href="#t_pointer">pointer</a> to a
1352 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1353 an <tt>i32</tt>.</td>
1358 <!-- _______________________________________________________________________ -->
1359 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1360 <div class="doc_text">
1362 <p>As in many languages, the pointer type represents a pointer or
1363 reference to another object, which must live in memory. Pointer types may have
1364 an optional address space attribute defining the target-specific numbered
1365 address space where the pointed-to object resides. The default address space is
1368 <pre> <type> *<br></pre>
1370 <table class="layout">
1372 <td class="left"><tt>[4x i32]*</tt></td>
1373 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1374 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1377 <td class="left"><tt>i32 (i32 *) *</tt></td>
1378 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
1379 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1383 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1384 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1385 that resides in address space #5.</td>
1390 <!-- _______________________________________________________________________ -->
1391 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1392 <div class="doc_text">
1396 <p>A vector type is a simple derived type that represents a vector
1397 of elements. Vector types are used when multiple primitive data
1398 are operated in parallel using a single instruction (SIMD).
1399 A vector type requires a size (number of
1400 elements) and an underlying primitive data type. Vectors must have a power
1401 of two length (1, 2, 4, 8, 16 ...). Vector types are
1402 considered <a href="#t_firstclass">first class</a>.</p>
1407 < <# elements> x <elementtype> >
1410 <p>The number of elements is a constant integer value; elementtype may
1411 be any integer or floating point type.</p>
1415 <table class="layout">
1417 <td class="left"><tt><4 x i32></tt></td>
1418 <td class="left">Vector of 4 32-bit integer values.</td>
1421 <td class="left"><tt><8 x float></tt></td>
1422 <td class="left">Vector of 8 32-bit floating-point values.</td>
1425 <td class="left"><tt><2 x i64></tt></td>
1426 <td class="left">Vector of 2 64-bit integer values.</td>
1431 <!-- _______________________________________________________________________ -->
1432 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1433 <div class="doc_text">
1437 <p>Opaque types are used to represent unknown types in the system. This
1438 corresponds (for example) to the C notion of a forward declared structure type.
1439 In LLVM, opaque types can eventually be resolved to any type (not just a
1440 structure type).</p>
1450 <table class="layout">
1452 <td class="left"><tt>opaque</tt></td>
1453 <td class="left">An opaque type.</td>
1459 <!-- *********************************************************************** -->
1460 <div class="doc_section"> <a name="constants">Constants</a> </div>
1461 <!-- *********************************************************************** -->
1463 <div class="doc_text">
1465 <p>LLVM has several different basic types of constants. This section describes
1466 them all and their syntax.</p>
1470 <!-- ======================================================================= -->
1471 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1473 <div class="doc_text">
1476 <dt><b>Boolean constants</b></dt>
1478 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1479 constants of the <tt><a href="#t_primitive">i1</a></tt> type.
1482 <dt><b>Integer constants</b></dt>
1484 <dd>Standard integers (such as '4') are constants of the <a
1485 href="#t_integer">integer</a> type. Negative numbers may be used with
1489 <dt><b>Floating point constants</b></dt>
1491 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1492 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1493 notation (see below). The assembler requires the exact decimal value of
1494 a floating-point constant. For example, the assembler accepts 1.25 but
1495 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
1496 constants must have a <a href="#t_floating">floating point</a> type. </dd>
1498 <dt><b>Null pointer constants</b></dt>
1500 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1501 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1505 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1506 of floating point constants. For example, the form '<tt>double
1507 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1508 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1509 (and the only time that they are generated by the disassembler) is when a
1510 floating point constant must be emitted but it cannot be represented as a
1511 decimal floating point number. For example, NaN's, infinities, and other
1512 special values are represented in their IEEE hexadecimal format so that
1513 assembly and disassembly do not cause any bits to change in the constants.</p>
1517 <!-- ======================================================================= -->
1518 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1521 <div class="doc_text">
1522 <p>Aggregate constants arise from aggregation of simple constants
1523 and smaller aggregate constants.</p>
1526 <dt><b>Structure constants</b></dt>
1528 <dd>Structure constants are represented with notation similar to structure
1529 type definitions (a comma separated list of elements, surrounded by braces
1530 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
1531 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>". Structure constants
1532 must have <a href="#t_struct">structure type</a>, and the number and
1533 types of elements must match those specified by the type.
1536 <dt><b>Array constants</b></dt>
1538 <dd>Array constants are represented with notation similar to array type
1539 definitions (a comma separated list of elements, surrounded by square brackets
1540 (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74 ]</tt>". Array
1541 constants must have <a href="#t_array">array type</a>, and the number and
1542 types of elements must match those specified by the type.
1545 <dt><b>Vector constants</b></dt>
1547 <dd>Vector constants are represented with notation similar to vector type
1548 definitions (a comma separated list of elements, surrounded by
1549 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32 42,
1550 i32 11, i32 74, i32 100 ></tt>". Vector constants must have <a
1551 href="#t_vector">vector type</a>, and the number and types of elements must
1552 match those specified by the type.
1555 <dt><b>Zero initialization</b></dt>
1557 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1558 value to zero of <em>any</em> type, including scalar and aggregate types.
1559 This is often used to avoid having to print large zero initializers (e.g. for
1560 large arrays) and is always exactly equivalent to using explicit zero
1567 <!-- ======================================================================= -->
1568 <div class="doc_subsection">
1569 <a name="globalconstants">Global Variable and Function Addresses</a>
1572 <div class="doc_text">
1574 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1575 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1576 constants. These constants are explicitly referenced when the <a
1577 href="#identifiers">identifier for the global</a> is used and always have <a
1578 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1581 <div class="doc_code">
1585 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1591 <!-- ======================================================================= -->
1592 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1593 <div class="doc_text">
1594 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1595 no specific value. Undefined values may be of any type and be used anywhere
1596 a constant is permitted.</p>
1598 <p>Undefined values indicate to the compiler that the program is well defined
1599 no matter what value is used, giving the compiler more freedom to optimize.
1603 <!-- ======================================================================= -->
1604 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1607 <div class="doc_text">
1609 <p>Constant expressions are used to allow expressions involving other constants
1610 to be used as constants. Constant expressions may be of any <a
1611 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1612 that does not have side effects (e.g. load and call are not supported). The
1613 following is the syntax for constant expressions:</p>
1616 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1617 <dd>Truncate a constant to another type. The bit size of CST must be larger
1618 than the bit size of TYPE. Both types must be integers.</dd>
1620 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1621 <dd>Zero extend a constant to another type. The bit size of CST must be
1622 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1624 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1625 <dd>Sign extend a constant to another type. The bit size of CST must be
1626 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1628 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1629 <dd>Truncate a floating point constant to another floating point type. The
1630 size of CST must be larger than the size of TYPE. Both types must be
1631 floating point.</dd>
1633 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1634 <dd>Floating point extend a constant to another type. The size of CST must be
1635 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1637 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
1638 <dd>Convert a floating point constant to the corresponding unsigned integer
1639 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1640 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1641 of the same number of elements. If the value won't fit in the integer type,
1642 the results are undefined.</dd>
1644 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
1645 <dd>Convert a floating point constant to the corresponding signed integer
1646 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1647 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1648 of the same number of elements. If the value won't fit in the integer type,
1649 the results are undefined.</dd>
1651 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
1652 <dd>Convert an unsigned integer constant to the corresponding floating point
1653 constant. TYPE must be a scalar or vector floating point type. CST must be of
1654 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1655 of the same number of elements. If the value won't fit in the floating point
1656 type, the results are undefined.</dd>
1658 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
1659 <dd>Convert a signed integer constant to the corresponding floating point
1660 constant. TYPE must be a scalar or vector floating point type. CST must be of
1661 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1662 of the same number of elements. If the value won't fit in the floating point
1663 type, the results are undefined.</dd>
1665 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
1666 <dd>Convert a pointer typed constant to the corresponding integer constant
1667 TYPE must be an integer type. CST must be of pointer type. The CST value is
1668 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1670 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
1671 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1672 pointer type. CST must be of integer type. The CST value is zero extended,
1673 truncated, or unchanged to make it fit in a pointer size. This one is
1674 <i>really</i> dangerous!</dd>
1676 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
1677 <dd>Convert a constant, CST, to another TYPE. The size of CST and TYPE must be
1678 identical (same number of bits). The conversion is done as if the CST value
1679 was stored to memory and read back as TYPE. In other words, no bits change
1680 with this operator, just the type. This can be used for conversion of
1681 vector types to any other type, as long as they have the same bit width. For
1682 pointers it is only valid to cast to another pointer type.
1685 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1687 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1688 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1689 instruction, the index list may have zero or more indexes, which are required
1690 to make sense for the type of "CSTPTR".</dd>
1692 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1694 <dd>Perform the <a href="#i_select">select operation</a> on
1697 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
1698 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
1700 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
1701 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
1703 <dt><b><tt>vicmp COND ( VAL1, VAL2 )</tt></b></dt>
1704 <dd>Performs the <a href="#i_vicmp">vicmp operation</a> on constants.</dd>
1706 <dt><b><tt>vfcmp COND ( VAL1, VAL2 )</tt></b></dt>
1707 <dd>Performs the <a href="#i_vfcmp">vfcmp operation</a> on constants.</dd>
1709 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1711 <dd>Perform the <a href="#i_extractelement">extractelement
1712 operation</a> on constants.
1714 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1716 <dd>Perform the <a href="#i_insertelement">insertelement
1717 operation</a> on constants.</dd>
1720 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
1722 <dd>Perform the <a href="#i_shufflevector">shufflevector
1723 operation</a> on constants.</dd>
1725 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1727 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1728 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1729 binary</a> operations. The constraints on operands are the same as those for
1730 the corresponding instruction (e.g. no bitwise operations on floating point
1731 values are allowed).</dd>
1735 <!-- *********************************************************************** -->
1736 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1737 <!-- *********************************************************************** -->
1739 <!-- ======================================================================= -->
1740 <div class="doc_subsection">
1741 <a name="inlineasm">Inline Assembler Expressions</a>
1744 <div class="doc_text">
1747 LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
1748 Module-Level Inline Assembly</a>) through the use of a special value. This
1749 value represents the inline assembler as a string (containing the instructions
1750 to emit), a list of operand constraints (stored as a string), and a flag that
1751 indicates whether or not the inline asm expression has side effects. An example
1752 inline assembler expression is:
1755 <div class="doc_code">
1757 i32 (i32) asm "bswap $0", "=r,r"
1762 Inline assembler expressions may <b>only</b> be used as the callee operand of
1763 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
1766 <div class="doc_code">
1768 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
1773 Inline asms with side effects not visible in the constraint list must be marked
1774 as having side effects. This is done through the use of the
1775 '<tt>sideeffect</tt>' keyword, like so:
1778 <div class="doc_code">
1780 call void asm sideeffect "eieio", ""()
1784 <p>TODO: The format of the asm and constraints string still need to be
1785 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1786 need to be documented).
1791 <!-- *********************************************************************** -->
1792 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1793 <!-- *********************************************************************** -->
1795 <div class="doc_text">
1797 <p>The LLVM instruction set consists of several different
1798 classifications of instructions: <a href="#terminators">terminator
1799 instructions</a>, <a href="#binaryops">binary instructions</a>,
1800 <a href="#bitwiseops">bitwise binary instructions</a>, <a
1801 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1802 instructions</a>.</p>
1806 <!-- ======================================================================= -->
1807 <div class="doc_subsection"> <a name="terminators">Terminator
1808 Instructions</a> </div>
1810 <div class="doc_text">
1812 <p>As mentioned <a href="#functionstructure">previously</a>, every
1813 basic block in a program ends with a "Terminator" instruction, which
1814 indicates which block should be executed after the current block is
1815 finished. These terminator instructions typically yield a '<tt>void</tt>'
1816 value: they produce control flow, not values (the one exception being
1817 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1818 <p>There are six different terminator instructions: the '<a
1819 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1820 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1821 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1822 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1823 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1827 <!-- _______________________________________________________________________ -->
1828 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1829 Instruction</a> </div>
1830 <div class="doc_text">
1832 <pre> ret <type> <value> <i>; Return a value from a non-void function</i>
1833 ret void <i>; Return from void function</i>
1834 ret <type> <value>, <type> <value> <i>; Return two values from a non-void function </i>
1839 <p>The '<tt>ret</tt>' instruction is used to return control flow (and a
1840 value) from a function back to the caller.</p>
1841 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
1842 returns value(s) and then causes control flow, and one that just causes
1843 control flow to occur.</p>
1847 <p>The '<tt>ret</tt>' instruction may return zero, one or multiple values.
1848 The type of each return value must be a '<a href="#t_firstclass">first
1849 class</a>' type. Note that a function is not <a href="#wellformed">well
1850 formed</a> if there exists a '<tt>ret</tt>' instruction inside of the
1851 function that returns values that do not match the return type of the
1856 <p>When the '<tt>ret</tt>' instruction is executed, control flow
1857 returns back to the calling function's context. If the caller is a "<a
1858 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1859 the instruction after the call. If the caller was an "<a
1860 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1861 at the beginning of the "normal" destination block. If the instruction
1862 returns a value, that value shall set the call or invoke instruction's
1863 return value. If the instruction returns multiple values then these
1864 values can only be accessed through a '<a href="#i_getresult"><tt>getresult</tt>
1865 </a>' instruction.</p>
1870 ret i32 5 <i>; Return an integer value of 5</i>
1871 ret void <i>; Return from a void function</i>
1872 ret i32 4, i8 2 <i>; Return two values 4 and 2 </i>
1875 <!-- _______________________________________________________________________ -->
1876 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1877 <div class="doc_text">
1879 <pre> br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1882 <p>The '<tt>br</tt>' instruction is used to cause control flow to
1883 transfer to a different basic block in the current function. There are
1884 two forms of this instruction, corresponding to a conditional branch
1885 and an unconditional branch.</p>
1887 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1888 single '<tt>i1</tt>' value and two '<tt>label</tt>' values. The
1889 unconditional form of the '<tt>br</tt>' instruction takes a single
1890 '<tt>label</tt>' value as a target.</p>
1892 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
1893 argument is evaluated. If the value is <tt>true</tt>, control flows
1894 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1895 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1897 <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
1898 href="#i_ret">ret</a> i32 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> i32 0<br></pre>
1900 <!-- _______________________________________________________________________ -->
1901 <div class="doc_subsubsection">
1902 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
1905 <div class="doc_text">
1909 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
1914 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
1915 several different places. It is a generalization of the '<tt>br</tt>'
1916 instruction, allowing a branch to occur to one of many possible
1922 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
1923 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
1924 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
1925 table is not allowed to contain duplicate constant entries.</p>
1929 <p>The <tt>switch</tt> instruction specifies a table of values and
1930 destinations. When the '<tt>switch</tt>' instruction is executed, this
1931 table is searched for the given value. If the value is found, control flow is
1932 transfered to the corresponding destination; otherwise, control flow is
1933 transfered to the default destination.</p>
1935 <h5>Implementation:</h5>
1937 <p>Depending on properties of the target machine and the particular
1938 <tt>switch</tt> instruction, this instruction may be code generated in different
1939 ways. For example, it could be generated as a series of chained conditional
1940 branches or with a lookup table.</p>
1945 <i>; Emulate a conditional br instruction</i>
1946 %Val = <a href="#i_zext">zext</a> i1 %value to i32
1947 switch i32 %Val, label %truedest [i32 0, label %falsedest ]
1949 <i>; Emulate an unconditional br instruction</i>
1950 switch i32 0, label %dest [ ]
1952 <i>; Implement a jump table:</i>
1953 switch i32 %val, label %otherwise [ i32 0, label %onzero
1955 i32 2, label %ontwo ]
1959 <!-- _______________________________________________________________________ -->
1960 <div class="doc_subsubsection">
1961 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
1964 <div class="doc_text">
1969 <result> = invoke [<a href="#callingconv">cconv</a>] <ptr to function ty> <function ptr val>(<function args>)
1970 to label <normal label> unwind label <exception label>
1975 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
1976 function, with the possibility of control flow transfer to either the
1977 '<tt>normal</tt>' label or the
1978 '<tt>exception</tt>' label. If the callee function returns with the
1979 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
1980 "normal" label. If the callee (or any indirect callees) returns with the "<a
1981 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
1982 continued at the dynamically nearest "exception" label. If the callee function
1983 returns multiple values then individual return values are only accessible through
1984 a '<tt><a href="#i_getresult">getresult</a></tt>' instruction.</p>
1988 <p>This instruction requires several arguments:</p>
1992 The optional "cconv" marker indicates which <a href="#callingconv">calling
1993 convention</a> the call should use. If none is specified, the call defaults
1994 to using C calling conventions.
1996 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
1997 function value being invoked. In most cases, this is a direct function
1998 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
1999 an arbitrary pointer to function value.
2002 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
2003 function to be invoked. </li>
2005 <li>'<tt>function args</tt>': argument list whose types match the function
2006 signature argument types. If the function signature indicates the function
2007 accepts a variable number of arguments, the extra arguments can be
2010 <li>'<tt>normal label</tt>': the label reached when the called function
2011 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
2013 <li>'<tt>exception label</tt>': the label reached when a callee returns with
2014 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
2020 <p>This instruction is designed to operate as a standard '<tt><a
2021 href="#i_call">call</a></tt>' instruction in most regards. The primary
2022 difference is that it establishes an association with a label, which is used by
2023 the runtime library to unwind the stack.</p>
2025 <p>This instruction is used in languages with destructors to ensure that proper
2026 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
2027 exception. Additionally, this is important for implementation of
2028 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
2032 %retval = invoke i32 @Test(i32 15) to label %Continue
2033 unwind label %TestCleanup <i>; {i32}:retval set</i>
2034 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
2035 unwind label %TestCleanup <i>; {i32}:retval set</i>
2040 <!-- _______________________________________________________________________ -->
2042 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
2043 Instruction</a> </div>
2045 <div class="doc_text">
2054 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
2055 at the first callee in the dynamic call stack which used an <a
2056 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
2057 primarily used to implement exception handling.</p>
2061 <p>The '<tt>unwind</tt>' instruction causes execution of the current function to
2062 immediately halt. The dynamic call stack is then searched for the first <a
2063 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
2064 execution continues at the "exceptional" destination block specified by the
2065 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
2066 dynamic call chain, undefined behavior results.</p>
2069 <!-- _______________________________________________________________________ -->
2071 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
2072 Instruction</a> </div>
2074 <div class="doc_text">
2083 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
2084 instruction is used to inform the optimizer that a particular portion of the
2085 code is not reachable. This can be used to indicate that the code after a
2086 no-return function cannot be reached, and other facts.</p>
2090 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
2095 <!-- ======================================================================= -->
2096 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
2097 <div class="doc_text">
2098 <p>Binary operators are used to do most of the computation in a
2099 program. They require two operands of the same type, execute an operation on them, and
2100 produce a single value. The operands might represent
2101 multiple data, as is the case with the <a href="#t_vector">vector</a> data type.
2102 The result value has the same type as its operands.</p>
2103 <p>There are several different binary operators:</p>
2105 <!-- _______________________________________________________________________ -->
2106 <div class="doc_subsubsection">
2107 <a name="i_add">'<tt>add</tt>' Instruction</a>
2110 <div class="doc_text">
2115 <result> = add <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2120 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
2124 <p>The two arguments to the '<tt>add</tt>' instruction must be <a
2125 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>, or
2126 <a href="#t_vector">vector</a> values. Both arguments must have identical
2131 <p>The value produced is the integer or floating point sum of the two
2134 <p>If an integer sum has unsigned overflow, the result returned is the
2135 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2138 <p>Because LLVM integers use a two's complement representation, this
2139 instruction is appropriate for both signed and unsigned integers.</p>
2144 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
2147 <!-- _______________________________________________________________________ -->
2148 <div class="doc_subsubsection">
2149 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
2152 <div class="doc_text">
2157 <result> = sub <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2162 <p>The '<tt>sub</tt>' instruction returns the difference of its two
2165 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
2166 '<tt>neg</tt>' instruction present in most other intermediate
2167 representations.</p>
2171 <p>The two arguments to the '<tt>sub</tt>' instruction must be <a
2172 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>,
2173 or <a href="#t_vector">vector</a> values. Both arguments must have identical
2178 <p>The value produced is the integer or floating point difference of
2179 the two operands.</p>
2181 <p>If an integer difference has unsigned overflow, the result returned is the
2182 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2185 <p>Because LLVM integers use a two's complement representation, this
2186 instruction is appropriate for both signed and unsigned integers.</p>
2190 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
2191 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
2195 <!-- _______________________________________________________________________ -->
2196 <div class="doc_subsubsection">
2197 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
2200 <div class="doc_text">
2203 <pre> <result> = mul <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2206 <p>The '<tt>mul</tt>' instruction returns the product of its two
2211 <p>The two arguments to the '<tt>mul</tt>' instruction must be <a
2212 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>,
2213 or <a href="#t_vector">vector</a> values. Both arguments must have identical
2218 <p>The value produced is the integer or floating point product of the
2221 <p>If the result of an integer multiplication has unsigned overflow,
2222 the result returned is the mathematical result modulo
2223 2<sup>n</sup>, where n is the bit width of the result.</p>
2224 <p>Because LLVM integers use a two's complement representation, and the
2225 result is the same width as the operands, this instruction returns the
2226 correct result for both signed and unsigned integers. If a full product
2227 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands
2228 should be sign-extended or zero-extended as appropriate to the
2229 width of the full product.</p>
2231 <pre> <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
2235 <!-- _______________________________________________________________________ -->
2236 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
2238 <div class="doc_text">
2240 <pre> <result> = udiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2243 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two
2248 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
2249 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2250 values. Both arguments must have identical types.</p>
2254 <p>The value produced is the unsigned integer quotient of the two operands.</p>
2255 <p>Note that unsigned integer division and signed integer division are distinct
2256 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
2257 <p>Division by zero leads to undefined behavior.</p>
2259 <pre> <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2262 <!-- _______________________________________________________________________ -->
2263 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
2265 <div class="doc_text">
2268 <result> = sdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2273 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
2278 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
2279 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2280 values. Both arguments must have identical types.</p>
2283 <p>The value produced is the signed integer quotient of the two operands rounded towards zero.</p>
2284 <p>Note that signed integer division and unsigned integer division are distinct
2285 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
2286 <p>Division by zero leads to undefined behavior. Overflow also leads to
2287 undefined behavior; this is a rare case, but can occur, for example,
2288 by doing a 32-bit division of -2147483648 by -1.</p>
2290 <pre> <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2293 <!-- _______________________________________________________________________ -->
2294 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
2295 Instruction</a> </div>
2296 <div class="doc_text">
2299 <result> = fdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2303 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
2308 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
2309 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a>
2310 of floating point values. Both arguments must have identical types.</p>
2314 <p>The value produced is the floating point quotient of the two operands.</p>
2319 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
2323 <!-- _______________________________________________________________________ -->
2324 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
2326 <div class="doc_text">
2328 <pre> <result> = urem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2331 <p>The '<tt>urem</tt>' instruction returns the remainder from the
2332 unsigned division of its two arguments.</p>
2334 <p>The two arguments to the '<tt>urem</tt>' instruction must be
2335 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2336 values. Both arguments must have identical types.</p>
2338 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
2339 This instruction always performs an unsigned division to get the remainder.</p>
2340 <p>Note that unsigned integer remainder and signed integer remainder are
2341 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
2342 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
2344 <pre> <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2348 <!-- _______________________________________________________________________ -->
2349 <div class="doc_subsubsection">
2350 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
2353 <div class="doc_text">
2358 <result> = srem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2363 <p>The '<tt>srem</tt>' instruction returns the remainder from the
2364 signed division of its two operands. This instruction can also take
2365 <a href="#t_vector">vector</a> versions of the values in which case
2366 the elements must be integers.</p>
2370 <p>The two arguments to the '<tt>srem</tt>' instruction must be
2371 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2372 values. Both arguments must have identical types.</p>
2376 <p>This instruction returns the <i>remainder</i> of a division (where the result
2377 has the same sign as the dividend, <tt>var1</tt>), not the <i>modulo</i>
2378 operator (where the result has the same sign as the divisor, <tt>var2</tt>) of
2379 a value. For more information about the difference, see <a
2380 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
2381 Math Forum</a>. For a table of how this is implemented in various languages,
2382 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
2383 Wikipedia: modulo operation</a>.</p>
2384 <p>Note that signed integer remainder and unsigned integer remainder are
2385 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
2386 <p>Taking the remainder of a division by zero leads to undefined behavior.
2387 Overflow also leads to undefined behavior; this is a rare case, but can occur,
2388 for example, by taking the remainder of a 32-bit division of -2147483648 by -1.
2389 (The remainder doesn't actually overflow, but this rule lets srem be
2390 implemented using instructions that return both the result of the division
2391 and the remainder.)</p>
2393 <pre> <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2397 <!-- _______________________________________________________________________ -->
2398 <div class="doc_subsubsection">
2399 <a name="i_frem">'<tt>frem</tt>' Instruction</a> </div>
2401 <div class="doc_text">
2404 <pre> <result> = frem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2407 <p>The '<tt>frem</tt>' instruction returns the remainder from the
2408 division of its two operands.</p>
2410 <p>The two arguments to the '<tt>frem</tt>' instruction must be
2411 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a>
2412 of floating point values. Both arguments must have identical types.</p>
2416 <p>This instruction returns the <i>remainder</i> of a division.
2417 The remainder has the same sign as the dividend.</p>
2422 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
2426 <!-- ======================================================================= -->
2427 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
2428 Operations</a> </div>
2429 <div class="doc_text">
2430 <p>Bitwise binary operators are used to do various forms of
2431 bit-twiddling in a program. They are generally very efficient
2432 instructions and can commonly be strength reduced from other
2433 instructions. They require two operands of the same type, execute an operation on them,
2434 and produce a single value. The resulting value is the same type as its operands.</p>
2437 <!-- _______________________________________________________________________ -->
2438 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
2439 Instruction</a> </div>
2440 <div class="doc_text">
2442 <pre> <result> = shl <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2447 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2448 the left a specified number of bits.</p>
2452 <p>Both arguments to the '<tt>shl</tt>' instruction must be the same <a
2453 href="#t_integer">integer</a> type. '<tt>var2</tt>' is treated as an
2454 unsigned value. This instruction does not support
2455 <a href="#t_vector">vector</a> operands.</p>
2459 <p>The value produced is <tt>var1</tt> * 2<sup><tt>var2</tt></sup> mod 2<sup>n</sup>,
2460 where n is the width of the result. If <tt>var2</tt> is (statically or dynamically) negative or
2461 equal to or larger than the number of bits in <tt>var1</tt>, the result is undefined.</p>
2463 <h5>Example:</h5><pre>
2464 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
2465 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
2466 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
2467 <result> = shl i32 1, 32 <i>; undefined</i>
2470 <!-- _______________________________________________________________________ -->
2471 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
2472 Instruction</a> </div>
2473 <div class="doc_text">
2475 <pre> <result> = lshr <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2479 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
2480 operand shifted to the right a specified number of bits with zero fill.</p>
2483 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
2484 <a href="#t_integer">integer</a> type. '<tt>var2</tt>' is treated as an
2485 unsigned value. This instruction does not support
2486 <a href="#t_vector">vector</a> operands.</p>
2490 <p>This instruction always performs a logical shift right operation. The most
2491 significant bits of the result will be filled with zero bits after the
2492 shift. If <tt>var2</tt> is (statically or dynamically) equal to or larger than
2493 the number of bits in <tt>var1</tt>, the result is undefined.</p>
2497 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
2498 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
2499 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
2500 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
2501 <result> = lshr i32 1, 32 <i>; undefined</i>
2505 <!-- _______________________________________________________________________ -->
2506 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
2507 Instruction</a> </div>
2508 <div class="doc_text">
2511 <pre> <result> = ashr <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2515 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
2516 operand shifted to the right a specified number of bits with sign extension.</p>
2519 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
2520 <a href="#t_integer">integer</a> type. '<tt>var2</tt>' is treated as an
2521 unsigned value. This instruction does not support
2522 <a href="#t_vector">vector</a> operands.</p>
2525 <p>This instruction always performs an arithmetic shift right operation,
2526 The most significant bits of the result will be filled with the sign bit
2527 of <tt>var1</tt>. If <tt>var2</tt> is (statically or dynamically) equal to or
2528 larger than the number of bits in <tt>var1</tt>, the result is undefined.
2533 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
2534 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
2535 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
2536 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
2537 <result> = ashr i32 1, 32 <i>; undefined</i>
2541 <!-- _______________________________________________________________________ -->
2542 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
2543 Instruction</a> </div>
2545 <div class="doc_text">
2550 <result> = and <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2555 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
2556 its two operands.</p>
2560 <p>The two arguments to the '<tt>and</tt>' instruction must be
2561 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2562 values. Both arguments must have identical types.</p>
2565 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
2567 <div style="align: center">
2568 <table border="1" cellspacing="0" cellpadding="4">
2600 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
2601 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
2602 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
2605 <!-- _______________________________________________________________________ -->
2606 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
2607 <div class="doc_text">
2609 <pre> <result> = or <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2612 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
2613 or of its two operands.</p>
2616 <p>The two arguments to the '<tt>or</tt>' instruction must be
2617 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2618 values. Both arguments must have identical types.</p>
2620 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
2622 <div style="align: center">
2623 <table border="1" cellspacing="0" cellpadding="4">
2654 <pre> <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
2655 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
2656 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
2659 <!-- _______________________________________________________________________ -->
2660 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
2661 Instruction</a> </div>
2662 <div class="doc_text">
2664 <pre> <result> = xor <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2667 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
2668 or of its two operands. The <tt>xor</tt> is used to implement the
2669 "one's complement" operation, which is the "~" operator in C.</p>
2671 <p>The two arguments to the '<tt>xor</tt>' instruction must be
2672 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2673 values. Both arguments must have identical types.</p>
2677 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
2679 <div style="align: center">
2680 <table border="1" cellspacing="0" cellpadding="4">
2712 <pre> <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
2713 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
2714 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
2715 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
2719 <!-- ======================================================================= -->
2720 <div class="doc_subsection">
2721 <a name="vectorops">Vector Operations</a>
2724 <div class="doc_text">
2726 <p>LLVM supports several instructions to represent vector operations in a
2727 target-independent manner. These instructions cover the element-access and
2728 vector-specific operations needed to process vectors effectively. While LLVM
2729 does directly support these vector operations, many sophisticated algorithms
2730 will want to use target-specific intrinsics to take full advantage of a specific
2735 <!-- _______________________________________________________________________ -->
2736 <div class="doc_subsubsection">
2737 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2740 <div class="doc_text">
2745 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
2751 The '<tt>extractelement</tt>' instruction extracts a single scalar
2752 element from a vector at a specified index.
2759 The first operand of an '<tt>extractelement</tt>' instruction is a
2760 value of <a href="#t_vector">vector</a> type. The second operand is
2761 an index indicating the position from which to extract the element.
2762 The index may be a variable.</p>
2767 The result is a scalar of the same type as the element type of
2768 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2769 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2770 results are undefined.
2776 %result = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
2781 <!-- _______________________________________________________________________ -->
2782 <div class="doc_subsubsection">
2783 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
2786 <div class="doc_text">
2791 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
2797 The '<tt>insertelement</tt>' instruction inserts a scalar
2798 element into a vector at a specified index.
2805 The first operand of an '<tt>insertelement</tt>' instruction is a
2806 value of <a href="#t_vector">vector</a> type. The second operand is a
2807 scalar value whose type must equal the element type of the first
2808 operand. The third operand is an index indicating the position at
2809 which to insert the value. The index may be a variable.</p>
2814 The result is a vector of the same type as <tt>val</tt>. Its
2815 element values are those of <tt>val</tt> except at position
2816 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
2817 exceeds the length of <tt>val</tt>, the results are undefined.
2823 %result = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
2827 <!-- _______________________________________________________________________ -->
2828 <div class="doc_subsubsection">
2829 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
2832 <div class="doc_text">
2837 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <n x i32> <mask> <i>; yields <n x <ty>></i>
2843 The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
2844 from two input vectors, returning a vector of the same type.
2850 The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
2851 with types that match each other and types that match the result of the
2852 instruction. The third argument is a shuffle mask, which has the same number
2853 of elements as the other vector type, but whose element type is always 'i32'.
2857 The shuffle mask operand is required to be a constant vector with either
2858 constant integer or undef values.
2864 The elements of the two input vectors are numbered from left to right across
2865 both of the vectors. The shuffle mask operand specifies, for each element of
2866 the result vector, which element of the two input registers the result element
2867 gets. The element selector may be undef (meaning "don't care") and the second
2868 operand may be undef if performing a shuffle from only one vector.
2874 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
2875 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
2876 %result = shufflevector <4 x i32> %v1, <4 x i32> undef,
2877 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
2882 <!-- ======================================================================= -->
2883 <div class="doc_subsection">
2884 <a name="aggregateops">Aggregate Operations</a>
2887 <div class="doc_text">
2889 <p>LLVM supports several instructions for working with aggregate values.
2894 <!-- _______________________________________________________________________ -->
2895 <div class="doc_subsubsection">
2896 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
2899 <div class="doc_text">
2904 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
2910 The '<tt>extractvalue</tt>' instruction extracts the value of a struct field
2911 or array element from an aggregate value.
2918 The first operand of an '<tt>extractvalue</tt>' instruction is a
2919 value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a>
2920 type. The operands are constant indices to specify which value to extract
2921 in a similar manner as indices in a
2922 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
2928 The result is the value at the position in the aggregate specified by
2935 %result = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
2940 <!-- _______________________________________________________________________ -->
2941 <div class="doc_subsubsection">
2942 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
2945 <div class="doc_text">
2950 <result> = insertvalue <aggregate type> <val>, <ty> <val>, <idx> <i>; yields <n x <ty>></i>
2956 The '<tt>insertvalue</tt>' instruction inserts a value
2957 into a struct field or array element in an aggregate.
2964 The first operand of an '<tt>insertvalue</tt>' instruction is a
2965 value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a> type.
2966 The second operand is a first-class value to insert.
2967 The following operands are constant indices
2968 indicating the position at which to insert the value in a similar manner as
2970 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
2971 The value to insert must have the same type as the value identified
2977 The result is an aggregate of the same type as <tt>val</tt>. Its
2978 value is that of <tt>val</tt> except that the value at the position
2979 specified by the indices is that of <tt>elt</tt>.
2985 %result = insertvalue {i32, float} %agg, 1, 0 <i>; yields {i32, float}</i>
2990 <!-- ======================================================================= -->
2991 <div class="doc_subsection">
2992 <a name="memoryops">Memory Access and Addressing Operations</a>
2995 <div class="doc_text">
2997 <p>A key design point of an SSA-based representation is how it
2998 represents memory. In LLVM, no memory locations are in SSA form, which
2999 makes things very simple. This section describes how to read, write,
3000 allocate, and free memory in LLVM.</p>
3004 <!-- _______________________________________________________________________ -->
3005 <div class="doc_subsubsection">
3006 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
3009 <div class="doc_text">
3014 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3019 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
3020 heap and returns a pointer to it. The object is always allocated in the generic
3021 address space (address space zero).</p>
3025 <p>The '<tt>malloc</tt>' instruction allocates
3026 <tt>sizeof(<type>)*NumElements</tt>
3027 bytes of memory from the operating system and returns a pointer of the
3028 appropriate type to the program. If "NumElements" is specified, it is the
3029 number of elements allocated, otherwise "NumElements" is defaulted to be one.
3030 If a constant alignment is specified, the value result of the allocation is guaranteed to
3031 be aligned to at least that boundary. If not specified, or if zero, the target can
3032 choose to align the allocation on any convenient boundary.</p>
3034 <p>'<tt>type</tt>' must be a sized type.</p>
3038 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
3039 a pointer is returned. The result of a zero byte allocattion is undefined. The
3040 result is null if there is insufficient memory available.</p>
3045 %array = malloc [4 x i8 ] <i>; yields {[%4 x i8]*}:array</i>
3047 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
3048 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
3049 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
3050 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
3051 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
3055 <!-- _______________________________________________________________________ -->
3056 <div class="doc_subsubsection">
3057 <a name="i_free">'<tt>free</tt>' Instruction</a>
3060 <div class="doc_text">
3065 free <type> <value> <i>; yields {void}</i>
3070 <p>The '<tt>free</tt>' instruction returns memory back to the unused
3071 memory heap to be reallocated in the future.</p>
3075 <p>'<tt>value</tt>' shall be a pointer value that points to a value
3076 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
3081 <p>Access to the memory pointed to by the pointer is no longer defined
3082 after this instruction executes. If the pointer is null, the operation
3088 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
3089 free [4 x i8]* %array
3093 <!-- _______________________________________________________________________ -->
3094 <div class="doc_subsubsection">
3095 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
3098 <div class="doc_text">
3103 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3108 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
3109 currently executing function, to be automatically released when this function
3110 returns to its caller. The object is always allocated in the generic address
3111 space (address space zero).</p>
3115 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
3116 bytes of memory on the runtime stack, returning a pointer of the
3117 appropriate type to the program. If "NumElements" is specified, it is the
3118 number of elements allocated, otherwise "NumElements" is defaulted to be one.
3119 If a constant alignment is specified, the value result of the allocation is guaranteed
3120 to be aligned to at least that boundary. If not specified, or if zero, the target
3121 can choose to align the allocation on any convenient boundary.</p>
3123 <p>'<tt>type</tt>' may be any sized type.</p>
3127 <p>Memory is allocated; a pointer is returned. The operation is undefiend if
3128 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
3129 memory is automatically released when the function returns. The '<tt>alloca</tt>'
3130 instruction is commonly used to represent automatic variables that must
3131 have an address available. When the function returns (either with the <tt><a
3132 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
3133 instructions), the memory is reclaimed. Allocating zero bytes
3134 is legal, but the result is undefined.</p>
3139 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
3140 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
3141 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
3142 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
3146 <!-- _______________________________________________________________________ -->
3147 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
3148 Instruction</a> </div>
3149 <div class="doc_text">
3151 <pre> <result> = load <ty>* <pointer>[, align <alignment>]<br> <result> = volatile load <ty>* <pointer>[, align <alignment>]<br></pre>
3153 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
3155 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
3156 address from which to load. The pointer must point to a <a
3157 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
3158 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
3159 the number or order of execution of this <tt>load</tt> with other
3160 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
3163 The optional constant "align" argument specifies the alignment of the operation
3164 (that is, the alignment of the memory address). A value of 0 or an
3165 omitted "align" argument means that the operation has the preferential
3166 alignment for the target. It is the responsibility of the code emitter
3167 to ensure that the alignment information is correct. Overestimating
3168 the alignment results in an undefined behavior. Underestimating the
3169 alignment may produce less efficient code. An alignment of 1 is always
3173 <p>The location of memory pointed to is loaded.</p>
3175 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3177 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
3178 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
3181 <!-- _______________________________________________________________________ -->
3182 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
3183 Instruction</a> </div>
3184 <div class="doc_text">
3186 <pre> store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3187 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3190 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
3192 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
3193 to store and an address at which to store it. The type of the '<tt><pointer></tt>'
3194 operand must be a pointer to the <a href="#t_firstclass">first class</a> type
3195 of the '<tt><value></tt>'
3196 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
3197 optimizer is not allowed to modify the number or order of execution of
3198 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
3199 href="#i_store">store</a></tt> instructions.</p>
3201 The optional constant "align" argument specifies the alignment of the operation
3202 (that is, the alignment of the memory address). A value of 0 or an
3203 omitted "align" argument means that the operation has the preferential
3204 alignment for the target. It is the responsibility of the code emitter
3205 to ensure that the alignment information is correct. Overestimating
3206 the alignment results in an undefined behavior. Underestimating the
3207 alignment may produce less efficient code. An alignment of 1 is always
3211 <p>The contents of memory are updated to contain '<tt><value></tt>'
3212 at the location specified by the '<tt><pointer></tt>' operand.</p>
3214 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3215 store i32 3, i32* %ptr <i>; yields {void}</i>
3216 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
3220 <!-- _______________________________________________________________________ -->
3221 <div class="doc_subsubsection">
3222 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
3225 <div class="doc_text">
3228 <result> = getelementptr <ty>* <ptrval>{, <ty> <idx>}*
3234 The '<tt>getelementptr</tt>' instruction is used to get the address of a
3235 subelement of an aggregate data structure.</p>
3239 <p>This instruction takes a list of integer operands that indicate what
3240 elements of the aggregate object to index to. The actual types of the arguments
3241 provided depend on the type of the first pointer argument. The
3242 '<tt>getelementptr</tt>' instruction is used to index down through the type
3243 levels of a structure or to a specific index in an array. When indexing into a
3244 structure, only <tt>i32</tt> integer constants are allowed. When indexing
3245 into an array or pointer, only integers of 32 or 64 bits are allowed; 32-bit
3246 values will be sign extended to 64-bits if required.</p>
3248 <p>For example, let's consider a C code fragment and how it gets
3249 compiled to LLVM:</p>
3251 <div class="doc_code">
3264 int *foo(struct ST *s) {
3265 return &s[1].Z.B[5][13];
3270 <p>The LLVM code generated by the GCC frontend is:</p>
3272 <div class="doc_code">
3274 %RT = type { i8 , [10 x [20 x i32]], i8 }
3275 %ST = type { i32, double, %RT }
3277 define i32* %foo(%ST* %s) {
3279 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
3287 <p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
3288 on the pointer type that is being indexed into. <a href="#t_pointer">Pointer</a>
3289 and <a href="#t_array">array</a> types can use a 32-bit or 64-bit
3290 <a href="#t_integer">integer</a> type but the value will always be sign extended
3291 to 64-bits. <a href="#t_struct">Structure</a> and <a href="#t_pstruct">packed
3292 structure</a> types require <tt>i32</tt> <b>constants</b>.</p>
3294 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
3295 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
3296 }</tt>' type, a structure. The second index indexes into the third element of
3297 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
3298 i8 }</tt>' type, another structure. The third index indexes into the second
3299 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
3300 array. The two dimensions of the array are subscripted into, yielding an
3301 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
3302 to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
3304 <p>Note that it is perfectly legal to index partially through a
3305 structure, returning a pointer to an inner element. Because of this,
3306 the LLVM code for the given testcase is equivalent to:</p>
3309 define i32* %foo(%ST* %s) {
3310 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
3311 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
3312 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
3313 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
3314 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
3319 <p>Note that it is undefined to access an array out of bounds: array and
3320 pointer indexes must always be within the defined bounds of the array type.
3321 The one exception for this rule is zero length arrays. These arrays are
3322 defined to be accessible as variable length arrays, which requires access
3323 beyond the zero'th element.</p>
3325 <p>The getelementptr instruction is often confusing. For some more insight
3326 into how it works, see <a href="GetElementPtr.html">the getelementptr
3332 <i>; yields [12 x i8]*:aptr</i>
3333 %aptr = getelementptr {i32, [12 x i8]}* %sptr, i64 0, i32 1
3337 <!-- ======================================================================= -->
3338 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
3340 <div class="doc_text">
3341 <p>The instructions in this category are the conversion instructions (casting)
3342 which all take a single operand and a type. They perform various bit conversions
3346 <!-- _______________________________________________________________________ -->
3347 <div class="doc_subsubsection">
3348 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
3350 <div class="doc_text">
3354 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
3359 The '<tt>trunc</tt>' instruction truncates its operand to the type <tt>ty2</tt>.
3364 The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
3365 be an <a href="#t_integer">integer</a> type, and a type that specifies the size
3366 and type of the result, which must be an <a href="#t_integer">integer</a>
3367 type. The bit size of <tt>value</tt> must be larger than the bit size of
3368 <tt>ty2</tt>. Equal sized types are not allowed.</p>
3372 The '<tt>trunc</tt>' instruction truncates the high order bits in <tt>value</tt>
3373 and converts the remaining bits to <tt>ty2</tt>. Since the source size must be
3374 larger than the destination size, <tt>trunc</tt> cannot be a <i>no-op cast</i>.
3375 It will always truncate bits.</p>
3379 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
3380 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
3381 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
3385 <!-- _______________________________________________________________________ -->
3386 <div class="doc_subsubsection">
3387 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
3389 <div class="doc_text">
3393 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
3397 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
3402 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
3403 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3404 also be of <a href="#t_integer">integer</a> type. The bit size of the
3405 <tt>value</tt> must be smaller than the bit size of the destination type,
3409 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
3410 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
3412 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
3416 %X = zext i32 257 to i64 <i>; yields i64:257</i>
3417 %Y = zext i1 true to i32 <i>; yields i32:1</i>
3421 <!-- _______________________________________________________________________ -->
3422 <div class="doc_subsubsection">
3423 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
3425 <div class="doc_text">
3429 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
3433 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
3437 The '<tt>sext</tt>' instruction takes a value to cast, which must be of
3438 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3439 also be of <a href="#t_integer">integer</a> type. The bit size of the
3440 <tt>value</tt> must be smaller than the bit size of the destination type,
3445 The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
3446 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size of
3447 the type <tt>ty2</tt>.</p>
3449 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
3453 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
3454 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
3458 <!-- _______________________________________________________________________ -->
3459 <div class="doc_subsubsection">
3460 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
3463 <div class="doc_text">
3468 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
3472 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
3477 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
3478 point</a> value to cast and a <a href="#t_floating">floating point</a> type to
3479 cast it to. The size of <tt>value</tt> must be larger than the size of
3480 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
3481 <i>no-op cast</i>.</p>
3484 <p> The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
3485 <a href="#t_floating">floating point</a> type to a smaller
3486 <a href="#t_floating">floating point</a> type. If the value cannot fit within
3487 the destination type, <tt>ty2</tt>, then the results are undefined.</p>
3491 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
3492 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
3496 <!-- _______________________________________________________________________ -->
3497 <div class="doc_subsubsection">
3498 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
3500 <div class="doc_text">
3504 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
3508 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
3509 floating point value.</p>
3512 <p>The '<tt>fpext</tt>' instruction takes a
3513 <a href="#t_floating">floating point</a> <tt>value</tt> to cast,
3514 and a <a href="#t_floating">floating point</a> type to cast it to. The source
3515 type must be smaller than the destination type.</p>
3518 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
3519 <a href="#t_floating">floating point</a> type to a larger
3520 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
3521 used to make a <i>no-op cast</i> because it always changes bits. Use
3522 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
3526 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
3527 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
3531 <!-- _______________________________________________________________________ -->
3532 <div class="doc_subsubsection">
3533 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
3535 <div class="doc_text">
3539 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
3543 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
3544 unsigned integer equivalent of type <tt>ty2</tt>.
3548 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
3549 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3550 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3551 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3552 vector integer type with the same number of elements as <tt>ty</tt></p>
3555 <p> The '<tt>fptoui</tt>' instruction converts its
3556 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3557 towards zero) unsigned integer value. If the value cannot fit in <tt>ty2</tt>,
3558 the results are undefined.</p>
3562 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
3563 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
3564 %X = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
3568 <!-- _______________________________________________________________________ -->
3569 <div class="doc_subsubsection">
3570 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
3572 <div class="doc_text">
3576 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
3580 <p>The '<tt>fptosi</tt>' instruction converts
3581 <a href="#t_floating">floating point</a> <tt>value</tt> to type <tt>ty2</tt>.
3585 <p> The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
3586 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3587 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3588 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3589 vector integer type with the same number of elements as <tt>ty</tt></p>
3592 <p>The '<tt>fptosi</tt>' instruction converts its
3593 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3594 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
3595 the results are undefined.</p>
3599 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
3600 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
3601 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
3605 <!-- _______________________________________________________________________ -->
3606 <div class="doc_subsubsection">
3607 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
3609 <div class="doc_text">
3613 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3617 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
3618 integer and converts that value to the <tt>ty2</tt> type.</p>
3621 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
3622 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3623 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3624 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3625 floating point type with the same number of elements as <tt>ty</tt></p>
3628 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
3629 integer quantity and converts it to the corresponding floating point value. If
3630 the value cannot fit in the floating point value, the results are undefined.</p>
3634 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
3635 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
3639 <!-- _______________________________________________________________________ -->
3640 <div class="doc_subsubsection">
3641 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
3643 <div class="doc_text">
3647 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3651 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed
3652 integer and converts that value to the <tt>ty2</tt> type.</p>
3655 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
3656 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3657 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3658 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3659 floating point type with the same number of elements as <tt>ty</tt></p>
3662 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed
3663 integer quantity and converts it to the corresponding floating point value. If
3664 the value cannot fit in the floating point value, the results are undefined.</p>
3668 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
3669 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
3673 <!-- _______________________________________________________________________ -->
3674 <div class="doc_subsubsection">
3675 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
3677 <div class="doc_text">
3681 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
3685 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
3686 the integer type <tt>ty2</tt>.</p>
3689 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
3690 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
3691 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.
3694 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
3695 <tt>ty2</tt> by interpreting the pointer value as an integer and either
3696 truncating or zero extending that value to the size of the integer type. If
3697 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
3698 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
3699 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
3704 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
3705 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
3709 <!-- _______________________________________________________________________ -->
3710 <div class="doc_subsubsection">
3711 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
3713 <div class="doc_text">
3717 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
3721 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to
3722 a pointer type, <tt>ty2</tt>.</p>
3725 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
3726 value to cast, and a type to cast it to, which must be a
3727 <a href="#t_pointer">pointer</a> type.
3730 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
3731 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
3732 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
3733 size of a pointer then a truncation is done. If <tt>value</tt> is smaller than
3734 the size of a pointer then a zero extension is done. If they are the same size,
3735 nothing is done (<i>no-op cast</i>).</p>
3739 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
3740 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
3741 %Y = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
3745 <!-- _______________________________________________________________________ -->
3746 <div class="doc_subsubsection">
3747 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
3749 <div class="doc_text">
3753 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
3758 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3759 <tt>ty2</tt> without changing any bits.</p>
3763 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be
3764 a first class value, and a type to cast it to, which must also be a <a
3765 href="#t_firstclass">first class</a> type. The bit sizes of <tt>value</tt>
3766 and the destination type, <tt>ty2</tt>, must be identical. If the source
3767 type is a pointer, the destination type must also be a pointer. This
3768 instruction supports bitwise conversion of vectors to integers and to vectors
3769 of other types (as long as they have the same size).</p>
3772 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3773 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
3774 this conversion. The conversion is done as if the <tt>value</tt> had been
3775 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only be
3776 converted to other pointer types with this instruction. To convert pointers to
3777 other types, use the <a href="#i_inttoptr">inttoptr</a> or
3778 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
3782 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
3783 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
3784 %Z = bitcast <2xint> %V to i64; <i>; yields i64: %V</i>
3788 <!-- ======================================================================= -->
3789 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
3790 <div class="doc_text">
3791 <p>The instructions in this category are the "miscellaneous"
3792 instructions, which defy better classification.</p>
3795 <!-- _______________________________________________________________________ -->
3796 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
3798 <div class="doc_text">
3800 <pre> <result> = icmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3803 <p>The '<tt>icmp</tt>' instruction returns a boolean value based on comparison
3804 of its two integer or pointer operands.</p>
3806 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
3807 the condition code indicating the kind of comparison to perform. It is not
3808 a value, just a keyword. The possible condition code are:
3810 <li><tt>eq</tt>: equal</li>
3811 <li><tt>ne</tt>: not equal </li>
3812 <li><tt>ugt</tt>: unsigned greater than</li>
3813 <li><tt>uge</tt>: unsigned greater or equal</li>
3814 <li><tt>ult</tt>: unsigned less than</li>
3815 <li><tt>ule</tt>: unsigned less or equal</li>
3816 <li><tt>sgt</tt>: signed greater than</li>
3817 <li><tt>sge</tt>: signed greater or equal</li>
3818 <li><tt>slt</tt>: signed less than</li>
3819 <li><tt>sle</tt>: signed less or equal</li>
3821 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
3822 <a href="#t_pointer">pointer</a> typed. They must also be identical types.</p>
3824 <p>The '<tt>icmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3825 the condition code given as <tt>cond</tt>. The comparison performed always
3826 yields a <a href="#t_primitive">i1</a> result, as follows:
3828 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
3829 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3831 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
3832 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3833 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
3834 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3835 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
3836 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3837 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
3838 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3839 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
3840 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3841 <li><tt>sgt</tt>: interprets the operands as signed values and yields
3842 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3843 <li><tt>sge</tt>: interprets the operands as signed values and yields
3844 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3845 <li><tt>slt</tt>: interprets the operands as signed values and yields
3846 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3847 <li><tt>sle</tt>: interprets the operands as signed values and yields
3848 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3850 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
3851 values are compared as if they were integers.</p>
3854 <pre> <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
3855 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
3856 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
3857 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
3858 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
3859 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
3863 <!-- _______________________________________________________________________ -->
3864 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
3866 <div class="doc_text">
3868 <pre> <result> = fcmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3871 <p>The '<tt>fcmp</tt>' instruction returns a boolean value based on comparison
3872 of its floating point operands.</p>
3874 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
3875 the condition code indicating the kind of comparison to perform. It is not
3876 a value, just a keyword. The possible condition code are:
3878 <li><tt>false</tt>: no comparison, always returns false</li>
3879 <li><tt>oeq</tt>: ordered and equal</li>
3880 <li><tt>ogt</tt>: ordered and greater than </li>
3881 <li><tt>oge</tt>: ordered and greater than or equal</li>
3882 <li><tt>olt</tt>: ordered and less than </li>
3883 <li><tt>ole</tt>: ordered and less than or equal</li>
3884 <li><tt>one</tt>: ordered and not equal</li>
3885 <li><tt>ord</tt>: ordered (no nans)</li>
3886 <li><tt>ueq</tt>: unordered or equal</li>
3887 <li><tt>ugt</tt>: unordered or greater than </li>
3888 <li><tt>uge</tt>: unordered or greater than or equal</li>
3889 <li><tt>ult</tt>: unordered or less than </li>
3890 <li><tt>ule</tt>: unordered or less than or equal</li>
3891 <li><tt>une</tt>: unordered or not equal</li>
3892 <li><tt>uno</tt>: unordered (either nans)</li>
3893 <li><tt>true</tt>: no comparison, always returns true</li>
3895 <p><i>Ordered</i> means that neither operand is a QNAN while
3896 <i>unordered</i> means that either operand may be a QNAN.</p>
3897 <p>The <tt>val1</tt> and <tt>val2</tt> arguments must be
3898 <a href="#t_floating">floating point</a> typed. They must have identical
3901 <p>The '<tt>fcmp</tt>' instruction compares <tt>var1</tt> and <tt>var2</tt>
3902 according to the condition code given as <tt>cond</tt>. The comparison performed
3903 always yields a <a href="#t_primitive">i1</a> result, as follows:
3905 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
3906 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3907 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3908 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3909 <tt>var1</tt> is greather than <tt>var2</tt>.</li>
3910 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3911 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3912 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3913 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3914 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3915 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3916 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3917 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3918 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
3919 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
3920 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3921 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
3922 <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3923 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
3924 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3925 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
3926 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3927 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
3928 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3929 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
3930 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3931 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
3932 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
3936 <pre> <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
3937 <result> = icmp one float 4.0, 5.0 <i>; yields: result=true</i>
3938 <result> = icmp olt float 4.0, 5.0 <i>; yields: result=true</i>
3939 <result> = icmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
3943 <!-- _______________________________________________________________________ -->
3944 <div class="doc_subsubsection">
3945 <a name="i_vicmp">'<tt>vicmp</tt>' Instruction</a>
3947 <div class="doc_text">
3949 <pre> <result> = vicmp <cond> <ty> <var1>, <var2> <i>; yields {ty}:result</i>
3952 <p>The '<tt>vicmp</tt>' instruction returns an integer vector value based on
3953 element-wise comparison of its two integer vector operands.</p>
3955 <p>The '<tt>vicmp</tt>' instruction takes three operands. The first operand is
3956 the condition code indicating the kind of comparison to perform. It is not
3957 a value, just a keyword. The possible condition code are:
3959 <li><tt>eq</tt>: equal</li>
3960 <li><tt>ne</tt>: not equal </li>
3961 <li><tt>ugt</tt>: unsigned greater than</li>
3962 <li><tt>uge</tt>: unsigned greater or equal</li>
3963 <li><tt>ult</tt>: unsigned less than</li>
3964 <li><tt>ule</tt>: unsigned less or equal</li>
3965 <li><tt>sgt</tt>: signed greater than</li>
3966 <li><tt>sge</tt>: signed greater or equal</li>
3967 <li><tt>slt</tt>: signed less than</li>
3968 <li><tt>sle</tt>: signed less or equal</li>
3970 <p>The remaining two arguments must be <a href="#t_vector">vector</a> of
3971 <a href="#t_integer">integer</a> typed. They must also be identical types.</p>
3973 <p>The '<tt>vicmp</tt>' instruction compares <tt>var1</tt> and <tt>var2</tt>
3974 according to the condition code given as <tt>cond</tt>. The comparison yields a
3975 <a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, of
3976 identical type as the values being compared. The most significant bit in each
3977 element is 1 if the element-wise comparison evaluates to true, and is 0
3978 otherwise. All other bits of the result are undefined. The condition codes
3979 are evaluated identically to the <a href="#i_icmp">'<tt>icmp</tt>'
3984 <result> = vicmp eq <2 x i32> < i32 4, i32 0>, < i32 5, i32 0> <i>; yields: result=<2 x i32> < i32 0, i32 -1 ></i>
3985 <result> = vicmp ult <2 x i8 > < i8 1, i8 2>, < i8 2, i8 2 > <i>; yields: result=<2 x i8> < i8 -1, i8 0 ></i>
3989 <!-- _______________________________________________________________________ -->
3990 <div class="doc_subsubsection">
3991 <a name="i_vfcmp">'<tt>vfcmp</tt>' Instruction</a>
3993 <div class="doc_text">
3995 <pre> <result> = vfcmp <cond> <ty> <var1>, <var2></pre>
3997 <p>The '<tt>vfcmp</tt>' instruction returns an integer vector value based on
3998 element-wise comparison of its two floating point vector operands. The output
3999 elements have the same width as the input elements.</p>
4001 <p>The '<tt>vfcmp</tt>' instruction takes three operands. The first operand is
4002 the condition code indicating the kind of comparison to perform. It is not
4003 a value, just a keyword. The possible condition code are:
4005 <li><tt>false</tt>: no comparison, always returns false</li>
4006 <li><tt>oeq</tt>: ordered and equal</li>
4007 <li><tt>ogt</tt>: ordered and greater than </li>
4008 <li><tt>oge</tt>: ordered and greater than or equal</li>
4009 <li><tt>olt</tt>: ordered and less than </li>
4010 <li><tt>ole</tt>: ordered and less than or equal</li>
4011 <li><tt>one</tt>: ordered and not equal</li>
4012 <li><tt>ord</tt>: ordered (no nans)</li>
4013 <li><tt>ueq</tt>: unordered or equal</li>
4014 <li><tt>ugt</tt>: unordered or greater than </li>
4015 <li><tt>uge</tt>: unordered or greater than or equal</li>
4016 <li><tt>ult</tt>: unordered or less than </li>
4017 <li><tt>ule</tt>: unordered or less than or equal</li>
4018 <li><tt>une</tt>: unordered or not equal</li>
4019 <li><tt>uno</tt>: unordered (either nans)</li>
4020 <li><tt>true</tt>: no comparison, always returns true</li>
4022 <p>The remaining two arguments must be <a href="#t_vector">vector</a> of
4023 <a href="#t_floating">floating point</a> typed. They must also be identical
4026 <p>The '<tt>vfcmp</tt>' instruction compares <tt>var1</tt> and <tt>var2</tt>
4027 according to the condition code given as <tt>cond</tt>. The comparison yields a
4028 <a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, with
4029 an identical number of elements as the values being compared, and each element
4030 having identical with to the width of the floating point elements. The most
4031 significant bit in each element is 1 if the element-wise comparison evaluates to
4032 true, and is 0 otherwise. All other bits of the result are undefined. The
4033 condition codes are evaluated identically to the
4034 <a href="#i_fcmp">'<tt>fcmp</tt>' instruction</a>.
4038 <result> = vfcmp oeq <2 x float> < float 4, float 0 >, < float 5, float 0 > <i>; yields: result=<2 x i32> < i32 0, i32 -1 ></i>
4039 <result> = vfcmp ult <2 x double> < double 1, double 2 >, < double 2, double 2> <i>; yields: result=<2 x i64> < i64 -1, i64 0 ></i>
4043 <!-- _______________________________________________________________________ -->
4044 <div class="doc_subsubsection">
4045 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
4048 <div class="doc_text">
4052 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
4054 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
4055 the SSA graph representing the function.</p>
4058 <p>The type of the incoming values is specified with the first type
4059 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
4060 as arguments, with one pair for each predecessor basic block of the
4061 current block. Only values of <a href="#t_firstclass">first class</a>
4062 type may be used as the value arguments to the PHI node. Only labels
4063 may be used as the label arguments.</p>
4065 <p>There must be no non-phi instructions between the start of a basic
4066 block and the PHI instructions: i.e. PHI instructions must be first in
4071 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
4072 specified by the pair corresponding to the predecessor basic block that executed
4073 just prior to the current block.</p>
4077 Loop: ; Infinite loop that counts from 0 on up...
4078 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
4079 %nextindvar = add i32 %indvar, 1
4084 <!-- _______________________________________________________________________ -->
4085 <div class="doc_subsubsection">
4086 <a name="i_select">'<tt>select</tt>' Instruction</a>
4089 <div class="doc_text">
4094 <result> = select i1 <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
4100 The '<tt>select</tt>' instruction is used to choose one value based on a
4101 condition, without branching.
4108 The '<tt>select</tt>' instruction requires an 'i1' value indicating the
4109 condition, and two values of the same <a href="#t_firstclass">first class</a>
4110 type. If the val1/val2 are vectors, the entire vectors are selected, not
4111 individual elements.
4117 If the i1 condition evaluates is 1, the instruction returns the first
4118 value argument; otherwise, it returns the second value argument.
4124 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
4129 <!-- _______________________________________________________________________ -->
4130 <div class="doc_subsubsection">
4131 <a name="i_call">'<tt>call</tt>' Instruction</a>
4134 <div class="doc_text">
4138 <result> = [tail] call [<a href="#callingconv">cconv</a>] <ty> [<fnty>*] <fnptrval>(<param list>)
4143 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
4147 <p>This instruction requires several arguments:</p>
4151 <p>The optional "tail" marker indicates whether the callee function accesses
4152 any allocas or varargs in the caller. If the "tail" marker is present, the
4153 function call is eligible for tail call optimization. Note that calls may
4154 be marked "tail" even if they do not occur before a <a
4155 href="#i_ret"><tt>ret</tt></a> instruction.
4158 <p>The optional "cconv" marker indicates which <a href="#callingconv">calling
4159 convention</a> the call should use. If none is specified, the call defaults
4160 to using C calling conventions.
4163 <p>'<tt>ty</tt>': the type of the call instruction itself which is also
4164 the type of the return value. Functions that return no value are marked
4165 <tt><a href="#t_void">void</a></tt>.</p>
4168 <p>'<tt>fnty</tt>': shall be the signature of the pointer to function
4169 value being invoked. The argument types must match the types implied by
4170 this signature. This type can be omitted if the function is not varargs
4171 and if the function type does not return a pointer to a function.</p>
4174 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
4175 be invoked. In most cases, this is a direct function invocation, but
4176 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
4177 to function value.</p>
4180 <p>'<tt>function args</tt>': argument list whose types match the
4181 function signature argument types. All arguments must be of
4182 <a href="#t_firstclass">first class</a> type. If the function signature
4183 indicates the function accepts a variable number of arguments, the extra
4184 arguments can be specified.</p>
4190 <p>The '<tt>call</tt>' instruction is used to cause control flow to
4191 transfer to a specified function, with its incoming arguments bound to
4192 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
4193 instruction in the called function, control flow continues with the
4194 instruction after the function call, and the return value of the
4195 function is bound to the result argument. If the callee returns multiple
4196 values then the return values of the function are only accessible through
4197 the '<tt><a href="#i_getresult">getresult</a></tt>' instruction.</p>
4202 %retval = call i32 @test(i32 %argc)
4203 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42) <i>; yields i32</i>
4204 %X = tail call i32 @foo() <i>; yields i32</i>
4205 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
4206 call void %foo(i8 97 signext)
4208 %struct.A = type { i32, i8 }
4209 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
4210 %gr = getresult %struct.A %r, 0 <i>; yields i32</i>
4211 %gr1 = getresult %struct.A %r, 1 <i>; yields i8</i>
4216 <!-- _______________________________________________________________________ -->
4217 <div class="doc_subsubsection">
4218 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
4221 <div class="doc_text">
4226 <resultval> = va_arg <va_list*> <arglist>, <argty>
4231 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
4232 the "variable argument" area of a function call. It is used to implement the
4233 <tt>va_arg</tt> macro in C.</p>
4237 <p>This instruction takes a <tt>va_list*</tt> value and the type of
4238 the argument. It returns a value of the specified argument type and
4239 increments the <tt>va_list</tt> to point to the next argument. The
4240 actual type of <tt>va_list</tt> is target specific.</p>
4244 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
4245 type from the specified <tt>va_list</tt> and causes the
4246 <tt>va_list</tt> to point to the next argument. For more information,
4247 see the variable argument handling <a href="#int_varargs">Intrinsic
4250 <p>It is legal for this instruction to be called in a function which does not
4251 take a variable number of arguments, for example, the <tt>vfprintf</tt>
4254 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
4255 href="#intrinsics">intrinsic function</a> because it takes a type as an
4260 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
4264 <!-- _______________________________________________________________________ -->
4265 <div class="doc_subsubsection">
4266 <a name="i_getresult">'<tt>getresult</tt>' Instruction</a>
4269 <div class="doc_text">
4273 <resultval> = getresult <type> <retval>, <index>
4278 <p> The '<tt>getresult</tt>' instruction is used to extract individual values
4279 from a '<tt><a href="#i_call">call</a></tt>'
4280 or '<tt><a href="#i_invoke">invoke</a></tt>' instruction that returns multiple
4285 <p>The '<tt>getresult</tt>' instruction takes a call or invoke value as its
4286 first argument, or an undef value. The value must have <a
4287 href="#t_struct">structure type</a>. The second argument is a constant
4288 unsigned index value which must be in range for the number of values returned
4293 <p>The '<tt>getresult</tt>' instruction extracts the element identified by
4294 '<tt>index</tt>' from the aggregate value.</p>
4299 %struct.A = type { i32, i8 }
4301 %r = call %struct.A @foo()
4302 %gr = getresult %struct.A %r, 0 <i>; yields i32:%gr</i>
4303 %gr1 = getresult %struct.A %r, 1 <i>; yields i8:%gr1</i>
4310 <!-- *********************************************************************** -->
4311 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
4312 <!-- *********************************************************************** -->
4314 <div class="doc_text">
4316 <p>LLVM supports the notion of an "intrinsic function". These functions have
4317 well known names and semantics and are required to follow certain restrictions.
4318 Overall, these intrinsics represent an extension mechanism for the LLVM
4319 language that does not require changing all of the transformations in LLVM when
4320 adding to the language (or the bitcode reader/writer, the parser, etc...).</p>
4322 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
4323 prefix is reserved in LLVM for intrinsic names; thus, function names may not
4324 begin with this prefix. Intrinsic functions must always be external functions:
4325 you cannot define the body of intrinsic functions. Intrinsic functions may
4326 only be used in call or invoke instructions: it is illegal to take the address
4327 of an intrinsic function. Additionally, because intrinsic functions are part
4328 of the LLVM language, it is required if any are added that they be documented
4331 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
4332 a family of functions that perform the same operation but on different data
4333 types. Because LLVM can represent over 8 million different integer types,
4334 overloading is used commonly to allow an intrinsic function to operate on any
4335 integer type. One or more of the argument types or the result type can be
4336 overloaded to accept any integer type. Argument types may also be defined as
4337 exactly matching a previous argument's type or the result type. This allows an
4338 intrinsic function which accepts multiple arguments, but needs all of them to
4339 be of the same type, to only be overloaded with respect to a single argument or
4342 <p>Overloaded intrinsics will have the names of its overloaded argument types
4343 encoded into its function name, each preceded by a period. Only those types
4344 which are overloaded result in a name suffix. Arguments whose type is matched
4345 against another type do not. For example, the <tt>llvm.ctpop</tt> function can
4346 take an integer of any width and returns an integer of exactly the same integer
4347 width. This leads to a family of functions such as
4348 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29 %val)</tt>.
4349 Only one type, the return type, is overloaded, and only one type suffix is
4350 required. Because the argument's type is matched against the return type, it
4351 does not require its own name suffix.</p>
4353 <p>To learn how to add an intrinsic function, please see the
4354 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.
4359 <!-- ======================================================================= -->
4360 <div class="doc_subsection">
4361 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
4364 <div class="doc_text">
4366 <p>Variable argument support is defined in LLVM with the <a
4367 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
4368 intrinsic functions. These functions are related to the similarly
4369 named macros defined in the <tt><stdarg.h></tt> header file.</p>
4371 <p>All of these functions operate on arguments that use a
4372 target-specific value type "<tt>va_list</tt>". The LLVM assembly
4373 language reference manual does not define what this type is, so all
4374 transformations should be prepared to handle these functions regardless of
4377 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
4378 instruction and the variable argument handling intrinsic functions are
4381 <div class="doc_code">
4383 define i32 @test(i32 %X, ...) {
4384 ; Initialize variable argument processing
4386 %ap2 = bitcast i8** %ap to i8*
4387 call void @llvm.va_start(i8* %ap2)
4389 ; Read a single integer argument
4390 %tmp = va_arg i8** %ap, i32
4392 ; Demonstrate usage of llvm.va_copy and llvm.va_end
4394 %aq2 = bitcast i8** %aq to i8*
4395 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
4396 call void @llvm.va_end(i8* %aq2)
4398 ; Stop processing of arguments.
4399 call void @llvm.va_end(i8* %ap2)
4403 declare void @llvm.va_start(i8*)
4404 declare void @llvm.va_copy(i8*, i8*)
4405 declare void @llvm.va_end(i8*)
4411 <!-- _______________________________________________________________________ -->
4412 <div class="doc_subsubsection">
4413 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
4417 <div class="doc_text">
4419 <pre> declare void %llvm.va_start(i8* <arglist>)<br></pre>
4421 <P>The '<tt>llvm.va_start</tt>' intrinsic initializes
4422 <tt>*<arglist></tt> for subsequent use by <tt><a
4423 href="#i_va_arg">va_arg</a></tt>.</p>
4427 <P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
4431 <P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
4432 macro available in C. In a target-dependent way, it initializes the
4433 <tt>va_list</tt> element to which the argument points, so that the next call to
4434 <tt>va_arg</tt> will produce the first variable argument passed to the function.
4435 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
4436 last argument of the function as the compiler can figure that out.</p>
4440 <!-- _______________________________________________________________________ -->
4441 <div class="doc_subsubsection">
4442 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
4445 <div class="doc_text">
4447 <pre> declare void @llvm.va_end(i8* <arglist>)<br></pre>
4450 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
4451 which has been initialized previously with <tt><a href="#int_va_start">llvm.va_start</a></tt>
4452 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
4456 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
4460 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
4461 macro available in C. In a target-dependent way, it destroys the
4462 <tt>va_list</tt> element to which the argument points. Calls to <a
4463 href="#int_va_start"><tt>llvm.va_start</tt></a> and <a href="#int_va_copy">
4464 <tt>llvm.va_copy</tt></a> must be matched exactly with calls to
4465 <tt>llvm.va_end</tt>.</p>
4469 <!-- _______________________________________________________________________ -->
4470 <div class="doc_subsubsection">
4471 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
4474 <div class="doc_text">
4479 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
4484 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
4485 from the source argument list to the destination argument list.</p>
4489 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
4490 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
4495 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
4496 macro available in C. In a target-dependent way, it copies the source
4497 <tt>va_list</tt> element into the destination <tt>va_list</tt> element. This
4498 intrinsic is necessary because the <tt><a href="#int_va_start">
4499 llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
4500 example, memory allocation.</p>
4504 <!-- ======================================================================= -->
4505 <div class="doc_subsection">
4506 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
4509 <div class="doc_text">
4512 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
4513 Collection</a> requires the implementation and generation of these intrinsics.
4514 These intrinsics allow identification of <a href="#int_gcroot">GC roots on the
4515 stack</a>, as well as garbage collector implementations that require <a
4516 href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a> barriers.
4517 Front-ends for type-safe garbage collected languages should generate these
4518 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
4519 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
4522 <p>The garbage collection intrinsics only operate on objects in the generic
4523 address space (address space zero).</p>
4527 <!-- _______________________________________________________________________ -->
4528 <div class="doc_subsubsection">
4529 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
4532 <div class="doc_text">
4537 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
4542 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
4543 the code generator, and allows some metadata to be associated with it.</p>
4547 <p>The first argument specifies the address of a stack object that contains the
4548 root pointer. The second pointer (which must be either a constant or a global
4549 value address) contains the meta-data to be associated with the root.</p>
4553 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
4554 location. At compile-time, the code generator generates information to allow
4555 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
4556 intrinsic may only be used in a function which <a href="#gc">specifies a GC
4562 <!-- _______________________________________________________________________ -->
4563 <div class="doc_subsubsection">
4564 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
4567 <div class="doc_text">
4572 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
4577 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
4578 locations, allowing garbage collector implementations that require read
4583 <p>The second argument is the address to read from, which should be an address
4584 allocated from the garbage collector. The first object is a pointer to the
4585 start of the referenced object, if needed by the language runtime (otherwise
4590 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
4591 instruction, but may be replaced with substantially more complex code by the
4592 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
4593 may only be used in a function which <a href="#gc">specifies a GC
4599 <!-- _______________________________________________________________________ -->
4600 <div class="doc_subsubsection">
4601 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
4604 <div class="doc_text">
4609 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
4614 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
4615 locations, allowing garbage collector implementations that require write
4616 barriers (such as generational or reference counting collectors).</p>
4620 <p>The first argument is the reference to store, the second is the start of the
4621 object to store it to, and the third is the address of the field of Obj to
4622 store to. If the runtime does not require a pointer to the object, Obj may be
4627 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
4628 instruction, but may be replaced with substantially more complex code by the
4629 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
4630 may only be used in a function which <a href="#gc">specifies a GC
4637 <!-- ======================================================================= -->
4638 <div class="doc_subsection">
4639 <a name="int_codegen">Code Generator Intrinsics</a>
4642 <div class="doc_text">
4644 These intrinsics are provided by LLVM to expose special features that may only
4645 be implemented with code generator support.
4650 <!-- _______________________________________________________________________ -->
4651 <div class="doc_subsubsection">
4652 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
4655 <div class="doc_text">
4659 declare i8 *@llvm.returnaddress(i32 <level>)
4665 The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
4666 target-specific value indicating the return address of the current function
4667 or one of its callers.
4673 The argument to this intrinsic indicates which function to return the address
4674 for. Zero indicates the calling function, one indicates its caller, etc. The
4675 argument is <b>required</b> to be a constant integer value.
4681 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
4682 the return address of the specified call frame, or zero if it cannot be
4683 identified. The value returned by this intrinsic is likely to be incorrect or 0
4684 for arguments other than zero, so it should only be used for debugging purposes.
4688 Note that calling this intrinsic does not prevent function inlining or other
4689 aggressive transformations, so the value returned may not be that of the obvious
4690 source-language caller.
4695 <!-- _______________________________________________________________________ -->
4696 <div class="doc_subsubsection">
4697 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
4700 <div class="doc_text">
4704 declare i8 *@llvm.frameaddress(i32 <level>)
4710 The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
4711 target-specific frame pointer value for the specified stack frame.
4717 The argument to this intrinsic indicates which function to return the frame
4718 pointer for. Zero indicates the calling function, one indicates its caller,
4719 etc. The argument is <b>required</b> to be a constant integer value.
4725 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
4726 the frame address of the specified call frame, or zero if it cannot be
4727 identified. The value returned by this intrinsic is likely to be incorrect or 0
4728 for arguments other than zero, so it should only be used for debugging purposes.
4732 Note that calling this intrinsic does not prevent function inlining or other
4733 aggressive transformations, so the value returned may not be that of the obvious
4734 source-language caller.
4738 <!-- _______________________________________________________________________ -->
4739 <div class="doc_subsubsection">
4740 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
4743 <div class="doc_text">
4747 declare i8 *@llvm.stacksave()
4753 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
4754 the function stack, for use with <a href="#int_stackrestore">
4755 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
4756 features like scoped automatic variable sized arrays in C99.
4762 This intrinsic returns a opaque pointer value that can be passed to <a
4763 href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
4764 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
4765 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
4766 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
4767 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
4768 that were allocated after the <tt>llvm.stacksave</tt> was executed.
4773 <!-- _______________________________________________________________________ -->
4774 <div class="doc_subsubsection">
4775 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
4778 <div class="doc_text">
4782 declare void @llvm.stackrestore(i8 * %ptr)
4788 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
4789 the function stack to the state it was in when the corresponding <a
4790 href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
4791 useful for implementing language features like scoped automatic variable sized
4798 See the description for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.
4804 <!-- _______________________________________________________________________ -->
4805 <div class="doc_subsubsection">
4806 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
4809 <div class="doc_text">
4813 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>)
4820 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
4821 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
4823 effect on the behavior of the program but can change its performance
4830 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
4831 determining if the fetch should be for a read (0) or write (1), and
4832 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
4833 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
4834 <tt>locality</tt> arguments must be constant integers.
4840 This intrinsic does not modify the behavior of the program. In particular,
4841 prefetches cannot trap and do not produce a value. On targets that support this
4842 intrinsic, the prefetch can provide hints to the processor cache for better
4848 <!-- _______________________________________________________________________ -->
4849 <div class="doc_subsubsection">
4850 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
4853 <div class="doc_text">
4857 declare void @llvm.pcmarker(i32 <id>)
4864 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
4866 code to simulators and other tools. The method is target specific, but it is
4867 expected that the marker will use exported symbols to transmit the PC of the marker.
4868 The marker makes no guarantees that it will remain with any specific instruction
4869 after optimizations. It is possible that the presence of a marker will inhibit
4870 optimizations. The intended use is to be inserted after optimizations to allow
4871 correlations of simulation runs.
4877 <tt>id</tt> is a numerical id identifying the marker.
4883 This intrinsic does not modify the behavior of the program. Backends that do not
4884 support this intrinisic may ignore it.
4889 <!-- _______________________________________________________________________ -->
4890 <div class="doc_subsubsection">
4891 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
4894 <div class="doc_text">
4898 declare i64 @llvm.readcyclecounter( )
4905 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
4906 counter register (or similar low latency, high accuracy clocks) on those targets
4907 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
4908 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
4909 should only be used for small timings.
4915 When directly supported, reading the cycle counter should not modify any memory.
4916 Implementations are allowed to either return a application specific value or a
4917 system wide value. On backends without support, this is lowered to a constant 0.
4922 <!-- ======================================================================= -->
4923 <div class="doc_subsection">
4924 <a name="int_libc">Standard C Library Intrinsics</a>
4927 <div class="doc_text">
4929 LLVM provides intrinsics for a few important standard C library functions.
4930 These intrinsics allow source-language front-ends to pass information about the
4931 alignment of the pointer arguments to the code generator, providing opportunity
4932 for more efficient code generation.
4937 <!-- _______________________________________________________________________ -->
4938 <div class="doc_subsubsection">
4939 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
4942 <div class="doc_text">
4946 declare void @llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
4947 i32 <len>, i32 <align>)
4948 declare void @llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
4949 i64 <len>, i32 <align>)
4955 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4956 location to the destination location.
4960 Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
4961 intrinsics do not return a value, and takes an extra alignment argument.
4967 The first argument is a pointer to the destination, the second is a pointer to
4968 the source. The third argument is an integer argument
4969 specifying the number of bytes to copy, and the fourth argument is the alignment
4970 of the source and destination locations.
4974 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4975 the caller guarantees that both the source and destination pointers are aligned
4982 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4983 location to the destination location, which are not allowed to overlap. It
4984 copies "len" bytes of memory over. If the argument is known to be aligned to
4985 some boundary, this can be specified as the fourth argument, otherwise it should
4991 <!-- _______________________________________________________________________ -->
4992 <div class="doc_subsubsection">
4993 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
4996 <div class="doc_text">
5000 declare void @llvm.memmove.i32(i8 * <dest>, i8 * <src>,
5001 i32 <len>, i32 <align>)
5002 declare void @llvm.memmove.i64(i8 * <dest>, i8 * <src>,
5003 i64 <len>, i32 <align>)
5009 The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
5010 location to the destination location. It is similar to the
5011 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to overlap.
5015 Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
5016 intrinsics do not return a value, and takes an extra alignment argument.
5022 The first argument is a pointer to the destination, the second is a pointer to
5023 the source. The third argument is an integer argument
5024 specifying the number of bytes to copy, and the fourth argument is the alignment
5025 of the source and destination locations.
5029 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5030 the caller guarantees that the source and destination pointers are aligned to
5037 The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
5038 location to the destination location, which may overlap. It
5039 copies "len" bytes of memory over. If the argument is known to be aligned to
5040 some boundary, this can be specified as the fourth argument, otherwise it should
5046 <!-- _______________________________________________________________________ -->
5047 <div class="doc_subsubsection">
5048 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
5051 <div class="doc_text">
5055 declare void @llvm.memset.i32(i8 * <dest>, i8 <val>,
5056 i32 <len>, i32 <align>)
5057 declare void @llvm.memset.i64(i8 * <dest>, i8 <val>,
5058 i64 <len>, i32 <align>)
5064 The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
5069 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
5070 does not return a value, and takes an extra alignment argument.
5076 The first argument is a pointer to the destination to fill, the second is the
5077 byte value to fill it with, the third argument is an integer
5078 argument specifying the number of bytes to fill, and the fourth argument is the
5079 known alignment of destination location.
5083 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5084 the caller guarantees that the destination pointer is aligned to that boundary.
5090 The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
5092 destination location. If the argument is known to be aligned to some boundary,
5093 this can be specified as the fourth argument, otherwise it should be set to 0 or
5099 <!-- _______________________________________________________________________ -->
5100 <div class="doc_subsubsection">
5101 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
5104 <div class="doc_text">
5107 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
5108 floating point or vector of floating point type. Not all targets support all
5111 declare float @llvm.sqrt.f32(float %Val)
5112 declare double @llvm.sqrt.f64(double %Val)
5113 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
5114 declare fp128 @llvm.sqrt.f128(fp128 %Val)
5115 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
5121 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
5122 returning the same value as the libm '<tt>sqrt</tt>' functions would. Unlike
5123 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
5124 negative numbers other than -0.0 (which allows for better optimization, because
5125 there is no need to worry about errno being set). <tt>llvm.sqrt(-0.0)</tt> is
5126 defined to return -0.0 like IEEE sqrt.
5132 The argument and return value are floating point numbers of the same type.
5138 This function returns the sqrt of the specified operand if it is a nonnegative
5139 floating point number.
5143 <!-- _______________________________________________________________________ -->
5144 <div class="doc_subsubsection">
5145 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
5148 <div class="doc_text">
5151 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
5152 floating point or vector of floating point type. Not all targets support all
5155 declare float @llvm.powi.f32(float %Val, i32 %power)
5156 declare double @llvm.powi.f64(double %Val, i32 %power)
5157 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
5158 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
5159 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
5165 The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
5166 specified (positive or negative) power. The order of evaluation of
5167 multiplications is not defined. When a vector of floating point type is
5168 used, the second argument remains a scalar integer value.
5174 The second argument is an integer power, and the first is a value to raise to
5181 This function returns the first value raised to the second power with an
5182 unspecified sequence of rounding operations.</p>
5185 <!-- _______________________________________________________________________ -->
5186 <div class="doc_subsubsection">
5187 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
5190 <div class="doc_text">
5193 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
5194 floating point or vector of floating point type. Not all targets support all
5197 declare float @llvm.sin.f32(float %Val)
5198 declare double @llvm.sin.f64(double %Val)
5199 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
5200 declare fp128 @llvm.sin.f128(fp128 %Val)
5201 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
5207 The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.
5213 The argument and return value are floating point numbers of the same type.
5219 This function returns the sine of the specified operand, returning the
5220 same values as the libm <tt>sin</tt> functions would, and handles error
5221 conditions in the same way.</p>
5224 <!-- _______________________________________________________________________ -->
5225 <div class="doc_subsubsection">
5226 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
5229 <div class="doc_text">
5232 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
5233 floating point or vector of floating point type. Not all targets support all
5236 declare float @llvm.cos.f32(float %Val)
5237 declare double @llvm.cos.f64(double %Val)
5238 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
5239 declare fp128 @llvm.cos.f128(fp128 %Val)
5240 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
5246 The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.
5252 The argument and return value are floating point numbers of the same type.
5258 This function returns the cosine of the specified operand, returning the
5259 same values as the libm <tt>cos</tt> functions would, and handles error
5260 conditions in the same way.</p>
5263 <!-- _______________________________________________________________________ -->
5264 <div class="doc_subsubsection">
5265 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
5268 <div class="doc_text">
5271 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
5272 floating point or vector of floating point type. Not all targets support all
5275 declare float @llvm.pow.f32(float %Val, float %Power)
5276 declare double @llvm.pow.f64(double %Val, double %Power)
5277 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
5278 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
5279 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
5285 The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
5286 specified (positive or negative) power.
5292 The second argument is a floating point power, and the first is a value to
5293 raise to that power.
5299 This function returns the first value raised to the second power,
5301 same values as the libm <tt>pow</tt> functions would, and handles error
5302 conditions in the same way.</p>
5306 <!-- ======================================================================= -->
5307 <div class="doc_subsection">
5308 <a name="int_manip">Bit Manipulation Intrinsics</a>
5311 <div class="doc_text">
5313 LLVM provides intrinsics for a few important bit manipulation operations.
5314 These allow efficient code generation for some algorithms.
5319 <!-- _______________________________________________________________________ -->
5320 <div class="doc_subsubsection">
5321 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
5324 <div class="doc_text">
5327 <p>This is an overloaded intrinsic function. You can use bswap on any integer
5328 type that is an even number of bytes (i.e. BitWidth % 16 == 0).
5330 declare i16 @llvm.bswap.i16(i16 <id>)
5331 declare i32 @llvm.bswap.i32(i32 <id>)
5332 declare i64 @llvm.bswap.i64(i64 <id>)
5338 The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
5339 values with an even number of bytes (positive multiple of 16 bits). These are
5340 useful for performing operations on data that is not in the target's native
5347 The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
5348 and low byte of the input i16 swapped. Similarly, the <tt>llvm.bswap.i32</tt>
5349 intrinsic returns an i32 value that has the four bytes of the input i32
5350 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the returned
5351 i32 will have its bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i48</tt>,
5352 <tt>llvm.bswap.i64</tt> and other intrinsics extend this concept to
5353 additional even-byte lengths (6 bytes, 8 bytes and more, respectively).
5358 <!-- _______________________________________________________________________ -->
5359 <div class="doc_subsubsection">
5360 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
5363 <div class="doc_text">
5366 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
5367 width. Not all targets support all bit widths however.
5369 declare i8 @llvm.ctpop.i8 (i8 <src>)
5370 declare i16 @llvm.ctpop.i16(i16 <src>)
5371 declare i32 @llvm.ctpop.i32(i32 <src>)
5372 declare i64 @llvm.ctpop.i64(i64 <src>)
5373 declare i256 @llvm.ctpop.i256(i256 <src>)
5379 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
5386 The only argument is the value to be counted. The argument may be of any
5387 integer type. The return type must match the argument type.
5393 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
5397 <!-- _______________________________________________________________________ -->
5398 <div class="doc_subsubsection">
5399 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
5402 <div class="doc_text">
5405 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
5406 integer bit width. Not all targets support all bit widths however.
5408 declare i8 @llvm.ctlz.i8 (i8 <src>)
5409 declare i16 @llvm.ctlz.i16(i16 <src>)
5410 declare i32 @llvm.ctlz.i32(i32 <src>)
5411 declare i64 @llvm.ctlz.i64(i64 <src>)
5412 declare i256 @llvm.ctlz.i256(i256 <src>)
5418 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
5419 leading zeros in a variable.
5425 The only argument is the value to be counted. The argument may be of any
5426 integer type. The return type must match the argument type.
5432 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
5433 in a variable. If the src == 0 then the result is the size in bits of the type
5434 of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.
5440 <!-- _______________________________________________________________________ -->
5441 <div class="doc_subsubsection">
5442 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
5445 <div class="doc_text">
5448 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
5449 integer bit width. Not all targets support all bit widths however.
5451 declare i8 @llvm.cttz.i8 (i8 <src>)
5452 declare i16 @llvm.cttz.i16(i16 <src>)
5453 declare i32 @llvm.cttz.i32(i32 <src>)
5454 declare i64 @llvm.cttz.i64(i64 <src>)
5455 declare i256 @llvm.cttz.i256(i256 <src>)
5461 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
5468 The only argument is the value to be counted. The argument may be of any
5469 integer type. The return type must match the argument type.
5475 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
5476 in a variable. If the src == 0 then the result is the size in bits of the type
5477 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
5481 <!-- _______________________________________________________________________ -->
5482 <div class="doc_subsubsection">
5483 <a name="int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic</a>
5486 <div class="doc_text">
5489 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.select</tt>
5490 on any integer bit width.
5492 declare i17 @llvm.part.select.i17 (i17 %val, i32 %loBit, i32 %hiBit)
5493 declare i29 @llvm.part.select.i29 (i29 %val, i32 %loBit, i32 %hiBit)
5497 <p>The '<tt>llvm.part.select</tt>' family of intrinsic functions selects a
5498 range of bits from an integer value and returns them in the same bit width as
5499 the original value.</p>
5502 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5503 any bit width but they must have the same bit width. The second and third
5504 arguments must be <tt>i32</tt> type since they specify only a bit index.</p>
5507 <p>The operation of the '<tt>llvm.part.select</tt>' intrinsic has two modes
5508 of operation: forwards and reverse. If <tt>%loBit</tt> is greater than
5509 <tt>%hiBits</tt> then the intrinsic operates in reverse mode. Otherwise it
5510 operates in forward mode.</p>
5511 <p>In forward mode, this intrinsic is the equivalent of shifting <tt>%val</tt>
5512 right by <tt>%loBit</tt> bits and then ANDing it with a mask with
5513 only the <tt>%hiBit - %loBit</tt> bits set, as follows:</p>
5515 <li>The <tt>%val</tt> is shifted right (LSHR) by the number of bits specified
5516 by <tt>%loBits</tt>. This normalizes the value to the low order bits.</li>
5517 <li>The <tt>%loBits</tt> value is subtracted from the <tt>%hiBits</tt> value
5518 to determine the number of bits to retain.</li>
5519 <li>A mask of the retained bits is created by shifting a -1 value.</li>
5520 <li>The mask is ANDed with <tt>%val</tt> to produce the result.
5522 <p>In reverse mode, a similar computation is made except that the bits are
5523 returned in the reverse order. So, for example, if <tt>X</tt> has the value
5524 <tt>i16 0x0ACF (101011001111)</tt> and we apply
5525 <tt>part.select(i16 X, 8, 3)</tt> to it, we get back the value
5526 <tt>i16 0x0026 (000000100110)</tt>.</p>
5529 <div class="doc_subsubsection">
5530 <a name="int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic</a>
5533 <div class="doc_text">
5536 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.set</tt>
5537 on any integer bit width.
5539 declare i17 @llvm.part.set.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
5540 declare i29 @llvm.part.set.i29.i9 (i29 %val, i9 %repl, i32 %lo, i32 %hi)
5544 <p>The '<tt>llvm.part.set</tt>' family of intrinsic functions replaces a range
5545 of bits in an integer value with another integer value. It returns the integer
5546 with the replaced bits.</p>
5549 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5550 any bit width but they must have the same bit width. <tt>%val</tt> is the value
5551 whose bits will be replaced. The second argument, <tt>%repl</tt> may be an
5552 integer of any bit width. The third and fourth arguments must be <tt>i32</tt>
5553 type since they specify only a bit index.</p>
5556 <p>The operation of the '<tt>llvm.part.set</tt>' intrinsic has two modes
5557 of operation: forwards and reverse. If <tt>%lo</tt> is greater than
5558 <tt>%hi</tt> then the intrinsic operates in reverse mode. Otherwise it
5559 operates in forward mode.</p>
5560 <p>For both modes, the <tt>%repl</tt> value is prepared for use by either
5561 truncating it down to the size of the replacement area or zero extending it
5562 up to that size.</p>
5563 <p>In forward mode, the bits between <tt>%lo</tt> and <tt>%hi</tt> (inclusive)
5564 are replaced with corresponding bits from <tt>%repl</tt>. That is the 0th bit
5565 in <tt>%repl</tt> replaces the <tt>%lo</tt>th bit in <tt>%val</tt> and etc. up
5566 to the <tt>%hi</tt>th bit.
5567 <p>In reverse mode, a similar computation is made except that the bits are
5568 reversed. That is, the <tt>0</tt>th bit in <tt>%repl</tt> replaces the
5569 <tt>%hi</tt> bit in <tt>%val</tt> and etc. down to the <tt>%lo</tt>th bit.
5572 llvm.part.set(0xFFFF, 0, 4, 7) -> 0xFF0F
5573 llvm.part.set(0xFFFF, 0, 7, 4) -> 0xFF0F
5574 llvm.part.set(0xFFFF, 1, 7, 4) -> 0xFF8F
5575 llvm.part.set(0xFFFF, F, 8, 3) -> 0xFFE7
5576 llvm.part.set(0xFFFF, 0, 3, 8) -> 0xFE07
5580 <!-- ======================================================================= -->
5581 <div class="doc_subsection">
5582 <a name="int_debugger">Debugger Intrinsics</a>
5585 <div class="doc_text">
5587 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
5588 are described in the <a
5589 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
5590 Debugging</a> document.
5595 <!-- ======================================================================= -->
5596 <div class="doc_subsection">
5597 <a name="int_eh">Exception Handling Intrinsics</a>
5600 <div class="doc_text">
5601 <p> The LLVM exception handling intrinsics (which all start with
5602 <tt>llvm.eh.</tt> prefix), are described in the <a
5603 href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
5604 Handling</a> document. </p>
5607 <!-- ======================================================================= -->
5608 <div class="doc_subsection">
5609 <a name="int_trampoline">Trampoline Intrinsic</a>
5612 <div class="doc_text">
5614 This intrinsic makes it possible to excise one parameter, marked with
5615 the <tt>nest</tt> attribute, from a function. The result is a callable
5616 function pointer lacking the nest parameter - the caller does not need
5617 to provide a value for it. Instead, the value to use is stored in
5618 advance in a "trampoline", a block of memory usually allocated
5619 on the stack, which also contains code to splice the nest value into the
5620 argument list. This is used to implement the GCC nested function address
5624 For example, if the function is
5625 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
5626 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as follows:</p>
5628 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
5629 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
5630 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
5631 %fp = bitcast i8* %p to i32 (i32, i32)*
5633 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
5634 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
5637 <!-- _______________________________________________________________________ -->
5638 <div class="doc_subsubsection">
5639 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
5641 <div class="doc_text">
5644 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
5648 This fills the memory pointed to by <tt>tramp</tt> with code
5649 and returns a function pointer suitable for executing it.
5653 The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
5654 pointers. The <tt>tramp</tt> argument must point to a sufficiently large
5655 and sufficiently aligned block of memory; this memory is written to by the
5656 intrinsic. Note that the size and the alignment are target-specific - LLVM
5657 currently provides no portable way of determining them, so a front-end that
5658 generates this intrinsic needs to have some target-specific knowledge.
5659 The <tt>func</tt> argument must hold a function bitcast to an <tt>i8*</tt>.
5663 The block of memory pointed to by <tt>tramp</tt> is filled with target
5664 dependent code, turning it into a function. A pointer to this function is
5665 returned, but needs to be bitcast to an
5666 <a href="#int_trampoline">appropriate function pointer type</a>
5667 before being called. The new function's signature is the same as that of
5668 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
5669 removed. At most one such <tt>nest</tt> argument is allowed, and it must be
5670 of pointer type. Calling the new function is equivalent to calling
5671 <tt>func</tt> with the same argument list, but with <tt>nval</tt> used for the
5672 missing <tt>nest</tt> argument. If, after calling
5673 <tt>llvm.init.trampoline</tt>, the memory pointed to by <tt>tramp</tt> is
5674 modified, then the effect of any later call to the returned function pointer is
5679 <!-- ======================================================================= -->
5680 <div class="doc_subsection">
5681 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
5684 <div class="doc_text">
5686 These intrinsic functions expand the "universal IR" of LLVM to represent
5687 hardware constructs for atomic operations and memory synchronization. This
5688 provides an interface to the hardware, not an interface to the programmer. It
5689 is aimed at a low enough level to allow any programming models or APIs which
5690 need atomic behaviors to map cleanly onto it. It is also modeled primarily on
5691 hardware behavior. Just as hardware provides a "universal IR" for source
5692 languages, it also provides a starting point for developing a "universal"
5693 atomic operation and synchronization IR.
5696 These do <em>not</em> form an API such as high-level threading libraries,
5697 software transaction memory systems, atomic primitives, and intrinsic
5698 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
5699 application libraries. The hardware interface provided by LLVM should allow
5700 a clean implementation of all of these APIs and parallel programming models.
5701 No one model or paradigm should be selected above others unless the hardware
5702 itself ubiquitously does so.
5707 <!-- _______________________________________________________________________ -->
5708 <div class="doc_subsubsection">
5709 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
5711 <div class="doc_text">
5714 declare void @llvm.memory.barrier( i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>,
5720 The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
5721 specific pairs of memory access types.
5725 The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
5726 The first four arguments enables a specific barrier as listed below. The fith
5727 argument specifies that the barrier applies to io or device or uncached memory.
5731 <li><tt>ll</tt>: load-load barrier</li>
5732 <li><tt>ls</tt>: load-store barrier</li>
5733 <li><tt>sl</tt>: store-load barrier</li>
5734 <li><tt>ss</tt>: store-store barrier</li>
5735 <li><tt>device</tt>: barrier applies to device and uncached memory also.
5739 This intrinsic causes the system to enforce some ordering constraints upon
5740 the loads and stores of the program. This barrier does not indicate
5741 <em>when</em> any events will occur, it only enforces an <em>order</em> in
5742 which they occur. For any of the specified pairs of load and store operations
5743 (f.ex. load-load, or store-load), all of the first operations preceding the
5744 barrier will complete before any of the second operations succeeding the
5745 barrier begin. Specifically the semantics for each pairing is as follows:
5748 <li><tt>ll</tt>: All loads before the barrier must complete before any load
5749 after the barrier begins.</li>
5751 <li><tt>ls</tt>: All loads before the barrier must complete before any
5752 store after the barrier begins.</li>
5753 <li><tt>ss</tt>: All stores before the barrier must complete before any
5754 store after the barrier begins.</li>
5755 <li><tt>sl</tt>: All stores before the barrier must complete before any
5756 load after the barrier begins.</li>
5759 These semantics are applied with a logical "and" behavior when more than one
5760 is enabled in a single memory barrier intrinsic.
5763 Backends may implement stronger barriers than those requested when they do not
5764 support as fine grained a barrier as requested. Some architectures do not
5765 need all types of barriers and on such architectures, these become noops.
5772 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
5773 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
5774 <i>; guarantee the above finishes</i>
5775 store i32 8, %ptr <i>; before this begins</i>
5779 <!-- _______________________________________________________________________ -->
5780 <div class="doc_subsubsection">
5781 <a name="int_atomic_lcs">'<tt>llvm.atomic.lcs.*</tt>' Intrinsic</a>
5783 <div class="doc_text">
5786 This is an overloaded intrinsic. You can use <tt>llvm.atomic.lcs</tt> on any
5787 integer bit width. Not all targets support all bit widths however.</p>
5790 declare i8 @llvm.atomic.lcs.i8( i8* <ptr>, i8 <cmp>, i8 <val> )
5791 declare i16 @llvm.atomic.lcs.i16( i16* <ptr>, i16 <cmp>, i16 <val> )
5792 declare i32 @llvm.atomic.lcs.i32( i32* <ptr>, i32 <cmp>, i32 <val> )
5793 declare i64 @llvm.atomic.lcs.i64( i64* <ptr>, i64 <cmp>, i64 <val> )
5798 This loads a value in memory and compares it to a given value. If they are
5799 equal, it stores a new value into the memory.
5803 The <tt>llvm.atomic.lcs</tt> intrinsic takes three arguments. The result as
5804 well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
5805 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
5806 this integer type. While any bit width integer may be used, targets may only
5807 lower representations they support in hardware.
5812 This entire intrinsic must be executed atomically. It first loads the value
5813 in memory pointed to by <tt>ptr</tt> and compares it with the value
5814 <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the memory. The
5815 loaded value is yielded in all cases. This provides the equivalent of an
5816 atomic compare-and-swap operation within the SSA framework.
5824 %val1 = add i32 4, 4
5825 %result1 = call i32 @llvm.atomic.lcs.i32( i32* %ptr, i32 4, %val1 )
5826 <i>; yields {i32}:result1 = 4</i>
5827 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
5828 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
5830 %val2 = add i32 1, 1
5831 %result2 = call i32 @llvm.atomic.lcs.i32( i32* %ptr, i32 5, %val2 )
5832 <i>; yields {i32}:result2 = 8</i>
5833 %stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
5835 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
5839 <!-- _______________________________________________________________________ -->
5840 <div class="doc_subsubsection">
5841 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
5843 <div class="doc_text">
5847 This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
5848 integer bit width. Not all targets support all bit widths however.</p>
5850 declare i8 @llvm.atomic.swap.i8( i8* <ptr>, i8 <val> )
5851 declare i16 @llvm.atomic.swap.i16( i16* <ptr>, i16 <val> )
5852 declare i32 @llvm.atomic.swap.i32( i32* <ptr>, i32 <val> )
5853 declare i64 @llvm.atomic.swap.i64( i64* <ptr>, i64 <val> )
5858 This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
5859 the value from memory. It then stores the value in <tt>val</tt> in the memory
5865 The <tt>llvm.atomic.ls</tt> intrinsic takes two arguments. Both the
5866 <tt>val</tt> argument and the result must be integers of the same bit width.
5867 The first argument, <tt>ptr</tt>, must be a pointer to a value of this
5868 integer type. The targets may only lower integer representations they
5873 This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
5874 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
5875 equivalent of an atomic swap operation within the SSA framework.
5883 %val1 = add i32 4, 4
5884 %result1 = call i32 @llvm.atomic.swap.i32( i32* %ptr, i32 %val1 )
5885 <i>; yields {i32}:result1 = 4</i>
5886 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
5887 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
5889 %val2 = add i32 1, 1
5890 %result2 = call i32 @llvm.atomic.swap.i32( i32* %ptr, i32 %val2 )
5891 <i>; yields {i32}:result2 = 8</i>
5893 %stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
5894 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
5898 <!-- _______________________________________________________________________ -->
5899 <div class="doc_subsubsection">
5900 <a name="int_atomic_las">'<tt>llvm.atomic.las.*</tt>' Intrinsic</a>
5903 <div class="doc_text">
5906 This is an overloaded intrinsic. You can use <tt>llvm.atomic.las</tt> on any
5907 integer bit width. Not all targets support all bit widths however.</p>
5909 declare i8 @llvm.atomic.las.i8.( i8* <ptr>, i8 <delta> )
5910 declare i16 @llvm.atomic.las.i16.( i16* <ptr>, i16 <delta> )
5911 declare i32 @llvm.atomic.las.i32.( i32* <ptr>, i32 <delta> )
5912 declare i64 @llvm.atomic.las.i64.( i64* <ptr>, i64 <delta> )
5917 This intrinsic adds <tt>delta</tt> to the value stored in memory at
5918 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
5923 The intrinsic takes two arguments, the first a pointer to an integer value
5924 and the second an integer value. The result is also an integer value. These
5925 integer types can have any bit width, but they must all have the same bit
5926 width. The targets may only lower integer representations they support.
5930 This intrinsic does a series of operations atomically. It first loads the
5931 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
5932 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
5939 %result1 = call i32 @llvm.atomic.las.i32( i32* %ptr, i32 4 )
5940 <i>; yields {i32}:result1 = 4</i>
5941 %result2 = call i32 @llvm.atomic.las.i32( i32* %ptr, i32 2 )
5942 <i>; yields {i32}:result2 = 8</i>
5943 %result3 = call i32 @llvm.atomic.las.i32( i32* %ptr, i32 5 )
5944 <i>; yields {i32}:result3 = 10</i>
5945 %memval = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
5950 <!-- ======================================================================= -->
5951 <div class="doc_subsection">
5952 <a name="int_general">General Intrinsics</a>
5955 <div class="doc_text">
5956 <p> This class of intrinsics is designed to be generic and has
5957 no specific purpose. </p>
5960 <!-- _______________________________________________________________________ -->
5961 <div class="doc_subsubsection">
5962 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
5965 <div class="doc_text">
5969 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
5975 The '<tt>llvm.var.annotation</tt>' intrinsic
5981 The first argument is a pointer to a value, the second is a pointer to a
5982 global string, the third is a pointer to a global string which is the source
5983 file name, and the last argument is the line number.
5989 This intrinsic allows annotation of local variables with arbitrary strings.
5990 This can be useful for special purpose optimizations that want to look for these
5991 annotations. These have no other defined use, they are ignored by code
5992 generation and optimization.
5996 <!-- _______________________________________________________________________ -->
5997 <div class="doc_subsubsection">
5998 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
6001 <div class="doc_text">
6004 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
6005 any integer bit width.
6008 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
6009 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
6010 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
6011 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
6012 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
6018 The '<tt>llvm.annotation</tt>' intrinsic.
6024 The first argument is an integer value (result of some expression),
6025 the second is a pointer to a global string, the third is a pointer to a global
6026 string which is the source file name, and the last argument is the line number.
6027 It returns the value of the first argument.
6033 This intrinsic allows annotations to be put on arbitrary expressions
6034 with arbitrary strings. This can be useful for special purpose optimizations
6035 that want to look for these annotations. These have no other defined use, they
6036 are ignored by code generation and optimization.
6039 <!-- _______________________________________________________________________ -->
6040 <div class="doc_subsubsection">
6041 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
6044 <div class="doc_text">
6048 declare void @llvm.trap()
6054 The '<tt>llvm.trap</tt>' intrinsic
6066 This intrinsics is lowered to the target dependent trap instruction. If the
6067 target does not have a trap instruction, this intrinsic will be lowered to the
6068 call of the abort() function.
6072 <!-- *********************************************************************** -->
6075 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
6076 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
6077 <a href="http://validator.w3.org/check/referer"><img
6078 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
6080 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
6081 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
6082 Last modified: $Date$