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="#namedtypes">Named Types</a></li>
26 <li><a href="#globalvars">Global Variables</a></li>
27 <li><a href="#functionstructure">Functions</a></li>
28 <li><a href="#aliasstructure">Aliases</a></li>
29 <li><a href="#paramattrs">Parameter Attributes</a></li>
30 <li><a href="#fnattrs">Function Attributes</a></li>
31 <li><a href="#gc">Garbage Collector Names</a></li>
32 <li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
33 <li><a href="#datalayout">Data Layout</a></li>
36 <li><a href="#typesystem">Type System</a>
38 <li><a href="#t_classifications">Type Classifications</a></li>
39 <li><a href="#t_primitive">Primitive Types</a>
41 <li><a href="#t_floating">Floating Point Types</a></li>
42 <li><a href="#t_void">Void Type</a></li>
43 <li><a href="#t_label">Label Type</a></li>
46 <li><a href="#t_derived">Derived Types</a>
48 <li><a href="#t_integer">Integer Type</a></li>
49 <li><a href="#t_array">Array Type</a></li>
50 <li><a href="#t_function">Function Type</a></li>
51 <li><a href="#t_pointer">Pointer Type</a></li>
52 <li><a href="#t_struct">Structure Type</a></li>
53 <li><a href="#t_pstruct">Packed Structure Type</a></li>
54 <li><a href="#t_vector">Vector Type</a></li>
55 <li><a href="#t_opaque">Opaque Type</a></li>
58 <li><a href="#t_uprefs">Type Up-references</a></li>
61 <li><a href="#constants">Constants</a>
63 <li><a href="#simpleconstants">Simple Constants</a></li>
64 <li><a href="#aggregateconstants">Aggregate Constants</a></li>
65 <li><a href="#globalconstants">Global Variable and Function Addresses</a></li>
66 <li><a href="#undefvalues">Undefined Values</a></li>
67 <li><a href="#constantexprs">Constant Expressions</a></li>
70 <li><a href="#othervalues">Other Values</a>
72 <li><a href="#inlineasm">Inline Assembler Expressions</a></li>
75 <li><a href="#instref">Instruction Reference</a>
77 <li><a href="#terminators">Terminator Instructions</a>
79 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
80 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
81 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
82 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
83 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
84 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
87 <li><a href="#binaryops">Binary Operations</a>
89 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
90 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
91 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
92 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
93 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
94 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
95 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
96 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
97 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
100 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
102 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
103 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
104 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
105 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
106 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
107 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
110 <li><a href="#vectorops">Vector Operations</a>
112 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
113 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
114 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
117 <li><a href="#aggregateops">Aggregate Operations</a>
119 <li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li>
120 <li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li>
123 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
125 <li><a href="#i_malloc">'<tt>malloc</tt>' Instruction</a></li>
126 <li><a href="#i_free">'<tt>free</tt>' Instruction</a></li>
127 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
128 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
129 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
130 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
133 <li><a href="#convertops">Conversion Operations</a>
135 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
136 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
137 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
138 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
139 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
140 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
141 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
142 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
143 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
144 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
145 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
146 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
149 <li><a href="#otherops">Other Operations</a>
151 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
152 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
153 <li><a href="#i_vicmp">'<tt>vicmp</tt>' Instruction</a></li>
154 <li><a href="#i_vfcmp">'<tt>vfcmp</tt>' Instruction</a></li>
155 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
156 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
157 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
158 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
163 <li><a href="#intrinsics">Intrinsic Functions</a>
165 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
167 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
168 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
169 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
172 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
174 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
175 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
176 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
179 <li><a href="#int_codegen">Code Generator Intrinsics</a>
181 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
182 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
183 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
184 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
185 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
186 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
187 <li><a href="#int_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
190 <li><a href="#int_libc">Standard C Library Intrinsics</a>
192 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
193 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
194 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
195 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
196 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
197 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
198 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
199 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
202 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
204 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
205 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
206 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
207 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
208 <li><a href="#int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic </a></li>
209 <li><a href="#int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic </a></li>
212 <li><a href="#int_overflow">Arithmetic with Overflow Intrinsics</a>
214 <li><a href="#int_sadd_ovf">'<tt>llvm.sadd.with.overflow.*</tt> Intrinsics</a></li>
215 <li><a href="#int_uadd_ovf">'<tt>llvm.uadd.with.overflow.*</tt> Intrinsics</a></li>
216 <li><a href="#int_ssub_ovf">'<tt>llvm.ssub.with.overflow.*</tt> Intrinsics</a></li>
217 <li><a href="#int_usub_ovf">'<tt>llvm.usub.with.overflow.*</tt> Intrinsics</a></li>
218 <li><a href="#int_smul_ovf">'<tt>llvm.smul.with.overflow.*</tt> Intrinsics</a></li>
221 <li><a href="#int_debugger">Debugger intrinsics</a></li>
222 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
223 <li><a href="#int_trampoline">Trampoline Intrinsic</a>
225 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
228 <li><a href="#int_atomics">Atomic intrinsics</a>
230 <li><a href="#int_memory_barrier"><tt>llvm.memory_barrier</tt></a></li>
231 <li><a href="#int_atomic_cmp_swap"><tt>llvm.atomic.cmp.swap</tt></a></li>
232 <li><a href="#int_atomic_swap"><tt>llvm.atomic.swap</tt></a></li>
233 <li><a href="#int_atomic_load_add"><tt>llvm.atomic.load.add</tt></a></li>
234 <li><a href="#int_atomic_load_sub"><tt>llvm.atomic.load.sub</tt></a></li>
235 <li><a href="#int_atomic_load_and"><tt>llvm.atomic.load.and</tt></a></li>
236 <li><a href="#int_atomic_load_nand"><tt>llvm.atomic.load.nand</tt></a></li>
237 <li><a href="#int_atomic_load_or"><tt>llvm.atomic.load.or</tt></a></li>
238 <li><a href="#int_atomic_load_xor"><tt>llvm.atomic.load.xor</tt></a></li>
239 <li><a href="#int_atomic_load_max"><tt>llvm.atomic.load.max</tt></a></li>
240 <li><a href="#int_atomic_load_min"><tt>llvm.atomic.load.min</tt></a></li>
241 <li><a href="#int_atomic_load_umax"><tt>llvm.atomic.load.umax</tt></a></li>
242 <li><a href="#int_atomic_load_umin"><tt>llvm.atomic.load.umin</tt></a></li>
245 <li><a href="#int_general">General intrinsics</a>
247 <li><a href="#int_var_annotation">
248 '<tt>llvm.var.annotation</tt>' Intrinsic</a></li>
249 <li><a href="#int_annotation">
250 '<tt>llvm.annotation.*</tt>' Intrinsic</a></li>
251 <li><a href="#int_trap">
252 '<tt>llvm.trap</tt>' Intrinsic</a></li>
253 <li><a href="#int_stackprotector">
254 '<tt>llvm.stackprotector</tt>' Intrinsic</a></li>
261 <div class="doc_author">
262 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
263 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
266 <!-- *********************************************************************** -->
267 <div class="doc_section"> <a name="abstract">Abstract </a></div>
268 <!-- *********************************************************************** -->
270 <div class="doc_text">
271 <p>This document is a reference manual for the LLVM assembly language.
272 LLVM is a Static Single Assignment (SSA) based representation that provides
273 type safety, low-level operations, flexibility, and the capability of
274 representing 'all' high-level languages cleanly. It is the common code
275 representation used throughout all phases of the LLVM compilation
279 <!-- *********************************************************************** -->
280 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
281 <!-- *********************************************************************** -->
283 <div class="doc_text">
285 <p>The LLVM code representation is designed to be used in three
286 different forms: as an in-memory compiler IR, as an on-disk bitcode
287 representation (suitable for fast loading by a Just-In-Time compiler),
288 and as a human readable assembly language representation. This allows
289 LLVM to provide a powerful intermediate representation for efficient
290 compiler transformations and analysis, while providing a natural means
291 to debug and visualize the transformations. The three different forms
292 of LLVM are all equivalent. This document describes the human readable
293 representation and notation.</p>
295 <p>The LLVM representation aims to be light-weight and low-level
296 while being expressive, typed, and extensible at the same time. It
297 aims to be a "universal IR" of sorts, by being at a low enough level
298 that high-level ideas may be cleanly mapped to it (similar to how
299 microprocessors are "universal IR's", allowing many source languages to
300 be mapped to them). By providing type information, LLVM can be used as
301 the target of optimizations: for example, through pointer analysis, it
302 can be proven that a C automatic variable is never accessed outside of
303 the current function... allowing it to be promoted to a simple SSA
304 value instead of a memory location.</p>
308 <!-- _______________________________________________________________________ -->
309 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
311 <div class="doc_text">
313 <p>It is important to note that this document describes 'well formed'
314 LLVM assembly language. There is a difference between what the parser
315 accepts and what is considered 'well formed'. For example, the
316 following instruction is syntactically okay, but not well formed:</p>
318 <div class="doc_code">
320 %x = <a href="#i_add">add</a> i32 1, %x
324 <p>...because the definition of <tt>%x</tt> does not dominate all of
325 its uses. The LLVM infrastructure provides a verification pass that may
326 be used to verify that an LLVM module is well formed. This pass is
327 automatically run by the parser after parsing input assembly and by
328 the optimizer before it outputs bitcode. The violations pointed out
329 by the verifier pass indicate bugs in transformation passes or input to
333 <!-- Describe the typesetting conventions here. -->
335 <!-- *********************************************************************** -->
336 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
337 <!-- *********************************************************************** -->
339 <div class="doc_text">
341 <p>LLVM identifiers come in two basic types: global and local. Global
342 identifiers (functions, global variables) begin with the @ character. Local
343 identifiers (register names, types) begin with the % character. Additionally,
344 there are three different formats for identifiers, for different purposes:</p>
347 <li>Named values are represented as a string of characters with their prefix.
348 For example, %foo, @DivisionByZero, %a.really.long.identifier. The actual
349 regular expression used is '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
350 Identifiers which require other characters in their names can be surrounded
351 with quotes. Special characters may be escaped using "\xx" where xx is the
352 ASCII code for the character in hexadecimal. In this way, any character can
353 be used in a name value, even quotes themselves.
355 <li>Unnamed values are represented as an unsigned numeric value with their
356 prefix. For example, %12, @2, %44.</li>
358 <li>Constants, which are described in a <a href="#constants">section about
359 constants</a>, below.</li>
362 <p>LLVM requires that values start with a prefix for two reasons: Compilers
363 don't need to worry about name clashes with reserved words, and the set of
364 reserved words may be expanded in the future without penalty. Additionally,
365 unnamed identifiers allow a compiler to quickly come up with a temporary
366 variable without having to avoid symbol table conflicts.</p>
368 <p>Reserved words in LLVM are very similar to reserved words in other
369 languages. There are keywords for different opcodes
370 ('<tt><a href="#i_add">add</a></tt>',
371 '<tt><a href="#i_bitcast">bitcast</a></tt>',
372 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
373 href="#t_void">void</a></tt>', '<tt><a href="#t_primitive">i32</a></tt>', etc...),
374 and others. These reserved words cannot conflict with variable names, because
375 none of them start with a prefix character ('%' or '@').</p>
377 <p>Here is an example of LLVM code to multiply the integer variable
378 '<tt>%X</tt>' by 8:</p>
382 <div class="doc_code">
384 %result = <a href="#i_mul">mul</a> i32 %X, 8
388 <p>After strength reduction:</p>
390 <div class="doc_code">
392 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
396 <p>And the hard way:</p>
398 <div class="doc_code">
400 <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
401 <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
402 %result = <a href="#i_add">add</a> i32 %1, %1
406 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
407 important lexical features of LLVM:</p>
411 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
414 <li>Unnamed temporaries are created when the result of a computation is not
415 assigned to a named value.</li>
417 <li>Unnamed temporaries are numbered sequentially</li>
421 <p>...and it also shows a convention that we follow in this document. When
422 demonstrating instructions, we will follow an instruction with a comment that
423 defines the type and name of value produced. Comments are shown in italic
428 <!-- *********************************************************************** -->
429 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
430 <!-- *********************************************************************** -->
432 <!-- ======================================================================= -->
433 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
436 <div class="doc_text">
438 <p>LLVM programs are composed of "Module"s, each of which is a
439 translation unit of the input programs. Each module consists of
440 functions, global variables, and symbol table entries. Modules may be
441 combined together with the LLVM linker, which merges function (and
442 global variable) definitions, resolves forward declarations, and merges
443 symbol table entries. Here is an example of the "hello world" module:</p>
445 <div class="doc_code">
446 <pre><i>; Declare the string constant as a global constant...</i>
447 <a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a
448 href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
450 <i>; External declaration of the puts function</i>
451 <a href="#functionstructure">declare</a> i32 @puts(i8 *) <i>; i32(i8 *)* </i>
453 <i>; Definition of main function</i>
454 define i32 @main() { <i>; i32()* </i>
455 <i>; Convert [13 x i8]* to i8 *...</i>
457 href="#i_getelementptr">getelementptr</a> [13 x i8]* @.LC0, i64 0, i64 0 <i>; i8 *</i>
459 <i>; Call puts function to write out the string to stdout...</i>
461 href="#i_call">call</a> i32 @puts(i8 * %cast210) <i>; i32</i>
463 href="#i_ret">ret</a> i32 0<br>}<br>
467 <p>This example is made up of a <a href="#globalvars">global variable</a>
468 named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
469 function, and a <a href="#functionstructure">function definition</a>
470 for "<tt>main</tt>".</p>
472 <p>In general, a module is made up of a list of global values,
473 where both functions and global variables are global values. Global values are
474 represented by a pointer to a memory location (in this case, a pointer to an
475 array of char, and a pointer to a function), and have one of the following <a
476 href="#linkage">linkage types</a>.</p>
480 <!-- ======================================================================= -->
481 <div class="doc_subsection">
482 <a name="linkage">Linkage Types</a>
485 <div class="doc_text">
488 All Global Variables and Functions have one of the following types of linkage:
493 <dt><tt><b><a name="linkage_private">private</a></b></tt>: </dt>
495 <dd>Global values with private linkage are only directly accessible by
496 objects in the current module. In particular, linking code into a module with
497 an private global value may cause the private to be renamed as necessary to
498 avoid collisions. Because the symbol is private to the module, all
499 references can be updated. This doesn't show up in any symbol table in the
503 <dt><tt><b><a name="linkage_internal">internal</a></b></tt>: </dt>
505 <dd> Similar to private, but the value shows as a local symbol (STB_LOCAL in
506 the case of ELF) in the object file. This corresponds to the notion of the
507 '<tt>static</tt>' keyword in C.
510 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
512 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
513 the same name when linkage occurs. This is typically used to implement
514 inline functions, templates, or other code which must be generated in each
515 translation unit that uses it. Unreferenced <tt>linkonce</tt> globals are
516 allowed to be discarded.
519 <dt><tt><b><a name="linkage_common">common</a></b></tt>: </dt>
521 <dd>"<tt>common</tt>" linkage is exactly the same as <tt>linkonce</tt>
522 linkage, except that unreferenced <tt>common</tt> globals may not be
523 discarded. This is used for globals that may be emitted in multiple
524 translation units, but that are not guaranteed to be emitted into every
525 translation unit that uses them. One example of this is tentative
526 definitions in C, such as "<tt>int X;</tt>" at global scope.
529 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
531 <dd>"<tt>weak</tt>" linkage is the same as <tt>common</tt> linkage, except
532 that some targets may choose to emit different assembly sequences for them
533 for target-dependent reasons. This is used for globals that are declared
534 "weak" in C source code.
537 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
539 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
540 pointer to array type. When two global variables with appending linkage are
541 linked together, the two global arrays are appended together. This is the
542 LLVM, typesafe, equivalent of having the system linker append together
543 "sections" with identical names when .o files are linked.
546 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt>: </dt>
547 <dd>The semantics of this linkage follow the ELF object file model: the
548 symbol is weak until linked, if not linked, the symbol becomes null instead
549 of being an undefined reference.
552 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
554 <dd>If none of the above identifiers are used, the global is externally
555 visible, meaning that it participates in linkage and can be used to resolve
556 external symbol references.
561 The next two types of linkage are targeted for Microsoft Windows platform
562 only. They are designed to support importing (exporting) symbols from (to)
563 DLLs (Dynamic Link Libraries).
567 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt>: </dt>
569 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
570 or variable via a global pointer to a pointer that is set up by the DLL
571 exporting the symbol. On Microsoft Windows targets, the pointer name is
572 formed by combining <code>__imp_</code> and the function or variable name.
575 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt>: </dt>
577 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
578 pointer to a pointer in a DLL, so that it can be referenced with the
579 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
580 name is formed by combining <code>__imp_</code> and the function or variable
586 <p>For example, since the "<tt>.LC0</tt>"
587 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
588 variable and was linked with this one, one of the two would be renamed,
589 preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
590 external (i.e., lacking any linkage declarations), they are accessible
591 outside of the current module.</p>
592 <p>It is illegal for a function <i>declaration</i>
593 to have any linkage type other than "externally visible", <tt>dllimport</tt>,
594 or <tt>extern_weak</tt>.</p>
595 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt> and <tt>weak</tt>
599 <!-- ======================================================================= -->
600 <div class="doc_subsection">
601 <a name="callingconv">Calling Conventions</a>
604 <div class="doc_text">
606 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
607 and <a href="#i_invoke">invokes</a> can all have an optional calling convention
608 specified for the call. The calling convention of any pair of dynamic
609 caller/callee must match, or the behavior of the program is undefined. The
610 following calling conventions are supported by LLVM, and more may be added in
614 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
616 <dd>This calling convention (the default if no other calling convention is
617 specified) matches the target C calling conventions. This calling convention
618 supports varargs function calls and tolerates some mismatch in the declared
619 prototype and implemented declaration of the function (as does normal C).
622 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
624 <dd>This calling convention attempts to make calls as fast as possible
625 (e.g. by passing things in registers). This calling convention allows the
626 target to use whatever tricks it wants to produce fast code for the target,
627 without having to conform to an externally specified ABI (Application Binary
628 Interface). Implementations of this convention should allow arbitrary
629 <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> to be
630 supported. This calling convention does not support varargs and requires the
631 prototype of all callees to exactly match the prototype of the function
635 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
637 <dd>This calling convention attempts to make code in the caller as efficient
638 as possible under the assumption that the call is not commonly executed. As
639 such, these calls often preserve all registers so that the call does not break
640 any live ranges in the caller side. This calling convention does not support
641 varargs and requires the prototype of all callees to exactly match the
642 prototype of the function definition.
645 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
647 <dd>Any calling convention may be specified by number, allowing
648 target-specific calling conventions to be used. Target specific calling
649 conventions start at 64.
653 <p>More calling conventions can be added/defined on an as-needed basis, to
654 support pascal conventions or any other well-known target-independent
659 <!-- ======================================================================= -->
660 <div class="doc_subsection">
661 <a name="visibility">Visibility Styles</a>
664 <div class="doc_text">
667 All Global Variables and Functions have one of the following visibility styles:
671 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
673 <dd>On targets that use the ELF object file format, default visibility means
674 that the declaration is visible to other
675 modules and, in shared libraries, means that the declared entity may be
676 overridden. On Darwin, default visibility means that the declaration is
677 visible to other modules. Default visibility corresponds to "external
678 linkage" in the language.
681 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
683 <dd>Two declarations of an object with hidden visibility refer to the same
684 object if they are in the same shared object. Usually, hidden visibility
685 indicates that the symbol will not be placed into the dynamic symbol table,
686 so no other module (executable or shared library) can reference it
690 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
692 <dd>On ELF, protected visibility indicates that the symbol will be placed in
693 the dynamic symbol table, but that references within the defining module will
694 bind to the local symbol. That is, the symbol cannot be overridden by another
701 <!-- ======================================================================= -->
702 <div class="doc_subsection">
703 <a name="namedtypes">Named Types</a>
706 <div class="doc_text">
708 <p>LLVM IR allows you to specify name aliases for certain types. This can make
709 it easier to read the IR and make the IR more condensed (particularly when
710 recursive types are involved). An example of a name specification is:
713 <div class="doc_code">
715 %mytype = type { %mytype*, i32 }
719 <p>You may give a name to any <a href="#typesystem">type</a> except "<a
720 href="t_void">void</a>". Type name aliases may be used anywhere a type is
721 expected with the syntax "%mytype".</p>
723 <p>Note that type names are aliases for the structural type that they indicate,
724 and that you can therefore specify multiple names for the same type. This often
725 leads to confusing behavior when dumping out a .ll file. Since LLVM IR uses
726 structural typing, the name is not part of the type. When printing out LLVM IR,
727 the printer will pick <em>one name</em> to render all types of a particular
728 shape. This means that if you have code where two different source types end up
729 having the same LLVM type, that the dumper will sometimes print the "wrong" or
730 unexpected type. This is an important design point and isn't going to
735 <!-- ======================================================================= -->
736 <div class="doc_subsection">
737 <a name="globalvars">Global Variables</a>
740 <div class="doc_text">
742 <p>Global variables define regions of memory allocated at compilation time
743 instead of run-time. Global variables may optionally be initialized, may have
744 an explicit section to be placed in, and may have an optional explicit alignment
745 specified. A variable may be defined as "thread_local", which means that it
746 will not be shared by threads (each thread will have a separated copy of the
747 variable). A variable may be defined as a global "constant," which indicates
748 that the contents of the variable will <b>never</b> be modified (enabling better
749 optimization, allowing the global data to be placed in the read-only section of
750 an executable, etc). Note that variables that need runtime initialization
751 cannot be marked "constant" as there is a store to the variable.</p>
754 LLVM explicitly allows <em>declarations</em> of global variables to be marked
755 constant, even if the final definition of the global is not. This capability
756 can be used to enable slightly better optimization of the program, but requires
757 the language definition to guarantee that optimizations based on the
758 'constantness' are valid for the translation units that do not include the
762 <p>As SSA values, global variables define pointer values that are in
763 scope (i.e. they dominate) all basic blocks in the program. Global
764 variables always define a pointer to their "content" type because they
765 describe a region of memory, and all memory objects in LLVM are
766 accessed through pointers.</p>
768 <p>A global variable may be declared to reside in a target-specifc numbered
769 address space. For targets that support them, address spaces may affect how
770 optimizations are performed and/or what target instructions are used to access
771 the variable. The default address space is zero. The address space qualifier
772 must precede any other attributes.</p>
774 <p>LLVM allows an explicit section to be specified for globals. If the target
775 supports it, it will emit globals to the section specified.</p>
777 <p>An explicit alignment may be specified for a global. If not present, or if
778 the alignment is set to zero, the alignment of the global is set by the target
779 to whatever it feels convenient. If an explicit alignment is specified, the
780 global is forced to have at least that much alignment. All alignments must be
783 <p>For example, the following defines a global in a numbered address space with
784 an initializer, section, and alignment:</p>
786 <div class="doc_code">
788 @G = addrspace(5) constant float 1.0, section "foo", align 4
795 <!-- ======================================================================= -->
796 <div class="doc_subsection">
797 <a name="functionstructure">Functions</a>
800 <div class="doc_text">
802 <p>LLVM function definitions consist of the "<tt>define</tt>" keyord,
803 an optional <a href="#linkage">linkage type</a>, an optional
804 <a href="#visibility">visibility style</a>, an optional
805 <a href="#callingconv">calling convention</a>, a return type, an optional
806 <a href="#paramattrs">parameter attribute</a> for the return type, a function
807 name, a (possibly empty) argument list (each with optional
808 <a href="#paramattrs">parameter attributes</a>), optional
809 <a href="#fnattrs">function attributes</a>, an optional section,
810 an optional alignment, an optional <a href="#gc">garbage collector name</a>,
811 an opening curly brace, a list of basic blocks, and a closing curly brace.
813 LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
814 optional <a href="#linkage">linkage type</a>, an optional
815 <a href="#visibility">visibility style</a>, an optional
816 <a href="#callingconv">calling convention</a>, a return type, an optional
817 <a href="#paramattrs">parameter attribute</a> for the return type, a function
818 name, a possibly empty list of arguments, an optional alignment, and an optional
819 <a href="#gc">garbage collector name</a>.</p>
821 <p>A function definition contains a list of basic blocks, forming the CFG
822 (Control Flow Graph) for
823 the function. Each basic block may optionally start with a label (giving the
824 basic block a symbol table entry), contains a list of instructions, and ends
825 with a <a href="#terminators">terminator</a> instruction (such as a branch or
826 function return).</p>
828 <p>The first basic block in a function is special in two ways: it is immediately
829 executed on entrance to the function, and it is not allowed to have predecessor
830 basic blocks (i.e. there can not be any branches to the entry block of a
831 function). Because the block can have no predecessors, it also cannot have any
832 <a href="#i_phi">PHI nodes</a>.</p>
834 <p>LLVM allows an explicit section to be specified for functions. If the target
835 supports it, it will emit functions to the section specified.</p>
837 <p>An explicit alignment may be specified for a function. If not present, or if
838 the alignment is set to zero, the alignment of the function is set by the target
839 to whatever it feels convenient. If an explicit alignment is specified, the
840 function is forced to have at least that much alignment. All alignments must be
845 <div class="doc_code">
847 define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
848 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
849 <ResultType> @<FunctionName> ([argument list])
850 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
851 [<a href="#gc">gc</a>] { ... }
858 <!-- ======================================================================= -->
859 <div class="doc_subsection">
860 <a name="aliasstructure">Aliases</a>
862 <div class="doc_text">
863 <p>Aliases act as "second name" for the aliasee value (which can be either
864 function, global variable, another alias or bitcast of global value). Aliases
865 may have an optional <a href="#linkage">linkage type</a>, and an
866 optional <a href="#visibility">visibility style</a>.</p>
870 <div class="doc_code">
872 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
880 <!-- ======================================================================= -->
881 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
882 <div class="doc_text">
883 <p>The return type and each parameter of a function type may have a set of
884 <i>parameter attributes</i> associated with them. Parameter attributes are
885 used to communicate additional information about the result or parameters of
886 a function. Parameter attributes are considered to be part of the function,
887 not of the function type, so functions with different parameter attributes
888 can have the same function type.</p>
890 <p>Parameter attributes are simple keywords that follow the type specified. If
891 multiple parameter attributes are needed, they are space separated. For
894 <div class="doc_code">
896 declare i32 @printf(i8* noalias , ...)
897 declare i32 @atoi(i8 zeroext)
898 declare signext i8 @returns_signed_char()
902 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
903 <tt>readonly</tt>) come immediately after the argument list.</p>
905 <p>Currently, only the following parameter attributes are defined:</p>
907 <dt><tt>zeroext</tt></dt>
908 <dd>This indicates to the code generator that the parameter or return value
909 should be zero-extended to a 32-bit value by the caller (for a parameter)
910 or the callee (for a return value).</dd>
912 <dt><tt>signext</tt></dt>
913 <dd>This indicates to the code generator that the parameter or return value
914 should be sign-extended to a 32-bit value by the caller (for a parameter)
915 or the callee (for a return value).</dd>
917 <dt><tt>inreg</tt></dt>
918 <dd>This indicates that this parameter or return value should be treated
919 in a special target-dependent fashion during while emitting code for a
920 function call or return (usually, by putting it in a register as opposed
921 to memory, though some targets use it to distinguish between two different
922 kinds of registers). Use of this attribute is target-specific.</dd>
924 <dt><tt><a name="byval">byval</a></tt></dt>
925 <dd>This indicates that the pointer parameter should really be passed by
926 value to the function. The attribute implies that a hidden copy of the
927 pointee is made between the caller and the callee, so the callee is unable
928 to modify the value in the callee. This attribute is only valid on LLVM
929 pointer arguments. It is generally used to pass structs and arrays by
930 value, but is also valid on pointers to scalars. The copy is considered to
931 belong to the caller not the callee (for example,
932 <tt><a href="#readonly">readonly</a></tt> functions should not write to
933 <tt>byval</tt> parameters). This is not a valid attribute for return
934 values. The byval attribute also supports specifying an alignment with the
935 align attribute. This has a target-specific effect on the code generator
936 that usually indicates a desired alignment for the synthesized stack
939 <dt><tt>sret</tt></dt>
940 <dd>This indicates that the pointer parameter specifies the address of a
941 structure that is the return value of the function in the source program.
942 This pointer must be guaranteed by the caller to be valid: loads and stores
943 to the structure may be assumed by the callee to not to trap. This may only
944 be applied to the first parameter. This is not a valid attribute for
947 <dt><tt>noalias</tt></dt>
948 <dd>This indicates that the pointer does not alias any global or any other
949 parameter. The caller is responsible for ensuring that this is the
950 case. On a function return value, <tt>noalias</tt> additionally indicates
951 that the pointer does not alias any other pointers visible to the
952 caller. For further details, please see the discussion of the NoAlias
954 <a href="http://llvm.org/docs/AliasAnalysis.html#MustMayNo">alias
957 <dt><tt>nocapture</tt></dt>
958 <dd>This indicates that the callee does not make any copies of the pointer
959 that outlive the callee itself. This is not a valid attribute for return
962 <dt><tt>nest</tt></dt>
963 <dd>This indicates that the pointer parameter can be excised using the
964 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
965 attribute for return values.</dd>
970 <!-- ======================================================================= -->
971 <div class="doc_subsection">
972 <a name="gc">Garbage Collector Names</a>
975 <div class="doc_text">
976 <p>Each function may specify a garbage collector name, which is simply a
979 <div class="doc_code"><pre
980 >define void @f() gc "name" { ...</pre></div>
982 <p>The compiler declares the supported values of <i>name</i>. Specifying a
983 collector which will cause the compiler to alter its output in order to support
984 the named garbage collection algorithm.</p>
987 <!-- ======================================================================= -->
988 <div class="doc_subsection">
989 <a name="fnattrs">Function Attributes</a>
992 <div class="doc_text">
994 <p>Function attributes are set to communicate additional information about
995 a function. Function attributes are considered to be part of the function,
996 not of the function type, so functions with different parameter attributes
997 can have the same function type.</p>
999 <p>Function attributes are simple keywords that follow the type specified. If
1000 multiple attributes are needed, they are space separated. For
1003 <div class="doc_code">
1005 define void @f() noinline { ... }
1006 define void @f() alwaysinline { ... }
1007 define void @f() alwaysinline optsize { ... }
1008 define void @f() optsize
1013 <dt><tt>alwaysinline</tt></dt>
1014 <dd>This attribute indicates that the inliner should attempt to inline this
1015 function into callers whenever possible, ignoring any active inlining size
1016 threshold for this caller.</dd>
1018 <dt><tt>noinline</tt></dt>
1019 <dd>This attribute indicates that the inliner should never inline this function
1020 in any situation. This attribute may not be used together with the
1021 <tt>alwaysinline</tt> attribute.</dd>
1023 <dt><tt>optsize</tt></dt>
1024 <dd>This attribute suggests that optimization passes and code generator passes
1025 make choices that keep the code size of this function low, and otherwise do
1026 optimizations specifically to reduce code size.</dd>
1028 <dt><tt>noreturn</tt></dt>
1029 <dd>This function attribute indicates that the function never returns normally.
1030 This produces undefined behavior at runtime if the function ever does
1031 dynamically return.</dd>
1033 <dt><tt>nounwind</tt></dt>
1034 <dd>This function attribute indicates that the function never returns with an
1035 unwind or exceptional control flow. If the function does unwind, its runtime
1036 behavior is undefined.</dd>
1038 <dt><tt>readnone</tt></dt>
1039 <dd>This attribute indicates that the function computes its result (or the
1040 exception it throws) based strictly on its arguments, without dereferencing any
1041 pointer arguments or otherwise accessing any mutable state (e.g. memory, control
1042 registers, etc) visible to caller functions. It does not write through any
1043 pointer arguments (including <tt><a href="#byval">byval</a></tt> arguments) and
1044 never changes any state visible to callers.</dd>
1046 <dt><tt><a name="readonly">readonly</a></tt></dt>
1047 <dd>This attribute indicates that the function does not write through any
1048 pointer arguments (including <tt><a href="#byval">byval</a></tt> arguments)
1049 or otherwise modify any state (e.g. memory, control registers, etc) visible to
1050 caller functions. It may dereference pointer arguments and read state that may
1051 be set in the caller. A readonly function always returns the same value (or
1052 throws the same exception) when called with the same set of arguments and global
1055 <dt><tt><a name="ssp">ssp</a></tt></dt>
1056 <dd>This attribute indicates that the function should emit a stack smashing
1057 protector. It is in the form of a "canary"—a random value placed on the
1058 stack before the local variables that's checked upon return from the function to
1059 see if it has been overwritten. A heuristic is used to determine if a function
1060 needs stack protectors or not.
1062 <p>If a function that has an <tt>ssp</tt> attribute is inlined into a function
1063 that doesn't have an <tt>ssp</tt> attribute, then the resulting function will
1064 have an <tt>ssp</tt> attribute.</p></dd>
1066 <dt><tt>sspreq</tt></dt>
1067 <dd>This attribute indicates that the function should <em>always</em> emit a
1068 stack smashing protector. This overrides the <tt><a href="#ssp">ssp</a></tt>
1071 <p>If a function that has an <tt>sspreq</tt> attribute is inlined into a
1072 function that doesn't have an <tt>sspreq</tt> attribute or which has
1073 an <tt>ssp</tt> attribute, then the resulting function will have
1074 an <tt>sspreq</tt> attribute.</p></dd>
1079 <!-- ======================================================================= -->
1080 <div class="doc_subsection">
1081 <a name="moduleasm">Module-Level Inline Assembly</a>
1084 <div class="doc_text">
1086 Modules may contain "module-level inline asm" blocks, which corresponds to the
1087 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
1088 LLVM and treated as a single unit, but may be separated in the .ll file if
1089 desired. The syntax is very simple:
1092 <div class="doc_code">
1094 module asm "inline asm code goes here"
1095 module asm "more can go here"
1099 <p>The strings can contain any character by escaping non-printable characters.
1100 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1105 The inline asm code is simply printed to the machine code .s file when
1106 assembly code is generated.
1110 <!-- ======================================================================= -->
1111 <div class="doc_subsection">
1112 <a name="datalayout">Data Layout</a>
1115 <div class="doc_text">
1116 <p>A module may specify a target specific data layout string that specifies how
1117 data is to be laid out in memory. The syntax for the data layout is simply:</p>
1118 <pre> target datalayout = "<i>layout specification</i>"</pre>
1119 <p>The <i>layout specification</i> consists of a list of specifications
1120 separated by the minus sign character ('-'). Each specification starts with a
1121 letter and may include other information after the letter to define some
1122 aspect of the data layout. The specifications accepted are as follows: </p>
1125 <dd>Specifies that the target lays out data in big-endian form. That is, the
1126 bits with the most significance have the lowest address location.</dd>
1128 <dd>Specifies that the target lays out data in little-endian form. That is,
1129 the bits with the least significance have the lowest address location.</dd>
1130 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1131 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1132 <i>preferred</i> alignments. All sizes are in bits. Specifying the <i>pref</i>
1133 alignment is optional. If omitted, the preceding <tt>:</tt> should be omitted
1135 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1136 <dd>This specifies the alignment for an integer type of a given bit
1137 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1138 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1139 <dd>This specifies the alignment for a vector type of a given bit
1141 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1142 <dd>This specifies the alignment for a floating point type of a given bit
1143 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
1145 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1146 <dd>This specifies the alignment for an aggregate type of a given bit
1149 <p>When constructing the data layout for a given target, LLVM starts with a
1150 default set of specifications which are then (possibly) overriden by the
1151 specifications in the <tt>datalayout</tt> keyword. The default specifications
1152 are given in this list:</p>
1154 <li><tt>E</tt> - big endian</li>
1155 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
1156 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1157 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1158 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1159 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1160 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1161 alignment of 64-bits</li>
1162 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1163 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1164 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1165 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1166 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1168 <p>When LLVM is determining the alignment for a given type, it uses the
1169 following rules:</p>
1171 <li>If the type sought is an exact match for one of the specifications, that
1172 specification is used.</li>
1173 <li>If no match is found, and the type sought is an integer type, then the
1174 smallest integer type that is larger than the bitwidth of the sought type is
1175 used. If none of the specifications are larger than the bitwidth then the the
1176 largest integer type is used. For example, given the default specifications
1177 above, the i7 type will use the alignment of i8 (next largest) while both
1178 i65 and i256 will use the alignment of i64 (largest specified).</li>
1179 <li>If no match is found, and the type sought is a vector type, then the
1180 largest vector type that is smaller than the sought vector type will be used
1181 as a fall back. This happens because <128 x double> can be implemented
1182 in terms of 64 <2 x double>, for example.</li>
1186 <!-- *********************************************************************** -->
1187 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
1188 <!-- *********************************************************************** -->
1190 <div class="doc_text">
1192 <p>The LLVM type system is one of the most important features of the
1193 intermediate representation. Being typed enables a number of
1194 optimizations to be performed on the intermediate representation directly,
1195 without having to do
1196 extra analyses on the side before the transformation. A strong type
1197 system makes it easier to read the generated code and enables novel
1198 analyses and transformations that are not feasible to perform on normal
1199 three address code representations.</p>
1203 <!-- ======================================================================= -->
1204 <div class="doc_subsection"> <a name="t_classifications">Type
1205 Classifications</a> </div>
1206 <div class="doc_text">
1207 <p>The types fall into a few useful
1208 classifications:</p>
1210 <table border="1" cellspacing="0" cellpadding="4">
1212 <tr><th>Classification</th><th>Types</th></tr>
1214 <td><a href="#t_integer">integer</a></td>
1215 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1218 <td><a href="#t_floating">floating point</a></td>
1219 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1222 <td><a name="t_firstclass">first class</a></td>
1223 <td><a href="#t_integer">integer</a>,
1224 <a href="#t_floating">floating point</a>,
1225 <a href="#t_pointer">pointer</a>,
1226 <a href="#t_vector">vector</a>,
1227 <a href="#t_struct">structure</a>,
1228 <a href="#t_array">array</a>,
1229 <a href="#t_label">label</a>.
1233 <td><a href="#t_primitive">primitive</a></td>
1234 <td><a href="#t_label">label</a>,
1235 <a href="#t_void">void</a>,
1236 <a href="#t_floating">floating point</a>.</td>
1239 <td><a href="#t_derived">derived</a></td>
1240 <td><a href="#t_integer">integer</a>,
1241 <a href="#t_array">array</a>,
1242 <a href="#t_function">function</a>,
1243 <a href="#t_pointer">pointer</a>,
1244 <a href="#t_struct">structure</a>,
1245 <a href="#t_pstruct">packed structure</a>,
1246 <a href="#t_vector">vector</a>,
1247 <a href="#t_opaque">opaque</a>.
1253 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
1254 most important. Values of these types are the only ones which can be
1255 produced by instructions, passed as arguments, or used as operands to
1259 <!-- ======================================================================= -->
1260 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
1262 <div class="doc_text">
1263 <p>The primitive types are the fundamental building blocks of the LLVM
1268 <!-- _______________________________________________________________________ -->
1269 <div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1271 <div class="doc_text">
1274 <tr><th>Type</th><th>Description</th></tr>
1275 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1276 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1277 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1278 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1279 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1284 <!-- _______________________________________________________________________ -->
1285 <div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1287 <div class="doc_text">
1289 <p>The void type does not represent any value and has no size.</p>
1298 <!-- _______________________________________________________________________ -->
1299 <div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1301 <div class="doc_text">
1303 <p>The label type represents code labels.</p>
1313 <!-- ======================================================================= -->
1314 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1316 <div class="doc_text">
1318 <p>The real power in LLVM comes from the derived types in the system.
1319 This is what allows a programmer to represent arrays, functions,
1320 pointers, and other useful types. Note that these derived types may be
1321 recursive: For example, it is possible to have a two dimensional array.</p>
1325 <!-- _______________________________________________________________________ -->
1326 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1328 <div class="doc_text">
1331 <p>The integer type is a very simple derived type that simply specifies an
1332 arbitrary bit width for the integer type desired. Any bit width from 1 bit to
1333 2^23-1 (about 8 million) can be specified.</p>
1341 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1345 <table class="layout">
1348 <td><tt>i1</tt></td>
1349 <td>a single-bit integer.</td>
1351 <td><tt>i32</tt></td>
1352 <td>a 32-bit integer.</td>
1354 <td><tt>i1942652</tt></td>
1355 <td>a really big integer of over 1 million bits.</td>
1360 <p>Note that the code generator does not yet support large integer types
1361 to be used as function return types. The specific limit on how large a
1362 return type the code generator can currently handle is target-dependent;
1363 currently it's often 64 bits for 32-bit targets and 128 bits for 64-bit
1368 <!-- _______________________________________________________________________ -->
1369 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1371 <div class="doc_text">
1375 <p>The array type is a very simple derived type that arranges elements
1376 sequentially in memory. The array type requires a size (number of
1377 elements) and an underlying data type.</p>
1382 [<# elements> x <elementtype>]
1385 <p>The number of elements is a constant integer value; elementtype may
1386 be any type with a size.</p>
1389 <table class="layout">
1391 <td class="left"><tt>[40 x i32]</tt></td>
1392 <td class="left">Array of 40 32-bit integer values.</td>
1395 <td class="left"><tt>[41 x i32]</tt></td>
1396 <td class="left">Array of 41 32-bit integer values.</td>
1399 <td class="left"><tt>[4 x i8]</tt></td>
1400 <td class="left">Array of 4 8-bit integer values.</td>
1403 <p>Here are some examples of multidimensional arrays:</p>
1404 <table class="layout">
1406 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1407 <td class="left">3x4 array of 32-bit integer values.</td>
1410 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1411 <td class="left">12x10 array of single precision floating point values.</td>
1414 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1415 <td class="left">2x3x4 array of 16-bit integer values.</td>
1419 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1420 length array. Normally, accesses past the end of an array are undefined in
1421 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
1422 As a special case, however, zero length arrays are recognized to be variable
1423 length. This allows implementation of 'pascal style arrays' with the LLVM
1424 type "{ i32, [0 x float]}", for example.</p>
1426 <p>Note that the code generator does not yet support large aggregate types
1427 to be used as function return types. The specific limit on how large an
1428 aggregate return type the code generator can currently handle is
1429 target-dependent, and also dependent on the aggregate element types.</p>
1433 <!-- _______________________________________________________________________ -->
1434 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1435 <div class="doc_text">
1439 <p>The function type can be thought of as a function signature. It
1440 consists of a return type and a list of formal parameter types. The
1441 return type of a function type is a scalar type, a void type, or a struct type.
1442 If the return type is a struct type then all struct elements must be of first
1443 class types, and the struct must have at least one element.</p>
1448 <returntype list> (<parameter list>)
1451 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1452 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1453 which indicates that the function takes a variable number of arguments.
1454 Variable argument functions can access their arguments with the <a
1455 href="#int_varargs">variable argument handling intrinsic</a> functions.
1456 '<tt><returntype list></tt>' is a comma-separated list of
1457 <a href="#t_firstclass">first class</a> type specifiers.</p>
1460 <table class="layout">
1462 <td class="left"><tt>i32 (i32)</tt></td>
1463 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1465 </tr><tr class="layout">
1466 <td class="left"><tt>float (i16 signext, i32 *) *
1468 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1469 an <tt>i16</tt> that should be sign extended and a
1470 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1473 </tr><tr class="layout">
1474 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1475 <td class="left">A vararg function that takes at least one
1476 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1477 which returns an integer. This is the signature for <tt>printf</tt> in
1480 </tr><tr class="layout">
1481 <td class="left"><tt>{i32, i32} (i32)</tt></td>
1482 <td class="left">A function taking an <tt>i32</tt>, returning two
1483 <tt>i32</tt> values as an aggregate of type <tt>{ i32, i32 }</tt>
1489 <!-- _______________________________________________________________________ -->
1490 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1491 <div class="doc_text">
1493 <p>The structure type is used to represent a collection of data members
1494 together in memory. The packing of the field types is defined to match
1495 the ABI of the underlying processor. The elements of a structure may
1496 be any type that has a size.</p>
1497 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1498 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1499 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1502 <pre> { <type list> }<br></pre>
1504 <table class="layout">
1506 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1507 <td class="left">A triple of three <tt>i32</tt> values</td>
1508 </tr><tr class="layout">
1509 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1510 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1511 second element is a <a href="#t_pointer">pointer</a> to a
1512 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1513 an <tt>i32</tt>.</td>
1517 <p>Note that the code generator does not yet support large aggregate types
1518 to be used as function return types. The specific limit on how large an
1519 aggregate return type the code generator can currently handle is
1520 target-dependent, and also dependent on the aggregate element types.</p>
1524 <!-- _______________________________________________________________________ -->
1525 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1527 <div class="doc_text">
1529 <p>The packed structure type is used to represent a collection of data members
1530 together in memory. There is no padding between fields. Further, the alignment
1531 of a packed structure is 1 byte. The elements of a packed structure may
1532 be any type that has a size.</p>
1533 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1534 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1535 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1538 <pre> < { <type list> } > <br></pre>
1540 <table class="layout">
1542 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1543 <td class="left">A triple of three <tt>i32</tt> values</td>
1544 </tr><tr class="layout">
1546 <tt>< { float, i32 (i32)* } ></tt></td>
1547 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1548 second element is a <a href="#t_pointer">pointer</a> to a
1549 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1550 an <tt>i32</tt>.</td>
1555 <!-- _______________________________________________________________________ -->
1556 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1557 <div class="doc_text">
1559 <p>As in many languages, the pointer type represents a pointer or
1560 reference to another object, which must live in memory. Pointer types may have
1561 an optional address space attribute defining the target-specific numbered
1562 address space where the pointed-to object resides. The default address space is
1565 <pre> <type> *<br></pre>
1567 <table class="layout">
1569 <td class="left"><tt>[4 x i32]*</tt></td>
1570 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1571 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1574 <td class="left"><tt>i32 (i32 *) *</tt></td>
1575 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
1576 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1580 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1581 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1582 that resides in address space #5.</td>
1587 <!-- _______________________________________________________________________ -->
1588 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1589 <div class="doc_text">
1593 <p>A vector type is a simple derived type that represents a vector
1594 of elements. Vector types are used when multiple primitive data
1595 are operated in parallel using a single instruction (SIMD).
1596 A vector type requires a size (number of
1597 elements) and an underlying primitive data type. Vectors must have a power
1598 of two length (1, 2, 4, 8, 16 ...). Vector types are
1599 considered <a href="#t_firstclass">first class</a>.</p>
1604 < <# elements> x <elementtype> >
1607 <p>The number of elements is a constant integer value; elementtype may
1608 be any integer or floating point type.</p>
1612 <table class="layout">
1614 <td class="left"><tt><4 x i32></tt></td>
1615 <td class="left">Vector of 4 32-bit integer values.</td>
1618 <td class="left"><tt><8 x float></tt></td>
1619 <td class="left">Vector of 8 32-bit floating-point values.</td>
1622 <td class="left"><tt><2 x i64></tt></td>
1623 <td class="left">Vector of 2 64-bit integer values.</td>
1627 <p>Note that the code generator does not yet support large vector types
1628 to be used as function return types. The specific limit on how large a
1629 vector return type codegen can currently handle is target-dependent;
1630 currently it's often a few times longer than a hardware vector register.</p>
1634 <!-- _______________________________________________________________________ -->
1635 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1636 <div class="doc_text">
1640 <p>Opaque types are used to represent unknown types in the system. This
1641 corresponds (for example) to the C notion of a forward declared structure type.
1642 In LLVM, opaque types can eventually be resolved to any type (not just a
1643 structure type).</p>
1653 <table class="layout">
1655 <td class="left"><tt>opaque</tt></td>
1656 <td class="left">An opaque type.</td>
1661 <!-- ======================================================================= -->
1662 <div class="doc_subsection">
1663 <a name="t_uprefs">Type Up-references</a>
1666 <div class="doc_text">
1669 An "up reference" allows you to refer to a lexically enclosing type without
1670 requiring it to have a name. For instance, a structure declaration may contain a
1671 pointer to any of the types it is lexically a member of. Example of up
1672 references (with their equivalent as named type declarations) include:</p>
1675 { \2 * } %x = type { %t* }
1676 { \2 }* %y = type { %y }*
1681 An up reference is needed by the asmprinter for printing out cyclic types when
1682 there is no declared name for a type in the cycle. Because the asmprinter does
1683 not want to print out an infinite type string, it needs a syntax to handle
1684 recursive types that have no names (all names are optional in llvm IR).
1693 The level is the count of the lexical type that is being referred to.
1698 <table class="layout">
1700 <td class="left"><tt>\1*</tt></td>
1701 <td class="left">Self-referential pointer.</td>
1704 <td class="left"><tt>{ { \3*, i8 }, i32 }</tt></td>
1705 <td class="left">Recursive structure where the upref refers to the out-most
1712 <!-- *********************************************************************** -->
1713 <div class="doc_section"> <a name="constants">Constants</a> </div>
1714 <!-- *********************************************************************** -->
1716 <div class="doc_text">
1718 <p>LLVM has several different basic types of constants. This section describes
1719 them all and their syntax.</p>
1723 <!-- ======================================================================= -->
1724 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1726 <div class="doc_text">
1729 <dt><b>Boolean constants</b></dt>
1731 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1732 constants of the <tt><a href="#t_primitive">i1</a></tt> type.
1735 <dt><b>Integer constants</b></dt>
1737 <dd>Standard integers (such as '4') are constants of the <a
1738 href="#t_integer">integer</a> type. Negative numbers may be used with
1742 <dt><b>Floating point constants</b></dt>
1744 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1745 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1746 notation (see below). The assembler requires the exact decimal value of
1747 a floating-point constant. For example, the assembler accepts 1.25 but
1748 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
1749 constants must have a <a href="#t_floating">floating point</a> type. </dd>
1751 <dt><b>Null pointer constants</b></dt>
1753 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1754 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1758 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1759 of floating point constants. For example, the form '<tt>double
1760 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1761 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1762 (and the only time that they are generated by the disassembler) is when a
1763 floating point constant must be emitted but it cannot be represented as a
1764 decimal floating point number. For example, NaN's, infinities, and other
1765 special values are represented in their IEEE hexadecimal format so that
1766 assembly and disassembly do not cause any bits to change in the constants.</p>
1770 <!-- ======================================================================= -->
1771 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1774 <div class="doc_text">
1775 <p>Aggregate constants arise from aggregation of simple constants
1776 and smaller aggregate constants.</p>
1779 <dt><b>Structure constants</b></dt>
1781 <dd>Structure constants are represented with notation similar to structure
1782 type definitions (a comma separated list of elements, surrounded by braces
1783 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
1784 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>". Structure constants
1785 must have <a href="#t_struct">structure type</a>, and the number and
1786 types of elements must match those specified by the type.
1789 <dt><b>Array constants</b></dt>
1791 <dd>Array constants are represented with notation similar to array type
1792 definitions (a comma separated list of elements, surrounded by square brackets
1793 (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74 ]</tt>". Array
1794 constants must have <a href="#t_array">array type</a>, and the number and
1795 types of elements must match those specified by the type.
1798 <dt><b>Vector constants</b></dt>
1800 <dd>Vector constants are represented with notation similar to vector type
1801 definitions (a comma separated list of elements, surrounded by
1802 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32 42,
1803 i32 11, i32 74, i32 100 ></tt>". Vector constants must have <a
1804 href="#t_vector">vector type</a>, and the number and types of elements must
1805 match those specified by the type.
1808 <dt><b>Zero initialization</b></dt>
1810 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1811 value to zero of <em>any</em> type, including scalar and aggregate types.
1812 This is often used to avoid having to print large zero initializers (e.g. for
1813 large arrays) and is always exactly equivalent to using explicit zero
1820 <!-- ======================================================================= -->
1821 <div class="doc_subsection">
1822 <a name="globalconstants">Global Variable and Function Addresses</a>
1825 <div class="doc_text">
1827 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1828 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1829 constants. These constants are explicitly referenced when the <a
1830 href="#identifiers">identifier for the global</a> is used and always have <a
1831 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1834 <div class="doc_code">
1838 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1844 <!-- ======================================================================= -->
1845 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1846 <div class="doc_text">
1847 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1848 no specific value. Undefined values may be of any type and be used anywhere
1849 a constant is permitted.</p>
1851 <p>Undefined values indicate to the compiler that the program is well defined
1852 no matter what value is used, giving the compiler more freedom to optimize.
1856 <!-- ======================================================================= -->
1857 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1860 <div class="doc_text">
1862 <p>Constant expressions are used to allow expressions involving other constants
1863 to be used as constants. Constant expressions may be of any <a
1864 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1865 that does not have side effects (e.g. load and call are not supported). The
1866 following is the syntax for constant expressions:</p>
1869 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1870 <dd>Truncate a constant to another type. The bit size of CST must be larger
1871 than the bit size of TYPE. Both types must be integers.</dd>
1873 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1874 <dd>Zero extend a constant to another type. The bit size of CST must be
1875 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1877 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1878 <dd>Sign extend a constant to another type. The bit size of CST must be
1879 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1881 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1882 <dd>Truncate a floating point constant to another floating point type. The
1883 size of CST must be larger than the size of TYPE. Both types must be
1884 floating point.</dd>
1886 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1887 <dd>Floating point extend a constant to another type. The size of CST must be
1888 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1890 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
1891 <dd>Convert a floating point constant to the corresponding unsigned integer
1892 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1893 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1894 of the same number of elements. If the value won't fit in the integer type,
1895 the results are undefined.</dd>
1897 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
1898 <dd>Convert a floating point constant to the corresponding signed integer
1899 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1900 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1901 of the same number of elements. If the value won't fit in the integer type,
1902 the results are undefined.</dd>
1904 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
1905 <dd>Convert an unsigned integer constant to the corresponding floating point
1906 constant. TYPE must be a scalar or vector floating point type. CST must be of
1907 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1908 of the same number of elements. If the value won't fit in the floating point
1909 type, the results are undefined.</dd>
1911 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
1912 <dd>Convert a signed integer constant to the corresponding floating point
1913 constant. TYPE must be a scalar or vector floating point type. CST must be of
1914 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1915 of the same number of elements. If the value won't fit in the floating point
1916 type, the results are undefined.</dd>
1918 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
1919 <dd>Convert a pointer typed constant to the corresponding integer constant
1920 TYPE must be an integer type. CST must be of pointer type. The CST value is
1921 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1923 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
1924 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1925 pointer type. CST must be of integer type. The CST value is zero extended,
1926 truncated, or unchanged to make it fit in a pointer size. This one is
1927 <i>really</i> dangerous!</dd>
1929 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
1930 <dd>Convert a constant, CST, to another TYPE. The size of CST and TYPE must be
1931 identical (same number of bits). The conversion is done as if the CST value
1932 was stored to memory and read back as TYPE. In other words, no bits change
1933 with this operator, just the type. This can be used for conversion of
1934 vector types to any other type, as long as they have the same bit width. For
1935 pointers it is only valid to cast to another pointer type. It is not valid
1936 to bitcast to or from an aggregate type.
1939 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1941 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1942 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1943 instruction, the index list may have zero or more indexes, which are required
1944 to make sense for the type of "CSTPTR".</dd>
1946 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1948 <dd>Perform the <a href="#i_select">select operation</a> on
1951 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
1952 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
1954 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
1955 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
1957 <dt><b><tt>vicmp COND ( VAL1, VAL2 )</tt></b></dt>
1958 <dd>Performs the <a href="#i_vicmp">vicmp operation</a> on constants.</dd>
1960 <dt><b><tt>vfcmp COND ( VAL1, VAL2 )</tt></b></dt>
1961 <dd>Performs the <a href="#i_vfcmp">vfcmp operation</a> on constants.</dd>
1963 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1965 <dd>Perform the <a href="#i_extractelement">extractelement
1966 operation</a> on constants.</dd>
1968 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1970 <dd>Perform the <a href="#i_insertelement">insertelement
1971 operation</a> on constants.</dd>
1974 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
1976 <dd>Perform the <a href="#i_shufflevector">shufflevector
1977 operation</a> on constants.</dd>
1979 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1981 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1982 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1983 binary</a> operations. The constraints on operands are the same as those for
1984 the corresponding instruction (e.g. no bitwise operations on floating point
1985 values are allowed).</dd>
1989 <!-- *********************************************************************** -->
1990 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1991 <!-- *********************************************************************** -->
1993 <!-- ======================================================================= -->
1994 <div class="doc_subsection">
1995 <a name="inlineasm">Inline Assembler Expressions</a>
1998 <div class="doc_text">
2001 LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
2002 Module-Level Inline Assembly</a>) through the use of a special value. This
2003 value represents the inline assembler as a string (containing the instructions
2004 to emit), a list of operand constraints (stored as a string), and a flag that
2005 indicates whether or not the inline asm expression has side effects. An example
2006 inline assembler expression is:
2009 <div class="doc_code">
2011 i32 (i32) asm "bswap $0", "=r,r"
2016 Inline assembler expressions may <b>only</b> be used as the callee operand of
2017 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
2020 <div class="doc_code">
2022 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
2027 Inline asms with side effects not visible in the constraint list must be marked
2028 as having side effects. This is done through the use of the
2029 '<tt>sideeffect</tt>' keyword, like so:
2032 <div class="doc_code">
2034 call void asm sideeffect "eieio", ""()
2038 <p>TODO: The format of the asm and constraints string still need to be
2039 documented here. Constraints on what can be done (e.g. duplication, moving, etc
2040 need to be documented). This is probably best done by reference to another
2041 document that covers inline asm from a holistic perspective.
2046 <!-- *********************************************************************** -->
2047 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
2048 <!-- *********************************************************************** -->
2050 <div class="doc_text">
2052 <p>The LLVM instruction set consists of several different
2053 classifications of instructions: <a href="#terminators">terminator
2054 instructions</a>, <a href="#binaryops">binary instructions</a>,
2055 <a href="#bitwiseops">bitwise binary instructions</a>, <a
2056 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
2057 instructions</a>.</p>
2061 <!-- ======================================================================= -->
2062 <div class="doc_subsection"> <a name="terminators">Terminator
2063 Instructions</a> </div>
2065 <div class="doc_text">
2067 <p>As mentioned <a href="#functionstructure">previously</a>, every
2068 basic block in a program ends with a "Terminator" instruction, which
2069 indicates which block should be executed after the current block is
2070 finished. These terminator instructions typically yield a '<tt>void</tt>'
2071 value: they produce control flow, not values (the one exception being
2072 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
2073 <p>There are six different terminator instructions: the '<a
2074 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
2075 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
2076 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
2077 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
2078 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
2082 <!-- _______________________________________________________________________ -->
2083 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
2084 Instruction</a> </div>
2085 <div class="doc_text">
2088 ret <type> <value> <i>; Return a value from a non-void function</i>
2089 ret void <i>; Return from void function</i>
2094 <p>The '<tt>ret</tt>' instruction is used to return control flow (and
2095 optionally a value) from a function back to the caller.</p>
2096 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
2097 returns a value and then causes control flow, and one that just causes
2098 control flow to occur.</p>
2102 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument,
2103 the return value. The type of the return value must be a
2104 '<a href="#t_firstclass">first class</a>' type.</p>
2106 <p>A function is not <a href="#wellformed">well formed</a> if
2107 it it has a non-void return type and contains a '<tt>ret</tt>'
2108 instruction with no return value or a return value with a type that
2109 does not match its type, or if it has a void return type and contains
2110 a '<tt>ret</tt>' instruction with a return value.</p>
2114 <p>When the '<tt>ret</tt>' instruction is executed, control flow
2115 returns back to the calling function's context. If the caller is a "<a
2116 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
2117 the instruction after the call. If the caller was an "<a
2118 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
2119 at the beginning of the "normal" destination block. If the instruction
2120 returns a value, that value shall set the call or invoke instruction's
2126 ret i32 5 <i>; Return an integer value of 5</i>
2127 ret void <i>; Return from a void function</i>
2128 ret { i32, i8 } { i32 4, i8 2 } <i>; Return an aggregate of values 4 and 2</i>
2131 <p>Note that the code generator does not yet fully support large
2132 return values. The specific sizes that are currently supported are
2133 dependent on the target. For integers, on 32-bit targets the limit
2134 is often 64 bits, and on 64-bit targets the limit is often 128 bits.
2135 For aggregate types, the current limits are dependent on the element
2136 types; for example targets are often limited to 2 total integer
2137 elements and 2 total floating-point elements.</p>
2140 <!-- _______________________________________________________________________ -->
2141 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
2142 <div class="doc_text">
2144 <pre> br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
2147 <p>The '<tt>br</tt>' instruction is used to cause control flow to
2148 transfer to a different basic block in the current function. There are
2149 two forms of this instruction, corresponding to a conditional branch
2150 and an unconditional branch.</p>
2152 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
2153 single '<tt>i1</tt>' value and two '<tt>label</tt>' values. The
2154 unconditional form of the '<tt>br</tt>' instruction takes a single
2155 '<tt>label</tt>' value as a target.</p>
2157 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
2158 argument is evaluated. If the value is <tt>true</tt>, control flows
2159 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
2160 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
2162 <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
2163 href="#i_ret">ret</a> i32 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> i32 0<br></pre>
2165 <!-- _______________________________________________________________________ -->
2166 <div class="doc_subsubsection">
2167 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
2170 <div class="doc_text">
2174 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
2179 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
2180 several different places. It is a generalization of the '<tt>br</tt>'
2181 instruction, allowing a branch to occur to one of many possible
2187 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
2188 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
2189 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
2190 table is not allowed to contain duplicate constant entries.</p>
2194 <p>The <tt>switch</tt> instruction specifies a table of values and
2195 destinations. When the '<tt>switch</tt>' instruction is executed, this
2196 table is searched for the given value. If the value is found, control flow is
2197 transfered to the corresponding destination; otherwise, control flow is
2198 transfered to the default destination.</p>
2200 <h5>Implementation:</h5>
2202 <p>Depending on properties of the target machine and the particular
2203 <tt>switch</tt> instruction, this instruction may be code generated in different
2204 ways. For example, it could be generated as a series of chained conditional
2205 branches or with a lookup table.</p>
2210 <i>; Emulate a conditional br instruction</i>
2211 %Val = <a href="#i_zext">zext</a> i1 %value to i32
2212 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
2214 <i>; Emulate an unconditional br instruction</i>
2215 switch i32 0, label %dest [ ]
2217 <i>; Implement a jump table:</i>
2218 switch i32 %val, label %otherwise [ i32 0, label %onzero
2220 i32 2, label %ontwo ]
2224 <!-- _______________________________________________________________________ -->
2225 <div class="doc_subsubsection">
2226 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
2229 <div class="doc_text">
2234 <result> = invoke [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>] <ptr to function ty> <function ptr val>(<function args>) [<a href="#fnattrs">fn attrs</a>]
2235 to label <normal label> unwind label <exception label>
2240 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
2241 function, with the possibility of control flow transfer to either the
2242 '<tt>normal</tt>' label or the
2243 '<tt>exception</tt>' label. If the callee function returns with the
2244 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
2245 "normal" label. If the callee (or any indirect callees) returns with the "<a
2246 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
2247 continued at the dynamically nearest "exception" label.</p>
2251 <p>This instruction requires several arguments:</p>
2255 The optional "cconv" marker indicates which <a href="#callingconv">calling
2256 convention</a> the call should use. If none is specified, the call defaults
2257 to using C calling conventions.
2260 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
2261 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>',
2262 and '<tt>inreg</tt>' attributes are valid here.</li>
2264 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
2265 function value being invoked. In most cases, this is a direct function
2266 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
2267 an arbitrary pointer to function value.
2270 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
2271 function to be invoked. </li>
2273 <li>'<tt>function args</tt>': argument list whose types match the function
2274 signature argument types. If the function signature indicates the function
2275 accepts a variable number of arguments, the extra arguments can be
2278 <li>'<tt>normal label</tt>': the label reached when the called function
2279 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
2281 <li>'<tt>exception label</tt>': the label reached when a callee returns with
2282 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
2284 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
2285 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
2286 '<tt>readnone</tt>' attributes are valid here.</li>
2291 <p>This instruction is designed to operate as a standard '<tt><a
2292 href="#i_call">call</a></tt>' instruction in most regards. The primary
2293 difference is that it establishes an association with a label, which is used by
2294 the runtime library to unwind the stack.</p>
2296 <p>This instruction is used in languages with destructors to ensure that proper
2297 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
2298 exception. Additionally, this is important for implementation of
2299 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
2303 %retval = invoke i32 @Test(i32 15) to label %Continue
2304 unwind label %TestCleanup <i>; {i32}:retval set</i>
2305 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
2306 unwind label %TestCleanup <i>; {i32}:retval set</i>
2311 <!-- _______________________________________________________________________ -->
2313 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
2314 Instruction</a> </div>
2316 <div class="doc_text">
2325 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
2326 at the first callee in the dynamic call stack which used an <a
2327 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
2328 primarily used to implement exception handling.</p>
2332 <p>The '<tt>unwind</tt>' instruction causes execution of the current function to
2333 immediately halt. The dynamic call stack is then searched for the first <a
2334 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
2335 execution continues at the "exceptional" destination block specified by the
2336 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
2337 dynamic call chain, undefined behavior results.</p>
2340 <!-- _______________________________________________________________________ -->
2342 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
2343 Instruction</a> </div>
2345 <div class="doc_text">
2354 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
2355 instruction is used to inform the optimizer that a particular portion of the
2356 code is not reachable. This can be used to indicate that the code after a
2357 no-return function cannot be reached, and other facts.</p>
2361 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
2366 <!-- ======================================================================= -->
2367 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
2368 <div class="doc_text">
2369 <p>Binary operators are used to do most of the computation in a
2370 program. They require two operands of the same type, execute an operation on them, and
2371 produce a single value. The operands might represent
2372 multiple data, as is the case with the <a href="#t_vector">vector</a> data type.
2373 The result value has the same type as its operands.</p>
2374 <p>There are several different binary operators:</p>
2376 <!-- _______________________________________________________________________ -->
2377 <div class="doc_subsubsection">
2378 <a name="i_add">'<tt>add</tt>' Instruction</a>
2381 <div class="doc_text">
2386 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2391 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
2395 <p>The two arguments to the '<tt>add</tt>' instruction must be <a
2396 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>, or
2397 <a href="#t_vector">vector</a> values. Both arguments must have identical
2402 <p>The value produced is the integer or floating point sum of the two
2405 <p>If an integer sum has unsigned overflow, the result returned is the
2406 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2409 <p>Because LLVM integers use a two's complement representation, this
2410 instruction is appropriate for both signed and unsigned integers.</p>
2415 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
2418 <!-- _______________________________________________________________________ -->
2419 <div class="doc_subsubsection">
2420 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
2423 <div class="doc_text">
2428 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2433 <p>The '<tt>sub</tt>' instruction returns the difference of its two
2436 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
2437 '<tt>neg</tt>' instruction present in most other intermediate
2438 representations.</p>
2442 <p>The two arguments to the '<tt>sub</tt>' instruction must be <a
2443 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>,
2444 or <a href="#t_vector">vector</a> values. Both arguments must have identical
2449 <p>The value produced is the integer or floating point difference of
2450 the two operands.</p>
2452 <p>If an integer difference has unsigned overflow, the result returned is the
2453 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2456 <p>Because LLVM integers use a two's complement representation, this
2457 instruction is appropriate for both signed and unsigned integers.</p>
2461 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
2462 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
2466 <!-- _______________________________________________________________________ -->
2467 <div class="doc_subsubsection">
2468 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
2471 <div class="doc_text">
2474 <pre> <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2477 <p>The '<tt>mul</tt>' instruction returns the product of its two
2482 <p>The two arguments to the '<tt>mul</tt>' instruction must be <a
2483 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>,
2484 or <a href="#t_vector">vector</a> values. Both arguments must have identical
2489 <p>The value produced is the integer or floating point product of the
2492 <p>If the result of an integer multiplication has unsigned overflow,
2493 the result returned is the mathematical result modulo
2494 2<sup>n</sup>, where n is the bit width of the result.</p>
2495 <p>Because LLVM integers use a two's complement representation, and the
2496 result is the same width as the operands, this instruction returns the
2497 correct result for both signed and unsigned integers. If a full product
2498 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands
2499 should be sign-extended or zero-extended as appropriate to the
2500 width of the full product.</p>
2502 <pre> <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
2506 <!-- _______________________________________________________________________ -->
2507 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
2509 <div class="doc_text">
2511 <pre> <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2514 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two
2519 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
2520 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2521 values. Both arguments must have identical types.</p>
2525 <p>The value produced is the unsigned integer quotient of the two operands.</p>
2526 <p>Note that unsigned integer division and signed integer division are distinct
2527 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
2528 <p>Division by zero leads to undefined behavior.</p>
2530 <pre> <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2533 <!-- _______________________________________________________________________ -->
2534 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
2536 <div class="doc_text">
2539 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2544 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
2549 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
2550 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2551 values. Both arguments must have identical types.</p>
2554 <p>The value produced is the signed integer quotient of the two operands rounded towards zero.</p>
2555 <p>Note that signed integer division and unsigned integer division are distinct
2556 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
2557 <p>Division by zero leads to undefined behavior. Overflow also leads to
2558 undefined behavior; this is a rare case, but can occur, for example,
2559 by doing a 32-bit division of -2147483648 by -1.</p>
2561 <pre> <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2564 <!-- _______________________________________________________________________ -->
2565 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
2566 Instruction</a> </div>
2567 <div class="doc_text">
2570 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2574 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
2579 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
2580 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a>
2581 of floating point values. Both arguments must have identical types.</p>
2585 <p>The value produced is the floating point quotient of the two operands.</p>
2590 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
2594 <!-- _______________________________________________________________________ -->
2595 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
2597 <div class="doc_text">
2599 <pre> <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2602 <p>The '<tt>urem</tt>' instruction returns the remainder from the
2603 unsigned division of its two arguments.</p>
2605 <p>The two arguments to the '<tt>urem</tt>' instruction must be
2606 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2607 values. Both arguments must have identical types.</p>
2609 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
2610 This instruction always performs an unsigned division to get the remainder.</p>
2611 <p>Note that unsigned integer remainder and signed integer remainder are
2612 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
2613 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
2615 <pre> <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2619 <!-- _______________________________________________________________________ -->
2620 <div class="doc_subsubsection">
2621 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
2624 <div class="doc_text">
2629 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2634 <p>The '<tt>srem</tt>' instruction returns the remainder from the
2635 signed division of its two operands. This instruction can also take
2636 <a href="#t_vector">vector</a> versions of the values in which case
2637 the elements must be integers.</p>
2641 <p>The two arguments to the '<tt>srem</tt>' instruction must be
2642 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2643 values. Both arguments must have identical types.</p>
2647 <p>This instruction returns the <i>remainder</i> of a division (where the result
2648 has the same sign as the dividend, <tt>op1</tt>), not the <i>modulo</i>
2649 operator (where the result has the same sign as the divisor, <tt>op2</tt>) of
2650 a value. For more information about the difference, see <a
2651 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
2652 Math Forum</a>. For a table of how this is implemented in various languages,
2653 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
2654 Wikipedia: modulo operation</a>.</p>
2655 <p>Note that signed integer remainder and unsigned integer remainder are
2656 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
2657 <p>Taking the remainder of a division by zero leads to undefined behavior.
2658 Overflow also leads to undefined behavior; this is a rare case, but can occur,
2659 for example, by taking the remainder of a 32-bit division of -2147483648 by -1.
2660 (The remainder doesn't actually overflow, but this rule lets srem be
2661 implemented using instructions that return both the result of the division
2662 and the remainder.)</p>
2664 <pre> <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2668 <!-- _______________________________________________________________________ -->
2669 <div class="doc_subsubsection">
2670 <a name="i_frem">'<tt>frem</tt>' Instruction</a> </div>
2672 <div class="doc_text">
2675 <pre> <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2678 <p>The '<tt>frem</tt>' instruction returns the remainder from the
2679 division of its two operands.</p>
2681 <p>The two arguments to the '<tt>frem</tt>' instruction must be
2682 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a>
2683 of floating point values. Both arguments must have identical types.</p>
2687 <p>This instruction returns the <i>remainder</i> of a division.
2688 The remainder has the same sign as the dividend.</p>
2693 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
2697 <!-- ======================================================================= -->
2698 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
2699 Operations</a> </div>
2700 <div class="doc_text">
2701 <p>Bitwise binary operators are used to do various forms of
2702 bit-twiddling in a program. They are generally very efficient
2703 instructions and can commonly be strength reduced from other
2704 instructions. They require two operands of the same type, execute an operation on them,
2705 and produce a single value. The resulting value is the same type as its operands.</p>
2708 <!-- _______________________________________________________________________ -->
2709 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
2710 Instruction</a> </div>
2711 <div class="doc_text">
2713 <pre> <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2718 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2719 the left a specified number of bits.</p>
2723 <p>Both arguments to the '<tt>shl</tt>' instruction must be the same <a
2724 href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2725 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
2729 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod 2<sup>n</sup>,
2730 where n is the width of the result. If <tt>op2</tt> is (statically or dynamically) negative or
2731 equal to or larger than the number of bits in <tt>op1</tt>, the result is undefined.
2732 If the arguments are vectors, each vector element of <tt>op1</tt> is shifted by the
2733 corresponding shift amount in <tt>op2</tt>.</p>
2735 <h5>Example:</h5><pre>
2736 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
2737 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
2738 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
2739 <result> = shl i32 1, 32 <i>; undefined</i>
2740 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
2743 <!-- _______________________________________________________________________ -->
2744 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
2745 Instruction</a> </div>
2746 <div class="doc_text">
2748 <pre> <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2752 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
2753 operand shifted to the right a specified number of bits with zero fill.</p>
2756 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
2757 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2758 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
2762 <p>This instruction always performs a logical shift right operation. The most
2763 significant bits of the result will be filled with zero bits after the
2764 shift. If <tt>op2</tt> is (statically or dynamically) equal to or larger than
2765 the number of bits in <tt>op1</tt>, the result is undefined. If the arguments are
2766 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding shift
2767 amount in <tt>op2</tt>.</p>
2771 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
2772 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
2773 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
2774 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
2775 <result> = lshr i32 1, 32 <i>; undefined</i>
2776 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
2780 <!-- _______________________________________________________________________ -->
2781 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
2782 Instruction</a> </div>
2783 <div class="doc_text">
2786 <pre> <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2790 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
2791 operand shifted to the right a specified number of bits with sign extension.</p>
2794 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
2795 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2796 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
2799 <p>This instruction always performs an arithmetic shift right operation,
2800 The most significant bits of the result will be filled with the sign bit
2801 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
2802 larger than the number of bits in <tt>op1</tt>, the result is undefined. If the
2803 arguments are vectors, each vector element of <tt>op1</tt> is shifted by the
2804 corresponding shift amount in <tt>op2</tt>.</p>
2808 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
2809 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
2810 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
2811 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
2812 <result> = ashr i32 1, 32 <i>; undefined</i>
2813 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
2817 <!-- _______________________________________________________________________ -->
2818 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
2819 Instruction</a> </div>
2821 <div class="doc_text">
2826 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2831 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
2832 its two operands.</p>
2836 <p>The two arguments to the '<tt>and</tt>' instruction must be
2837 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2838 values. Both arguments must have identical types.</p>
2841 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
2844 <table border="1" cellspacing="0" cellpadding="4">
2876 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
2877 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
2878 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
2881 <!-- _______________________________________________________________________ -->
2882 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
2883 <div class="doc_text">
2885 <pre> <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2888 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
2889 or of its two operands.</p>
2892 <p>The two arguments to the '<tt>or</tt>' instruction must be
2893 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2894 values. Both arguments must have identical types.</p>
2896 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
2899 <table border="1" cellspacing="0" cellpadding="4">
2930 <pre> <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
2931 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
2932 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
2935 <!-- _______________________________________________________________________ -->
2936 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
2937 Instruction</a> </div>
2938 <div class="doc_text">
2940 <pre> <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2943 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
2944 or of its two operands. The <tt>xor</tt> is used to implement the
2945 "one's complement" operation, which is the "~" operator in C.</p>
2947 <p>The two arguments to the '<tt>xor</tt>' instruction must be
2948 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2949 values. Both arguments must have identical types.</p>
2953 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
2956 <table border="1" cellspacing="0" cellpadding="4">
2988 <pre> <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
2989 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
2990 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
2991 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
2995 <!-- ======================================================================= -->
2996 <div class="doc_subsection">
2997 <a name="vectorops">Vector Operations</a>
3000 <div class="doc_text">
3002 <p>LLVM supports several instructions to represent vector operations in a
3003 target-independent manner. These instructions cover the element-access and
3004 vector-specific operations needed to process vectors effectively. While LLVM
3005 does directly support these vector operations, many sophisticated algorithms
3006 will want to use target-specific intrinsics to take full advantage of a specific
3011 <!-- _______________________________________________________________________ -->
3012 <div class="doc_subsubsection">
3013 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
3016 <div class="doc_text">
3021 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
3027 The '<tt>extractelement</tt>' instruction extracts a single scalar
3028 element from a vector at a specified index.
3035 The first operand of an '<tt>extractelement</tt>' instruction is a
3036 value of <a href="#t_vector">vector</a> type. The second operand is
3037 an index indicating the position from which to extract the element.
3038 The index may be a variable.</p>
3043 The result is a scalar of the same type as the element type of
3044 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
3045 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
3046 results are undefined.
3052 %result = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
3057 <!-- _______________________________________________________________________ -->
3058 <div class="doc_subsubsection">
3059 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
3062 <div class="doc_text">
3067 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
3073 The '<tt>insertelement</tt>' instruction inserts a scalar
3074 element into a vector at a specified index.
3081 The first operand of an '<tt>insertelement</tt>' instruction is a
3082 value of <a href="#t_vector">vector</a> type. The second operand is a
3083 scalar value whose type must equal the element type of the first
3084 operand. The third operand is an index indicating the position at
3085 which to insert the value. The index may be a variable.</p>
3090 The result is a vector of the same type as <tt>val</tt>. Its
3091 element values are those of <tt>val</tt> except at position
3092 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
3093 exceeds the length of <tt>val</tt>, the results are undefined.
3099 %result = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
3103 <!-- _______________________________________________________________________ -->
3104 <div class="doc_subsubsection">
3105 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
3108 <div class="doc_text">
3113 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
3119 The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
3120 from two input vectors, returning a vector with the same element type as
3121 the input and length that is the same as the shuffle mask.
3127 The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
3128 with types that match each other. The third argument is a shuffle mask whose
3129 element type is always 'i32'. The result of the instruction is a vector whose
3130 length is the same as the shuffle mask and whose element type is the same as
3131 the element type of the first two operands.
3135 The shuffle mask operand is required to be a constant vector with either
3136 constant integer or undef values.
3142 The elements of the two input vectors are numbered from left to right across
3143 both of the vectors. The shuffle mask operand specifies, for each element of
3144 the result vector, which element of the two input vectors the result element
3145 gets. The element selector may be undef (meaning "don't care") and the second
3146 operand may be undef if performing a shuffle from only one vector.
3152 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
3153 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
3154 %result = shufflevector <4 x i32> %v1, <4 x i32> undef,
3155 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
3156 %result = shufflevector <8 x i32> %v1, <8 x i32> undef,
3157 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i>
3158 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
3159 <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7 > <i>; yields <8 x i32></i>
3164 <!-- ======================================================================= -->
3165 <div class="doc_subsection">
3166 <a name="aggregateops">Aggregate Operations</a>
3169 <div class="doc_text">
3171 <p>LLVM supports several instructions for working with aggregate values.
3176 <!-- _______________________________________________________________________ -->
3177 <div class="doc_subsubsection">
3178 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
3181 <div class="doc_text">
3186 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
3192 The '<tt>extractvalue</tt>' instruction extracts the value of a struct field
3193 or array element from an aggregate value.
3200 The first operand of an '<tt>extractvalue</tt>' instruction is a
3201 value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a>
3202 type. The operands are constant indices to specify which value to extract
3203 in a similar manner as indices in a
3204 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
3210 The result is the value at the position in the aggregate specified by
3217 %result = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
3222 <!-- _______________________________________________________________________ -->
3223 <div class="doc_subsubsection">
3224 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
3227 <div class="doc_text">
3232 <result> = insertvalue <aggregate type> <val>, <ty> <val>, <idx> <i>; yields <n x <ty>></i>
3238 The '<tt>insertvalue</tt>' instruction inserts a value
3239 into a struct field or array element in an aggregate.
3246 The first operand of an '<tt>insertvalue</tt>' instruction is a
3247 value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a> type.
3248 The second operand is a first-class value to insert.
3249 The following operands are constant indices
3250 indicating the position at which to insert the value in a similar manner as
3252 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
3253 The value to insert must have the same type as the value identified
3260 The result is an aggregate of the same type as <tt>val</tt>. Its
3261 value is that of <tt>val</tt> except that the value at the position
3262 specified by the indices is that of <tt>elt</tt>.
3268 %result = insertvalue {i32, float} %agg, i32 1, 0 <i>; yields {i32, float}</i>
3273 <!-- ======================================================================= -->
3274 <div class="doc_subsection">
3275 <a name="memoryops">Memory Access and Addressing Operations</a>
3278 <div class="doc_text">
3280 <p>A key design point of an SSA-based representation is how it
3281 represents memory. In LLVM, no memory locations are in SSA form, which
3282 makes things very simple. This section describes how to read, write,
3283 allocate, and free memory in LLVM.</p>
3287 <!-- _______________________________________________________________________ -->
3288 <div class="doc_subsubsection">
3289 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
3292 <div class="doc_text">
3297 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3302 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
3303 heap and returns a pointer to it. The object is always allocated in the generic
3304 address space (address space zero).</p>
3308 <p>The '<tt>malloc</tt>' instruction allocates
3309 <tt>sizeof(<type>)*NumElements</tt>
3310 bytes of memory from the operating system and returns a pointer of the
3311 appropriate type to the program. If "NumElements" is specified, it is the
3312 number of elements allocated, otherwise "NumElements" is defaulted to be one.
3313 If a constant alignment is specified, the value result of the allocation is guaranteed to
3314 be aligned to at least that boundary. If not specified, or if zero, the target can
3315 choose to align the allocation on any convenient boundary.</p>
3317 <p>'<tt>type</tt>' must be a sized type.</p>
3321 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
3322 a pointer is returned. The result of a zero byte allocation is undefined. The
3323 result is null if there is insufficient memory available.</p>
3328 %array = malloc [4 x i8] <i>; yields {[%4 x i8]*}:array</i>
3330 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
3331 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
3332 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
3333 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
3334 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
3337 <p>Note that the code generator does not yet respect the
3338 alignment value.</p>
3342 <!-- _______________________________________________________________________ -->
3343 <div class="doc_subsubsection">
3344 <a name="i_free">'<tt>free</tt>' Instruction</a>
3347 <div class="doc_text">
3352 free <type> <value> <i>; yields {void}</i>
3357 <p>The '<tt>free</tt>' instruction returns memory back to the unused
3358 memory heap to be reallocated in the future.</p>
3362 <p>'<tt>value</tt>' shall be a pointer value that points to a value
3363 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
3368 <p>Access to the memory pointed to by the pointer is no longer defined
3369 after this instruction executes. If the pointer is null, the operation
3375 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
3376 free [4 x i8]* %array
3380 <!-- _______________________________________________________________________ -->
3381 <div class="doc_subsubsection">
3382 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
3385 <div class="doc_text">
3390 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3395 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
3396 currently executing function, to be automatically released when this function
3397 returns to its caller. The object is always allocated in the generic address
3398 space (address space zero).</p>
3402 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
3403 bytes of memory on the runtime stack, returning a pointer of the
3404 appropriate type to the program. If "NumElements" is specified, it is the
3405 number of elements allocated, otherwise "NumElements" is defaulted to be one.
3406 If a constant alignment is specified, the value result of the allocation is guaranteed
3407 to be aligned to at least that boundary. If not specified, or if zero, the target
3408 can choose to align the allocation on any convenient boundary.</p>
3410 <p>'<tt>type</tt>' may be any sized type.</p>
3414 <p>Memory is allocated; a pointer is returned. The operation is undefiend if
3415 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
3416 memory is automatically released when the function returns. The '<tt>alloca</tt>'
3417 instruction is commonly used to represent automatic variables that must
3418 have an address available. When the function returns (either with the <tt><a
3419 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
3420 instructions), the memory is reclaimed. Allocating zero bytes
3421 is legal, but the result is undefined.</p>
3426 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
3427 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
3428 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
3429 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
3433 <!-- _______________________________________________________________________ -->
3434 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
3435 Instruction</a> </div>
3436 <div class="doc_text">
3438 <pre> <result> = load <ty>* <pointer>[, align <alignment>]<br> <result> = volatile load <ty>* <pointer>[, align <alignment>]<br></pre>
3440 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
3442 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
3443 address from which to load. The pointer must point to a <a
3444 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
3445 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
3446 the number or order of execution of this <tt>load</tt> with other
3447 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
3450 The optional constant "align" argument specifies the alignment of the operation
3451 (that is, the alignment of the memory address). A value of 0 or an
3452 omitted "align" argument means that the operation has the preferential
3453 alignment for the target. It is the responsibility of the code emitter
3454 to ensure that the alignment information is correct. Overestimating
3455 the alignment results in an undefined behavior. Underestimating the
3456 alignment may produce less efficient code. An alignment of 1 is always
3460 <p>The location of memory pointed to is loaded.</p>
3462 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3464 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
3465 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
3468 <!-- _______________________________________________________________________ -->
3469 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
3470 Instruction</a> </div>
3471 <div class="doc_text">
3473 <pre> store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3474 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3477 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
3479 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
3480 to store and an address at which to store it. The type of the '<tt><pointer></tt>'
3481 operand must be a pointer to the <a href="#t_firstclass">first class</a> type
3482 of the '<tt><value></tt>'
3483 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
3484 optimizer is not allowed to modify the number or order of execution of
3485 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
3486 href="#i_store">store</a></tt> instructions.</p>
3488 The optional constant "align" argument specifies the alignment of the operation
3489 (that is, the alignment of the memory address). A value of 0 or an
3490 omitted "align" argument means that the operation has the preferential
3491 alignment for the target. It is the responsibility of the code emitter
3492 to ensure that the alignment information is correct. Overestimating
3493 the alignment results in an undefined behavior. Underestimating the
3494 alignment may produce less efficient code. An alignment of 1 is always
3498 <p>The contents of memory are updated to contain '<tt><value></tt>'
3499 at the location specified by the '<tt><pointer></tt>' operand.</p>
3501 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3502 store i32 3, i32* %ptr <i>; yields {void}</i>
3503 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
3507 <!-- _______________________________________________________________________ -->
3508 <div class="doc_subsubsection">
3509 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
3512 <div class="doc_text">
3515 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
3521 The '<tt>getelementptr</tt>' instruction is used to get the address of a
3522 subelement of an aggregate data structure. It performs address calculation only
3523 and does not access memory.</p>
3527 <p>The first argument is always a pointer, and forms the basis of the
3528 calculation. The remaining arguments are indices, that indicate which of the
3529 elements of the aggregate object are indexed. The interpretation of each index
3530 is dependent on the type being indexed into. The first index always indexes the
3531 pointer value given as the first argument, the second index indexes a value of
3532 the type pointed to (not necessarily the value directly pointed to, since the
3533 first index can be non-zero), etc. The first type indexed into must be a pointer
3534 value, subsequent types can be arrays, vectors and structs. Note that subsequent
3535 types being indexed into can never be pointers, since that would require loading
3536 the pointer before continuing calculation.</p>
3538 <p>The type of each index argument depends on the type it is indexing into.
3539 When indexing into a (packed) structure, only <tt>i32</tt> integer
3540 <b>constants</b> are allowed. When indexing into an array, pointer or vector,
3541 only integers of 32 or 64 bits are allowed (also non-constants). 32-bit values
3542 will be sign extended to 64-bits if required.</p>
3544 <p>For example, let's consider a C code fragment and how it gets
3545 compiled to LLVM:</p>
3547 <div class="doc_code">
3560 int *foo(struct ST *s) {
3561 return &s[1].Z.B[5][13];
3566 <p>The LLVM code generated by the GCC frontend is:</p>
3568 <div class="doc_code">
3570 %RT = <a href="#namedtypes">type</a> { i8 , [10 x [20 x i32]], i8 }
3571 %ST = <a href="#namedtypes">type</a> { i32, double, %RT }
3573 define i32* %foo(%ST* %s) {
3575 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
3583 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
3584 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
3585 }</tt>' type, a structure. The second index indexes into the third element of
3586 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
3587 i8 }</tt>' type, another structure. The third index indexes into the second
3588 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
3589 array. The two dimensions of the array are subscripted into, yielding an
3590 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
3591 to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
3593 <p>Note that it is perfectly legal to index partially through a
3594 structure, returning a pointer to an inner element. Because of this,
3595 the LLVM code for the given testcase is equivalent to:</p>
3598 define i32* %foo(%ST* %s) {
3599 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
3600 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
3601 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
3602 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
3603 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
3608 <p>Note that it is undefined to access an array out of bounds: array and
3609 pointer indexes must always be within the defined bounds of the array type.
3610 The one exception for this rule is zero length arrays. These arrays are
3611 defined to be accessible as variable length arrays, which requires access
3612 beyond the zero'th element.</p>
3614 <p>The getelementptr instruction is often confusing. For some more insight
3615 into how it works, see <a href="GetElementPtr.html">the getelementptr
3621 <i>; yields [12 x i8]*:aptr</i>
3622 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
3623 <i>; yields i8*:vptr</i>
3624 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
3625 <i>; yields i8*:eptr</i>
3626 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
3630 <!-- ======================================================================= -->
3631 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
3633 <div class="doc_text">
3634 <p>The instructions in this category are the conversion instructions (casting)
3635 which all take a single operand and a type. They perform various bit conversions
3639 <!-- _______________________________________________________________________ -->
3640 <div class="doc_subsubsection">
3641 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
3643 <div class="doc_text">
3647 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
3652 The '<tt>trunc</tt>' instruction truncates its operand to the type <tt>ty2</tt>.
3657 The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
3658 be an <a href="#t_integer">integer</a> type, and a type that specifies the size
3659 and type of the result, which must be an <a href="#t_integer">integer</a>
3660 type. The bit size of <tt>value</tt> must be larger than the bit size of
3661 <tt>ty2</tt>. Equal sized types are not allowed.</p>
3665 The '<tt>trunc</tt>' instruction truncates the high order bits in <tt>value</tt>
3666 and converts the remaining bits to <tt>ty2</tt>. Since the source size must be
3667 larger than the destination size, <tt>trunc</tt> cannot be a <i>no-op cast</i>.
3668 It will always truncate bits.</p>
3672 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
3673 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
3674 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
3678 <!-- _______________________________________________________________________ -->
3679 <div class="doc_subsubsection">
3680 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
3682 <div class="doc_text">
3686 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
3690 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
3695 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
3696 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3697 also be of <a href="#t_integer">integer</a> type. The bit size of the
3698 <tt>value</tt> must be smaller than the bit size of the destination type,
3702 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
3703 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
3705 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
3709 %X = zext i32 257 to i64 <i>; yields i64:257</i>
3710 %Y = zext i1 true to i32 <i>; yields i32:1</i>
3714 <!-- _______________________________________________________________________ -->
3715 <div class="doc_subsubsection">
3716 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
3718 <div class="doc_text">
3722 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
3726 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
3730 The '<tt>sext</tt>' instruction takes a value to cast, which must be of
3731 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3732 also be of <a href="#t_integer">integer</a> type. The bit size of the
3733 <tt>value</tt> must be smaller than the bit size of the destination type,
3738 The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
3739 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size of
3740 the type <tt>ty2</tt>.</p>
3742 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
3746 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
3747 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
3751 <!-- _______________________________________________________________________ -->
3752 <div class="doc_subsubsection">
3753 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
3756 <div class="doc_text">
3761 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
3765 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
3770 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
3771 point</a> value to cast and a <a href="#t_floating">floating point</a> type to
3772 cast it to. The size of <tt>value</tt> must be larger than the size of
3773 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
3774 <i>no-op cast</i>.</p>
3777 <p> The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
3778 <a href="#t_floating">floating point</a> type to a smaller
3779 <a href="#t_floating">floating point</a> type. If the value cannot fit within
3780 the destination type, <tt>ty2</tt>, then the results are undefined.</p>
3784 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
3785 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
3789 <!-- _______________________________________________________________________ -->
3790 <div class="doc_subsubsection">
3791 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
3793 <div class="doc_text">
3797 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
3801 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
3802 floating point value.</p>
3805 <p>The '<tt>fpext</tt>' instruction takes a
3806 <a href="#t_floating">floating point</a> <tt>value</tt> to cast,
3807 and a <a href="#t_floating">floating point</a> type to cast it to. The source
3808 type must be smaller than the destination type.</p>
3811 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
3812 <a href="#t_floating">floating point</a> type to a larger
3813 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
3814 used to make a <i>no-op cast</i> because it always changes bits. Use
3815 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
3819 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
3820 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
3824 <!-- _______________________________________________________________________ -->
3825 <div class="doc_subsubsection">
3826 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
3828 <div class="doc_text">
3832 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
3836 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
3837 unsigned integer equivalent of type <tt>ty2</tt>.
3841 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
3842 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3843 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3844 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3845 vector integer type with the same number of elements as <tt>ty</tt></p>
3848 <p> The '<tt>fptoui</tt>' instruction converts its
3849 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3850 towards zero) unsigned integer value. If the value cannot fit in <tt>ty2</tt>,
3851 the results are undefined.</p>
3855 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
3856 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
3857 %X = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
3861 <!-- _______________________________________________________________________ -->
3862 <div class="doc_subsubsection">
3863 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
3865 <div class="doc_text">
3869 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
3873 <p>The '<tt>fptosi</tt>' instruction converts
3874 <a href="#t_floating">floating point</a> <tt>value</tt> to type <tt>ty2</tt>.
3878 <p> The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
3879 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3880 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3881 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3882 vector integer type with the same number of elements as <tt>ty</tt></p>
3885 <p>The '<tt>fptosi</tt>' instruction converts its
3886 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3887 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
3888 the results are undefined.</p>
3892 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
3893 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
3894 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
3898 <!-- _______________________________________________________________________ -->
3899 <div class="doc_subsubsection">
3900 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
3902 <div class="doc_text">
3906 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3910 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
3911 integer and converts that value to the <tt>ty2</tt> type.</p>
3914 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
3915 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3916 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3917 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3918 floating point type with the same number of elements as <tt>ty</tt></p>
3921 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
3922 integer quantity and converts it to the corresponding floating point value. If
3923 the value cannot fit in the floating point value, the results are undefined.</p>
3927 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
3928 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
3932 <!-- _______________________________________________________________________ -->
3933 <div class="doc_subsubsection">
3934 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
3936 <div class="doc_text">
3940 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3944 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed
3945 integer and converts that value to the <tt>ty2</tt> type.</p>
3948 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
3949 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3950 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3951 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3952 floating point type with the same number of elements as <tt>ty</tt></p>
3955 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed
3956 integer quantity and converts it to the corresponding floating point value. If
3957 the value cannot fit in the floating point value, the results are undefined.</p>
3961 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
3962 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
3966 <!-- _______________________________________________________________________ -->
3967 <div class="doc_subsubsection">
3968 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
3970 <div class="doc_text">
3974 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
3978 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
3979 the integer type <tt>ty2</tt>.</p>
3982 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
3983 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
3984 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.</p>
3987 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
3988 <tt>ty2</tt> by interpreting the pointer value as an integer and either
3989 truncating or zero extending that value to the size of the integer type. If
3990 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
3991 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
3992 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
3997 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
3998 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
4002 <!-- _______________________________________________________________________ -->
4003 <div class="doc_subsubsection">
4004 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
4006 <div class="doc_text">
4010 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
4014 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to
4015 a pointer type, <tt>ty2</tt>.</p>
4018 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
4019 value to cast, and a type to cast it to, which must be a
4020 <a href="#t_pointer">pointer</a> type.</p>
4023 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
4024 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
4025 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
4026 size of a pointer then a truncation is done. If <tt>value</tt> is smaller than
4027 the size of a pointer then a zero extension is done. If they are the same size,
4028 nothing is done (<i>no-op cast</i>).</p>
4032 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
4033 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
4034 %Y = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
4038 <!-- _______________________________________________________________________ -->
4039 <div class="doc_subsubsection">
4040 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
4042 <div class="doc_text">
4046 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
4051 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
4052 <tt>ty2</tt> without changing any bits.</p>
4056 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be
4057 a non-aggregate first class value, and a type to cast it to, which must also be
4058 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes of
4060 and the destination type, <tt>ty2</tt>, must be identical. If the source
4061 type is a pointer, the destination type must also be a pointer. This
4062 instruction supports bitwise conversion of vectors to integers and to vectors
4063 of other types (as long as they have the same size).</p>
4066 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
4067 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
4068 this conversion. The conversion is done as if the <tt>value</tt> had been
4069 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only be
4070 converted to other pointer types with this instruction. To convert pointers to
4071 other types, use the <a href="#i_inttoptr">inttoptr</a> or
4072 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
4076 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
4077 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
4078 %Z = bitcast <2 x int> %V to i64; <i>; yields i64: %V</i>
4082 <!-- ======================================================================= -->
4083 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
4084 <div class="doc_text">
4085 <p>The instructions in this category are the "miscellaneous"
4086 instructions, which defy better classification.</p>
4089 <!-- _______________________________________________________________________ -->
4090 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
4092 <div class="doc_text">
4094 <pre> <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4097 <p>The '<tt>icmp</tt>' instruction returns a boolean value or
4098 a vector of boolean values based on comparison
4099 of its two integer, integer vector, or pointer operands.</p>
4101 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
4102 the condition code indicating the kind of comparison to perform. It is not
4103 a value, just a keyword. The possible condition code are:
4106 <li><tt>eq</tt>: equal</li>
4107 <li><tt>ne</tt>: not equal </li>
4108 <li><tt>ugt</tt>: unsigned greater than</li>
4109 <li><tt>uge</tt>: unsigned greater or equal</li>
4110 <li><tt>ult</tt>: unsigned less than</li>
4111 <li><tt>ule</tt>: unsigned less or equal</li>
4112 <li><tt>sgt</tt>: signed greater than</li>
4113 <li><tt>sge</tt>: signed greater or equal</li>
4114 <li><tt>slt</tt>: signed less than</li>
4115 <li><tt>sle</tt>: signed less or equal</li>
4117 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
4118 <a href="#t_pointer">pointer</a>
4119 or integer <a href="#t_vector">vector</a> typed.
4120 They must also be identical types.</p>
4122 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to
4123 the condition code given as <tt>cond</tt>. The comparison performed always
4124 yields either an <a href="#t_primitive"><tt>i1</tt></a> or vector of <tt>i1</tt> result, as follows:
4127 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
4128 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
4130 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
4131 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.</li>
4132 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
4133 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4134 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
4135 <tt>true</tt> if <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4136 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
4137 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
4138 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
4139 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4140 <li><tt>sgt</tt>: interprets the operands as signed values and yields
4141 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4142 <li><tt>sge</tt>: interprets the operands as signed values and yields
4143 <tt>true</tt> if <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4144 <li><tt>slt</tt>: interprets the operands as signed values and yields
4145 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
4146 <li><tt>sle</tt>: interprets the operands as signed values and yields
4147 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4149 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
4150 values are compared as if they were integers.</p>
4151 <p>If the operands are integer vectors, then they are compared
4152 element by element. The result is an <tt>i1</tt> vector with
4153 the same number of elements as the values being compared.
4154 Otherwise, the result is an <tt>i1</tt>.
4158 <pre> <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
4159 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
4160 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
4161 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
4162 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
4163 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
4166 <p>Note that the code generator does not yet support vector types with
4167 the <tt>icmp</tt> instruction.</p>
4171 <!-- _______________________________________________________________________ -->
4172 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
4174 <div class="doc_text">
4176 <pre> <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4179 <p>The '<tt>fcmp</tt>' instruction returns a boolean value
4180 or vector of boolean values based on comparison
4181 of its operands.</p>
4183 If the operands are floating point scalars, then the result
4184 type is a boolean (<a href="#t_primitive"><tt>i1</tt></a>).
4186 <p>If the operands are floating point vectors, then the result type
4187 is a vector of boolean with the same number of elements as the
4188 operands being compared.</p>
4190 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
4191 the condition code indicating the kind of comparison to perform. It is not
4192 a value, just a keyword. The possible condition code are:</p>
4194 <li><tt>false</tt>: no comparison, always returns false</li>
4195 <li><tt>oeq</tt>: ordered and equal</li>
4196 <li><tt>ogt</tt>: ordered and greater than </li>
4197 <li><tt>oge</tt>: ordered and greater than or equal</li>
4198 <li><tt>olt</tt>: ordered and less than </li>
4199 <li><tt>ole</tt>: ordered and less than or equal</li>
4200 <li><tt>one</tt>: ordered and not equal</li>
4201 <li><tt>ord</tt>: ordered (no nans)</li>
4202 <li><tt>ueq</tt>: unordered or equal</li>
4203 <li><tt>ugt</tt>: unordered or greater than </li>
4204 <li><tt>uge</tt>: unordered or greater than or equal</li>
4205 <li><tt>ult</tt>: unordered or less than </li>
4206 <li><tt>ule</tt>: unordered or less than or equal</li>
4207 <li><tt>une</tt>: unordered or not equal</li>
4208 <li><tt>uno</tt>: unordered (either nans)</li>
4209 <li><tt>true</tt>: no comparison, always returns true</li>
4211 <p><i>Ordered</i> means that neither operand is a QNAN while
4212 <i>unordered</i> means that either operand may be a QNAN.</p>
4213 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be
4214 either a <a href="#t_floating">floating point</a> type
4215 or a <a href="#t_vector">vector</a> of floating point type.
4216 They must have identical types.</p>
4218 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4219 according to the condition code given as <tt>cond</tt>.
4220 If the operands are vectors, then the vectors are compared
4222 Each comparison performed
4223 always yields an <a href="#t_primitive">i1</a> result, as follows:</p>
4225 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
4226 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4227 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
4228 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4229 <tt>op1</tt> is greather than <tt>op2</tt>.</li>
4230 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4231 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4232 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4233 <tt>op1</tt> is less than <tt>op2</tt>.</li>
4234 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4235 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4236 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4237 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
4238 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
4239 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
4240 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
4241 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
4242 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4243 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
4244 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4245 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
4246 <tt>op1</tt> is less than <tt>op2</tt>.</li>
4247 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
4248 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4249 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
4250 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
4251 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
4252 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
4256 <pre> <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
4257 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
4258 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
4259 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
4262 <p>Note that the code generator does not yet support vector types with
4263 the <tt>fcmp</tt> instruction.</p>
4267 <!-- _______________________________________________________________________ -->
4268 <div class="doc_subsubsection">
4269 <a name="i_vicmp">'<tt>vicmp</tt>' Instruction</a>
4271 <div class="doc_text">
4273 <pre> <result> = vicmp <cond> <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4276 <p>The '<tt>vicmp</tt>' instruction returns an integer vector value based on
4277 element-wise comparison of its two integer vector operands.</p>
4279 <p>The '<tt>vicmp</tt>' instruction takes three operands. The first operand is
4280 the condition code indicating the kind of comparison to perform. It is not
4281 a value, just a keyword. The possible condition code are:</p>
4283 <li><tt>eq</tt>: equal</li>
4284 <li><tt>ne</tt>: not equal </li>
4285 <li><tt>ugt</tt>: unsigned greater than</li>
4286 <li><tt>uge</tt>: unsigned greater or equal</li>
4287 <li><tt>ult</tt>: unsigned less than</li>
4288 <li><tt>ule</tt>: unsigned less or equal</li>
4289 <li><tt>sgt</tt>: signed greater than</li>
4290 <li><tt>sge</tt>: signed greater or equal</li>
4291 <li><tt>slt</tt>: signed less than</li>
4292 <li><tt>sle</tt>: signed less or equal</li>
4294 <p>The remaining two arguments must be <a href="#t_vector">vector</a> or
4295 <a href="#t_integer">integer</a> typed. They must also be identical types.</p>
4297 <p>The '<tt>vicmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4298 according to the condition code given as <tt>cond</tt>. The comparison yields a
4299 <a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, of
4300 identical type as the values being compared. The most significant bit in each
4301 element is 1 if the element-wise comparison evaluates to true, and is 0
4302 otherwise. All other bits of the result are undefined. The condition codes
4303 are evaluated identically to the <a href="#i_icmp">'<tt>icmp</tt>'
4304 instruction</a>.</p>
4308 <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>
4309 <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>
4313 <!-- _______________________________________________________________________ -->
4314 <div class="doc_subsubsection">
4315 <a name="i_vfcmp">'<tt>vfcmp</tt>' Instruction</a>
4317 <div class="doc_text">
4319 <pre> <result> = vfcmp <cond> <ty> <op1>, <op2></pre>
4321 <p>The '<tt>vfcmp</tt>' instruction returns an integer vector value based on
4322 element-wise comparison of its two floating point vector operands. The output
4323 elements have the same width as the input elements.</p>
4325 <p>The '<tt>vfcmp</tt>' instruction takes three operands. The first operand is
4326 the condition code indicating the kind of comparison to perform. It is not
4327 a value, just a keyword. The possible condition code are:</p>
4329 <li><tt>false</tt>: no comparison, always returns false</li>
4330 <li><tt>oeq</tt>: ordered and equal</li>
4331 <li><tt>ogt</tt>: ordered and greater than </li>
4332 <li><tt>oge</tt>: ordered and greater than or equal</li>
4333 <li><tt>olt</tt>: ordered and less than </li>
4334 <li><tt>ole</tt>: ordered and less than or equal</li>
4335 <li><tt>one</tt>: ordered and not equal</li>
4336 <li><tt>ord</tt>: ordered (no nans)</li>
4337 <li><tt>ueq</tt>: unordered or equal</li>
4338 <li><tt>ugt</tt>: unordered or greater than </li>
4339 <li><tt>uge</tt>: unordered or greater than or equal</li>
4340 <li><tt>ult</tt>: unordered or less than </li>
4341 <li><tt>ule</tt>: unordered or less than or equal</li>
4342 <li><tt>une</tt>: unordered or not equal</li>
4343 <li><tt>uno</tt>: unordered (either nans)</li>
4344 <li><tt>true</tt>: no comparison, always returns true</li>
4346 <p>The remaining two arguments must be <a href="#t_vector">vector</a> of
4347 <a href="#t_floating">floating point</a> typed. They must also be identical
4350 <p>The '<tt>vfcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4351 according to the condition code given as <tt>cond</tt>. The comparison yields a
4352 <a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, with
4353 an identical number of elements as the values being compared, and each element
4354 having identical with to the width of the floating point elements. The most
4355 significant bit in each element is 1 if the element-wise comparison evaluates to
4356 true, and is 0 otherwise. All other bits of the result are undefined. The
4357 condition codes are evaluated identically to the
4358 <a href="#i_fcmp">'<tt>fcmp</tt>' instruction</a>.</p>
4362 <i>; yields: result=<2 x i32> < i32 0, i32 -1 ></i>
4363 <result> = vfcmp oeq <2 x float> < float 4, float 0 >, < float 5, float 0 >
4365 <i>; yields: result=<2 x i64> < i64 -1, i64 0 ></i>
4366 <result> = vfcmp ult <2 x double> < double 1, double 2 >, < double 2, double 2>
4370 <!-- _______________________________________________________________________ -->
4371 <div class="doc_subsubsection">
4372 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
4375 <div class="doc_text">
4379 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
4381 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
4382 the SSA graph representing the function.</p>
4385 <p>The type of the incoming values is specified with the first type
4386 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
4387 as arguments, with one pair for each predecessor basic block of the
4388 current block. Only values of <a href="#t_firstclass">first class</a>
4389 type may be used as the value arguments to the PHI node. Only labels
4390 may be used as the label arguments.</p>
4392 <p>There must be no non-phi instructions between the start of a basic
4393 block and the PHI instructions: i.e. PHI instructions must be first in
4398 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
4399 specified by the pair corresponding to the predecessor basic block that executed
4400 just prior to the current block.</p>
4404 Loop: ; Infinite loop that counts from 0 on up...
4405 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
4406 %nextindvar = add i32 %indvar, 1
4411 <!-- _______________________________________________________________________ -->
4412 <div class="doc_subsubsection">
4413 <a name="i_select">'<tt>select</tt>' Instruction</a>
4416 <div class="doc_text">
4421 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
4423 <i>selty</i> is either i1 or {<N x i1>}
4429 The '<tt>select</tt>' instruction is used to choose one value based on a
4430 condition, without branching.
4437 The '<tt>select</tt>' instruction requires an 'i1' value or
4438 a vector of 'i1' values indicating the
4439 condition, and two values of the same <a href="#t_firstclass">first class</a>
4440 type. If the val1/val2 are vectors and
4441 the condition is a scalar, then entire vectors are selected, not
4442 individual elements.
4448 If the condition is an i1 and it evaluates to 1, the instruction returns the first
4449 value argument; otherwise, it returns the second value argument.
4452 If the condition is a vector of i1, then the value arguments must
4453 be vectors of the same size, and the selection is done element
4460 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
4463 <p>Note that the code generator does not yet support conditions
4464 with vector type.</p>
4469 <!-- _______________________________________________________________________ -->
4470 <div class="doc_subsubsection">
4471 <a name="i_call">'<tt>call</tt>' Instruction</a>
4474 <div class="doc_text">
4478 <result> = [tail] call [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>] <ty> [<fnty>*] <fnptrval>(<function args>) [<a href="#fnattrs">fn attrs</a>]
4483 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
4487 <p>This instruction requires several arguments:</p>
4491 <p>The optional "tail" marker indicates whether the callee function accesses
4492 any allocas or varargs in the caller. If the "tail" marker is present, the
4493 function call is eligible for tail call optimization. Note that calls may
4494 be marked "tail" even if they do not occur before a <a
4495 href="#i_ret"><tt>ret</tt></a> instruction.</p>
4498 <p>The optional "cconv" marker indicates which <a href="#callingconv">calling
4499 convention</a> the call should use. If none is specified, the call defaults
4500 to using C calling conventions.</p>
4504 <p>The optional <a href="#paramattrs">Parameter Attributes</a> list for
4505 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>',
4506 and '<tt>inreg</tt>' attributes are valid here.</p>
4510 <p>'<tt>ty</tt>': the type of the call instruction itself which is also
4511 the type of the return value. Functions that return no value are marked
4512 <tt><a href="#t_void">void</a></tt>.</p>
4515 <p>'<tt>fnty</tt>': shall be the signature of the pointer to function
4516 value being invoked. The argument types must match the types implied by
4517 this signature. This type can be omitted if the function is not varargs
4518 and if the function type does not return a pointer to a function.</p>
4521 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
4522 be invoked. In most cases, this is a direct function invocation, but
4523 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
4524 to function value.</p>
4527 <p>'<tt>function args</tt>': argument list whose types match the
4528 function signature argument types. All arguments must be of
4529 <a href="#t_firstclass">first class</a> type. If the function signature
4530 indicates the function accepts a variable number of arguments, the extra
4531 arguments can be specified.</p>
4534 <p>The optional <a href="#fnattrs">function attributes</a> list. Only
4535 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
4536 '<tt>readnone</tt>' attributes are valid here.</p>
4542 <p>The '<tt>call</tt>' instruction is used to cause control flow to
4543 transfer to a specified function, with its incoming arguments bound to
4544 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
4545 instruction in the called function, control flow continues with the
4546 instruction after the function call, and the return value of the
4547 function is bound to the result argument.</p>
4552 %retval = call i32 @test(i32 %argc)
4553 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42) <i>; yields i32</i>
4554 %X = tail call i32 @foo() <i>; yields i32</i>
4555 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
4556 call void %foo(i8 97 signext)
4558 %struct.A = type { i32, i8 }
4559 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
4560 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
4561 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
4562 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
4563 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
4568 <!-- _______________________________________________________________________ -->
4569 <div class="doc_subsubsection">
4570 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
4573 <div class="doc_text">
4578 <resultval> = va_arg <va_list*> <arglist>, <argty>
4583 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
4584 the "variable argument" area of a function call. It is used to implement the
4585 <tt>va_arg</tt> macro in C.</p>
4589 <p>This instruction takes a <tt>va_list*</tt> value and the type of
4590 the argument. It returns a value of the specified argument type and
4591 increments the <tt>va_list</tt> to point to the next argument. The
4592 actual type of <tt>va_list</tt> is target specific.</p>
4596 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
4597 type from the specified <tt>va_list</tt> and causes the
4598 <tt>va_list</tt> to point to the next argument. For more information,
4599 see the variable argument handling <a href="#int_varargs">Intrinsic
4602 <p>It is legal for this instruction to be called in a function which does not
4603 take a variable number of arguments, for example, the <tt>vfprintf</tt>
4606 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
4607 href="#intrinsics">intrinsic function</a> because it takes a type as an
4612 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
4614 <p>Note that the code generator does not yet fully support va_arg
4615 on many targets. Also, it does not currently support va_arg with
4616 aggregate types on any target.</p>
4620 <!-- *********************************************************************** -->
4621 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
4622 <!-- *********************************************************************** -->
4624 <div class="doc_text">
4626 <p>LLVM supports the notion of an "intrinsic function". These functions have
4627 well known names and semantics and are required to follow certain restrictions.
4628 Overall, these intrinsics represent an extension mechanism for the LLVM
4629 language that does not require changing all of the transformations in LLVM when
4630 adding to the language (or the bitcode reader/writer, the parser, etc...).</p>
4632 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
4633 prefix is reserved in LLVM for intrinsic names; thus, function names may not
4634 begin with this prefix. Intrinsic functions must always be external functions:
4635 you cannot define the body of intrinsic functions. Intrinsic functions may
4636 only be used in call or invoke instructions: it is illegal to take the address
4637 of an intrinsic function. Additionally, because intrinsic functions are part
4638 of the LLVM language, it is required if any are added that they be documented
4641 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
4642 a family of functions that perform the same operation but on different data
4643 types. Because LLVM can represent over 8 million different integer types,
4644 overloading is used commonly to allow an intrinsic function to operate on any
4645 integer type. One or more of the argument types or the result type can be
4646 overloaded to accept any integer type. Argument types may also be defined as
4647 exactly matching a previous argument's type or the result type. This allows an
4648 intrinsic function which accepts multiple arguments, but needs all of them to
4649 be of the same type, to only be overloaded with respect to a single argument or
4652 <p>Overloaded intrinsics will have the names of its overloaded argument types
4653 encoded into its function name, each preceded by a period. Only those types
4654 which are overloaded result in a name suffix. Arguments whose type is matched
4655 against another type do not. For example, the <tt>llvm.ctpop</tt> function can
4656 take an integer of any width and returns an integer of exactly the same integer
4657 width. This leads to a family of functions such as
4658 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29 %val)</tt>.
4659 Only one type, the return type, is overloaded, and only one type suffix is
4660 required. Because the argument's type is matched against the return type, it
4661 does not require its own name suffix.</p>
4663 <p>To learn how to add an intrinsic function, please see the
4664 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.
4669 <!-- ======================================================================= -->
4670 <div class="doc_subsection">
4671 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
4674 <div class="doc_text">
4676 <p>Variable argument support is defined in LLVM with the <a
4677 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
4678 intrinsic functions. These functions are related to the similarly
4679 named macros defined in the <tt><stdarg.h></tt> header file.</p>
4681 <p>All of these functions operate on arguments that use a
4682 target-specific value type "<tt>va_list</tt>". The LLVM assembly
4683 language reference manual does not define what this type is, so all
4684 transformations should be prepared to handle these functions regardless of
4687 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
4688 instruction and the variable argument handling intrinsic functions are
4691 <div class="doc_code">
4693 define i32 @test(i32 %X, ...) {
4694 ; Initialize variable argument processing
4696 %ap2 = bitcast i8** %ap to i8*
4697 call void @llvm.va_start(i8* %ap2)
4699 ; Read a single integer argument
4700 %tmp = va_arg i8** %ap, i32
4702 ; Demonstrate usage of llvm.va_copy and llvm.va_end
4704 %aq2 = bitcast i8** %aq to i8*
4705 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
4706 call void @llvm.va_end(i8* %aq2)
4708 ; Stop processing of arguments.
4709 call void @llvm.va_end(i8* %ap2)
4713 declare void @llvm.va_start(i8*)
4714 declare void @llvm.va_copy(i8*, i8*)
4715 declare void @llvm.va_end(i8*)
4721 <!-- _______________________________________________________________________ -->
4722 <div class="doc_subsubsection">
4723 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
4727 <div class="doc_text">
4729 <pre> declare void %llvm.va_start(i8* <arglist>)<br></pre>
4731 <p>The '<tt>llvm.va_start</tt>' intrinsic initializes
4732 <tt>*<arglist></tt> for subsequent use by <tt><a
4733 href="#i_va_arg">va_arg</a></tt>.</p>
4737 <p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
4741 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
4742 macro available in C. In a target-dependent way, it initializes the
4743 <tt>va_list</tt> element to which the argument points, so that the next call to
4744 <tt>va_arg</tt> will produce the first variable argument passed to the function.
4745 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
4746 last argument of the function as the compiler can figure that out.</p>
4750 <!-- _______________________________________________________________________ -->
4751 <div class="doc_subsubsection">
4752 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
4755 <div class="doc_text">
4757 <pre> declare void @llvm.va_end(i8* <arglist>)<br></pre>
4760 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
4761 which has been initialized previously with <tt><a href="#int_va_start">llvm.va_start</a></tt>
4762 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
4766 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
4770 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
4771 macro available in C. In a target-dependent way, it destroys the
4772 <tt>va_list</tt> element to which the argument points. Calls to <a
4773 href="#int_va_start"><tt>llvm.va_start</tt></a> and <a href="#int_va_copy">
4774 <tt>llvm.va_copy</tt></a> must be matched exactly with calls to
4775 <tt>llvm.va_end</tt>.</p>
4779 <!-- _______________________________________________________________________ -->
4780 <div class="doc_subsubsection">
4781 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
4784 <div class="doc_text">
4789 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
4794 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
4795 from the source argument list to the destination argument list.</p>
4799 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
4800 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
4805 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
4806 macro available in C. In a target-dependent way, it copies the source
4807 <tt>va_list</tt> element into the destination <tt>va_list</tt> element. This
4808 intrinsic is necessary because the <tt><a href="#int_va_start">
4809 llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
4810 example, memory allocation.</p>
4814 <!-- ======================================================================= -->
4815 <div class="doc_subsection">
4816 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
4819 <div class="doc_text">
4822 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
4823 Collection</a> (GC) requires the implementation and generation of these
4825 These intrinsics allow identification of <a href="#int_gcroot">GC roots on the
4826 stack</a>, as well as garbage collector implementations that require <a
4827 href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a> barriers.
4828 Front-ends for type-safe garbage collected languages should generate these
4829 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
4830 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
4833 <p>The garbage collection intrinsics only operate on objects in the generic
4834 address space (address space zero).</p>
4838 <!-- _______________________________________________________________________ -->
4839 <div class="doc_subsubsection">
4840 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
4843 <div class="doc_text">
4848 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
4853 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
4854 the code generator, and allows some metadata to be associated with it.</p>
4858 <p>The first argument specifies the address of a stack object that contains the
4859 root pointer. The second pointer (which must be either a constant or a global
4860 value address) contains the meta-data to be associated with the root.</p>
4864 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
4865 location. At compile-time, the code generator generates information to allow
4866 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
4867 intrinsic may only be used in a function which <a href="#gc">specifies a GC
4873 <!-- _______________________________________________________________________ -->
4874 <div class="doc_subsubsection">
4875 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
4878 <div class="doc_text">
4883 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
4888 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
4889 locations, allowing garbage collector implementations that require read
4894 <p>The second argument is the address to read from, which should be an address
4895 allocated from the garbage collector. The first object is a pointer to the
4896 start of the referenced object, if needed by the language runtime (otherwise
4901 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
4902 instruction, but may be replaced with substantially more complex code by the
4903 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
4904 may only be used in a function which <a href="#gc">specifies a GC
4910 <!-- _______________________________________________________________________ -->
4911 <div class="doc_subsubsection">
4912 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
4915 <div class="doc_text">
4920 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
4925 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
4926 locations, allowing garbage collector implementations that require write
4927 barriers (such as generational or reference counting collectors).</p>
4931 <p>The first argument is the reference to store, the second is the start of the
4932 object to store it to, and the third is the address of the field of Obj to
4933 store to. If the runtime does not require a pointer to the object, Obj may be
4938 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
4939 instruction, but may be replaced with substantially more complex code by the
4940 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
4941 may only be used in a function which <a href="#gc">specifies a GC
4948 <!-- ======================================================================= -->
4949 <div class="doc_subsection">
4950 <a name="int_codegen">Code Generator Intrinsics</a>
4953 <div class="doc_text">
4955 These intrinsics are provided by LLVM to expose special features that may only
4956 be implemented with code generator support.
4961 <!-- _______________________________________________________________________ -->
4962 <div class="doc_subsubsection">
4963 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
4966 <div class="doc_text">
4970 declare i8 *@llvm.returnaddress(i32 <level>)
4976 The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
4977 target-specific value indicating the return address of the current function
4978 or one of its callers.
4984 The argument to this intrinsic indicates which function to return the address
4985 for. Zero indicates the calling function, one indicates its caller, etc. The
4986 argument is <b>required</b> to be a constant integer value.
4992 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
4993 the return address of the specified call frame, or zero if it cannot be
4994 identified. The value returned by this intrinsic is likely to be incorrect or 0
4995 for arguments other than zero, so it should only be used for debugging purposes.
4999 Note that calling this intrinsic does not prevent function inlining or other
5000 aggressive transformations, so the value returned may not be that of the obvious
5001 source-language caller.
5006 <!-- _______________________________________________________________________ -->
5007 <div class="doc_subsubsection">
5008 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
5011 <div class="doc_text">
5015 declare i8 *@llvm.frameaddress(i32 <level>)
5021 The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
5022 target-specific frame pointer value for the specified stack frame.
5028 The argument to this intrinsic indicates which function to return the frame
5029 pointer for. Zero indicates the calling function, one indicates its caller,
5030 etc. The argument is <b>required</b> to be a constant integer value.
5036 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
5037 the frame address of the specified call frame, or zero if it cannot be
5038 identified. The value returned by this intrinsic is likely to be incorrect or 0
5039 for arguments other than zero, so it should only be used for debugging purposes.
5043 Note that calling this intrinsic does not prevent function inlining or other
5044 aggressive transformations, so the value returned may not be that of the obvious
5045 source-language caller.
5049 <!-- _______________________________________________________________________ -->
5050 <div class="doc_subsubsection">
5051 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
5054 <div class="doc_text">
5058 declare i8 *@llvm.stacksave()
5064 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
5065 the function stack, for use with <a href="#int_stackrestore">
5066 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
5067 features like scoped automatic variable sized arrays in C99.
5073 This intrinsic returns a opaque pointer value that can be passed to <a
5074 href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
5075 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
5076 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
5077 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
5078 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
5079 that were allocated after the <tt>llvm.stacksave</tt> was executed.
5084 <!-- _______________________________________________________________________ -->
5085 <div class="doc_subsubsection">
5086 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
5089 <div class="doc_text">
5093 declare void @llvm.stackrestore(i8 * %ptr)
5099 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
5100 the function stack to the state it was in when the corresponding <a
5101 href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
5102 useful for implementing language features like scoped automatic variable sized
5109 See the description for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.
5115 <!-- _______________________________________________________________________ -->
5116 <div class="doc_subsubsection">
5117 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
5120 <div class="doc_text">
5124 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>)
5131 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
5132 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
5134 effect on the behavior of the program but can change its performance
5141 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
5142 determining if the fetch should be for a read (0) or write (1), and
5143 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
5144 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
5145 <tt>locality</tt> arguments must be constant integers.
5151 This intrinsic does not modify the behavior of the program. In particular,
5152 prefetches cannot trap and do not produce a value. On targets that support this
5153 intrinsic, the prefetch can provide hints to the processor cache for better
5159 <!-- _______________________________________________________________________ -->
5160 <div class="doc_subsubsection">
5161 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
5164 <div class="doc_text">
5168 declare void @llvm.pcmarker(i32 <id>)
5175 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
5177 code to simulators and other tools. The method is target specific, but it is
5178 expected that the marker will use exported symbols to transmit the PC of the
5180 The marker makes no guarantees that it will remain with any specific instruction
5181 after optimizations. It is possible that the presence of a marker will inhibit
5182 optimizations. The intended use is to be inserted after optimizations to allow
5183 correlations of simulation runs.
5189 <tt>id</tt> is a numerical id identifying the marker.
5195 This intrinsic does not modify the behavior of the program. Backends that do not
5196 support this intrinisic may ignore it.
5201 <!-- _______________________________________________________________________ -->
5202 <div class="doc_subsubsection">
5203 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
5206 <div class="doc_text">
5210 declare i64 @llvm.readcyclecounter( )
5217 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
5218 counter register (or similar low latency, high accuracy clocks) on those targets
5219 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
5220 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
5221 should only be used for small timings.
5227 When directly supported, reading the cycle counter should not modify any memory.
5228 Implementations are allowed to either return a application specific value or a
5229 system wide value. On backends without support, this is lowered to a constant 0.
5234 <!-- ======================================================================= -->
5235 <div class="doc_subsection">
5236 <a name="int_libc">Standard C Library Intrinsics</a>
5239 <div class="doc_text">
5241 LLVM provides intrinsics for a few important standard C library functions.
5242 These intrinsics allow source-language front-ends to pass information about the
5243 alignment of the pointer arguments to the code generator, providing opportunity
5244 for more efficient code generation.
5249 <!-- _______________________________________________________________________ -->
5250 <div class="doc_subsubsection">
5251 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
5254 <div class="doc_text">
5257 <p>This is an overloaded intrinsic. You can use llvm.memcpy on any integer bit
5258 width. Not all targets support all bit widths however.</p>
5260 declare void @llvm.memcpy.i8(i8 * <dest>, i8 * <src>,
5261 i8 <len>, i32 <align>)
5262 declare void @llvm.memcpy.i16(i8 * <dest>, i8 * <src>,
5263 i16 <len>, i32 <align>)
5264 declare void @llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
5265 i32 <len>, i32 <align>)
5266 declare void @llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
5267 i64 <len>, i32 <align>)
5273 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
5274 location to the destination location.
5278 Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
5279 intrinsics do not return a value, and takes an extra alignment argument.
5285 The first argument is a pointer to the destination, the second is a pointer to
5286 the source. The third argument is an integer argument
5287 specifying the number of bytes to copy, and the fourth argument is the alignment
5288 of the source and destination locations.
5292 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5293 the caller guarantees that both the source and destination pointers are aligned
5300 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
5301 location to the destination location, which are not allowed to overlap. It
5302 copies "len" bytes of memory over. If the argument is known to be aligned to
5303 some boundary, this can be specified as the fourth argument, otherwise it should
5309 <!-- _______________________________________________________________________ -->
5310 <div class="doc_subsubsection">
5311 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
5314 <div class="doc_text">
5317 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
5318 width. Not all targets support all bit widths however.</p>
5320 declare void @llvm.memmove.i8(i8 * <dest>, i8 * <src>,
5321 i8 <len>, i32 <align>)
5322 declare void @llvm.memmove.i16(i8 * <dest>, i8 * <src>,
5323 i16 <len>, i32 <align>)
5324 declare void @llvm.memmove.i32(i8 * <dest>, i8 * <src>,
5325 i32 <len>, i32 <align>)
5326 declare void @llvm.memmove.i64(i8 * <dest>, i8 * <src>,
5327 i64 <len>, i32 <align>)
5333 The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
5334 location to the destination location. It is similar to the
5335 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to overlap.
5339 Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
5340 intrinsics do not return a value, and takes an extra alignment argument.
5346 The first argument is a pointer to the destination, the second is a pointer to
5347 the source. The third argument is an integer argument
5348 specifying the number of bytes to copy, and the fourth argument is the alignment
5349 of the source and destination locations.
5353 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5354 the caller guarantees that the source and destination pointers are aligned to
5361 The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
5362 location to the destination location, which may overlap. It
5363 copies "len" bytes of memory over. If the argument is known to be aligned to
5364 some boundary, this can be specified as the fourth argument, otherwise it should
5370 <!-- _______________________________________________________________________ -->
5371 <div class="doc_subsubsection">
5372 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
5375 <div class="doc_text">
5378 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
5379 width. Not all targets support all bit widths however.</p>
5381 declare void @llvm.memset.i8(i8 * <dest>, i8 <val>,
5382 i8 <len>, i32 <align>)
5383 declare void @llvm.memset.i16(i8 * <dest>, i8 <val>,
5384 i16 <len>, i32 <align>)
5385 declare void @llvm.memset.i32(i8 * <dest>, i8 <val>,
5386 i32 <len>, i32 <align>)
5387 declare void @llvm.memset.i64(i8 * <dest>, i8 <val>,
5388 i64 <len>, i32 <align>)
5394 The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
5399 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
5400 does not return a value, and takes an extra alignment argument.
5406 The first argument is a pointer to the destination to fill, the second is the
5407 byte value to fill it with, the third argument is an integer
5408 argument specifying the number of bytes to fill, and the fourth argument is the
5409 known alignment of destination location.
5413 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5414 the caller guarantees that the destination pointer is aligned to that boundary.
5420 The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
5422 destination location. If the argument is known to be aligned to some boundary,
5423 this can be specified as the fourth argument, otherwise it should be set to 0 or
5429 <!-- _______________________________________________________________________ -->
5430 <div class="doc_subsubsection">
5431 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
5434 <div class="doc_text">
5437 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
5438 floating point or vector of floating point type. Not all targets support all
5441 declare float @llvm.sqrt.f32(float %Val)
5442 declare double @llvm.sqrt.f64(double %Val)
5443 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
5444 declare fp128 @llvm.sqrt.f128(fp128 %Val)
5445 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
5451 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
5452 returning the same value as the libm '<tt>sqrt</tt>' functions would. Unlike
5453 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
5454 negative numbers other than -0.0 (which allows for better optimization, because
5455 there is no need to worry about errno being set). <tt>llvm.sqrt(-0.0)</tt> is
5456 defined to return -0.0 like IEEE sqrt.
5462 The argument and return value are floating point numbers of the same type.
5468 This function returns the sqrt of the specified operand if it is a nonnegative
5469 floating point number.
5473 <!-- _______________________________________________________________________ -->
5474 <div class="doc_subsubsection">
5475 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
5478 <div class="doc_text">
5481 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
5482 floating point or vector of floating point type. Not all targets support all
5485 declare float @llvm.powi.f32(float %Val, i32 %power)
5486 declare double @llvm.powi.f64(double %Val, i32 %power)
5487 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
5488 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
5489 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
5495 The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
5496 specified (positive or negative) power. The order of evaluation of
5497 multiplications is not defined. When a vector of floating point type is
5498 used, the second argument remains a scalar integer value.
5504 The second argument is an integer power, and the first is a value to raise to
5511 This function returns the first value raised to the second power with an
5512 unspecified sequence of rounding operations.</p>
5515 <!-- _______________________________________________________________________ -->
5516 <div class="doc_subsubsection">
5517 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
5520 <div class="doc_text">
5523 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
5524 floating point or vector of floating point type. Not all targets support all
5527 declare float @llvm.sin.f32(float %Val)
5528 declare double @llvm.sin.f64(double %Val)
5529 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
5530 declare fp128 @llvm.sin.f128(fp128 %Val)
5531 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
5537 The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.
5543 The argument and return value are floating point numbers of the same type.
5549 This function returns the sine of the specified operand, returning the
5550 same values as the libm <tt>sin</tt> functions would, and handles error
5551 conditions in the same way.</p>
5554 <!-- _______________________________________________________________________ -->
5555 <div class="doc_subsubsection">
5556 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
5559 <div class="doc_text">
5562 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
5563 floating point or vector of floating point type. Not all targets support all
5566 declare float @llvm.cos.f32(float %Val)
5567 declare double @llvm.cos.f64(double %Val)
5568 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
5569 declare fp128 @llvm.cos.f128(fp128 %Val)
5570 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
5576 The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.
5582 The argument and return value are floating point numbers of the same type.
5588 This function returns the cosine of the specified operand, returning the
5589 same values as the libm <tt>cos</tt> functions would, and handles error
5590 conditions in the same way.</p>
5593 <!-- _______________________________________________________________________ -->
5594 <div class="doc_subsubsection">
5595 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
5598 <div class="doc_text">
5601 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
5602 floating point or vector of floating point type. Not all targets support all
5605 declare float @llvm.pow.f32(float %Val, float %Power)
5606 declare double @llvm.pow.f64(double %Val, double %Power)
5607 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
5608 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
5609 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
5615 The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
5616 specified (positive or negative) power.
5622 The second argument is a floating point power, and the first is a value to
5623 raise to that power.
5629 This function returns the first value raised to the second power,
5631 same values as the libm <tt>pow</tt> functions would, and handles error
5632 conditions in the same way.</p>
5636 <!-- ======================================================================= -->
5637 <div class="doc_subsection">
5638 <a name="int_manip">Bit Manipulation Intrinsics</a>
5641 <div class="doc_text">
5643 LLVM provides intrinsics for a few important bit manipulation operations.
5644 These allow efficient code generation for some algorithms.
5649 <!-- _______________________________________________________________________ -->
5650 <div class="doc_subsubsection">
5651 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
5654 <div class="doc_text">
5657 <p>This is an overloaded intrinsic function. You can use bswap on any integer
5658 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
5660 declare i16 @llvm.bswap.i16(i16 <id>)
5661 declare i32 @llvm.bswap.i32(i32 <id>)
5662 declare i64 @llvm.bswap.i64(i64 <id>)
5668 The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
5669 values with an even number of bytes (positive multiple of 16 bits). These are
5670 useful for performing operations on data that is not in the target's native
5677 The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
5678 and low byte of the input i16 swapped. Similarly, the <tt>llvm.bswap.i32</tt>
5679 intrinsic returns an i32 value that has the four bytes of the input i32
5680 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the returned
5681 i32 will have its bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i48</tt>,
5682 <tt>llvm.bswap.i64</tt> and other intrinsics extend this concept to
5683 additional even-byte lengths (6 bytes, 8 bytes and more, respectively).
5688 <!-- _______________________________________________________________________ -->
5689 <div class="doc_subsubsection">
5690 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
5693 <div class="doc_text">
5696 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
5697 width. Not all targets support all bit widths however.</p>
5699 declare i8 @llvm.ctpop.i8(i8 <src>)
5700 declare i16 @llvm.ctpop.i16(i16 <src>)
5701 declare i32 @llvm.ctpop.i32(i32 <src>)
5702 declare i64 @llvm.ctpop.i64(i64 <src>)
5703 declare i256 @llvm.ctpop.i256(i256 <src>)
5709 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
5716 The only argument is the value to be counted. The argument may be of any
5717 integer type. The return type must match the argument type.
5723 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
5727 <!-- _______________________________________________________________________ -->
5728 <div class="doc_subsubsection">
5729 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
5732 <div class="doc_text">
5735 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
5736 integer bit width. Not all targets support all bit widths however.</p>
5738 declare i8 @llvm.ctlz.i8 (i8 <src>)
5739 declare i16 @llvm.ctlz.i16(i16 <src>)
5740 declare i32 @llvm.ctlz.i32(i32 <src>)
5741 declare i64 @llvm.ctlz.i64(i64 <src>)
5742 declare i256 @llvm.ctlz.i256(i256 <src>)
5748 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
5749 leading zeros in a variable.
5755 The only argument is the value to be counted. The argument may be of any
5756 integer type. The return type must match the argument type.
5762 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
5763 in a variable. If the src == 0 then the result is the size in bits of the type
5764 of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.
5770 <!-- _______________________________________________________________________ -->
5771 <div class="doc_subsubsection">
5772 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
5775 <div class="doc_text">
5778 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
5779 integer bit width. Not all targets support all bit widths however.</p>
5781 declare i8 @llvm.cttz.i8 (i8 <src>)
5782 declare i16 @llvm.cttz.i16(i16 <src>)
5783 declare i32 @llvm.cttz.i32(i32 <src>)
5784 declare i64 @llvm.cttz.i64(i64 <src>)
5785 declare i256 @llvm.cttz.i256(i256 <src>)
5791 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
5798 The only argument is the value to be counted. The argument may be of any
5799 integer type. The return type must match the argument type.
5805 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
5806 in a variable. If the src == 0 then the result is the size in bits of the type
5807 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
5811 <!-- _______________________________________________________________________ -->
5812 <div class="doc_subsubsection">
5813 <a name="int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic</a>
5816 <div class="doc_text">
5819 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.select</tt>
5820 on any integer bit width.</p>
5822 declare i17 @llvm.part.select.i17 (i17 %val, i32 %loBit, i32 %hiBit)
5823 declare i29 @llvm.part.select.i29 (i29 %val, i32 %loBit, i32 %hiBit)
5827 <p>The '<tt>llvm.part.select</tt>' family of intrinsic functions selects a
5828 range of bits from an integer value and returns them in the same bit width as
5829 the original value.</p>
5832 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5833 any bit width but they must have the same bit width. The second and third
5834 arguments must be <tt>i32</tt> type since they specify only a bit index.</p>
5837 <p>The operation of the '<tt>llvm.part.select</tt>' intrinsic has two modes
5838 of operation: forwards and reverse. If <tt>%loBit</tt> is greater than
5839 <tt>%hiBits</tt> then the intrinsic operates in reverse mode. Otherwise it
5840 operates in forward mode.</p>
5841 <p>In forward mode, this intrinsic is the equivalent of shifting <tt>%val</tt>
5842 right by <tt>%loBit</tt> bits and then ANDing it with a mask with
5843 only the <tt>%hiBit - %loBit</tt> bits set, as follows:</p>
5845 <li>The <tt>%val</tt> is shifted right (LSHR) by the number of bits specified
5846 by <tt>%loBits</tt>. This normalizes the value to the low order bits.</li>
5847 <li>The <tt>%loBits</tt> value is subtracted from the <tt>%hiBits</tt> value
5848 to determine the number of bits to retain.</li>
5849 <li>A mask of the retained bits is created by shifting a -1 value.</li>
5850 <li>The mask is ANDed with <tt>%val</tt> to produce the result.</li>
5852 <p>In reverse mode, a similar computation is made except that the bits are
5853 returned in the reverse order. So, for example, if <tt>X</tt> has the value
5854 <tt>i16 0x0ACF (101011001111)</tt> and we apply
5855 <tt>part.select(i16 X, 8, 3)</tt> to it, we get back the value
5856 <tt>i16 0x0026 (000000100110)</tt>.</p>
5859 <div class="doc_subsubsection">
5860 <a name="int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic</a>
5863 <div class="doc_text">
5866 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.set</tt>
5867 on any integer bit width.</p>
5869 declare i17 @llvm.part.set.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
5870 declare i29 @llvm.part.set.i29.i9 (i29 %val, i9 %repl, i32 %lo, i32 %hi)
5874 <p>The '<tt>llvm.part.set</tt>' family of intrinsic functions replaces a range
5875 of bits in an integer value with another integer value. It returns the integer
5876 with the replaced bits.</p>
5879 <p>The first argument, <tt>%val</tt>, and the result may be integer types of
5880 any bit width, but they must have the same bit width. <tt>%val</tt> is the value
5881 whose bits will be replaced. The second argument, <tt>%repl</tt> may be an
5882 integer of any bit width. The third and fourth arguments must be <tt>i32</tt>
5883 type since they specify only a bit index.</p>
5886 <p>The operation of the '<tt>llvm.part.set</tt>' intrinsic has two modes
5887 of operation: forwards and reverse. If <tt>%lo</tt> is greater than
5888 <tt>%hi</tt> then the intrinsic operates in reverse mode. Otherwise it
5889 operates in forward mode.</p>
5891 <p>For both modes, the <tt>%repl</tt> value is prepared for use by either
5892 truncating it down to the size of the replacement area or zero extending it
5893 up to that size.</p>
5895 <p>In forward mode, the bits between <tt>%lo</tt> and <tt>%hi</tt> (inclusive)
5896 are replaced with corresponding bits from <tt>%repl</tt>. That is the 0th bit
5897 in <tt>%repl</tt> replaces the <tt>%lo</tt>th bit in <tt>%val</tt> and etc. up
5898 to the <tt>%hi</tt>th bit.</p>
5900 <p>In reverse mode, a similar computation is made except that the bits are
5901 reversed. That is, the <tt>0</tt>th bit in <tt>%repl</tt> replaces the
5902 <tt>%hi</tt> bit in <tt>%val</tt> and etc. down to the <tt>%lo</tt>th bit.</p>
5907 llvm.part.set(0xFFFF, 0, 4, 7) -> 0xFF0F
5908 llvm.part.set(0xFFFF, 0, 7, 4) -> 0xFF0F
5909 llvm.part.set(0xFFFF, 1, 7, 4) -> 0xFF8F
5910 llvm.part.set(0xFFFF, F, 8, 3) -> 0xFFE7
5911 llvm.part.set(0xFFFF, 0, 3, 8) -> 0xFE07
5916 <!-- _______________________________________________________________________ -->
5917 <div class="doc_subsubsection">
5918 <a name="int_sadd_ovf">'<tt>llvm.sadd.with.overflow.*</tt>' Intrinsics</a>
5921 <div class="doc_text">
5925 <p>This is an overloaded intrinsic. You can use <tt>llvm.sadd.with.overflow</tt>
5926 on any integer bit width. However, not all targets support all bit widths.</p>
5929 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
5930 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
5931 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
5936 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
5937 a signed addition of the two arguments, and indicate whether an overflow
5938 occurred during the signed summation.</p>
5942 <p>The arguments (%a and %b) and the first element of the result structure may
5943 be of integer types of any bit width, but they must have the same bit width. The
5944 second element of the result structure must be of type <tt>i1</tt>. <tt>%a</tt>
5945 and <tt>%b</tt> are the two values that will undergo signed addition.</p>
5949 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
5950 a signed addition of the two variables. They return a structure — the
5951 first element of which is the signed summation, and the second element of which
5952 is a bit specifying if the signed summation resulted in an overflow.</p>
5956 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
5957 %sum = extractvalue {i32, i1} %res, 0
5958 %obit = extractvalue {i32, i1} %res, 1
5959 br i1 %obit, label %overflow, label %normal
5964 <!-- _______________________________________________________________________ -->
5965 <div class="doc_subsubsection">
5966 <a name="int_uadd_ovf">'<tt>llvm.uadd.with.overflow.*</tt>' Intrinsics</a>
5969 <div class="doc_text">
5973 <p>This is an overloaded intrinsic. You can use <tt>llvm.uadd.with.overflow</tt>
5974 on any integer bit width. However, not all targets support all bit widths.</p>
5977 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
5978 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
5979 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
5984 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
5985 an unsigned addition of the two arguments, and indicate whether a carry occurred
5986 during the unsigned summation.</p>
5990 <p>The arguments (%a and %b) and the first element of the result structure may
5991 be of integer types of any bit width, but they must have the same bit width. The
5992 second element of the result structure must be of type <tt>i1</tt>. <tt>%a</tt>
5993 and <tt>%b</tt> are the two values that will undergo unsigned addition.</p>
5997 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
5998 an unsigned addition of the two arguments. They return a structure — the
5999 first element of which is the sum, and the second element of which is a bit
6000 specifying if the unsigned summation resulted in a carry.</p>
6004 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6005 %sum = extractvalue {i32, i1} %res, 0
6006 %obit = extractvalue {i32, i1} %res, 1
6007 br i1 %obit, label %carry, label %normal
6012 <!-- _______________________________________________________________________ -->
6013 <div class="doc_subsubsection">
6014 <a name="int_ssub_ovf">'<tt>llvm.ssub.with.overflow.*</tt>' Intrinsics</a>
6017 <div class="doc_text">
6021 <p>This is an overloaded intrinsic. You can use <tt>llvm.ssub.with.overflow</tt>
6022 on any integer bit width. However, not all targets support all bit widths.</p>
6025 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
6026 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6027 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
6032 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
6033 a signed subtraction of the two arguments, and indicate whether an overflow
6034 occurred during the signed subtraction.</p>
6038 <p>The arguments (%a and %b) and the first element of the result structure may
6039 be of integer types of any bit width, but they must have the same bit width. The
6040 second element of the result structure must be of type <tt>i1</tt>. <tt>%a</tt>
6041 and <tt>%b</tt> are the two values that will undergo signed subtraction.</p>
6045 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
6046 a signed subtraction of the two arguments. They return a structure — the
6047 first element of which is the subtraction, and the second element of which is a bit
6048 specifying if the signed subtraction resulted in an overflow.</p>
6052 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6053 %sum = extractvalue {i32, i1} %res, 0
6054 %obit = extractvalue {i32, i1} %res, 1
6055 br i1 %obit, label %overflow, label %normal
6060 <!-- _______________________________________________________________________ -->
6061 <div class="doc_subsubsection">
6062 <a name="int_usub_ovf">'<tt>llvm.usub.with.overflow.*</tt>' Intrinsics</a>
6065 <div class="doc_text">
6069 <p>This is an overloaded intrinsic. You can use <tt>llvm.usub.with.overflow</tt>
6070 on any integer bit width. However, not all targets support all bit widths.</p>
6073 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
6074 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6075 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
6080 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
6081 an unsigned subtraction of the two arguments, and indicate whether an overflow
6082 occurred during the unsigned subtraction.</p>
6086 <p>The arguments (%a and %b) and the first element of the result structure may
6087 be of integer types of any bit width, but they must have the same bit width. The
6088 second element of the result structure must be of type <tt>i1</tt>. <tt>%a</tt>
6089 and <tt>%b</tt> are the two values that will undergo unsigned subtraction.</p>
6093 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
6094 an unsigned subtraction of the two arguments. They return a structure — the
6095 first element of which is the subtraction, and the second element of which is a bit
6096 specifying if the unsigned subtraction resulted in an overflow.</p>
6100 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6101 %sum = extractvalue {i32, i1} %res, 0
6102 %obit = extractvalue {i32, i1} %res, 1
6103 br i1 %obit, label %overflow, label %normal
6108 <!-- _______________________________________________________________________ -->
6109 <div class="doc_subsubsection">
6110 <a name="int_smul_ovf">'<tt>llvm.smul.with.overflow.*</tt>' Intrinsics</a>
6113 <div class="doc_text">
6117 <p>This is an overloaded intrinsic. You can use <tt>llvm.smul.with.overflow</tt>
6118 on any integer bit width. However, not all targets support all bit widths.</p>
6121 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
6122 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6123 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
6128 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
6129 a signed multiplication of the two arguments, and indicate whether an overflow
6130 occurred during the signed multiplication.</p>
6134 <p>The arguments (%a and %b) and the first element of the result structure may
6135 be of integer types of any bit width, but they must have the same bit width. The
6136 second element of the result structure must be of type <tt>i1</tt>. <tt>%a</tt>
6137 and <tt>%b</tt> are the two values that will undergo signed multiplication.</p>
6141 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
6142 a signed multiplication of the two arguments. They return a structure —
6143 the first element of which is the multiplication, and the second element of
6144 which is a bit specifying if the signed multiplication resulted in an
6149 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6150 %sum = extractvalue {i32, i1} %res, 0
6151 %obit = extractvalue {i32, i1} %res, 1
6152 br i1 %obit, label %overflow, label %normal
6157 <!-- ======================================================================= -->
6158 <div class="doc_subsection">
6159 <a name="int_debugger">Debugger Intrinsics</a>
6162 <div class="doc_text">
6164 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
6165 are described in the <a
6166 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
6167 Debugging</a> document.
6172 <!-- ======================================================================= -->
6173 <div class="doc_subsection">
6174 <a name="int_eh">Exception Handling Intrinsics</a>
6177 <div class="doc_text">
6178 <p> The LLVM exception handling intrinsics (which all start with
6179 <tt>llvm.eh.</tt> prefix), are described in the <a
6180 href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
6181 Handling</a> document. </p>
6184 <!-- ======================================================================= -->
6185 <div class="doc_subsection">
6186 <a name="int_trampoline">Trampoline Intrinsic</a>
6189 <div class="doc_text">
6191 This intrinsic makes it possible to excise one parameter, marked with
6192 the <tt>nest</tt> attribute, from a function. The result is a callable
6193 function pointer lacking the nest parameter - the caller does not need
6194 to provide a value for it. Instead, the value to use is stored in
6195 advance in a "trampoline", a block of memory usually allocated
6196 on the stack, which also contains code to splice the nest value into the
6197 argument list. This is used to implement the GCC nested function address
6201 For example, if the function is
6202 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
6203 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as follows:</p>
6205 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
6206 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
6207 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
6208 %fp = bitcast i8* %p to i32 (i32, i32)*
6210 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
6211 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
6214 <!-- _______________________________________________________________________ -->
6215 <div class="doc_subsubsection">
6216 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
6218 <div class="doc_text">
6221 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
6225 This fills the memory pointed to by <tt>tramp</tt> with code
6226 and returns a function pointer suitable for executing it.
6230 The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
6231 pointers. The <tt>tramp</tt> argument must point to a sufficiently large
6232 and sufficiently aligned block of memory; this memory is written to by the
6233 intrinsic. Note that the size and the alignment are target-specific - LLVM
6234 currently provides no portable way of determining them, so a front-end that
6235 generates this intrinsic needs to have some target-specific knowledge.
6236 The <tt>func</tt> argument must hold a function bitcast to an <tt>i8*</tt>.
6240 The block of memory pointed to by <tt>tramp</tt> is filled with target
6241 dependent code, turning it into a function. A pointer to this function is
6242 returned, but needs to be bitcast to an
6243 <a href="#int_trampoline">appropriate function pointer type</a>
6244 before being called. The new function's signature is the same as that of
6245 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
6246 removed. At most one such <tt>nest</tt> argument is allowed, and it must be
6247 of pointer type. Calling the new function is equivalent to calling
6248 <tt>func</tt> with the same argument list, but with <tt>nval</tt> used for the
6249 missing <tt>nest</tt> argument. If, after calling
6250 <tt>llvm.init.trampoline</tt>, the memory pointed to by <tt>tramp</tt> is
6251 modified, then the effect of any later call to the returned function pointer is
6256 <!-- ======================================================================= -->
6257 <div class="doc_subsection">
6258 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
6261 <div class="doc_text">
6263 These intrinsic functions expand the "universal IR" of LLVM to represent
6264 hardware constructs for atomic operations and memory synchronization. This
6265 provides an interface to the hardware, not an interface to the programmer. It
6266 is aimed at a low enough level to allow any programming models or APIs
6267 (Application Programming Interfaces) which
6268 need atomic behaviors to map cleanly onto it. It is also modeled primarily on
6269 hardware behavior. Just as hardware provides a "universal IR" for source
6270 languages, it also provides a starting point for developing a "universal"
6271 atomic operation and synchronization IR.
6274 These do <em>not</em> form an API such as high-level threading libraries,
6275 software transaction memory systems, atomic primitives, and intrinsic
6276 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
6277 application libraries. The hardware interface provided by LLVM should allow
6278 a clean implementation of all of these APIs and parallel programming models.
6279 No one model or paradigm should be selected above others unless the hardware
6280 itself ubiquitously does so.
6285 <!-- _______________________________________________________________________ -->
6286 <div class="doc_subsubsection">
6287 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
6289 <div class="doc_text">
6292 declare void @llvm.memory.barrier( i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>,
6298 The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
6299 specific pairs of memory access types.
6303 The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
6304 The first four arguments enables a specific barrier as listed below. The fith
6305 argument specifies that the barrier applies to io or device or uncached memory.
6309 <li><tt>ll</tt>: load-load barrier</li>
6310 <li><tt>ls</tt>: load-store barrier</li>
6311 <li><tt>sl</tt>: store-load barrier</li>
6312 <li><tt>ss</tt>: store-store barrier</li>
6313 <li><tt>device</tt>: barrier applies to device and uncached memory also.</li>
6317 This intrinsic causes the system to enforce some ordering constraints upon
6318 the loads and stores of the program. This barrier does not indicate
6319 <em>when</em> any events will occur, it only enforces an <em>order</em> in
6320 which they occur. For any of the specified pairs of load and store operations
6321 (f.ex. load-load, or store-load), all of the first operations preceding the
6322 barrier will complete before any of the second operations succeeding the
6323 barrier begin. Specifically the semantics for each pairing is as follows:
6326 <li><tt>ll</tt>: All loads before the barrier must complete before any load
6327 after the barrier begins.</li>
6329 <li><tt>ls</tt>: All loads before the barrier must complete before any
6330 store after the barrier begins.</li>
6331 <li><tt>ss</tt>: All stores before the barrier must complete before any
6332 store after the barrier begins.</li>
6333 <li><tt>sl</tt>: All stores before the barrier must complete before any
6334 load after the barrier begins.</li>
6337 These semantics are applied with a logical "and" behavior when more than one
6338 is enabled in a single memory barrier intrinsic.
6341 Backends may implement stronger barriers than those requested when they do not
6342 support as fine grained a barrier as requested. Some architectures do not
6343 need all types of barriers and on such architectures, these become noops.
6350 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
6351 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
6352 <i>; guarantee the above finishes</i>
6353 store i32 8, %ptr <i>; before this begins</i>
6357 <!-- _______________________________________________________________________ -->
6358 <div class="doc_subsubsection">
6359 <a name="int_atomic_cmp_swap">'<tt>llvm.atomic.cmp.swap.*</tt>' Intrinsic</a>
6361 <div class="doc_text">
6364 This is an overloaded intrinsic. You can use <tt>llvm.atomic.cmp.swap</tt> on
6365 any integer bit width and for different address spaces. Not all targets
6366 support all bit widths however.</p>
6369 declare i8 @llvm.atomic.cmp.swap.i8.p0i8( i8* <ptr>, i8 <cmp>, i8 <val> )
6370 declare i16 @llvm.atomic.cmp.swap.i16.p0i16( i16* <ptr>, i16 <cmp>, i16 <val> )
6371 declare i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* <ptr>, i32 <cmp>, i32 <val> )
6372 declare i64 @llvm.atomic.cmp.swap.i64.p0i64( i64* <ptr>, i64 <cmp>, i64 <val> )
6377 This loads a value in memory and compares it to a given value. If they are
6378 equal, it stores a new value into the memory.
6382 The <tt>llvm.atomic.cmp.swap</tt> intrinsic takes three arguments. The result as
6383 well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
6384 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
6385 this integer type. While any bit width integer may be used, targets may only
6386 lower representations they support in hardware.
6391 This entire intrinsic must be executed atomically. It first loads the value
6392 in memory pointed to by <tt>ptr</tt> and compares it with the value
6393 <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the memory. The
6394 loaded value is yielded in all cases. This provides the equivalent of an
6395 atomic compare-and-swap operation within the SSA framework.
6403 %val1 = add i32 4, 4
6404 %result1 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 4, %val1 )
6405 <i>; yields {i32}:result1 = 4</i>
6406 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
6407 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
6409 %val2 = add i32 1, 1
6410 %result2 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 5, %val2 )
6411 <i>; yields {i32}:result2 = 8</i>
6412 %stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
6414 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
6418 <!-- _______________________________________________________________________ -->
6419 <div class="doc_subsubsection">
6420 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
6422 <div class="doc_text">
6426 This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
6427 integer bit width. Not all targets support all bit widths however.</p>
6429 declare i8 @llvm.atomic.swap.i8.p0i8( i8* <ptr>, i8 <val> )
6430 declare i16 @llvm.atomic.swap.i16.p0i16( i16* <ptr>, i16 <val> )
6431 declare i32 @llvm.atomic.swap.i32.p0i32( i32* <ptr>, i32 <val> )
6432 declare i64 @llvm.atomic.swap.i64.p0i64( i64* <ptr>, i64 <val> )
6437 This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
6438 the value from memory. It then stores the value in <tt>val</tt> in the memory
6444 The <tt>llvm.atomic.swap</tt> intrinsic takes two arguments. Both the
6445 <tt>val</tt> argument and the result must be integers of the same bit width.
6446 The first argument, <tt>ptr</tt>, must be a pointer to a value of this
6447 integer type. The targets may only lower integer representations they
6452 This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
6453 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
6454 equivalent of an atomic swap operation within the SSA framework.
6462 %val1 = add i32 4, 4
6463 %result1 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val1 )
6464 <i>; yields {i32}:result1 = 4</i>
6465 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
6466 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
6468 %val2 = add i32 1, 1
6469 %result2 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val2 )
6470 <i>; yields {i32}:result2 = 8</i>
6472 %stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
6473 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
6477 <!-- _______________________________________________________________________ -->
6478 <div class="doc_subsubsection">
6479 <a name="int_atomic_load_add">'<tt>llvm.atomic.load.add.*</tt>' Intrinsic</a>
6482 <div class="doc_text">
6485 This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.add</tt> on any
6486 integer bit width. Not all targets support all bit widths however.</p>
6488 declare i8 @llvm.atomic.load.add.i8..p0i8( i8* <ptr>, i8 <delta> )
6489 declare i16 @llvm.atomic.load.add.i16..p0i16( i16* <ptr>, i16 <delta> )
6490 declare i32 @llvm.atomic.load.add.i32..p0i32( i32* <ptr>, i32 <delta> )
6491 declare i64 @llvm.atomic.load.add.i64..p0i64( i64* <ptr>, i64 <delta> )
6496 This intrinsic adds <tt>delta</tt> to the value stored in memory at
6497 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
6502 The intrinsic takes two arguments, the first a pointer to an integer value
6503 and the second an integer value. The result is also an integer value. These
6504 integer types can have any bit width, but they must all have the same bit
6505 width. The targets may only lower integer representations they support.
6509 This intrinsic does a series of operations atomically. It first loads the
6510 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
6511 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
6518 %result1 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 4 )
6519 <i>; yields {i32}:result1 = 4</i>
6520 %result2 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 2 )
6521 <i>; yields {i32}:result2 = 8</i>
6522 %result3 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 5 )
6523 <i>; yields {i32}:result3 = 10</i>
6524 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
6528 <!-- _______________________________________________________________________ -->
6529 <div class="doc_subsubsection">
6530 <a name="int_atomic_load_sub">'<tt>llvm.atomic.load.sub.*</tt>' Intrinsic</a>
6533 <div class="doc_text">
6536 This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.sub</tt> on
6537 any integer bit width and for different address spaces. Not all targets
6538 support all bit widths however.</p>
6540 declare i8 @llvm.atomic.load.sub.i8.p0i32( i8* <ptr>, i8 <delta> )
6541 declare i16 @llvm.atomic.load.sub.i16.p0i32( i16* <ptr>, i16 <delta> )
6542 declare i32 @llvm.atomic.load.sub.i32.p0i32( i32* <ptr>, i32 <delta> )
6543 declare i64 @llvm.atomic.load.sub.i64.p0i32( i64* <ptr>, i64 <delta> )
6548 This intrinsic subtracts <tt>delta</tt> to the value stored in memory at
6549 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
6554 The intrinsic takes two arguments, the first a pointer to an integer value
6555 and the second an integer value. The result is also an integer value. These
6556 integer types can have any bit width, but they must all have the same bit
6557 width. The targets may only lower integer representations they support.
6561 This intrinsic does a series of operations atomically. It first loads the
6562 value stored at <tt>ptr</tt>. It then subtracts <tt>delta</tt>, stores the
6563 result to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
6570 %result1 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 4 )
6571 <i>; yields {i32}:result1 = 8</i>
6572 %result2 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 2 )
6573 <i>; yields {i32}:result2 = 4</i>
6574 %result3 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 5 )
6575 <i>; yields {i32}:result3 = 2</i>
6576 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = -3</i>
6580 <!-- _______________________________________________________________________ -->
6581 <div class="doc_subsubsection">
6582 <a name="int_atomic_load_and">'<tt>llvm.atomic.load.and.*</tt>' Intrinsic</a><br>
6583 <a name="int_atomic_load_nand">'<tt>llvm.atomic.load.nand.*</tt>' Intrinsic</a><br>
6584 <a name="int_atomic_load_or">'<tt>llvm.atomic.load.or.*</tt>' Intrinsic</a><br>
6585 <a name="int_atomic_load_xor">'<tt>llvm.atomic.load.xor.*</tt>' Intrinsic</a><br>
6588 <div class="doc_text">
6591 These are overloaded intrinsics. You can use <tt>llvm.atomic.load_and</tt>,
6592 <tt>llvm.atomic.load_nand</tt>, <tt>llvm.atomic.load_or</tt>, and
6593 <tt>llvm.atomic.load_xor</tt> on any integer bit width and for different
6594 address spaces. Not all targets support all bit widths however.</p>
6596 declare i8 @llvm.atomic.load.and.i8.p0i8( i8* <ptr>, i8 <delta> )
6597 declare i16 @llvm.atomic.load.and.i16.p0i16( i16* <ptr>, i16 <delta> )
6598 declare i32 @llvm.atomic.load.and.i32.p0i32( i32* <ptr>, i32 <delta> )
6599 declare i64 @llvm.atomic.load.and.i64.p0i64( i64* <ptr>, i64 <delta> )
6604 declare i8 @llvm.atomic.load.or.i8.p0i8( i8* <ptr>, i8 <delta> )
6605 declare i16 @llvm.atomic.load.or.i16.p0i16( i16* <ptr>, i16 <delta> )
6606 declare i32 @llvm.atomic.load.or.i32.p0i32( i32* <ptr>, i32 <delta> )
6607 declare i64 @llvm.atomic.load.or.i64.p0i64( i64* <ptr>, i64 <delta> )
6612 declare i8 @llvm.atomic.load.nand.i8.p0i32( i8* <ptr>, i8 <delta> )
6613 declare i16 @llvm.atomic.load.nand.i16.p0i32( i16* <ptr>, i16 <delta> )
6614 declare i32 @llvm.atomic.load.nand.i32.p0i32( i32* <ptr>, i32 <delta> )
6615 declare i64 @llvm.atomic.load.nand.i64.p0i32( i64* <ptr>, i64 <delta> )
6620 declare i8 @llvm.atomic.load.xor.i8.p0i32( i8* <ptr>, i8 <delta> )
6621 declare i16 @llvm.atomic.load.xor.i16.p0i32( i16* <ptr>, i16 <delta> )
6622 declare i32 @llvm.atomic.load.xor.i32.p0i32( i32* <ptr>, i32 <delta> )
6623 declare i64 @llvm.atomic.load.xor.i64.p0i32( i64* <ptr>, i64 <delta> )
6628 These intrinsics bitwise the operation (and, nand, or, xor) <tt>delta</tt> to
6629 the value stored in memory at <tt>ptr</tt>. It yields the original value
6635 These intrinsics take two arguments, the first a pointer to an integer value
6636 and the second an integer value. The result is also an integer value. These
6637 integer types can have any bit width, but they must all have the same bit
6638 width. The targets may only lower integer representations they support.
6642 These intrinsics does a series of operations atomically. They first load the
6643 value stored at <tt>ptr</tt>. They then do the bitwise operation
6644 <tt>delta</tt>, store the result to <tt>ptr</tt>. They yield the original
6645 value stored at <tt>ptr</tt>.
6651 store i32 0x0F0F, %ptr
6652 %result0 = call i32 @llvm.atomic.load.nand.i32.p0i32( i32* %ptr, i32 0xFF )
6653 <i>; yields {i32}:result0 = 0x0F0F</i>
6654 %result1 = call i32 @llvm.atomic.load.and.i32.p0i32( i32* %ptr, i32 0xFF )
6655 <i>; yields {i32}:result1 = 0xFFFFFFF0</i>
6656 %result2 = call i32 @llvm.atomic.load.or.i32.p0i32( i32* %ptr, i32 0F )
6657 <i>; yields {i32}:result2 = 0xF0</i>
6658 %result3 = call i32 @llvm.atomic.load.xor.i32.p0i32( i32* %ptr, i32 0F )
6659 <i>; yields {i32}:result3 = FF</i>
6660 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = F0</i>
6665 <!-- _______________________________________________________________________ -->
6666 <div class="doc_subsubsection">
6667 <a name="int_atomic_load_max">'<tt>llvm.atomic.load.max.*</tt>' Intrinsic</a><br>
6668 <a name="int_atomic_load_min">'<tt>llvm.atomic.load.min.*</tt>' Intrinsic</a><br>
6669 <a name="int_atomic_load_umax">'<tt>llvm.atomic.load.umax.*</tt>' Intrinsic</a><br>
6670 <a name="int_atomic_load_umin">'<tt>llvm.atomic.load.umin.*</tt>' Intrinsic</a><br>
6673 <div class="doc_text">
6676 These are overloaded intrinsics. You can use <tt>llvm.atomic.load_max</tt>,
6677 <tt>llvm.atomic.load_min</tt>, <tt>llvm.atomic.load_umax</tt>, and
6678 <tt>llvm.atomic.load_umin</tt> on any integer bit width and for different
6679 address spaces. Not all targets
6680 support all bit widths however.</p>
6682 declare i8 @llvm.atomic.load.max.i8.p0i8( i8* <ptr>, i8 <delta> )
6683 declare i16 @llvm.atomic.load.max.i16.p0i16( i16* <ptr>, i16 <delta> )
6684 declare i32 @llvm.atomic.load.max.i32.p0i32( i32* <ptr>, i32 <delta> )
6685 declare i64 @llvm.atomic.load.max.i64.p0i64( i64* <ptr>, i64 <delta> )
6690 declare i8 @llvm.atomic.load.min.i8.p0i8( i8* <ptr>, i8 <delta> )
6691 declare i16 @llvm.atomic.load.min.i16.p0i16( i16* <ptr>, i16 <delta> )
6692 declare i32 @llvm.atomic.load.min.i32..p0i32( i32* <ptr>, i32 <delta> )
6693 declare i64 @llvm.atomic.load.min.i64..p0i64( i64* <ptr>, i64 <delta> )
6698 declare i8 @llvm.atomic.load.umax.i8.p0i8( i8* <ptr>, i8 <delta> )
6699 declare i16 @llvm.atomic.load.umax.i16.p0i16( i16* <ptr>, i16 <delta> )
6700 declare i32 @llvm.atomic.load.umax.i32.p0i32( i32* <ptr>, i32 <delta> )
6701 declare i64 @llvm.atomic.load.umax.i64.p0i64( i64* <ptr>, i64 <delta> )
6706 declare i8 @llvm.atomic.load.umin.i8..p0i8( i8* <ptr>, i8 <delta> )
6707 declare i16 @llvm.atomic.load.umin.i16.p0i16( i16* <ptr>, i16 <delta> )
6708 declare i32 @llvm.atomic.load.umin.i32..p0i32( i32* <ptr>, i32 <delta> )
6709 declare i64 @llvm.atomic.load.umin.i64..p0i64( i64* <ptr>, i64 <delta> )
6714 These intrinsics takes the signed or unsigned minimum or maximum of
6715 <tt>delta</tt> and the value stored in memory at <tt>ptr</tt>. It yields the
6716 original value at <tt>ptr</tt>.
6721 These intrinsics take two arguments, the first a pointer to an integer value
6722 and the second an integer value. The result is also an integer value. These
6723 integer types can have any bit width, but they must all have the same bit
6724 width. The targets may only lower integer representations they support.
6728 These intrinsics does a series of operations atomically. They first load the
6729 value stored at <tt>ptr</tt>. They then do the signed or unsigned min or max
6730 <tt>delta</tt> and the value, store the result to <tt>ptr</tt>. They yield
6731 the original value stored at <tt>ptr</tt>.
6738 %result0 = call i32 @llvm.atomic.load.min.i32.p0i32( i32* %ptr, i32 -2 )
6739 <i>; yields {i32}:result0 = 7</i>
6740 %result1 = call i32 @llvm.atomic.load.max.i32.p0i32( i32* %ptr, i32 8 )
6741 <i>; yields {i32}:result1 = -2</i>
6742 %result2 = call i32 @llvm.atomic.load.umin.i32.p0i32( i32* %ptr, i32 10 )
6743 <i>; yields {i32}:result2 = 8</i>
6744 %result3 = call i32 @llvm.atomic.load.umax.i32.p0i32( i32* %ptr, i32 30 )
6745 <i>; yields {i32}:result3 = 8</i>
6746 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 30</i>
6750 <!-- ======================================================================= -->
6751 <div class="doc_subsection">
6752 <a name="int_general">General Intrinsics</a>
6755 <div class="doc_text">
6756 <p> This class of intrinsics is designed to be generic and has
6757 no specific purpose. </p>
6760 <!-- _______________________________________________________________________ -->
6761 <div class="doc_subsubsection">
6762 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
6765 <div class="doc_text">
6769 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
6775 The '<tt>llvm.var.annotation</tt>' intrinsic
6781 The first argument is a pointer to a value, the second is a pointer to a
6782 global string, the third is a pointer to a global string which is the source
6783 file name, and the last argument is the line number.
6789 This intrinsic allows annotation of local variables with arbitrary strings.
6790 This can be useful for special purpose optimizations that want to look for these
6791 annotations. These have no other defined use, they are ignored by code
6792 generation and optimization.
6796 <!-- _______________________________________________________________________ -->
6797 <div class="doc_subsubsection">
6798 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
6801 <div class="doc_text">
6804 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
6805 any integer bit width.
6808 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
6809 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
6810 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
6811 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
6812 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
6818 The '<tt>llvm.annotation</tt>' intrinsic.
6824 The first argument is an integer value (result of some expression),
6825 the second is a pointer to a global string, the third is a pointer to a global
6826 string which is the source file name, and the last argument is the line number.
6827 It returns the value of the first argument.
6833 This intrinsic allows annotations to be put on arbitrary expressions
6834 with arbitrary strings. This can be useful for special purpose optimizations
6835 that want to look for these annotations. These have no other defined use, they
6836 are ignored by code generation and optimization.
6840 <!-- _______________________________________________________________________ -->
6841 <div class="doc_subsubsection">
6842 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
6845 <div class="doc_text">
6849 declare void @llvm.trap()
6855 The '<tt>llvm.trap</tt>' intrinsic
6867 This intrinsics is lowered to the target dependent trap instruction. If the
6868 target does not have a trap instruction, this intrinsic will be lowered to the
6869 call of the abort() function.
6873 <!-- _______________________________________________________________________ -->
6874 <div class="doc_subsubsection">
6875 <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a>
6877 <div class="doc_text">
6880 declare void @llvm.stackprotector( i8* <guard>, i8** <slot> )
6885 The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and stores
6886 it onto the stack at <tt>slot</tt>. The stack slot is adjusted to ensure that
6887 it is placed on the stack before local variables.
6891 The <tt>llvm.stackprotector</tt> intrinsic requires two pointer arguments. The
6892 first argument is the value loaded from the stack guard
6893 <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt> that
6894 has enough space to hold the value of the guard.
6898 This intrinsic causes the prologue/epilogue inserter to force the position of
6899 the <tt>AllocaInst</tt> stack slot to be before local variables on the
6900 stack. This is to ensure that if a local variable on the stack is overwritten,
6901 it will destroy the value of the guard. When the function exits, the guard on
6902 the stack is checked against the original guard. If they're different, then
6903 the program aborts by calling the <tt>__stack_chk_fail()</tt> function.
6907 <!-- *********************************************************************** -->
6910 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
6911 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
6912 <a href="http://validator.w3.org/check/referer"><img
6913 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
6915 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
6916 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
6917 Last modified: $Date$