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="#complexconstants">Complex 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>
68 <li><a href="#metadata">Embedded Metadata</a></li>
71 <li><a href="#othervalues">Other Values</a>
73 <li><a href="#inlineasm">Inline Assembler Expressions</a></li>
76 <li><a href="#instref">Instruction Reference</a>
78 <li><a href="#terminators">Terminator Instructions</a>
80 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
81 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
82 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
83 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
84 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
85 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
88 <li><a href="#binaryops">Binary Operations</a>
90 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
91 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
92 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
93 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
94 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
95 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
96 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
97 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
98 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
101 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
103 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
104 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
105 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
106 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
107 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
108 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
111 <li><a href="#vectorops">Vector Operations</a>
113 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
114 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
115 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
118 <li><a href="#aggregateops">Aggregate Operations</a>
120 <li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li>
121 <li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li>
124 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
126 <li><a href="#i_malloc">'<tt>malloc</tt>' Instruction</a></li>
127 <li><a href="#i_free">'<tt>free</tt>' Instruction</a></li>
128 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
129 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
130 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
131 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
134 <li><a href="#convertops">Conversion Operations</a>
136 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
137 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
138 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
139 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
140 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
141 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
142 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
143 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
144 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
145 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
146 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
147 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
150 <li><a href="#otherops">Other Operations</a>
152 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
153 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
154 <li><a href="#i_vicmp">'<tt>vicmp</tt>' Instruction</a></li>
155 <li><a href="#i_vfcmp">'<tt>vfcmp</tt>' Instruction</a></li>
156 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
157 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
158 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
159 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
164 <li><a href="#intrinsics">Intrinsic Functions</a>
166 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
168 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
169 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
170 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
173 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
175 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
176 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
177 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
180 <li><a href="#int_codegen">Code Generator Intrinsics</a>
182 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
183 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
184 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
185 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
186 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
187 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
188 <li><a href="#int_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
191 <li><a href="#int_libc">Standard C Library Intrinsics</a>
193 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
194 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
195 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
196 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
197 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
198 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
199 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
200 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
203 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
205 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
206 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
207 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
208 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
209 <li><a href="#int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic </a></li>
210 <li><a href="#int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic </a></li>
213 <li><a href="#int_overflow">Arithmetic with Overflow Intrinsics</a>
215 <li><a href="#int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt> Intrinsics</a></li>
216 <li><a href="#int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt> Intrinsics</a></li>
217 <li><a href="#int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt> Intrinsics</a></li>
218 <li><a href="#int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt> Intrinsics</a></li>
219 <li><a href="#int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt> Intrinsics</a></li>
220 <li><a href="#int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt> Intrinsics</a></li>
223 <li><a href="#int_debugger">Debugger intrinsics</a></li>
224 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
225 <li><a href="#int_trampoline">Trampoline Intrinsic</a>
227 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
230 <li><a href="#int_atomics">Atomic intrinsics</a>
232 <li><a href="#int_memory_barrier"><tt>llvm.memory_barrier</tt></a></li>
233 <li><a href="#int_atomic_cmp_swap"><tt>llvm.atomic.cmp.swap</tt></a></li>
234 <li><a href="#int_atomic_swap"><tt>llvm.atomic.swap</tt></a></li>
235 <li><a href="#int_atomic_load_add"><tt>llvm.atomic.load.add</tt></a></li>
236 <li><a href="#int_atomic_load_sub"><tt>llvm.atomic.load.sub</tt></a></li>
237 <li><a href="#int_atomic_load_and"><tt>llvm.atomic.load.and</tt></a></li>
238 <li><a href="#int_atomic_load_nand"><tt>llvm.atomic.load.nand</tt></a></li>
239 <li><a href="#int_atomic_load_or"><tt>llvm.atomic.load.or</tt></a></li>
240 <li><a href="#int_atomic_load_xor"><tt>llvm.atomic.load.xor</tt></a></li>
241 <li><a href="#int_atomic_load_max"><tt>llvm.atomic.load.max</tt></a></li>
242 <li><a href="#int_atomic_load_min"><tt>llvm.atomic.load.min</tt></a></li>
243 <li><a href="#int_atomic_load_umax"><tt>llvm.atomic.load.umax</tt></a></li>
244 <li><a href="#int_atomic_load_umin"><tt>llvm.atomic.load.umin</tt></a></li>
247 <li><a href="#int_general">General intrinsics</a>
249 <li><a href="#int_var_annotation">
250 '<tt>llvm.var.annotation</tt>' Intrinsic</a></li>
251 <li><a href="#int_annotation">
252 '<tt>llvm.annotation.*</tt>' Intrinsic</a></li>
253 <li><a href="#int_trap">
254 '<tt>llvm.trap</tt>' Intrinsic</a></li>
255 <li><a href="#int_stackprotector">
256 '<tt>llvm.stackprotector</tt>' Intrinsic</a></li>
263 <div class="doc_author">
264 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
265 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
268 <!-- *********************************************************************** -->
269 <div class="doc_section"> <a name="abstract">Abstract </a></div>
270 <!-- *********************************************************************** -->
272 <div class="doc_text">
273 <p>This document is a reference manual for the LLVM assembly language.
274 LLVM is a Static Single Assignment (SSA) based representation that provides
275 type safety, low-level operations, flexibility, and the capability of
276 representing 'all' high-level languages cleanly. It is the common code
277 representation used throughout all phases of the LLVM compilation
281 <!-- *********************************************************************** -->
282 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
283 <!-- *********************************************************************** -->
285 <div class="doc_text">
287 <p>The LLVM code representation is designed to be used in three
288 different forms: as an in-memory compiler IR, as an on-disk bitcode
289 representation (suitable for fast loading by a Just-In-Time compiler),
290 and as a human readable assembly language representation. This allows
291 LLVM to provide a powerful intermediate representation for efficient
292 compiler transformations and analysis, while providing a natural means
293 to debug and visualize the transformations. The three different forms
294 of LLVM are all equivalent. This document describes the human readable
295 representation and notation.</p>
297 <p>The LLVM representation aims to be light-weight and low-level
298 while being expressive, typed, and extensible at the same time. It
299 aims to be a "universal IR" of sorts, by being at a low enough level
300 that high-level ideas may be cleanly mapped to it (similar to how
301 microprocessors are "universal IR's", allowing many source languages to
302 be mapped to them). By providing type information, LLVM can be used as
303 the target of optimizations: for example, through pointer analysis, it
304 can be proven that a C automatic variable is never accessed outside of
305 the current function... allowing it to be promoted to a simple SSA
306 value instead of a memory location.</p>
310 <!-- _______________________________________________________________________ -->
311 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
313 <div class="doc_text">
315 <p>It is important to note that this document describes 'well formed'
316 LLVM assembly language. There is a difference between what the parser
317 accepts and what is considered 'well formed'. For example, the
318 following instruction is syntactically okay, but not well formed:</p>
320 <div class="doc_code">
322 %x = <a href="#i_add">add</a> i32 1, %x
326 <p>...because the definition of <tt>%x</tt> does not dominate all of
327 its uses. The LLVM infrastructure provides a verification pass that may
328 be used to verify that an LLVM module is well formed. This pass is
329 automatically run by the parser after parsing input assembly and by
330 the optimizer before it outputs bitcode. The violations pointed out
331 by the verifier pass indicate bugs in transformation passes or input to
335 <!-- Describe the typesetting conventions here. -->
337 <!-- *********************************************************************** -->
338 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
339 <!-- *********************************************************************** -->
341 <div class="doc_text">
343 <p>LLVM identifiers come in two basic types: global and local. Global
344 identifiers (functions, global variables) begin with the @ character. Local
345 identifiers (register names, types) begin with the % character. Additionally,
346 there are three different formats for identifiers, for different purposes:</p>
349 <li>Named values are represented as a string of characters with their prefix.
350 For example, %foo, @DivisionByZero, %a.really.long.identifier. The actual
351 regular expression used is '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
352 Identifiers which require other characters in their names can be surrounded
353 with quotes. Special characters may be escaped using "\xx" where xx is the
354 ASCII code for the character in hexadecimal. In this way, any character can
355 be used in a name value, even quotes themselves.
357 <li>Unnamed values are represented as an unsigned numeric value with their
358 prefix. For example, %12, @2, %44.</li>
360 <li>Constants, which are described in a <a href="#constants">section about
361 constants</a>, below.</li>
364 <p>LLVM requires that values start with a prefix for two reasons: Compilers
365 don't need to worry about name clashes with reserved words, and the set of
366 reserved words may be expanded in the future without penalty. Additionally,
367 unnamed identifiers allow a compiler to quickly come up with a temporary
368 variable without having to avoid symbol table conflicts.</p>
370 <p>Reserved words in LLVM are very similar to reserved words in other
371 languages. There are keywords for different opcodes
372 ('<tt><a href="#i_add">add</a></tt>',
373 '<tt><a href="#i_bitcast">bitcast</a></tt>',
374 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
375 href="#t_void">void</a></tt>', '<tt><a href="#t_primitive">i32</a></tt>', etc...),
376 and others. These reserved words cannot conflict with variable names, because
377 none of them start with a prefix character ('%' or '@').</p>
379 <p>Here is an example of LLVM code to multiply the integer variable
380 '<tt>%X</tt>' by 8:</p>
384 <div class="doc_code">
386 %result = <a href="#i_mul">mul</a> i32 %X, 8
390 <p>After strength reduction:</p>
392 <div class="doc_code">
394 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
398 <p>And the hard way:</p>
400 <div class="doc_code">
402 <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
403 <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
404 %result = <a href="#i_add">add</a> i32 %1, %1
408 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
409 important lexical features of LLVM:</p>
413 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
416 <li>Unnamed temporaries are created when the result of a computation is not
417 assigned to a named value.</li>
419 <li>Unnamed temporaries are numbered sequentially</li>
423 <p>...and it also shows a convention that we follow in this document. When
424 demonstrating instructions, we will follow an instruction with a comment that
425 defines the type and name of value produced. Comments are shown in italic
430 <!-- *********************************************************************** -->
431 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
432 <!-- *********************************************************************** -->
434 <!-- ======================================================================= -->
435 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
438 <div class="doc_text">
440 <p>LLVM programs are composed of "Module"s, each of which is a
441 translation unit of the input programs. Each module consists of
442 functions, global variables, and symbol table entries. Modules may be
443 combined together with the LLVM linker, which merges function (and
444 global variable) definitions, resolves forward declarations, and merges
445 symbol table entries. Here is an example of the "hello world" module:</p>
447 <div class="doc_code">
448 <pre><i>; Declare the string constant as a global constant...</i>
449 <a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a
450 href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
452 <i>; External declaration of the puts function</i>
453 <a href="#functionstructure">declare</a> i32 @puts(i8 *) <i>; i32(i8 *)* </i>
455 <i>; Definition of main function</i>
456 define i32 @main() { <i>; i32()* </i>
457 <i>; Convert [13 x i8]* to i8 *...</i>
459 href="#i_getelementptr">getelementptr</a> [13 x i8]* @.LC0, i64 0, i64 0 <i>; i8 *</i>
461 <i>; Call puts function to write out the string to stdout...</i>
463 href="#i_call">call</a> i32 @puts(i8 * %cast210) <i>; i32</i>
465 href="#i_ret">ret</a> i32 0<br>}<br>
469 <p>This example is made up of a <a href="#globalvars">global variable</a>
470 named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
471 function, and a <a href="#functionstructure">function definition</a>
472 for "<tt>main</tt>".</p>
474 <p>In general, a module is made up of a list of global values,
475 where both functions and global variables are global values. Global values are
476 represented by a pointer to a memory location (in this case, a pointer to an
477 array of char, and a pointer to a function), and have one of the following <a
478 href="#linkage">linkage types</a>.</p>
482 <!-- ======================================================================= -->
483 <div class="doc_subsection">
484 <a name="linkage">Linkage Types</a>
487 <div class="doc_text">
490 All Global Variables and Functions have one of the following types of linkage:
495 <dt><tt><b><a name="linkage_private">private</a></b></tt>: </dt>
497 <dd>Global values with private linkage are only directly accessible by
498 objects in the current module. In particular, linking code into a module with
499 an private global value may cause the private to be renamed as necessary to
500 avoid collisions. Because the symbol is private to the module, all
501 references can be updated. This doesn't show up in any symbol table in the
505 <dt><tt><b><a name="linkage_internal">internal</a></b></tt>: </dt>
507 <dd> Similar to private, but the value shows as a local symbol (STB_LOCAL in
508 the case of ELF) in the object file. This corresponds to the notion of the
509 '<tt>static</tt>' keyword in C.
512 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
514 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
515 the same name when linkage occurs. This is typically used to implement
516 inline functions, templates, or other code which must be generated in each
517 translation unit that uses it. Unreferenced <tt>linkonce</tt> globals are
518 allowed to be discarded.
521 <dt><tt><b><a name="linkage_common">common</a></b></tt>: </dt>
523 <dd>"<tt>common</tt>" linkage is exactly the same as <tt>linkonce</tt>
524 linkage, except that unreferenced <tt>common</tt> globals may not be
525 discarded. This is used for globals that may be emitted in multiple
526 translation units, but that are not guaranteed to be emitted into every
527 translation unit that uses them. One example of this is tentative
528 definitions in C, such as "<tt>int X;</tt>" at global scope.
531 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
533 <dd>"<tt>weak</tt>" linkage is the same as <tt>common</tt> linkage, except
534 that some targets may choose to emit different assembly sequences for them
535 for target-dependent reasons. This is used for globals that are declared
536 "weak" in C source code.
539 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
541 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
542 pointer to array type. When two global variables with appending linkage are
543 linked together, the two global arrays are appended together. This is the
544 LLVM, typesafe, equivalent of having the system linker append together
545 "sections" with identical names when .o files are linked.
548 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt>: </dt>
550 <dd>The semantics of this linkage follow the ELF object file model: the
551 symbol is weak until linked, if not linked, the symbol becomes null instead
552 of being an undefined reference.
555 <dt><tt><b><a name="linkage_linkonce">linkonce_odr</a></b></tt>: </dt>
556 <dt><tt><b><a name="linkage_weak">weak_odr</a></b></tt>: </dt>
557 <dd>Some languages allow inequivalent globals to be merged, such as two
558 functions with different semantics. Other languages, such as <tt>C++</tt>,
559 ensure that only equivalent globals are ever merged (the "one definition
560 rule" - <tt>odr</tt>). Such languages can use the <tt>linkonce_odr</tt>
561 and <tt>weak_odr</tt> linkage types to indicate that the global will only
562 be merged with equivalent globals. These linkage types are otherwise the
563 same as their non-<tt>odr</tt> versions.
566 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
568 <dd>If none of the above identifiers are used, the global is externally
569 visible, meaning that it participates in linkage and can be used to resolve
570 external symbol references.
575 The next two types of linkage are targeted for Microsoft Windows platform
576 only. They are designed to support importing (exporting) symbols from (to)
577 DLLs (Dynamic Link Libraries).
581 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt>: </dt>
583 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
584 or variable via a global pointer to a pointer that is set up by the DLL
585 exporting the symbol. On Microsoft Windows targets, the pointer name is
586 formed by combining <code>__imp_</code> and the function or variable name.
589 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt>: </dt>
591 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
592 pointer to a pointer in a DLL, so that it can be referenced with the
593 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
594 name is formed by combining <code>__imp_</code> and the function or variable
600 <p>For example, since the "<tt>.LC0</tt>"
601 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
602 variable and was linked with this one, one of the two would be renamed,
603 preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
604 external (i.e., lacking any linkage declarations), they are accessible
605 outside of the current module.</p>
606 <p>It is illegal for a function <i>declaration</i>
607 to have any linkage type other than "externally visible", <tt>dllimport</tt>
608 or <tt>extern_weak</tt>.</p>
609 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt>, <tt>weak</tt>
610 or <tt>weak_odr</tt> linkages.</p>
613 <!-- ======================================================================= -->
614 <div class="doc_subsection">
615 <a name="callingconv">Calling Conventions</a>
618 <div class="doc_text">
620 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
621 and <a href="#i_invoke">invokes</a> can all have an optional calling convention
622 specified for the call. The calling convention of any pair of dynamic
623 caller/callee must match, or the behavior of the program is undefined. The
624 following calling conventions are supported by LLVM, and more may be added in
628 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
630 <dd>This calling convention (the default if no other calling convention is
631 specified) matches the target C calling conventions. This calling convention
632 supports varargs function calls and tolerates some mismatch in the declared
633 prototype and implemented declaration of the function (as does normal C).
636 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
638 <dd>This calling convention attempts to make calls as fast as possible
639 (e.g. by passing things in registers). This calling convention allows the
640 target to use whatever tricks it wants to produce fast code for the target,
641 without having to conform to an externally specified ABI (Application Binary
642 Interface). Implementations of this convention should allow arbitrary
643 <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> to be
644 supported. This calling convention does not support varargs and requires the
645 prototype of all callees to exactly match the prototype of the function
649 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
651 <dd>This calling convention attempts to make code in the caller as efficient
652 as possible under the assumption that the call is not commonly executed. As
653 such, these calls often preserve all registers so that the call does not break
654 any live ranges in the caller side. This calling convention does not support
655 varargs and requires the prototype of all callees to exactly match the
656 prototype of the function definition.
659 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
661 <dd>Any calling convention may be specified by number, allowing
662 target-specific calling conventions to be used. Target specific calling
663 conventions start at 64.
667 <p>More calling conventions can be added/defined on an as-needed basis, to
668 support pascal conventions or any other well-known target-independent
673 <!-- ======================================================================= -->
674 <div class="doc_subsection">
675 <a name="visibility">Visibility Styles</a>
678 <div class="doc_text">
681 All Global Variables and Functions have one of the following visibility styles:
685 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
687 <dd>On targets that use the ELF object file format, default visibility means
688 that the declaration is visible to other
689 modules and, in shared libraries, means that the declared entity may be
690 overridden. On Darwin, default visibility means that the declaration is
691 visible to other modules. Default visibility corresponds to "external
692 linkage" in the language.
695 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
697 <dd>Two declarations of an object with hidden visibility refer to the same
698 object if they are in the same shared object. Usually, hidden visibility
699 indicates that the symbol will not be placed into the dynamic symbol table,
700 so no other module (executable or shared library) can reference it
704 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
706 <dd>On ELF, protected visibility indicates that the symbol will be placed in
707 the dynamic symbol table, but that references within the defining module will
708 bind to the local symbol. That is, the symbol cannot be overridden by another
715 <!-- ======================================================================= -->
716 <div class="doc_subsection">
717 <a name="namedtypes">Named Types</a>
720 <div class="doc_text">
722 <p>LLVM IR allows you to specify name aliases for certain types. This can make
723 it easier to read the IR and make the IR more condensed (particularly when
724 recursive types are involved). An example of a name specification is:
727 <div class="doc_code">
729 %mytype = type { %mytype*, i32 }
733 <p>You may give a name to any <a href="#typesystem">type</a> except "<a
734 href="t_void">void</a>". Type name aliases may be used anywhere a type is
735 expected with the syntax "%mytype".</p>
737 <p>Note that type names are aliases for the structural type that they indicate,
738 and that you can therefore specify multiple names for the same type. This often
739 leads to confusing behavior when dumping out a .ll file. Since LLVM IR uses
740 structural typing, the name is not part of the type. When printing out LLVM IR,
741 the printer will pick <em>one name</em> to render all types of a particular
742 shape. This means that if you have code where two different source types end up
743 having the same LLVM type, that the dumper will sometimes print the "wrong" or
744 unexpected type. This is an important design point and isn't going to
749 <!-- ======================================================================= -->
750 <div class="doc_subsection">
751 <a name="globalvars">Global Variables</a>
754 <div class="doc_text">
756 <p>Global variables define regions of memory allocated at compilation time
757 instead of run-time. Global variables may optionally be initialized, may have
758 an explicit section to be placed in, and may have an optional explicit alignment
759 specified. A variable may be defined as "thread_local", which means that it
760 will not be shared by threads (each thread will have a separated copy of the
761 variable). A variable may be defined as a global "constant," which indicates
762 that the contents of the variable will <b>never</b> be modified (enabling better
763 optimization, allowing the global data to be placed in the read-only section of
764 an executable, etc). Note that variables that need runtime initialization
765 cannot be marked "constant" as there is a store to the variable.</p>
768 LLVM explicitly allows <em>declarations</em> of global variables to be marked
769 constant, even if the final definition of the global is not. This capability
770 can be used to enable slightly better optimization of the program, but requires
771 the language definition to guarantee that optimizations based on the
772 'constantness' are valid for the translation units that do not include the
776 <p>As SSA values, global variables define pointer values that are in
777 scope (i.e. they dominate) all basic blocks in the program. Global
778 variables always define a pointer to their "content" type because they
779 describe a region of memory, and all memory objects in LLVM are
780 accessed through pointers.</p>
782 <p>A global variable may be declared to reside in a target-specifc numbered
783 address space. For targets that support them, address spaces may affect how
784 optimizations are performed and/or what target instructions are used to access
785 the variable. The default address space is zero. The address space qualifier
786 must precede any other attributes.</p>
788 <p>LLVM allows an explicit section to be specified for globals. If the target
789 supports it, it will emit globals to the section specified.</p>
791 <p>An explicit alignment may be specified for a global. If not present, or if
792 the alignment is set to zero, the alignment of the global is set by the target
793 to whatever it feels convenient. If an explicit alignment is specified, the
794 global is forced to have at least that much alignment. All alignments must be
797 <p>For example, the following defines a global in a numbered address space with
798 an initializer, section, and alignment:</p>
800 <div class="doc_code">
802 @G = addrspace(5) constant float 1.0, section "foo", align 4
809 <!-- ======================================================================= -->
810 <div class="doc_subsection">
811 <a name="functionstructure">Functions</a>
814 <div class="doc_text">
816 <p>LLVM function definitions consist of the "<tt>define</tt>" keyord,
817 an optional <a href="#linkage">linkage type</a>, an optional
818 <a href="#visibility">visibility style</a>, an optional
819 <a href="#callingconv">calling convention</a>, a return type, an optional
820 <a href="#paramattrs">parameter attribute</a> for the return type, a function
821 name, a (possibly empty) argument list (each with optional
822 <a href="#paramattrs">parameter attributes</a>), optional
823 <a href="#fnattrs">function attributes</a>, an optional section,
824 an optional alignment, an optional <a href="#gc">garbage collector name</a>,
825 an opening curly brace, a list of basic blocks, and a closing curly brace.
827 LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
828 optional <a href="#linkage">linkage type</a>, an optional
829 <a href="#visibility">visibility style</a>, an optional
830 <a href="#callingconv">calling convention</a>, a return type, an optional
831 <a href="#paramattrs">parameter attribute</a> for the return type, a function
832 name, a possibly empty list of arguments, an optional alignment, and an optional
833 <a href="#gc">garbage collector name</a>.</p>
835 <p>A function definition contains a list of basic blocks, forming the CFG
836 (Control Flow Graph) for
837 the function. Each basic block may optionally start with a label (giving the
838 basic block a symbol table entry), contains a list of instructions, and ends
839 with a <a href="#terminators">terminator</a> instruction (such as a branch or
840 function return).</p>
842 <p>The first basic block in a function is special in two ways: it is immediately
843 executed on entrance to the function, and it is not allowed to have predecessor
844 basic blocks (i.e. there can not be any branches to the entry block of a
845 function). Because the block can have no predecessors, it also cannot have any
846 <a href="#i_phi">PHI nodes</a>.</p>
848 <p>LLVM allows an explicit section to be specified for functions. If the target
849 supports it, it will emit functions to the section specified.</p>
851 <p>An explicit alignment may be specified for a function. If not present, or if
852 the alignment is set to zero, the alignment of the function is set by the target
853 to whatever it feels convenient. If an explicit alignment is specified, the
854 function is forced to have at least that much alignment. All alignments must be
859 <div class="doc_code">
861 define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
862 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
863 <ResultType> @<FunctionName> ([argument list])
864 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
865 [<a href="#gc">gc</a>] { ... }
872 <!-- ======================================================================= -->
873 <div class="doc_subsection">
874 <a name="aliasstructure">Aliases</a>
876 <div class="doc_text">
877 <p>Aliases act as "second name" for the aliasee value (which can be either
878 function, global variable, another alias or bitcast of global value). Aliases
879 may have an optional <a href="#linkage">linkage type</a>, and an
880 optional <a href="#visibility">visibility style</a>.</p>
884 <div class="doc_code">
886 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
894 <!-- ======================================================================= -->
895 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
896 <div class="doc_text">
897 <p>The return type and each parameter of a function type may have a set of
898 <i>parameter attributes</i> associated with them. Parameter attributes are
899 used to communicate additional information about the result or parameters of
900 a function. Parameter attributes are considered to be part of the function,
901 not of the function type, so functions with different parameter attributes
902 can have the same function type.</p>
904 <p>Parameter attributes are simple keywords that follow the type specified. If
905 multiple parameter attributes are needed, they are space separated. For
908 <div class="doc_code">
910 declare i32 @printf(i8* noalias nocapture, ...)
911 declare i32 @atoi(i8 zeroext)
912 declare signext i8 @returns_signed_char()
916 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
917 <tt>readonly</tt>) come immediately after the argument list.</p>
919 <p>Currently, only the following parameter attributes are defined:</p>
921 <dt><tt>zeroext</tt></dt>
922 <dd>This indicates to the code generator that the parameter or return value
923 should be zero-extended to a 32-bit value by the caller (for a parameter)
924 or the callee (for a return value).</dd>
926 <dt><tt>signext</tt></dt>
927 <dd>This indicates to the code generator that the parameter or return value
928 should be sign-extended to a 32-bit value by the caller (for a parameter)
929 or the callee (for a return value).</dd>
931 <dt><tt>inreg</tt></dt>
932 <dd>This indicates that this parameter or return value should be treated
933 in a special target-dependent fashion during while emitting code for a
934 function call or return (usually, by putting it in a register as opposed
935 to memory, though some targets use it to distinguish between two different
936 kinds of registers). Use of this attribute is target-specific.</dd>
938 <dt><tt><a name="byval">byval</a></tt></dt>
939 <dd>This indicates that the pointer parameter should really be passed by
940 value to the function. The attribute implies that a hidden copy of the
941 pointee is made between the caller and the callee, so the callee is unable
942 to modify the value in the callee. This attribute is only valid on LLVM
943 pointer arguments. It is generally used to pass structs and arrays by
944 value, but is also valid on pointers to scalars. The copy is considered to
945 belong to the caller not the callee (for example,
946 <tt><a href="#readonly">readonly</a></tt> functions should not write to
947 <tt>byval</tt> parameters). This is not a valid attribute for return
948 values. The byval attribute also supports specifying an alignment with the
949 align attribute. This has a target-specific effect on the code generator
950 that usually indicates a desired alignment for the synthesized stack
953 <dt><tt>sret</tt></dt>
954 <dd>This indicates that the pointer parameter specifies the address of a
955 structure that is the return value of the function in the source program.
956 This pointer must be guaranteed by the caller to be valid: loads and stores
957 to the structure may be assumed by the callee to not to trap. This may only
958 be applied to the first parameter. This is not a valid attribute for
961 <dt><tt>noalias</tt></dt>
962 <dd>This indicates that the pointer does not alias any global or any other
963 parameter. The caller is responsible for ensuring that this is the
964 case. On a function return value, <tt>noalias</tt> additionally indicates
965 that the pointer does not alias any other pointers visible to the
966 caller. For further details, please see the discussion of the NoAlias
968 <a href="http://llvm.org/docs/AliasAnalysis.html#MustMayNo">alias
971 <dt><tt>nocapture</tt></dt>
972 <dd>This indicates that the callee does not make any copies of the pointer
973 that outlive the callee itself. This is not a valid attribute for return
976 <dt><tt>nest</tt></dt>
977 <dd>This indicates that the pointer parameter can be excised using the
978 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
979 attribute for return values.</dd>
984 <!-- ======================================================================= -->
985 <div class="doc_subsection">
986 <a name="gc">Garbage Collector Names</a>
989 <div class="doc_text">
990 <p>Each function may specify a garbage collector name, which is simply a
993 <div class="doc_code"><pre
994 >define void @f() gc "name" { ...</pre></div>
996 <p>The compiler declares the supported values of <i>name</i>. Specifying a
997 collector which will cause the compiler to alter its output in order to support
998 the named garbage collection algorithm.</p>
1001 <!-- ======================================================================= -->
1002 <div class="doc_subsection">
1003 <a name="fnattrs">Function Attributes</a>
1006 <div class="doc_text">
1008 <p>Function attributes are set to communicate additional information about
1009 a function. Function attributes are considered to be part of the function,
1010 not of the function type, so functions with different parameter attributes
1011 can have the same function type.</p>
1013 <p>Function attributes are simple keywords that follow the type specified. If
1014 multiple attributes are needed, they are space separated. For
1017 <div class="doc_code">
1019 define void @f() noinline { ... }
1020 define void @f() alwaysinline { ... }
1021 define void @f() alwaysinline optsize { ... }
1022 define void @f() optsize
1027 <dt><tt>alwaysinline</tt></dt>
1028 <dd>This attribute indicates that the inliner should attempt to inline this
1029 function into callers whenever possible, ignoring any active inlining size
1030 threshold for this caller.</dd>
1032 <dt><tt>noinline</tt></dt>
1033 <dd>This attribute indicates that the inliner should never inline this function
1034 in any situation. This attribute may not be used together with the
1035 <tt>alwaysinline</tt> attribute.</dd>
1037 <dt><tt>optsize</tt></dt>
1038 <dd>This attribute suggests that optimization passes and code generator passes
1039 make choices that keep the code size of this function low, and otherwise do
1040 optimizations specifically to reduce code size.</dd>
1042 <dt><tt>noreturn</tt></dt>
1043 <dd>This function attribute indicates that the function never returns normally.
1044 This produces undefined behavior at runtime if the function ever does
1045 dynamically return.</dd>
1047 <dt><tt>nounwind</tt></dt>
1048 <dd>This function attribute indicates that the function never returns with an
1049 unwind or exceptional control flow. If the function does unwind, its runtime
1050 behavior is undefined.</dd>
1052 <dt><tt>readnone</tt></dt>
1053 <dd>This attribute indicates that the function computes its result (or the
1054 exception it throws) based strictly on its arguments, without dereferencing any
1055 pointer arguments or otherwise accessing any mutable state (e.g. memory, control
1056 registers, etc) visible to caller functions. It does not write through any
1057 pointer arguments (including <tt><a href="#byval">byval</a></tt> arguments) and
1058 never changes any state visible to callers.</dd>
1060 <dt><tt><a name="readonly">readonly</a></tt></dt>
1061 <dd>This attribute indicates that the function does not write through any
1062 pointer arguments (including <tt><a href="#byval">byval</a></tt> arguments)
1063 or otherwise modify any state (e.g. memory, control registers, etc) visible to
1064 caller functions. It may dereference pointer arguments and read state that may
1065 be set in the caller. A readonly function always returns the same value (or
1066 throws the same exception) when called with the same set of arguments and global
1069 <dt><tt><a name="ssp">ssp</a></tt></dt>
1070 <dd>This attribute indicates that the function should emit a stack smashing
1071 protector. It is in the form of a "canary"—a random value placed on the
1072 stack before the local variables that's checked upon return from the function to
1073 see if it has been overwritten. A heuristic is used to determine if a function
1074 needs stack protectors or not.
1076 <p>If a function that has an <tt>ssp</tt> attribute is inlined into a function
1077 that doesn't have an <tt>ssp</tt> attribute, then the resulting function will
1078 have an <tt>ssp</tt> attribute.</p></dd>
1080 <dt><tt>sspreq</tt></dt>
1081 <dd>This attribute indicates that the function should <em>always</em> emit a
1082 stack smashing protector. This overrides the <tt><a href="#ssp">ssp</a></tt>
1085 <p>If a function that has an <tt>sspreq</tt> attribute is inlined into a
1086 function that doesn't have an <tt>sspreq</tt> attribute or which has
1087 an <tt>ssp</tt> attribute, then the resulting function will have
1088 an <tt>sspreq</tt> attribute.</p></dd>
1093 <!-- ======================================================================= -->
1094 <div class="doc_subsection">
1095 <a name="moduleasm">Module-Level Inline Assembly</a>
1098 <div class="doc_text">
1100 Modules may contain "module-level inline asm" blocks, which corresponds to the
1101 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
1102 LLVM and treated as a single unit, but may be separated in the .ll file if
1103 desired. The syntax is very simple:
1106 <div class="doc_code">
1108 module asm "inline asm code goes here"
1109 module asm "more can go here"
1113 <p>The strings can contain any character by escaping non-printable characters.
1114 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1119 The inline asm code is simply printed to the machine code .s file when
1120 assembly code is generated.
1124 <!-- ======================================================================= -->
1125 <div class="doc_subsection">
1126 <a name="datalayout">Data Layout</a>
1129 <div class="doc_text">
1130 <p>A module may specify a target specific data layout string that specifies how
1131 data is to be laid out in memory. The syntax for the data layout is simply:</p>
1132 <pre> target datalayout = "<i>layout specification</i>"</pre>
1133 <p>The <i>layout specification</i> consists of a list of specifications
1134 separated by the minus sign character ('-'). Each specification starts with a
1135 letter and may include other information after the letter to define some
1136 aspect of the data layout. The specifications accepted are as follows: </p>
1139 <dd>Specifies that the target lays out data in big-endian form. That is, the
1140 bits with the most significance have the lowest address location.</dd>
1142 <dd>Specifies that the target lays out data in little-endian form. That is,
1143 the bits with the least significance have the lowest address location.</dd>
1144 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1145 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1146 <i>preferred</i> alignments. All sizes are in bits. Specifying the <i>pref</i>
1147 alignment is optional. If omitted, the preceding <tt>:</tt> should be omitted
1149 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1150 <dd>This specifies the alignment for an integer type of a given bit
1151 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1152 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1153 <dd>This specifies the alignment for a vector type of a given bit
1155 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1156 <dd>This specifies the alignment for a floating point type of a given bit
1157 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
1159 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1160 <dd>This specifies the alignment for an aggregate type of a given bit
1163 <p>When constructing the data layout for a given target, LLVM starts with a
1164 default set of specifications which are then (possibly) overriden by the
1165 specifications in the <tt>datalayout</tt> keyword. The default specifications
1166 are given in this list:</p>
1168 <li><tt>E</tt> - big endian</li>
1169 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
1170 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1171 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1172 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1173 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1174 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1175 alignment of 64-bits</li>
1176 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1177 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1178 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1179 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1180 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1182 <p>When LLVM is determining the alignment for a given type, it uses the
1183 following rules:</p>
1185 <li>If the type sought is an exact match for one of the specifications, that
1186 specification is used.</li>
1187 <li>If no match is found, and the type sought is an integer type, then the
1188 smallest integer type that is larger than the bitwidth of the sought type is
1189 used. If none of the specifications are larger than the bitwidth then the the
1190 largest integer type is used. For example, given the default specifications
1191 above, the i7 type will use the alignment of i8 (next largest) while both
1192 i65 and i256 will use the alignment of i64 (largest specified).</li>
1193 <li>If no match is found, and the type sought is a vector type, then the
1194 largest vector type that is smaller than the sought vector type will be used
1195 as a fall back. This happens because <128 x double> can be implemented
1196 in terms of 64 <2 x double>, for example.</li>
1200 <!-- *********************************************************************** -->
1201 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
1202 <!-- *********************************************************************** -->
1204 <div class="doc_text">
1206 <p>The LLVM type system is one of the most important features of the
1207 intermediate representation. Being typed enables a number of
1208 optimizations to be performed on the intermediate representation directly,
1209 without having to do
1210 extra analyses on the side before the transformation. A strong type
1211 system makes it easier to read the generated code and enables novel
1212 analyses and transformations that are not feasible to perform on normal
1213 three address code representations.</p>
1217 <!-- ======================================================================= -->
1218 <div class="doc_subsection"> <a name="t_classifications">Type
1219 Classifications</a> </div>
1220 <div class="doc_text">
1221 <p>The types fall into a few useful
1222 classifications:</p>
1224 <table border="1" cellspacing="0" cellpadding="4">
1226 <tr><th>Classification</th><th>Types</th></tr>
1228 <td><a href="#t_integer">integer</a></td>
1229 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1232 <td><a href="#t_floating">floating point</a></td>
1233 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1236 <td><a name="t_firstclass">first class</a></td>
1237 <td><a href="#t_integer">integer</a>,
1238 <a href="#t_floating">floating point</a>,
1239 <a href="#t_pointer">pointer</a>,
1240 <a href="#t_vector">vector</a>,
1241 <a href="#t_struct">structure</a>,
1242 <a href="#t_array">array</a>,
1243 <a href="#t_label">label</a>.
1247 <td><a href="#t_primitive">primitive</a></td>
1248 <td><a href="#t_label">label</a>,
1249 <a href="#t_void">void</a>,
1250 <a href="#t_floating">floating point</a>.</td>
1253 <td><a href="#t_derived">derived</a></td>
1254 <td><a href="#t_integer">integer</a>,
1255 <a href="#t_array">array</a>,
1256 <a href="#t_function">function</a>,
1257 <a href="#t_pointer">pointer</a>,
1258 <a href="#t_struct">structure</a>,
1259 <a href="#t_pstruct">packed structure</a>,
1260 <a href="#t_vector">vector</a>,
1261 <a href="#t_opaque">opaque</a>.
1267 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
1268 most important. Values of these types are the only ones which can be
1269 produced by instructions, passed as arguments, or used as operands to
1273 <!-- ======================================================================= -->
1274 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
1276 <div class="doc_text">
1277 <p>The primitive types are the fundamental building blocks of the LLVM
1282 <!-- _______________________________________________________________________ -->
1283 <div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1285 <div class="doc_text">
1288 <tr><th>Type</th><th>Description</th></tr>
1289 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1290 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1291 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1292 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1293 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1298 <!-- _______________________________________________________________________ -->
1299 <div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1301 <div class="doc_text">
1303 <p>The void type does not represent any value and has no size.</p>
1312 <!-- _______________________________________________________________________ -->
1313 <div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1315 <div class="doc_text">
1317 <p>The label type represents code labels.</p>
1327 <!-- ======================================================================= -->
1328 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1330 <div class="doc_text">
1332 <p>The real power in LLVM comes from the derived types in the system.
1333 This is what allows a programmer to represent arrays, functions,
1334 pointers, and other useful types. Note that these derived types may be
1335 recursive: For example, it is possible to have a two dimensional array.</p>
1339 <!-- _______________________________________________________________________ -->
1340 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1342 <div class="doc_text">
1345 <p>The integer type is a very simple derived type that simply specifies an
1346 arbitrary bit width for the integer type desired. Any bit width from 1 bit to
1347 2^23-1 (about 8 million) can be specified.</p>
1355 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1359 <table class="layout">
1362 <td><tt>i1</tt></td>
1363 <td>a single-bit integer.</td>
1365 <td><tt>i32</tt></td>
1366 <td>a 32-bit integer.</td>
1368 <td><tt>i1942652</tt></td>
1369 <td>a really big integer of over 1 million bits.</td>
1374 <p>Note that the code generator does not yet support large integer types
1375 to be used as function return types. The specific limit on how large a
1376 return type the code generator can currently handle is target-dependent;
1377 currently it's often 64 bits for 32-bit targets and 128 bits for 64-bit
1382 <!-- _______________________________________________________________________ -->
1383 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1385 <div class="doc_text">
1389 <p>The array type is a very simple derived type that arranges elements
1390 sequentially in memory. The array type requires a size (number of
1391 elements) and an underlying data type.</p>
1396 [<# elements> x <elementtype>]
1399 <p>The number of elements is a constant integer value; elementtype may
1400 be any type with a size.</p>
1403 <table class="layout">
1405 <td class="left"><tt>[40 x i32]</tt></td>
1406 <td class="left">Array of 40 32-bit integer values.</td>
1409 <td class="left"><tt>[41 x i32]</tt></td>
1410 <td class="left">Array of 41 32-bit integer values.</td>
1413 <td class="left"><tt>[4 x i8]</tt></td>
1414 <td class="left">Array of 4 8-bit integer values.</td>
1417 <p>Here are some examples of multidimensional arrays:</p>
1418 <table class="layout">
1420 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1421 <td class="left">3x4 array of 32-bit integer values.</td>
1424 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1425 <td class="left">12x10 array of single precision floating point values.</td>
1428 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1429 <td class="left">2x3x4 array of 16-bit integer values.</td>
1433 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1434 length array. Normally, accesses past the end of an array are undefined in
1435 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
1436 As a special case, however, zero length arrays are recognized to be variable
1437 length. This allows implementation of 'pascal style arrays' with the LLVM
1438 type "{ i32, [0 x float]}", for example.</p>
1440 <p>Note that the code generator does not yet support large aggregate types
1441 to be used as function return types. The specific limit on how large an
1442 aggregate return type the code generator can currently handle is
1443 target-dependent, and also dependent on the aggregate element types.</p>
1447 <!-- _______________________________________________________________________ -->
1448 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1449 <div class="doc_text">
1453 <p>The function type can be thought of as a function signature. It
1454 consists of a return type and a list of formal parameter types. The
1455 return type of a function type is a scalar type, a void type, or a struct type.
1456 If the return type is a struct type then all struct elements must be of first
1457 class types, and the struct must have at least one element.</p>
1462 <returntype list> (<parameter list>)
1465 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1466 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1467 which indicates that the function takes a variable number of arguments.
1468 Variable argument functions can access their arguments with the <a
1469 href="#int_varargs">variable argument handling intrinsic</a> functions.
1470 '<tt><returntype list></tt>' is a comma-separated list of
1471 <a href="#t_firstclass">first class</a> type specifiers.</p>
1474 <table class="layout">
1476 <td class="left"><tt>i32 (i32)</tt></td>
1477 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1479 </tr><tr class="layout">
1480 <td class="left"><tt>float (i16 signext, i32 *) *
1482 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1483 an <tt>i16</tt> that should be sign extended and a
1484 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1487 </tr><tr class="layout">
1488 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1489 <td class="left">A vararg function that takes at least one
1490 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1491 which returns an integer. This is the signature for <tt>printf</tt> in
1494 </tr><tr class="layout">
1495 <td class="left"><tt>{i32, i32} (i32)</tt></td>
1496 <td class="left">A function taking an <tt>i32</tt>, returning two
1497 <tt>i32</tt> values as an aggregate of type <tt>{ i32, i32 }</tt>
1503 <!-- _______________________________________________________________________ -->
1504 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1505 <div class="doc_text">
1507 <p>The structure type is used to represent a collection of data members
1508 together in memory. The packing of the field types is defined to match
1509 the ABI of the underlying processor. The elements of a structure may
1510 be any type that has a size.</p>
1511 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1512 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1513 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1516 <pre> { <type list> }<br></pre>
1518 <table class="layout">
1520 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1521 <td class="left">A triple of three <tt>i32</tt> values</td>
1522 </tr><tr class="layout">
1523 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1524 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1525 second element is a <a href="#t_pointer">pointer</a> to a
1526 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1527 an <tt>i32</tt>.</td>
1531 <p>Note that the code generator does not yet support large aggregate types
1532 to be used as function return types. The specific limit on how large an
1533 aggregate return type the code generator can currently handle is
1534 target-dependent, and also dependent on the aggregate element types.</p>
1538 <!-- _______________________________________________________________________ -->
1539 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1541 <div class="doc_text">
1543 <p>The packed structure type is used to represent a collection of data members
1544 together in memory. There is no padding between fields. Further, the alignment
1545 of a packed structure is 1 byte. The elements of a packed structure may
1546 be any type that has a size.</p>
1547 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1548 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1549 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1552 <pre> < { <type list> } > <br></pre>
1554 <table class="layout">
1556 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1557 <td class="left">A triple of three <tt>i32</tt> values</td>
1558 </tr><tr class="layout">
1560 <tt>< { float, i32 (i32)* } ></tt></td>
1561 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1562 second element is a <a href="#t_pointer">pointer</a> to a
1563 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1564 an <tt>i32</tt>.</td>
1569 <!-- _______________________________________________________________________ -->
1570 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1571 <div class="doc_text">
1573 <p>As in many languages, the pointer type represents a pointer or
1574 reference to another object, which must live in memory. Pointer types may have
1575 an optional address space attribute defining the target-specific numbered
1576 address space where the pointed-to object resides. The default address space is
1579 <p>Note that LLVM does not permit pointers to void (<tt>void*</tt>) nor does
1580 it permit pointers to labels (<tt>label*</tt>). Use <tt>i8*</tt> instead.</p>
1583 <pre> <type> *<br></pre>
1585 <table class="layout">
1587 <td class="left"><tt>[4 x i32]*</tt></td>
1588 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1589 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1592 <td class="left"><tt>i32 (i32 *) *</tt></td>
1593 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
1594 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1598 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1599 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1600 that resides in address space #5.</td>
1605 <!-- _______________________________________________________________________ -->
1606 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1607 <div class="doc_text">
1611 <p>A vector type is a simple derived type that represents a vector
1612 of elements. Vector types are used when multiple primitive data
1613 are operated in parallel using a single instruction (SIMD).
1614 A vector type requires a size (number of
1615 elements) and an underlying primitive data type. Vectors must have a power
1616 of two length (1, 2, 4, 8, 16 ...). Vector types are
1617 considered <a href="#t_firstclass">first class</a>.</p>
1622 < <# elements> x <elementtype> >
1625 <p>The number of elements is a constant integer value; elementtype may
1626 be any integer or floating point type.</p>
1630 <table class="layout">
1632 <td class="left"><tt><4 x i32></tt></td>
1633 <td class="left">Vector of 4 32-bit integer values.</td>
1636 <td class="left"><tt><8 x float></tt></td>
1637 <td class="left">Vector of 8 32-bit floating-point values.</td>
1640 <td class="left"><tt><2 x i64></tt></td>
1641 <td class="left">Vector of 2 64-bit integer values.</td>
1645 <p>Note that the code generator does not yet support large vector types
1646 to be used as function return types. The specific limit on how large a
1647 vector return type codegen can currently handle is target-dependent;
1648 currently it's often a few times longer than a hardware vector register.</p>
1652 <!-- _______________________________________________________________________ -->
1653 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1654 <div class="doc_text">
1658 <p>Opaque types are used to represent unknown types in the system. This
1659 corresponds (for example) to the C notion of a forward declared structure type.
1660 In LLVM, opaque types can eventually be resolved to any type (not just a
1661 structure type).</p>
1671 <table class="layout">
1673 <td class="left"><tt>opaque</tt></td>
1674 <td class="left">An opaque type.</td>
1679 <!-- ======================================================================= -->
1680 <div class="doc_subsection">
1681 <a name="t_uprefs">Type Up-references</a>
1684 <div class="doc_text">
1687 An "up reference" allows you to refer to a lexically enclosing type without
1688 requiring it to have a name. For instance, a structure declaration may contain a
1689 pointer to any of the types it is lexically a member of. Example of up
1690 references (with their equivalent as named type declarations) include:</p>
1693 { \2 * } %x = type { %x* }
1694 { \2 }* %y = type { %y }*
1699 An up reference is needed by the asmprinter for printing out cyclic types when
1700 there is no declared name for a type in the cycle. Because the asmprinter does
1701 not want to print out an infinite type string, it needs a syntax to handle
1702 recursive types that have no names (all names are optional in llvm IR).
1711 The level is the count of the lexical type that is being referred to.
1716 <table class="layout">
1718 <td class="left"><tt>\1*</tt></td>
1719 <td class="left">Self-referential pointer.</td>
1722 <td class="left"><tt>{ { \3*, i8 }, i32 }</tt></td>
1723 <td class="left">Recursive structure where the upref refers to the out-most
1730 <!-- *********************************************************************** -->
1731 <div class="doc_section"> <a name="constants">Constants</a> </div>
1732 <!-- *********************************************************************** -->
1734 <div class="doc_text">
1736 <p>LLVM has several different basic types of constants. This section describes
1737 them all and their syntax.</p>
1741 <!-- ======================================================================= -->
1742 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1744 <div class="doc_text">
1747 <dt><b>Boolean constants</b></dt>
1749 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1750 constants of the <tt><a href="#t_primitive">i1</a></tt> type.
1753 <dt><b>Integer constants</b></dt>
1755 <dd>Standard integers (such as '4') are constants of the <a
1756 href="#t_integer">integer</a> type. Negative numbers may be used with
1760 <dt><b>Floating point constants</b></dt>
1762 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1763 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1764 notation (see below). The assembler requires the exact decimal value of
1765 a floating-point constant. For example, the assembler accepts 1.25 but
1766 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
1767 constants must have a <a href="#t_floating">floating point</a> type. </dd>
1769 <dt><b>Null pointer constants</b></dt>
1771 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1772 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1776 <p>The one non-intuitive notation for constants is the hexadecimal form
1777 of floating point constants. For example, the form '<tt>double
1778 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1779 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1780 (and the only time that they are generated by the disassembler) is when a
1781 floating point constant must be emitted but it cannot be represented as a
1782 decimal floating point number in a reasonable number of digits. For example,
1783 NaN's, infinities, and other
1784 special values are represented in their IEEE hexadecimal format so that
1785 assembly and disassembly do not cause any bits to change in the constants.</p>
1786 <p>When using the hexadecimal form, constants of types float and double are
1787 represented using the 16-digit form shown above (which matches the IEEE754
1788 representation for double); float values must, however, be exactly representable
1789 as IEE754 single precision.
1790 Hexadecimal format is always used for long
1791 double, and there are three forms of long double. The 80-bit
1792 format used by x86 is represented as <tt>0xK</tt>
1793 followed by 20 hexadecimal digits.
1794 The 128-bit format used by PowerPC (two adjacent doubles) is represented
1795 by <tt>0xM</tt> followed by 32 hexadecimal digits. The IEEE 128-bit
1796 format is represented
1797 by <tt>0xL</tt> followed by 32 hexadecimal digits; no currently supported
1798 target uses this format. Long doubles will only work if they match
1799 the long double format on your target. All hexadecimal formats are big-endian
1800 (sign bit at the left).</p>
1803 <!-- ======================================================================= -->
1804 <div class="doc_subsection">
1805 <a name="aggregateconstants"> <!-- old anchor -->
1806 <a name="complexconstants">Complex Constants</a></a>
1809 <div class="doc_text">
1810 <p>Complex constants are a (potentially recursive) combination of simple
1811 constants and smaller complex constants.</p>
1814 <dt><b>Structure constants</b></dt>
1816 <dd>Structure constants are represented with notation similar to structure
1817 type definitions (a comma separated list of elements, surrounded by braces
1818 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
1819 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>". Structure constants
1820 must have <a href="#t_struct">structure type</a>, and the number and
1821 types of elements must match those specified by the type.
1824 <dt><b>Array constants</b></dt>
1826 <dd>Array constants are represented with notation similar to array type
1827 definitions (a comma separated list of elements, surrounded by square brackets
1828 (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74 ]</tt>". Array
1829 constants must have <a href="#t_array">array type</a>, and the number and
1830 types of elements must match those specified by the type.
1833 <dt><b>Vector constants</b></dt>
1835 <dd>Vector constants are represented with notation similar to vector type
1836 definitions (a comma separated list of elements, surrounded by
1837 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32 42,
1838 i32 11, i32 74, i32 100 ></tt>". Vector constants must have <a
1839 href="#t_vector">vector type</a>, and the number and types of elements must
1840 match those specified by the type.
1843 <dt><b>Zero initialization</b></dt>
1845 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1846 value to zero of <em>any</em> type, including scalar and aggregate types.
1847 This is often used to avoid having to print large zero initializers (e.g. for
1848 large arrays) and is always exactly equivalent to using explicit zero
1852 <dt><b>Metadata node</b></dt>
1854 <dd>A metadata node is a structure-like constant with the type of an empty
1855 struct. For example: "<tt>{ } !{ i32 0, { } !"test" }</tt>". Unlike other
1856 constants that are meant to be interpreted as part of the instruction stream,
1857 metadata is a place to attach additional information such as debug info.
1863 <!-- ======================================================================= -->
1864 <div class="doc_subsection">
1865 <a name="globalconstants">Global Variable and Function Addresses</a>
1868 <div class="doc_text">
1870 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1871 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1872 constants. These constants are explicitly referenced when the <a
1873 href="#identifiers">identifier for the global</a> is used and always have <a
1874 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1877 <div class="doc_code">
1881 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1887 <!-- ======================================================================= -->
1888 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1889 <div class="doc_text">
1890 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1891 no specific value. Undefined values may be of any type and be used anywhere
1892 a constant is permitted.</p>
1894 <p>Undefined values indicate to the compiler that the program is well defined
1895 no matter what value is used, giving the compiler more freedom to optimize.
1899 <!-- ======================================================================= -->
1900 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1903 <div class="doc_text">
1905 <p>Constant expressions are used to allow expressions involving other constants
1906 to be used as constants. Constant expressions may be of any <a
1907 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1908 that does not have side effects (e.g. load and call are not supported). The
1909 following is the syntax for constant expressions:</p>
1912 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1913 <dd>Truncate a constant to another type. The bit size of CST must be larger
1914 than the bit size of TYPE. Both types must be integers.</dd>
1916 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1917 <dd>Zero extend a constant to another type. The bit size of CST must be
1918 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1920 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1921 <dd>Sign extend a constant to another type. The bit size of CST must be
1922 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1924 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1925 <dd>Truncate a floating point constant to another floating point type. The
1926 size of CST must be larger than the size of TYPE. Both types must be
1927 floating point.</dd>
1929 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1930 <dd>Floating point extend a constant to another type. The size of CST must be
1931 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1933 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
1934 <dd>Convert a floating point constant to the corresponding unsigned integer
1935 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1936 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1937 of the same number of elements. If the value won't fit in the integer type,
1938 the results are undefined.</dd>
1940 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
1941 <dd>Convert a floating point constant to the corresponding signed integer
1942 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1943 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1944 of the same number of elements. If the value won't fit in the integer type,
1945 the results are undefined.</dd>
1947 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
1948 <dd>Convert an unsigned integer constant to the corresponding floating point
1949 constant. TYPE must be a scalar or vector floating point type. CST must be of
1950 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1951 of the same number of elements. If the value won't fit in the floating point
1952 type, the results are undefined.</dd>
1954 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
1955 <dd>Convert a signed integer constant to the corresponding floating point
1956 constant. TYPE must be a scalar or vector floating point type. CST must be of
1957 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1958 of the same number of elements. If the value won't fit in the floating point
1959 type, the results are undefined.</dd>
1961 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
1962 <dd>Convert a pointer typed constant to the corresponding integer constant
1963 TYPE must be an integer type. CST must be of pointer type. The CST value is
1964 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1966 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
1967 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1968 pointer type. CST must be of integer type. The CST value is zero extended,
1969 truncated, or unchanged to make it fit in a pointer size. This one is
1970 <i>really</i> dangerous!</dd>
1972 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
1973 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
1974 are the same as those for the <a href="#i_bitcast">bitcast
1975 instruction</a>.</dd>
1977 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1979 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1980 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1981 instruction, the index list may have zero or more indexes, which are required
1982 to make sense for the type of "CSTPTR".</dd>
1984 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1986 <dd>Perform the <a href="#i_select">select operation</a> on
1989 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
1990 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
1992 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
1993 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
1995 <dt><b><tt>vicmp COND ( VAL1, VAL2 )</tt></b></dt>
1996 <dd>Performs the <a href="#i_vicmp">vicmp operation</a> on constants.</dd>
1998 <dt><b><tt>vfcmp COND ( VAL1, VAL2 )</tt></b></dt>
1999 <dd>Performs the <a href="#i_vfcmp">vfcmp operation</a> on constants.</dd>
2001 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
2003 <dd>Perform the <a href="#i_extractelement">extractelement
2004 operation</a> on constants.</dd>
2006 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
2008 <dd>Perform the <a href="#i_insertelement">insertelement
2009 operation</a> on constants.</dd>
2012 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
2014 <dd>Perform the <a href="#i_shufflevector">shufflevector
2015 operation</a> on constants.</dd>
2017 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
2019 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2020 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
2021 binary</a> operations. The constraints on operands are the same as those for
2022 the corresponding instruction (e.g. no bitwise operations on floating point
2023 values are allowed).</dd>
2027 <!-- ======================================================================= -->
2028 <div class="doc_subsection"><a name="metadata">Embedded Metadata</a>
2031 <div class="doc_text">
2033 <p>Embedded metadata provides a way to attach arbitrary data to the
2034 instruction stream without affecting the behaviour of the program. There are
2035 two metadata primitives, strings and nodes. All metadata has the type of an
2036 empty struct and is identified in syntax by a preceding exclamation point
2040 <p>A metadata string is a string surrounded by double quotes. It can contain
2041 any character by escaping non-printable characters with "\xx" where "xx" is
2042 the two digit hex code. For example: "<tt>!"test\00"</tt>".
2045 <p>Metadata nodes are represented with notation similar to structure constants
2046 (a comma separated list of elements, surrounded by braces and preceeded by an
2047 exclamation point). For example: "<tt>!{ { } !"test\00", i32 10}</tt>".
2050 <p>Optimizations may rely on metadata to provide additional information about
2051 the program that isn't available in the instructions, or that isn't easily
2052 computable. Similarly, the code generator may expect a certain metadata format
2053 to be used to express debugging information.</p>
2056 <!-- *********************************************************************** -->
2057 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
2058 <!-- *********************************************************************** -->
2060 <!-- ======================================================================= -->
2061 <div class="doc_subsection">
2062 <a name="inlineasm">Inline Assembler Expressions</a>
2065 <div class="doc_text">
2068 LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
2069 Module-Level Inline Assembly</a>) through the use of a special value. This
2070 value represents the inline assembler as a string (containing the instructions
2071 to emit), a list of operand constraints (stored as a string), and a flag that
2072 indicates whether or not the inline asm expression has side effects. An example
2073 inline assembler expression is:
2076 <div class="doc_code">
2078 i32 (i32) asm "bswap $0", "=r,r"
2083 Inline assembler expressions may <b>only</b> be used as the callee operand of
2084 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
2087 <div class="doc_code">
2089 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
2094 Inline asms with side effects not visible in the constraint list must be marked
2095 as having side effects. This is done through the use of the
2096 '<tt>sideeffect</tt>' keyword, like so:
2099 <div class="doc_code">
2101 call void asm sideeffect "eieio", ""()
2105 <p>TODO: The format of the asm and constraints string still need to be
2106 documented here. Constraints on what can be done (e.g. duplication, moving, etc
2107 need to be documented). This is probably best done by reference to another
2108 document that covers inline asm from a holistic perspective.
2113 <!-- *********************************************************************** -->
2114 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
2115 <!-- *********************************************************************** -->
2117 <div class="doc_text">
2119 <p>The LLVM instruction set consists of several different
2120 classifications of instructions: <a href="#terminators">terminator
2121 instructions</a>, <a href="#binaryops">binary instructions</a>,
2122 <a href="#bitwiseops">bitwise binary instructions</a>, <a
2123 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
2124 instructions</a>.</p>
2128 <!-- ======================================================================= -->
2129 <div class="doc_subsection"> <a name="terminators">Terminator
2130 Instructions</a> </div>
2132 <div class="doc_text">
2134 <p>As mentioned <a href="#functionstructure">previously</a>, every
2135 basic block in a program ends with a "Terminator" instruction, which
2136 indicates which block should be executed after the current block is
2137 finished. These terminator instructions typically yield a '<tt>void</tt>'
2138 value: they produce control flow, not values (the one exception being
2139 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
2140 <p>There are six different terminator instructions: the '<a
2141 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
2142 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
2143 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
2144 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
2145 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
2149 <!-- _______________________________________________________________________ -->
2150 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
2151 Instruction</a> </div>
2152 <div class="doc_text">
2155 ret <type> <value> <i>; Return a value from a non-void function</i>
2156 ret void <i>; Return from void function</i>
2161 <p>The '<tt>ret</tt>' instruction is used to return control flow (and
2162 optionally a value) from a function back to the caller.</p>
2163 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
2164 returns a value and then causes control flow, and one that just causes
2165 control flow to occur.</p>
2169 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument,
2170 the return value. The type of the return value must be a
2171 '<a href="#t_firstclass">first class</a>' type.</p>
2173 <p>A function is not <a href="#wellformed">well formed</a> if
2174 it it has a non-void return type and contains a '<tt>ret</tt>'
2175 instruction with no return value or a return value with a type that
2176 does not match its type, or if it has a void return type and contains
2177 a '<tt>ret</tt>' instruction with a return value.</p>
2181 <p>When the '<tt>ret</tt>' instruction is executed, control flow
2182 returns back to the calling function's context. If the caller is a "<a
2183 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
2184 the instruction after the call. If the caller was an "<a
2185 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
2186 at the beginning of the "normal" destination block. If the instruction
2187 returns a value, that value shall set the call or invoke instruction's
2193 ret i32 5 <i>; Return an integer value of 5</i>
2194 ret void <i>; Return from a void function</i>
2195 ret { i32, i8 } { i32 4, i8 2 } <i>; Return a struct of values 4 and 2</i>
2198 <p>Note that the code generator does not yet fully support large
2199 return values. The specific sizes that are currently supported are
2200 dependent on the target. For integers, on 32-bit targets the limit
2201 is often 64 bits, and on 64-bit targets the limit is often 128 bits.
2202 For aggregate types, the current limits are dependent on the element
2203 types; for example targets are often limited to 2 total integer
2204 elements and 2 total floating-point elements.</p>
2207 <!-- _______________________________________________________________________ -->
2208 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
2209 <div class="doc_text">
2211 <pre> br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
2214 <p>The '<tt>br</tt>' instruction is used to cause control flow to
2215 transfer to a different basic block in the current function. There are
2216 two forms of this instruction, corresponding to a conditional branch
2217 and an unconditional branch.</p>
2219 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
2220 single '<tt>i1</tt>' value and two '<tt>label</tt>' values. The
2221 unconditional form of the '<tt>br</tt>' instruction takes a single
2222 '<tt>label</tt>' value as a target.</p>
2224 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
2225 argument is evaluated. If the value is <tt>true</tt>, control flows
2226 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
2227 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
2229 <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
2230 href="#i_ret">ret</a> i32 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> i32 0<br></pre>
2232 <!-- _______________________________________________________________________ -->
2233 <div class="doc_subsubsection">
2234 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
2237 <div class="doc_text">
2241 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
2246 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
2247 several different places. It is a generalization of the '<tt>br</tt>'
2248 instruction, allowing a branch to occur to one of many possible
2254 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
2255 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
2256 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
2257 table is not allowed to contain duplicate constant entries.</p>
2261 <p>The <tt>switch</tt> instruction specifies a table of values and
2262 destinations. When the '<tt>switch</tt>' instruction is executed, this
2263 table is searched for the given value. If the value is found, control flow is
2264 transfered to the corresponding destination; otherwise, control flow is
2265 transfered to the default destination.</p>
2267 <h5>Implementation:</h5>
2269 <p>Depending on properties of the target machine and the particular
2270 <tt>switch</tt> instruction, this instruction may be code generated in different
2271 ways. For example, it could be generated as a series of chained conditional
2272 branches or with a lookup table.</p>
2277 <i>; Emulate a conditional br instruction</i>
2278 %Val = <a href="#i_zext">zext</a> i1 %value to i32
2279 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
2281 <i>; Emulate an unconditional br instruction</i>
2282 switch i32 0, label %dest [ ]
2284 <i>; Implement a jump table:</i>
2285 switch i32 %val, label %otherwise [ i32 0, label %onzero
2287 i32 2, label %ontwo ]
2291 <!-- _______________________________________________________________________ -->
2292 <div class="doc_subsubsection">
2293 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
2296 <div class="doc_text">
2301 <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>]
2302 to label <normal label> unwind label <exception label>
2307 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
2308 function, with the possibility of control flow transfer to either the
2309 '<tt>normal</tt>' label or the
2310 '<tt>exception</tt>' label. If the callee function returns with the
2311 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
2312 "normal" label. If the callee (or any indirect callees) returns with the "<a
2313 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
2314 continued at the dynamically nearest "exception" label.</p>
2318 <p>This instruction requires several arguments:</p>
2322 The optional "cconv" marker indicates which <a href="#callingconv">calling
2323 convention</a> the call should use. If none is specified, the call defaults
2324 to using C calling conventions.
2327 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
2328 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>',
2329 and '<tt>inreg</tt>' attributes are valid here.</li>
2331 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
2332 function value being invoked. In most cases, this is a direct function
2333 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
2334 an arbitrary pointer to function value.
2337 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
2338 function to be invoked. </li>
2340 <li>'<tt>function args</tt>': argument list whose types match the function
2341 signature argument types. If the function signature indicates the function
2342 accepts a variable number of arguments, the extra arguments can be
2345 <li>'<tt>normal label</tt>': the label reached when the called function
2346 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
2348 <li>'<tt>exception label</tt>': the label reached when a callee returns with
2349 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
2351 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
2352 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
2353 '<tt>readnone</tt>' attributes are valid here.</li>
2358 <p>This instruction is designed to operate as a standard '<tt><a
2359 href="#i_call">call</a></tt>' instruction in most regards. The primary
2360 difference is that it establishes an association with a label, which is used by
2361 the runtime library to unwind the stack.</p>
2363 <p>This instruction is used in languages with destructors to ensure that proper
2364 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
2365 exception. Additionally, this is important for implementation of
2366 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
2370 %retval = invoke i32 @Test(i32 15) to label %Continue
2371 unwind label %TestCleanup <i>; {i32}:retval set</i>
2372 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
2373 unwind label %TestCleanup <i>; {i32}:retval set</i>
2378 <!-- _______________________________________________________________________ -->
2380 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
2381 Instruction</a> </div>
2383 <div class="doc_text">
2392 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
2393 at the first callee in the dynamic call stack which used an <a
2394 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
2395 primarily used to implement exception handling.</p>
2399 <p>The '<tt>unwind</tt>' instruction causes execution of the current function to
2400 immediately halt. The dynamic call stack is then searched for the first <a
2401 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
2402 execution continues at the "exceptional" destination block specified by the
2403 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
2404 dynamic call chain, undefined behavior results.</p>
2407 <!-- _______________________________________________________________________ -->
2409 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
2410 Instruction</a> </div>
2412 <div class="doc_text">
2421 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
2422 instruction is used to inform the optimizer that a particular portion of the
2423 code is not reachable. This can be used to indicate that the code after a
2424 no-return function cannot be reached, and other facts.</p>
2428 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
2433 <!-- ======================================================================= -->
2434 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
2435 <div class="doc_text">
2436 <p>Binary operators are used to do most of the computation in a
2437 program. They require two operands of the same type, execute an operation on them, and
2438 produce a single value. The operands might represent
2439 multiple data, as is the case with the <a href="#t_vector">vector</a> data type.
2440 The result value has the same type as its operands.</p>
2441 <p>There are several different binary operators:</p>
2443 <!-- _______________________________________________________________________ -->
2444 <div class="doc_subsubsection">
2445 <a name="i_add">'<tt>add</tt>' Instruction</a>
2448 <div class="doc_text">
2453 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2458 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
2462 <p>The two arguments to the '<tt>add</tt>' instruction must be <a
2463 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>, or
2464 <a href="#t_vector">vector</a> values. Both arguments must have identical
2469 <p>The value produced is the integer or floating point sum of the two
2472 <p>If an integer sum has unsigned overflow, the result returned is the
2473 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2476 <p>Because LLVM integers use a two's complement representation, this
2477 instruction is appropriate for both signed and unsigned integers.</p>
2482 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
2485 <!-- _______________________________________________________________________ -->
2486 <div class="doc_subsubsection">
2487 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
2490 <div class="doc_text">
2495 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2500 <p>The '<tt>sub</tt>' instruction returns the difference of its two
2503 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
2504 '<tt>neg</tt>' instruction present in most other intermediate
2505 representations.</p>
2509 <p>The two arguments to the '<tt>sub</tt>' instruction must be <a
2510 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>,
2511 or <a href="#t_vector">vector</a> values. Both arguments must have identical
2516 <p>The value produced is the integer or floating point difference of
2517 the two operands.</p>
2519 <p>If an integer difference has unsigned overflow, the result returned is the
2520 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2523 <p>Because LLVM integers use a two's complement representation, this
2524 instruction is appropriate for both signed and unsigned integers.</p>
2528 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
2529 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
2533 <!-- _______________________________________________________________________ -->
2534 <div class="doc_subsubsection">
2535 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
2538 <div class="doc_text">
2541 <pre> <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2544 <p>The '<tt>mul</tt>' instruction returns the product of its two
2549 <p>The two arguments to the '<tt>mul</tt>' instruction must be <a
2550 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>,
2551 or <a href="#t_vector">vector</a> values. Both arguments must have identical
2556 <p>The value produced is the integer or floating point product of the
2559 <p>If the result of an integer multiplication has unsigned overflow,
2560 the result returned is the mathematical result modulo
2561 2<sup>n</sup>, where n is the bit width of the result.</p>
2562 <p>Because LLVM integers use a two's complement representation, and the
2563 result is the same width as the operands, this instruction returns the
2564 correct result for both signed and unsigned integers. If a full product
2565 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands
2566 should be sign-extended or zero-extended as appropriate to the
2567 width of the full product.</p>
2569 <pre> <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
2573 <!-- _______________________________________________________________________ -->
2574 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
2576 <div class="doc_text">
2578 <pre> <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2581 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two
2586 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
2587 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2588 values. Both arguments must have identical types.</p>
2592 <p>The value produced is the unsigned integer quotient of the two operands.</p>
2593 <p>Note that unsigned integer division and signed integer division are distinct
2594 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
2595 <p>Division by zero leads to undefined behavior.</p>
2597 <pre> <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2600 <!-- _______________________________________________________________________ -->
2601 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
2603 <div class="doc_text">
2606 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2611 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
2616 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
2617 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2618 values. Both arguments must have identical types.</p>
2621 <p>The value produced is the signed integer quotient of the two operands rounded towards zero.</p>
2622 <p>Note that signed integer division and unsigned integer division are distinct
2623 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
2624 <p>Division by zero leads to undefined behavior. Overflow also leads to
2625 undefined behavior; this is a rare case, but can occur, for example,
2626 by doing a 32-bit division of -2147483648 by -1.</p>
2628 <pre> <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2631 <!-- _______________________________________________________________________ -->
2632 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
2633 Instruction</a> </div>
2634 <div class="doc_text">
2637 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2641 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
2646 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
2647 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a>
2648 of floating point values. Both arguments must have identical types.</p>
2652 <p>The value produced is the floating point quotient of the two operands.</p>
2657 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
2661 <!-- _______________________________________________________________________ -->
2662 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
2664 <div class="doc_text">
2666 <pre> <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2669 <p>The '<tt>urem</tt>' instruction returns the remainder from the
2670 unsigned division of its two arguments.</p>
2672 <p>The two arguments to the '<tt>urem</tt>' instruction must be
2673 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2674 values. Both arguments must have identical types.</p>
2676 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
2677 This instruction always performs an unsigned division to get the remainder.</p>
2678 <p>Note that unsigned integer remainder and signed integer remainder are
2679 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
2680 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
2682 <pre> <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2686 <!-- _______________________________________________________________________ -->
2687 <div class="doc_subsubsection">
2688 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
2691 <div class="doc_text">
2696 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2701 <p>The '<tt>srem</tt>' instruction returns the remainder from the
2702 signed division of its two operands. This instruction can also take
2703 <a href="#t_vector">vector</a> versions of the values in which case
2704 the elements must be integers.</p>
2708 <p>The two arguments to the '<tt>srem</tt>' instruction must be
2709 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2710 values. Both arguments must have identical types.</p>
2714 <p>This instruction returns the <i>remainder</i> of a division (where the result
2715 has the same sign as the dividend, <tt>op1</tt>), not the <i>modulo</i>
2716 operator (where the result has the same sign as the divisor, <tt>op2</tt>) of
2717 a value. For more information about the difference, see <a
2718 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
2719 Math Forum</a>. For a table of how this is implemented in various languages,
2720 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
2721 Wikipedia: modulo operation</a>.</p>
2722 <p>Note that signed integer remainder and unsigned integer remainder are
2723 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
2724 <p>Taking the remainder of a division by zero leads to undefined behavior.
2725 Overflow also leads to undefined behavior; this is a rare case, but can occur,
2726 for example, by taking the remainder of a 32-bit division of -2147483648 by -1.
2727 (The remainder doesn't actually overflow, but this rule lets srem be
2728 implemented using instructions that return both the result of the division
2729 and the remainder.)</p>
2731 <pre> <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2735 <!-- _______________________________________________________________________ -->
2736 <div class="doc_subsubsection">
2737 <a name="i_frem">'<tt>frem</tt>' Instruction</a> </div>
2739 <div class="doc_text">
2742 <pre> <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2745 <p>The '<tt>frem</tt>' instruction returns the remainder from the
2746 division of its two operands.</p>
2748 <p>The two arguments to the '<tt>frem</tt>' instruction must be
2749 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a>
2750 of floating point values. Both arguments must have identical types.</p>
2754 <p>This instruction returns the <i>remainder</i> of a division.
2755 The remainder has the same sign as the dividend.</p>
2760 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
2764 <!-- ======================================================================= -->
2765 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
2766 Operations</a> </div>
2767 <div class="doc_text">
2768 <p>Bitwise binary operators are used to do various forms of
2769 bit-twiddling in a program. They are generally very efficient
2770 instructions and can commonly be strength reduced from other
2771 instructions. They require two operands of the same type, execute an operation on them,
2772 and produce a single value. The resulting value is the same type as its operands.</p>
2775 <!-- _______________________________________________________________________ -->
2776 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
2777 Instruction</a> </div>
2778 <div class="doc_text">
2780 <pre> <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2785 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2786 the left a specified number of bits.</p>
2790 <p>Both arguments to the '<tt>shl</tt>' instruction must be the same <a
2791 href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2792 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
2796 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod 2<sup>n</sup>,
2797 where n is the width of the result. If <tt>op2</tt> is (statically or dynamically) negative or
2798 equal to or larger than the number of bits in <tt>op1</tt>, the result is undefined.
2799 If the arguments are vectors, each vector element of <tt>op1</tt> is shifted by the
2800 corresponding shift amount in <tt>op2</tt>.</p>
2802 <h5>Example:</h5><pre>
2803 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
2804 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
2805 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
2806 <result> = shl i32 1, 32 <i>; undefined</i>
2807 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
2810 <!-- _______________________________________________________________________ -->
2811 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
2812 Instruction</a> </div>
2813 <div class="doc_text">
2815 <pre> <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2819 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
2820 operand shifted to the right a specified number of bits with zero fill.</p>
2823 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
2824 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2825 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
2829 <p>This instruction always performs a logical shift right operation. The most
2830 significant bits of the result will be filled with zero bits after the
2831 shift. If <tt>op2</tt> is (statically or dynamically) equal to or larger than
2832 the number of bits in <tt>op1</tt>, the result is undefined. If the arguments are
2833 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding shift
2834 amount in <tt>op2</tt>.</p>
2838 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
2839 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
2840 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
2841 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
2842 <result> = lshr i32 1, 32 <i>; undefined</i>
2843 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
2847 <!-- _______________________________________________________________________ -->
2848 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
2849 Instruction</a> </div>
2850 <div class="doc_text">
2853 <pre> <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2857 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
2858 operand shifted to the right a specified number of bits with sign extension.</p>
2861 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
2862 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2863 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
2866 <p>This instruction always performs an arithmetic shift right operation,
2867 The most significant bits of the result will be filled with the sign bit
2868 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
2869 larger than the number of bits in <tt>op1</tt>, the result is undefined. If the
2870 arguments are vectors, each vector element of <tt>op1</tt> is shifted by the
2871 corresponding shift amount in <tt>op2</tt>.</p>
2875 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
2876 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
2877 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
2878 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
2879 <result> = ashr i32 1, 32 <i>; undefined</i>
2880 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
2884 <!-- _______________________________________________________________________ -->
2885 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
2886 Instruction</a> </div>
2888 <div class="doc_text">
2893 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2898 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
2899 its two operands.</p>
2903 <p>The two arguments to the '<tt>and</tt>' instruction must be
2904 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2905 values. Both arguments must have identical types.</p>
2908 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
2911 <table border="1" cellspacing="0" cellpadding="4">
2943 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
2944 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
2945 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
2948 <!-- _______________________________________________________________________ -->
2949 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
2950 <div class="doc_text">
2952 <pre> <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2955 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
2956 or of its two operands.</p>
2959 <p>The two arguments to the '<tt>or</tt>' instruction must be
2960 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2961 values. Both arguments must have identical types.</p>
2963 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
2966 <table border="1" cellspacing="0" cellpadding="4">
2997 <pre> <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
2998 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
2999 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
3002 <!-- _______________________________________________________________________ -->
3003 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
3004 Instruction</a> </div>
3005 <div class="doc_text">
3007 <pre> <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3010 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
3011 or of its two operands. The <tt>xor</tt> is used to implement the
3012 "one's complement" operation, which is the "~" operator in C.</p>
3014 <p>The two arguments to the '<tt>xor</tt>' instruction must be
3015 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3016 values. Both arguments must have identical types.</p>
3020 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
3023 <table border="1" cellspacing="0" cellpadding="4">
3055 <pre> <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
3056 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
3057 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
3058 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
3062 <!-- ======================================================================= -->
3063 <div class="doc_subsection">
3064 <a name="vectorops">Vector Operations</a>
3067 <div class="doc_text">
3069 <p>LLVM supports several instructions to represent vector operations in a
3070 target-independent manner. These instructions cover the element-access and
3071 vector-specific operations needed to process vectors effectively. While LLVM
3072 does directly support these vector operations, many sophisticated algorithms
3073 will want to use target-specific intrinsics to take full advantage of a specific
3078 <!-- _______________________________________________________________________ -->
3079 <div class="doc_subsubsection">
3080 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
3083 <div class="doc_text">
3088 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
3094 The '<tt>extractelement</tt>' instruction extracts a single scalar
3095 element from a vector at a specified index.
3102 The first operand of an '<tt>extractelement</tt>' instruction is a
3103 value of <a href="#t_vector">vector</a> type. The second operand is
3104 an index indicating the position from which to extract the element.
3105 The index may be a variable.</p>
3110 The result is a scalar of the same type as the element type of
3111 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
3112 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
3113 results are undefined.
3119 %result = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
3124 <!-- _______________________________________________________________________ -->
3125 <div class="doc_subsubsection">
3126 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
3129 <div class="doc_text">
3134 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
3140 The '<tt>insertelement</tt>' instruction inserts a scalar
3141 element into a vector at a specified index.
3148 The first operand of an '<tt>insertelement</tt>' instruction is a
3149 value of <a href="#t_vector">vector</a> type. The second operand is a
3150 scalar value whose type must equal the element type of the first
3151 operand. The third operand is an index indicating the position at
3152 which to insert the value. The index may be a variable.</p>
3157 The result is a vector of the same type as <tt>val</tt>. Its
3158 element values are those of <tt>val</tt> except at position
3159 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
3160 exceeds the length of <tt>val</tt>, the results are undefined.
3166 %result = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
3170 <!-- _______________________________________________________________________ -->
3171 <div class="doc_subsubsection">
3172 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
3175 <div class="doc_text">
3180 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
3186 The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
3187 from two input vectors, returning a vector with the same element type as
3188 the input and length that is the same as the shuffle mask.
3194 The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
3195 with types that match each other. The third argument is a shuffle mask whose
3196 element type is always 'i32'. The result of the instruction is a vector whose
3197 length is the same as the shuffle mask and whose element type is the same as
3198 the element type of the first two operands.
3202 The shuffle mask operand is required to be a constant vector with either
3203 constant integer or undef values.
3209 The elements of the two input vectors are numbered from left to right across
3210 both of the vectors. The shuffle mask operand specifies, for each element of
3211 the result vector, which element of the two input vectors the result element
3212 gets. The element selector may be undef (meaning "don't care") and the second
3213 operand may be undef if performing a shuffle from only one vector.
3219 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
3220 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
3221 %result = shufflevector <4 x i32> %v1, <4 x i32> undef,
3222 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
3223 %result = shufflevector <8 x i32> %v1, <8 x i32> undef,
3224 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i>
3225 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
3226 <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>
3231 <!-- ======================================================================= -->
3232 <div class="doc_subsection">
3233 <a name="aggregateops">Aggregate Operations</a>
3236 <div class="doc_text">
3238 <p>LLVM supports several instructions for working with aggregate values.
3243 <!-- _______________________________________________________________________ -->
3244 <div class="doc_subsubsection">
3245 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
3248 <div class="doc_text">
3253 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
3259 The '<tt>extractvalue</tt>' instruction extracts the value of a struct field
3260 or array element from an aggregate value.
3267 The first operand of an '<tt>extractvalue</tt>' instruction is a
3268 value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a>
3269 type. The operands are constant indices to specify which value to extract
3270 in a similar manner as indices in a
3271 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
3277 The result is the value at the position in the aggregate specified by
3284 %result = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
3289 <!-- _______________________________________________________________________ -->
3290 <div class="doc_subsubsection">
3291 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
3294 <div class="doc_text">
3299 <result> = insertvalue <aggregate type> <val>, <ty> <val>, <idx> <i>; yields <n x <ty>></i>
3305 The '<tt>insertvalue</tt>' instruction inserts a value
3306 into a struct field or array element in an aggregate.
3313 The first operand of an '<tt>insertvalue</tt>' instruction is a
3314 value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a> type.
3315 The second operand is a first-class value to insert.
3316 The following operands are constant indices
3317 indicating the position at which to insert the value in a similar manner as
3319 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
3320 The value to insert must have the same type as the value identified
3327 The result is an aggregate of the same type as <tt>val</tt>. Its
3328 value is that of <tt>val</tt> except that the value at the position
3329 specified by the indices is that of <tt>elt</tt>.
3335 %result = insertvalue {i32, float} %agg, i32 1, 0 <i>; yields {i32, float}</i>
3340 <!-- ======================================================================= -->
3341 <div class="doc_subsection">
3342 <a name="memoryops">Memory Access and Addressing Operations</a>
3345 <div class="doc_text">
3347 <p>A key design point of an SSA-based representation is how it
3348 represents memory. In LLVM, no memory locations are in SSA form, which
3349 makes things very simple. This section describes how to read, write,
3350 allocate, and free memory in LLVM.</p>
3354 <!-- _______________________________________________________________________ -->
3355 <div class="doc_subsubsection">
3356 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
3359 <div class="doc_text">
3364 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3369 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
3370 heap and returns a pointer to it. The object is always allocated in the generic
3371 address space (address space zero).</p>
3375 <p>The '<tt>malloc</tt>' instruction allocates
3376 <tt>sizeof(<type>)*NumElements</tt>
3377 bytes of memory from the operating system and returns a pointer of the
3378 appropriate type to the program. If "NumElements" is specified, it is the
3379 number of elements allocated, otherwise "NumElements" is defaulted to be one.
3380 If a constant alignment is specified, the value result of the allocation is guaranteed to
3381 be aligned to at least that boundary. If not specified, or if zero, the target can
3382 choose to align the allocation on any convenient boundary.</p>
3384 <p>'<tt>type</tt>' must be a sized type.</p>
3388 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
3389 a pointer is returned. The result of a zero byte allocation is undefined. The
3390 result is null if there is insufficient memory available.</p>
3395 %array = malloc [4 x i8] <i>; yields {[%4 x i8]*}:array</i>
3397 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
3398 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
3399 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
3400 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
3401 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
3404 <p>Note that the code generator does not yet respect the
3405 alignment value.</p>
3409 <!-- _______________________________________________________________________ -->
3410 <div class="doc_subsubsection">
3411 <a name="i_free">'<tt>free</tt>' Instruction</a>
3414 <div class="doc_text">
3419 free <type> <value> <i>; yields {void}</i>
3424 <p>The '<tt>free</tt>' instruction returns memory back to the unused
3425 memory heap to be reallocated in the future.</p>
3429 <p>'<tt>value</tt>' shall be a pointer value that points to a value
3430 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
3435 <p>Access to the memory pointed to by the pointer is no longer defined
3436 after this instruction executes. If the pointer is null, the operation
3442 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
3443 free [4 x i8]* %array
3447 <!-- _______________________________________________________________________ -->
3448 <div class="doc_subsubsection">
3449 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
3452 <div class="doc_text">
3457 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3462 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
3463 currently executing function, to be automatically released when this function
3464 returns to its caller. The object is always allocated in the generic address
3465 space (address space zero).</p>
3469 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
3470 bytes of memory on the runtime stack, returning a pointer of the
3471 appropriate type to the program. If "NumElements" is specified, it is the
3472 number of elements allocated, otherwise "NumElements" is defaulted to be one.
3473 If a constant alignment is specified, the value result of the allocation is guaranteed
3474 to be aligned to at least that boundary. If not specified, or if zero, the target
3475 can choose to align the allocation on any convenient boundary.</p>
3477 <p>'<tt>type</tt>' may be any sized type.</p>
3481 <p>Memory is allocated; a pointer is returned. The operation is undefiend if
3482 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
3483 memory is automatically released when the function returns. The '<tt>alloca</tt>'
3484 instruction is commonly used to represent automatic variables that must
3485 have an address available. When the function returns (either with the <tt><a
3486 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
3487 instructions), the memory is reclaimed. Allocating zero bytes
3488 is legal, but the result is undefined.</p>
3493 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
3494 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
3495 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
3496 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
3500 <!-- _______________________________________________________________________ -->
3501 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
3502 Instruction</a> </div>
3503 <div class="doc_text">
3505 <pre> <result> = load <ty>* <pointer>[, align <alignment>]<br> <result> = volatile load <ty>* <pointer>[, align <alignment>]<br></pre>
3507 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
3509 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
3510 address from which to load. The pointer must point to a <a
3511 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
3512 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
3513 the number or order of execution of this <tt>load</tt> with other
3514 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
3517 The optional constant "align" argument specifies the alignment of the operation
3518 (that is, the alignment of the memory address). A value of 0 or an
3519 omitted "align" argument means that the operation has the preferential
3520 alignment for the target. It is the responsibility of the code emitter
3521 to ensure that the alignment information is correct. Overestimating
3522 the alignment results in an undefined behavior. Underestimating the
3523 alignment may produce less efficient code. An alignment of 1 is always
3527 <p>The location of memory pointed to is loaded. If the value being loaded
3528 is of scalar type then the number of bytes read does not exceed the minimum
3529 number of bytes needed to hold all bits of the type. For example, loading an
3530 <tt>i24</tt> reads at most three bytes. When loading a value of a type like
3531 <tt>i20</tt> with a size that is not an integral number of bytes, the result
3532 is undefined if the value was not originally written using a store of the
3535 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3537 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
3538 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
3541 <!-- _______________________________________________________________________ -->
3542 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
3543 Instruction</a> </div>
3544 <div class="doc_text">
3546 <pre> store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3547 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3550 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
3552 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
3553 to store and an address at which to store it. The type of the '<tt><pointer></tt>'
3554 operand must be a pointer to the <a href="#t_firstclass">first class</a> type
3555 of the '<tt><value></tt>'
3556 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
3557 optimizer is not allowed to modify the number or order of execution of
3558 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
3559 href="#i_store">store</a></tt> instructions.</p>
3561 The optional constant "align" argument specifies the alignment of the operation
3562 (that is, the alignment of the memory address). A value of 0 or an
3563 omitted "align" argument means that the operation has the preferential
3564 alignment for the target. It is the responsibility of the code emitter
3565 to ensure that the alignment information is correct. Overestimating
3566 the alignment results in an undefined behavior. Underestimating the
3567 alignment may produce less efficient code. An alignment of 1 is always
3571 <p>The contents of memory are updated to contain '<tt><value></tt>'
3572 at the location specified by the '<tt><pointer></tt>' operand.
3573 If '<tt><value></tt>' is of scalar type then the number of bytes
3574 written does not exceed the minimum number of bytes needed to hold all
3575 bits of the type. For example, storing an <tt>i24</tt> writes at most
3576 three bytes. When writing a value of a type like <tt>i20</tt> with a
3577 size that is not an integral number of bytes, it is unspecified what
3578 happens to the extra bits that do not belong to the type, but they will
3579 typically be overwritten.</p>
3581 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3582 store i32 3, i32* %ptr <i>; yields {void}</i>
3583 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
3587 <!-- _______________________________________________________________________ -->
3588 <div class="doc_subsubsection">
3589 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
3592 <div class="doc_text">
3595 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
3601 The '<tt>getelementptr</tt>' instruction is used to get the address of a
3602 subelement of an aggregate data structure. It performs address calculation only
3603 and does not access memory.</p>
3607 <p>The first argument is always a pointer, and forms the basis of the
3608 calculation. The remaining arguments are indices, that indicate which of the
3609 elements of the aggregate object are indexed. The interpretation of each index
3610 is dependent on the type being indexed into. The first index always indexes the
3611 pointer value given as the first argument, the second index indexes a value of
3612 the type pointed to (not necessarily the value directly pointed to, since the
3613 first index can be non-zero), etc. The first type indexed into must be a pointer
3614 value, subsequent types can be arrays, vectors and structs. Note that subsequent
3615 types being indexed into can never be pointers, since that would require loading
3616 the pointer before continuing calculation.</p>
3618 <p>The type of each index argument depends on the type it is indexing into.
3619 When indexing into a (packed) structure, only <tt>i32</tt> integer
3620 <b>constants</b> are allowed. When indexing into an array, pointer or vector,
3621 only integers of 32 or 64 bits are allowed (also non-constants). 32-bit values
3622 will be sign extended to 64-bits if required.</p>
3624 <p>For example, let's consider a C code fragment and how it gets
3625 compiled to LLVM:</p>
3627 <div class="doc_code">
3640 int *foo(struct ST *s) {
3641 return &s[1].Z.B[5][13];
3646 <p>The LLVM code generated by the GCC frontend is:</p>
3648 <div class="doc_code">
3650 %RT = <a href="#namedtypes">type</a> { i8 , [10 x [20 x i32]], i8 }
3651 %ST = <a href="#namedtypes">type</a> { i32, double, %RT }
3653 define i32* %foo(%ST* %s) {
3655 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
3663 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
3664 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
3665 }</tt>' type, a structure. The second index indexes into the third element of
3666 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
3667 i8 }</tt>' type, another structure. The third index indexes into the second
3668 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
3669 array. The two dimensions of the array are subscripted into, yielding an
3670 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
3671 to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
3673 <p>Note that it is perfectly legal to index partially through a
3674 structure, returning a pointer to an inner element. Because of this,
3675 the LLVM code for the given testcase is equivalent to:</p>
3678 define i32* %foo(%ST* %s) {
3679 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
3680 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
3681 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
3682 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
3683 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
3688 <p>Note that it is undefined to access an array out of bounds: array
3689 and pointer indexes must always be within the defined bounds of the
3690 array type when accessed with an instruction that dereferences the
3691 pointer (e.g. a load or store instruction). The one exception for
3692 this rule is zero length arrays. These arrays are defined to be
3693 accessible as variable length arrays, which requires access beyond the
3694 zero'th element.</p>
3696 <p>The getelementptr instruction is often confusing. For some more insight
3697 into how it works, see <a href="GetElementPtr.html">the getelementptr
3703 <i>; yields [12 x i8]*:aptr</i>
3704 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
3705 <i>; yields i8*:vptr</i>
3706 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
3707 <i>; yields i8*:eptr</i>
3708 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
3712 <!-- ======================================================================= -->
3713 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
3715 <div class="doc_text">
3716 <p>The instructions in this category are the conversion instructions (casting)
3717 which all take a single operand and a type. They perform various bit conversions
3721 <!-- _______________________________________________________________________ -->
3722 <div class="doc_subsubsection">
3723 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
3725 <div class="doc_text">
3729 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
3734 The '<tt>trunc</tt>' instruction truncates its operand to the type <tt>ty2</tt>.
3739 The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
3740 be an <a href="#t_integer">integer</a> type, and a type that specifies the size
3741 and type of the result, which must be an <a href="#t_integer">integer</a>
3742 type. The bit size of <tt>value</tt> must be larger than the bit size of
3743 <tt>ty2</tt>. Equal sized types are not allowed.</p>
3747 The '<tt>trunc</tt>' instruction truncates the high order bits in <tt>value</tt>
3748 and converts the remaining bits to <tt>ty2</tt>. Since the source size must be
3749 larger than the destination size, <tt>trunc</tt> cannot be a <i>no-op cast</i>.
3750 It will always truncate bits.</p>
3754 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
3755 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
3756 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
3760 <!-- _______________________________________________________________________ -->
3761 <div class="doc_subsubsection">
3762 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
3764 <div class="doc_text">
3768 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
3772 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
3777 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
3778 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3779 also be of <a href="#t_integer">integer</a> type. The bit size of the
3780 <tt>value</tt> must be smaller than the bit size of the destination type,
3784 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
3785 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
3787 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
3791 %X = zext i32 257 to i64 <i>; yields i64:257</i>
3792 %Y = zext i1 true to i32 <i>; yields i32:1</i>
3796 <!-- _______________________________________________________________________ -->
3797 <div class="doc_subsubsection">
3798 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
3800 <div class="doc_text">
3804 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
3808 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
3812 The '<tt>sext</tt>' instruction takes a value to cast, which must be of
3813 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3814 also be of <a href="#t_integer">integer</a> type. The bit size of the
3815 <tt>value</tt> must be smaller than the bit size of the destination type,
3820 The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
3821 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size of
3822 the type <tt>ty2</tt>.</p>
3824 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
3828 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
3829 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
3833 <!-- _______________________________________________________________________ -->
3834 <div class="doc_subsubsection">
3835 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
3838 <div class="doc_text">
3843 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
3847 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
3852 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
3853 point</a> value to cast and a <a href="#t_floating">floating point</a> type to
3854 cast it to. The size of <tt>value</tt> must be larger than the size of
3855 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
3856 <i>no-op cast</i>.</p>
3859 <p> The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
3860 <a href="#t_floating">floating point</a> type to a smaller
3861 <a href="#t_floating">floating point</a> type. If the value cannot fit within
3862 the destination type, <tt>ty2</tt>, then the results are undefined.</p>
3866 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
3867 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
3871 <!-- _______________________________________________________________________ -->
3872 <div class="doc_subsubsection">
3873 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
3875 <div class="doc_text">
3879 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
3883 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
3884 floating point value.</p>
3887 <p>The '<tt>fpext</tt>' instruction takes a
3888 <a href="#t_floating">floating point</a> <tt>value</tt> to cast,
3889 and a <a href="#t_floating">floating point</a> type to cast it to. The source
3890 type must be smaller than the destination type.</p>
3893 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
3894 <a href="#t_floating">floating point</a> type to a larger
3895 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
3896 used to make a <i>no-op cast</i> because it always changes bits. Use
3897 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
3901 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
3902 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
3906 <!-- _______________________________________________________________________ -->
3907 <div class="doc_subsubsection">
3908 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
3910 <div class="doc_text">
3914 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
3918 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
3919 unsigned integer equivalent of type <tt>ty2</tt>.
3923 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
3924 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3925 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3926 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3927 vector integer type with the same number of elements as <tt>ty</tt></p>
3930 <p> The '<tt>fptoui</tt>' instruction converts its
3931 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3932 towards zero) unsigned integer value. If the value cannot fit in <tt>ty2</tt>,
3933 the results are undefined.</p>
3937 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
3938 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
3939 %X = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
3943 <!-- _______________________________________________________________________ -->
3944 <div class="doc_subsubsection">
3945 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
3947 <div class="doc_text">
3951 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
3955 <p>The '<tt>fptosi</tt>' instruction converts
3956 <a href="#t_floating">floating point</a> <tt>value</tt> to type <tt>ty2</tt>.
3960 <p> The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
3961 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3962 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3963 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3964 vector integer type with the same number of elements as <tt>ty</tt></p>
3967 <p>The '<tt>fptosi</tt>' instruction converts its
3968 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3969 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
3970 the results are undefined.</p>
3974 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
3975 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
3976 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
3980 <!-- _______________________________________________________________________ -->
3981 <div class="doc_subsubsection">
3982 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
3984 <div class="doc_text">
3988 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3992 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
3993 integer and converts that value to the <tt>ty2</tt> type.</p>
3996 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
3997 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3998 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3999 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
4000 floating point type with the same number of elements as <tt>ty</tt></p>
4003 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
4004 integer quantity and converts it to the corresponding floating point value. If
4005 the value cannot fit in the floating point value, the results are undefined.</p>
4009 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
4010 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
4014 <!-- _______________________________________________________________________ -->
4015 <div class="doc_subsubsection">
4016 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
4018 <div class="doc_text">
4022 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
4026 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed
4027 integer and converts that value to the <tt>ty2</tt> type.</p>
4030 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
4031 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
4032 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
4033 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
4034 floating point type with the same number of elements as <tt>ty</tt></p>
4037 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed
4038 integer quantity and converts it to the corresponding floating point value. If
4039 the value cannot fit in the floating point value, the results are undefined.</p>
4043 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
4044 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
4048 <!-- _______________________________________________________________________ -->
4049 <div class="doc_subsubsection">
4050 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
4052 <div class="doc_text">
4056 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
4060 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
4061 the integer type <tt>ty2</tt>.</p>
4064 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
4065 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
4066 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.</p>
4069 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
4070 <tt>ty2</tt> by interpreting the pointer value as an integer and either
4071 truncating or zero extending that value to the size of the integer type. If
4072 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
4073 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
4074 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
4079 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
4080 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
4084 <!-- _______________________________________________________________________ -->
4085 <div class="doc_subsubsection">
4086 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
4088 <div class="doc_text">
4092 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
4096 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to
4097 a pointer type, <tt>ty2</tt>.</p>
4100 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
4101 value to cast, and a type to cast it to, which must be a
4102 <a href="#t_pointer">pointer</a> type.</p>
4105 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
4106 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
4107 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
4108 size of a pointer then a truncation is done. If <tt>value</tt> is smaller than
4109 the size of a pointer then a zero extension is done. If they are the same size,
4110 nothing is done (<i>no-op cast</i>).</p>
4114 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
4115 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
4116 %Y = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
4120 <!-- _______________________________________________________________________ -->
4121 <div class="doc_subsubsection">
4122 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
4124 <div class="doc_text">
4128 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
4133 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
4134 <tt>ty2</tt> without changing any bits.</p>
4138 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be
4139 a non-aggregate first class value, and a type to cast it to, which must also be
4140 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes of
4142 and the destination type, <tt>ty2</tt>, must be identical. If the source
4143 type is a pointer, the destination type must also be a pointer. This
4144 instruction supports bitwise conversion of vectors to integers and to vectors
4145 of other types (as long as they have the same size).</p>
4148 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
4149 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
4150 this conversion. The conversion is done as if the <tt>value</tt> had been
4151 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only be
4152 converted to other pointer types with this instruction. To convert pointers to
4153 other types, use the <a href="#i_inttoptr">inttoptr</a> or
4154 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
4158 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
4159 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
4160 %Z = bitcast <2 x int> %V to i64; <i>; yields i64: %V</i>
4164 <!-- ======================================================================= -->
4165 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
4166 <div class="doc_text">
4167 <p>The instructions in this category are the "miscellaneous"
4168 instructions, which defy better classification.</p>
4171 <!-- _______________________________________________________________________ -->
4172 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
4174 <div class="doc_text">
4176 <pre> <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4179 <p>The '<tt>icmp</tt>' instruction returns a boolean value or
4180 a vector of boolean values based on comparison
4181 of its two integer, integer vector, or pointer operands.</p>
4183 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
4184 the condition code indicating the kind of comparison to perform. It is not
4185 a value, just a keyword. The possible condition code are:
4188 <li><tt>eq</tt>: equal</li>
4189 <li><tt>ne</tt>: not equal </li>
4190 <li><tt>ugt</tt>: unsigned greater than</li>
4191 <li><tt>uge</tt>: unsigned greater or equal</li>
4192 <li><tt>ult</tt>: unsigned less than</li>
4193 <li><tt>ule</tt>: unsigned less or equal</li>
4194 <li><tt>sgt</tt>: signed greater than</li>
4195 <li><tt>sge</tt>: signed greater or equal</li>
4196 <li><tt>slt</tt>: signed less than</li>
4197 <li><tt>sle</tt>: signed less or equal</li>
4199 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
4200 <a href="#t_pointer">pointer</a>
4201 or integer <a href="#t_vector">vector</a> typed.
4202 They must also be identical types.</p>
4204 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to
4205 the condition code given as <tt>cond</tt>. The comparison performed always
4206 yields either an <a href="#t_primitive"><tt>i1</tt></a> or vector of <tt>i1</tt> result, as follows:
4209 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
4210 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
4212 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
4213 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.</li>
4214 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
4215 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4216 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
4217 <tt>true</tt> if <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4218 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
4219 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
4220 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
4221 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4222 <li><tt>sgt</tt>: interprets the operands as signed values and yields
4223 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4224 <li><tt>sge</tt>: interprets the operands as signed values and yields
4225 <tt>true</tt> if <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4226 <li><tt>slt</tt>: interprets the operands as signed values and yields
4227 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
4228 <li><tt>sle</tt>: interprets the operands as signed values and yields
4229 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4231 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
4232 values are compared as if they were integers.</p>
4233 <p>If the operands are integer vectors, then they are compared
4234 element by element. The result is an <tt>i1</tt> vector with
4235 the same number of elements as the values being compared.
4236 Otherwise, the result is an <tt>i1</tt>.
4240 <pre> <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
4241 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
4242 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
4243 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
4244 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
4245 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
4248 <p>Note that the code generator does not yet support vector types with
4249 the <tt>icmp</tt> instruction.</p>
4253 <!-- _______________________________________________________________________ -->
4254 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
4256 <div class="doc_text">
4258 <pre> <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4261 <p>The '<tt>fcmp</tt>' instruction returns a boolean value
4262 or vector of boolean values based on comparison
4263 of its operands.</p>
4265 If the operands are floating point scalars, then the result
4266 type is a boolean (<a href="#t_primitive"><tt>i1</tt></a>).
4268 <p>If the operands are floating point vectors, then the result type
4269 is a vector of boolean with the same number of elements as the
4270 operands being compared.</p>
4272 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
4273 the condition code indicating the kind of comparison to perform. It is not
4274 a value, just a keyword. The possible condition code are:</p>
4276 <li><tt>false</tt>: no comparison, always returns false</li>
4277 <li><tt>oeq</tt>: ordered and equal</li>
4278 <li><tt>ogt</tt>: ordered and greater than </li>
4279 <li><tt>oge</tt>: ordered and greater than or equal</li>
4280 <li><tt>olt</tt>: ordered and less than </li>
4281 <li><tt>ole</tt>: ordered and less than or equal</li>
4282 <li><tt>one</tt>: ordered and not equal</li>
4283 <li><tt>ord</tt>: ordered (no nans)</li>
4284 <li><tt>ueq</tt>: unordered or equal</li>
4285 <li><tt>ugt</tt>: unordered or greater than </li>
4286 <li><tt>uge</tt>: unordered or greater than or equal</li>
4287 <li><tt>ult</tt>: unordered or less than </li>
4288 <li><tt>ule</tt>: unordered or less than or equal</li>
4289 <li><tt>une</tt>: unordered or not equal</li>
4290 <li><tt>uno</tt>: unordered (either nans)</li>
4291 <li><tt>true</tt>: no comparison, always returns true</li>
4293 <p><i>Ordered</i> means that neither operand is a QNAN while
4294 <i>unordered</i> means that either operand may be a QNAN.</p>
4295 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be
4296 either a <a href="#t_floating">floating point</a> type
4297 or a <a href="#t_vector">vector</a> of floating point type.
4298 They must have identical types.</p>
4300 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4301 according to the condition code given as <tt>cond</tt>.
4302 If the operands are vectors, then the vectors are compared
4304 Each comparison performed
4305 always yields an <a href="#t_primitive">i1</a> result, as follows:</p>
4307 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
4308 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4309 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
4310 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4311 <tt>op1</tt> is greather than <tt>op2</tt>.</li>
4312 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4313 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4314 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4315 <tt>op1</tt> is less than <tt>op2</tt>.</li>
4316 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4317 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4318 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4319 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
4320 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
4321 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
4322 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
4323 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
4324 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4325 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
4326 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4327 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
4328 <tt>op1</tt> is less than <tt>op2</tt>.</li>
4329 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
4330 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4331 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
4332 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
4333 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
4334 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
4338 <pre> <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
4339 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
4340 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
4341 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
4344 <p>Note that the code generator does not yet support vector types with
4345 the <tt>fcmp</tt> instruction.</p>
4349 <!-- _______________________________________________________________________ -->
4350 <div class="doc_subsubsection">
4351 <a name="i_vicmp">'<tt>vicmp</tt>' Instruction</a>
4353 <div class="doc_text">
4355 <pre> <result> = vicmp <cond> <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4358 <p>The '<tt>vicmp</tt>' instruction returns an integer vector value based on
4359 element-wise comparison of its two integer vector operands.</p>
4361 <p>The '<tt>vicmp</tt>' instruction takes three operands. The first operand is
4362 the condition code indicating the kind of comparison to perform. It is not
4363 a value, just a keyword. The possible condition code are:</p>
4365 <li><tt>eq</tt>: equal</li>
4366 <li><tt>ne</tt>: not equal </li>
4367 <li><tt>ugt</tt>: unsigned greater than</li>
4368 <li><tt>uge</tt>: unsigned greater or equal</li>
4369 <li><tt>ult</tt>: unsigned less than</li>
4370 <li><tt>ule</tt>: unsigned less or equal</li>
4371 <li><tt>sgt</tt>: signed greater than</li>
4372 <li><tt>sge</tt>: signed greater or equal</li>
4373 <li><tt>slt</tt>: signed less than</li>
4374 <li><tt>sle</tt>: signed less or equal</li>
4376 <p>The remaining two arguments must be <a href="#t_vector">vector</a> or
4377 <a href="#t_integer">integer</a> typed. They must also be identical types.</p>
4379 <p>The '<tt>vicmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4380 according to the condition code given as <tt>cond</tt>. The comparison yields a
4381 <a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, of
4382 identical type as the values being compared. The most significant bit in each
4383 element is 1 if the element-wise comparison evaluates to true, and is 0
4384 otherwise. All other bits of the result are undefined. The condition codes
4385 are evaluated identically to the <a href="#i_icmp">'<tt>icmp</tt>'
4386 instruction</a>.</p>
4390 <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>
4391 <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>
4395 <!-- _______________________________________________________________________ -->
4396 <div class="doc_subsubsection">
4397 <a name="i_vfcmp">'<tt>vfcmp</tt>' Instruction</a>
4399 <div class="doc_text">
4401 <pre> <result> = vfcmp <cond> <ty> <op1>, <op2></pre>
4403 <p>The '<tt>vfcmp</tt>' instruction returns an integer vector value based on
4404 element-wise comparison of its two floating point vector operands. The output
4405 elements have the same width as the input elements.</p>
4407 <p>The '<tt>vfcmp</tt>' instruction takes three operands. The first operand is
4408 the condition code indicating the kind of comparison to perform. It is not
4409 a value, just a keyword. The possible condition code are:</p>
4411 <li><tt>false</tt>: no comparison, always returns false</li>
4412 <li><tt>oeq</tt>: ordered and equal</li>
4413 <li><tt>ogt</tt>: ordered and greater than </li>
4414 <li><tt>oge</tt>: ordered and greater than or equal</li>
4415 <li><tt>olt</tt>: ordered and less than </li>
4416 <li><tt>ole</tt>: ordered and less than or equal</li>
4417 <li><tt>one</tt>: ordered and not equal</li>
4418 <li><tt>ord</tt>: ordered (no nans)</li>
4419 <li><tt>ueq</tt>: unordered or equal</li>
4420 <li><tt>ugt</tt>: unordered or greater than </li>
4421 <li><tt>uge</tt>: unordered or greater than or equal</li>
4422 <li><tt>ult</tt>: unordered or less than </li>
4423 <li><tt>ule</tt>: unordered or less than or equal</li>
4424 <li><tt>une</tt>: unordered or not equal</li>
4425 <li><tt>uno</tt>: unordered (either nans)</li>
4426 <li><tt>true</tt>: no comparison, always returns true</li>
4428 <p>The remaining two arguments must be <a href="#t_vector">vector</a> of
4429 <a href="#t_floating">floating point</a> typed. They must also be identical
4432 <p>The '<tt>vfcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4433 according to the condition code given as <tt>cond</tt>. The comparison yields a
4434 <a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, with
4435 an identical number of elements as the values being compared, and each element
4436 having identical with to the width of the floating point elements. The most
4437 significant bit in each element is 1 if the element-wise comparison evaluates to
4438 true, and is 0 otherwise. All other bits of the result are undefined. The
4439 condition codes are evaluated identically to the
4440 <a href="#i_fcmp">'<tt>fcmp</tt>' instruction</a>.</p>
4444 <i>; yields: result=<2 x i32> < i32 0, i32 -1 ></i>
4445 <result> = vfcmp oeq <2 x float> < float 4, float 0 >, < float 5, float 0 >
4447 <i>; yields: result=<2 x i64> < i64 -1, i64 0 ></i>
4448 <result> = vfcmp ult <2 x double> < double 1, double 2 >, < double 2, double 2>
4452 <!-- _______________________________________________________________________ -->
4453 <div class="doc_subsubsection">
4454 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
4457 <div class="doc_text">
4461 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
4463 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
4464 the SSA graph representing the function.</p>
4467 <p>The type of the incoming values is specified with the first type
4468 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
4469 as arguments, with one pair for each predecessor basic block of the
4470 current block. Only values of <a href="#t_firstclass">first class</a>
4471 type may be used as the value arguments to the PHI node. Only labels
4472 may be used as the label arguments.</p>
4474 <p>There must be no non-phi instructions between the start of a basic
4475 block and the PHI instructions: i.e. PHI instructions must be first in
4480 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
4481 specified by the pair corresponding to the predecessor basic block that executed
4482 just prior to the current block.</p>
4486 Loop: ; Infinite loop that counts from 0 on up...
4487 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
4488 %nextindvar = add i32 %indvar, 1
4493 <!-- _______________________________________________________________________ -->
4494 <div class="doc_subsubsection">
4495 <a name="i_select">'<tt>select</tt>' Instruction</a>
4498 <div class="doc_text">
4503 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
4505 <i>selty</i> is either i1 or {<N x i1>}
4511 The '<tt>select</tt>' instruction is used to choose one value based on a
4512 condition, without branching.
4519 The '<tt>select</tt>' instruction requires an 'i1' value or
4520 a vector of 'i1' values indicating the
4521 condition, and two values of the same <a href="#t_firstclass">first class</a>
4522 type. If the val1/val2 are vectors and
4523 the condition is a scalar, then entire vectors are selected, not
4524 individual elements.
4530 If the condition is an i1 and it evaluates to 1, the instruction returns the first
4531 value argument; otherwise, it returns the second value argument.
4534 If the condition is a vector of i1, then the value arguments must
4535 be vectors of the same size, and the selection is done element
4542 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
4545 <p>Note that the code generator does not yet support conditions
4546 with vector type.</p>
4551 <!-- _______________________________________________________________________ -->
4552 <div class="doc_subsubsection">
4553 <a name="i_call">'<tt>call</tt>' Instruction</a>
4556 <div class="doc_text">
4560 <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>]
4565 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
4569 <p>This instruction requires several arguments:</p>
4573 <p>The optional "tail" marker indicates whether the callee function accesses
4574 any allocas or varargs in the caller. If the "tail" marker is present, the
4575 function call is eligible for tail call optimization. Note that calls may
4576 be marked "tail" even if they do not occur before a <a
4577 href="#i_ret"><tt>ret</tt></a> instruction.</p>
4580 <p>The optional "cconv" marker indicates which <a href="#callingconv">calling
4581 convention</a> the call should use. If none is specified, the call defaults
4582 to using C calling conventions.</p>
4586 <p>The optional <a href="#paramattrs">Parameter Attributes</a> list for
4587 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>',
4588 and '<tt>inreg</tt>' attributes are valid here.</p>
4592 <p>'<tt>ty</tt>': the type of the call instruction itself which is also
4593 the type of the return value. Functions that return no value are marked
4594 <tt><a href="#t_void">void</a></tt>.</p>
4597 <p>'<tt>fnty</tt>': shall be the signature of the pointer to function
4598 value being invoked. The argument types must match the types implied by
4599 this signature. This type can be omitted if the function is not varargs
4600 and if the function type does not return a pointer to a function.</p>
4603 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
4604 be invoked. In most cases, this is a direct function invocation, but
4605 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
4606 to function value.</p>
4609 <p>'<tt>function args</tt>': argument list whose types match the
4610 function signature argument types. All arguments must be of
4611 <a href="#t_firstclass">first class</a> type. If the function signature
4612 indicates the function accepts a variable number of arguments, the extra
4613 arguments can be specified.</p>
4616 <p>The optional <a href="#fnattrs">function attributes</a> list. Only
4617 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
4618 '<tt>readnone</tt>' attributes are valid here.</p>
4624 <p>The '<tt>call</tt>' instruction is used to cause control flow to
4625 transfer to a specified function, with its incoming arguments bound to
4626 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
4627 instruction in the called function, control flow continues with the
4628 instruction after the function call, and the return value of the
4629 function is bound to the result argument.</p>
4634 %retval = call i32 @test(i32 %argc)
4635 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42) <i>; yields i32</i>
4636 %X = tail call i32 @foo() <i>; yields i32</i>
4637 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
4638 call void %foo(i8 97 signext)
4640 %struct.A = type { i32, i8 }
4641 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
4642 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
4643 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
4644 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
4645 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
4650 <!-- _______________________________________________________________________ -->
4651 <div class="doc_subsubsection">
4652 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
4655 <div class="doc_text">
4660 <resultval> = va_arg <va_list*> <arglist>, <argty>
4665 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
4666 the "variable argument" area of a function call. It is used to implement the
4667 <tt>va_arg</tt> macro in C.</p>
4671 <p>This instruction takes a <tt>va_list*</tt> value and the type of
4672 the argument. It returns a value of the specified argument type and
4673 increments the <tt>va_list</tt> to point to the next argument. The
4674 actual type of <tt>va_list</tt> is target specific.</p>
4678 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
4679 type from the specified <tt>va_list</tt> and causes the
4680 <tt>va_list</tt> to point to the next argument. For more information,
4681 see the variable argument handling <a href="#int_varargs">Intrinsic
4684 <p>It is legal for this instruction to be called in a function which does not
4685 take a variable number of arguments, for example, the <tt>vfprintf</tt>
4688 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
4689 href="#intrinsics">intrinsic function</a> because it takes a type as an
4694 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
4696 <p>Note that the code generator does not yet fully support va_arg
4697 on many targets. Also, it does not currently support va_arg with
4698 aggregate types on any target.</p>
4702 <!-- *********************************************************************** -->
4703 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
4704 <!-- *********************************************************************** -->
4706 <div class="doc_text">
4708 <p>LLVM supports the notion of an "intrinsic function". These functions have
4709 well known names and semantics and are required to follow certain restrictions.
4710 Overall, these intrinsics represent an extension mechanism for the LLVM
4711 language that does not require changing all of the transformations in LLVM when
4712 adding to the language (or the bitcode reader/writer, the parser, etc...).</p>
4714 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
4715 prefix is reserved in LLVM for intrinsic names; thus, function names may not
4716 begin with this prefix. Intrinsic functions must always be external functions:
4717 you cannot define the body of intrinsic functions. Intrinsic functions may
4718 only be used in call or invoke instructions: it is illegal to take the address
4719 of an intrinsic function. Additionally, because intrinsic functions are part
4720 of the LLVM language, it is required if any are added that they be documented
4723 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
4724 a family of functions that perform the same operation but on different data
4725 types. Because LLVM can represent over 8 million different integer types,
4726 overloading is used commonly to allow an intrinsic function to operate on any
4727 integer type. One or more of the argument types or the result type can be
4728 overloaded to accept any integer type. Argument types may also be defined as
4729 exactly matching a previous argument's type or the result type. This allows an
4730 intrinsic function which accepts multiple arguments, but needs all of them to
4731 be of the same type, to only be overloaded with respect to a single argument or
4734 <p>Overloaded intrinsics will have the names of its overloaded argument types
4735 encoded into its function name, each preceded by a period. Only those types
4736 which are overloaded result in a name suffix. Arguments whose type is matched
4737 against another type do not. For example, the <tt>llvm.ctpop</tt> function can
4738 take an integer of any width and returns an integer of exactly the same integer
4739 width. This leads to a family of functions such as
4740 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29 %val)</tt>.
4741 Only one type, the return type, is overloaded, and only one type suffix is
4742 required. Because the argument's type is matched against the return type, it
4743 does not require its own name suffix.</p>
4745 <p>To learn how to add an intrinsic function, please see the
4746 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.
4751 <!-- ======================================================================= -->
4752 <div class="doc_subsection">
4753 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
4756 <div class="doc_text">
4758 <p>Variable argument support is defined in LLVM with the <a
4759 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
4760 intrinsic functions. These functions are related to the similarly
4761 named macros defined in the <tt><stdarg.h></tt> header file.</p>
4763 <p>All of these functions operate on arguments that use a
4764 target-specific value type "<tt>va_list</tt>". The LLVM assembly
4765 language reference manual does not define what this type is, so all
4766 transformations should be prepared to handle these functions regardless of
4769 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
4770 instruction and the variable argument handling intrinsic functions are
4773 <div class="doc_code">
4775 define i32 @test(i32 %X, ...) {
4776 ; Initialize variable argument processing
4778 %ap2 = bitcast i8** %ap to i8*
4779 call void @llvm.va_start(i8* %ap2)
4781 ; Read a single integer argument
4782 %tmp = va_arg i8** %ap, i32
4784 ; Demonstrate usage of llvm.va_copy and llvm.va_end
4786 %aq2 = bitcast i8** %aq to i8*
4787 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
4788 call void @llvm.va_end(i8* %aq2)
4790 ; Stop processing of arguments.
4791 call void @llvm.va_end(i8* %ap2)
4795 declare void @llvm.va_start(i8*)
4796 declare void @llvm.va_copy(i8*, i8*)
4797 declare void @llvm.va_end(i8*)
4803 <!-- _______________________________________________________________________ -->
4804 <div class="doc_subsubsection">
4805 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
4809 <div class="doc_text">
4811 <pre> declare void %llvm.va_start(i8* <arglist>)<br></pre>
4813 <p>The '<tt>llvm.va_start</tt>' intrinsic initializes
4814 <tt>*<arglist></tt> for subsequent use by <tt><a
4815 href="#i_va_arg">va_arg</a></tt>.</p>
4819 <p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
4823 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
4824 macro available in C. In a target-dependent way, it initializes the
4825 <tt>va_list</tt> element to which the argument points, so that the next call to
4826 <tt>va_arg</tt> will produce the first variable argument passed to the function.
4827 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
4828 last argument of the function as the compiler can figure that out.</p>
4832 <!-- _______________________________________________________________________ -->
4833 <div class="doc_subsubsection">
4834 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
4837 <div class="doc_text">
4839 <pre> declare void @llvm.va_end(i8* <arglist>)<br></pre>
4842 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
4843 which has been initialized previously with <tt><a href="#int_va_start">llvm.va_start</a></tt>
4844 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
4848 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
4852 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
4853 macro available in C. In a target-dependent way, it destroys the
4854 <tt>va_list</tt> element to which the argument points. Calls to <a
4855 href="#int_va_start"><tt>llvm.va_start</tt></a> and <a href="#int_va_copy">
4856 <tt>llvm.va_copy</tt></a> must be matched exactly with calls to
4857 <tt>llvm.va_end</tt>.</p>
4861 <!-- _______________________________________________________________________ -->
4862 <div class="doc_subsubsection">
4863 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
4866 <div class="doc_text">
4871 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
4876 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
4877 from the source argument list to the destination argument list.</p>
4881 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
4882 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
4887 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
4888 macro available in C. In a target-dependent way, it copies the source
4889 <tt>va_list</tt> element into the destination <tt>va_list</tt> element. This
4890 intrinsic is necessary because the <tt><a href="#int_va_start">
4891 llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
4892 example, memory allocation.</p>
4896 <!-- ======================================================================= -->
4897 <div class="doc_subsection">
4898 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
4901 <div class="doc_text">
4904 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
4905 Collection</a> (GC) requires the implementation and generation of these
4907 These intrinsics allow identification of <a href="#int_gcroot">GC roots on the
4908 stack</a>, as well as garbage collector implementations that require <a
4909 href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a> barriers.
4910 Front-ends for type-safe garbage collected languages should generate these
4911 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
4912 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
4915 <p>The garbage collection intrinsics only operate on objects in the generic
4916 address space (address space zero).</p>
4920 <!-- _______________________________________________________________________ -->
4921 <div class="doc_subsubsection">
4922 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
4925 <div class="doc_text">
4930 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
4935 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
4936 the code generator, and allows some metadata to be associated with it.</p>
4940 <p>The first argument specifies the address of a stack object that contains the
4941 root pointer. The second pointer (which must be either a constant or a global
4942 value address) contains the meta-data to be associated with the root.</p>
4946 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
4947 location. At compile-time, the code generator generates information to allow
4948 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
4949 intrinsic may only be used in a function which <a href="#gc">specifies a GC
4955 <!-- _______________________________________________________________________ -->
4956 <div class="doc_subsubsection">
4957 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
4960 <div class="doc_text">
4965 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
4970 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
4971 locations, allowing garbage collector implementations that require read
4976 <p>The second argument is the address to read from, which should be an address
4977 allocated from the garbage collector. The first object is a pointer to the
4978 start of the referenced object, if needed by the language runtime (otherwise
4983 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
4984 instruction, but may be replaced with substantially more complex code by the
4985 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
4986 may only be used in a function which <a href="#gc">specifies a GC
4992 <!-- _______________________________________________________________________ -->
4993 <div class="doc_subsubsection">
4994 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
4997 <div class="doc_text">
5002 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
5007 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
5008 locations, allowing garbage collector implementations that require write
5009 barriers (such as generational or reference counting collectors).</p>
5013 <p>The first argument is the reference to store, the second is the start of the
5014 object to store it to, and the third is the address of the field of Obj to
5015 store to. If the runtime does not require a pointer to the object, Obj may be
5020 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
5021 instruction, but may be replaced with substantially more complex code by the
5022 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
5023 may only be used in a function which <a href="#gc">specifies a GC
5030 <!-- ======================================================================= -->
5031 <div class="doc_subsection">
5032 <a name="int_codegen">Code Generator Intrinsics</a>
5035 <div class="doc_text">
5037 These intrinsics are provided by LLVM to expose special features that may only
5038 be implemented with code generator support.
5043 <!-- _______________________________________________________________________ -->
5044 <div class="doc_subsubsection">
5045 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
5048 <div class="doc_text">
5052 declare i8 *@llvm.returnaddress(i32 <level>)
5058 The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
5059 target-specific value indicating the return address of the current function
5060 or one of its callers.
5066 The argument to this intrinsic indicates which function to return the address
5067 for. Zero indicates the calling function, one indicates its caller, etc. The
5068 argument is <b>required</b> to be a constant integer value.
5074 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
5075 the return address of the specified call frame, or zero if it cannot be
5076 identified. The value returned by this intrinsic is likely to be incorrect or 0
5077 for arguments other than zero, so it should only be used for debugging purposes.
5081 Note that calling this intrinsic does not prevent function inlining or other
5082 aggressive transformations, so the value returned may not be that of the obvious
5083 source-language caller.
5088 <!-- _______________________________________________________________________ -->
5089 <div class="doc_subsubsection">
5090 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
5093 <div class="doc_text">
5097 declare i8 *@llvm.frameaddress(i32 <level>)
5103 The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
5104 target-specific frame pointer value for the specified stack frame.
5110 The argument to this intrinsic indicates which function to return the frame
5111 pointer for. Zero indicates the calling function, one indicates its caller,
5112 etc. The argument is <b>required</b> to be a constant integer value.
5118 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
5119 the frame address of the specified call frame, or zero if it cannot be
5120 identified. The value returned by this intrinsic is likely to be incorrect or 0
5121 for arguments other than zero, so it should only be used for debugging purposes.
5125 Note that calling this intrinsic does not prevent function inlining or other
5126 aggressive transformations, so the value returned may not be that of the obvious
5127 source-language caller.
5131 <!-- _______________________________________________________________________ -->
5132 <div class="doc_subsubsection">
5133 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
5136 <div class="doc_text">
5140 declare i8 *@llvm.stacksave()
5146 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
5147 the function stack, for use with <a href="#int_stackrestore">
5148 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
5149 features like scoped automatic variable sized arrays in C99.
5155 This intrinsic returns a opaque pointer value that can be passed to <a
5156 href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
5157 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
5158 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
5159 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
5160 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
5161 that were allocated after the <tt>llvm.stacksave</tt> was executed.
5166 <!-- _______________________________________________________________________ -->
5167 <div class="doc_subsubsection">
5168 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
5171 <div class="doc_text">
5175 declare void @llvm.stackrestore(i8 * %ptr)
5181 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
5182 the function stack to the state it was in when the corresponding <a
5183 href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
5184 useful for implementing language features like scoped automatic variable sized
5191 See the description for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.
5197 <!-- _______________________________________________________________________ -->
5198 <div class="doc_subsubsection">
5199 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
5202 <div class="doc_text">
5206 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>)
5213 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
5214 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
5216 effect on the behavior of the program but can change its performance
5223 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
5224 determining if the fetch should be for a read (0) or write (1), and
5225 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
5226 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
5227 <tt>locality</tt> arguments must be constant integers.
5233 This intrinsic does not modify the behavior of the program. In particular,
5234 prefetches cannot trap and do not produce a value. On targets that support this
5235 intrinsic, the prefetch can provide hints to the processor cache for better
5241 <!-- _______________________________________________________________________ -->
5242 <div class="doc_subsubsection">
5243 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
5246 <div class="doc_text">
5250 declare void @llvm.pcmarker(i32 <id>)
5257 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
5259 code to simulators and other tools. The method is target specific, but it is
5260 expected that the marker will use exported symbols to transmit the PC of the
5262 The marker makes no guarantees that it will remain with any specific instruction
5263 after optimizations. It is possible that the presence of a marker will inhibit
5264 optimizations. The intended use is to be inserted after optimizations to allow
5265 correlations of simulation runs.
5271 <tt>id</tt> is a numerical id identifying the marker.
5277 This intrinsic does not modify the behavior of the program. Backends that do not
5278 support this intrinisic may ignore it.
5283 <!-- _______________________________________________________________________ -->
5284 <div class="doc_subsubsection">
5285 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
5288 <div class="doc_text">
5292 declare i64 @llvm.readcyclecounter( )
5299 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
5300 counter register (or similar low latency, high accuracy clocks) on those targets
5301 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
5302 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
5303 should only be used for small timings.
5309 When directly supported, reading the cycle counter should not modify any memory.
5310 Implementations are allowed to either return a application specific value or a
5311 system wide value. On backends without support, this is lowered to a constant 0.
5316 <!-- ======================================================================= -->
5317 <div class="doc_subsection">
5318 <a name="int_libc">Standard C Library Intrinsics</a>
5321 <div class="doc_text">
5323 LLVM provides intrinsics for a few important standard C library functions.
5324 These intrinsics allow source-language front-ends to pass information about the
5325 alignment of the pointer arguments to the code generator, providing opportunity
5326 for more efficient code generation.
5331 <!-- _______________________________________________________________________ -->
5332 <div class="doc_subsubsection">
5333 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
5336 <div class="doc_text">
5339 <p>This is an overloaded intrinsic. You can use llvm.memcpy on any integer bit
5340 width. Not all targets support all bit widths however.</p>
5342 declare void @llvm.memcpy.i8(i8 * <dest>, i8 * <src>,
5343 i8 <len>, i32 <align>)
5344 declare void @llvm.memcpy.i16(i8 * <dest>, i8 * <src>,
5345 i16 <len>, i32 <align>)
5346 declare void @llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
5347 i32 <len>, i32 <align>)
5348 declare void @llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
5349 i64 <len>, i32 <align>)
5355 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
5356 location to the destination location.
5360 Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
5361 intrinsics do not return a value, and takes an extra alignment argument.
5367 The first argument is a pointer to the destination, the second is a pointer to
5368 the source. The third argument is an integer argument
5369 specifying the number of bytes to copy, and the fourth argument is the alignment
5370 of the source and destination locations.
5374 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5375 the caller guarantees that both the source and destination pointers are aligned
5382 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
5383 location to the destination location, which are not allowed to overlap. It
5384 copies "len" bytes of memory over. If the argument is known to be aligned to
5385 some boundary, this can be specified as the fourth argument, otherwise it should
5391 <!-- _______________________________________________________________________ -->
5392 <div class="doc_subsubsection">
5393 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
5396 <div class="doc_text">
5399 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
5400 width. Not all targets support all bit widths however.</p>
5402 declare void @llvm.memmove.i8(i8 * <dest>, i8 * <src>,
5403 i8 <len>, i32 <align>)
5404 declare void @llvm.memmove.i16(i8 * <dest>, i8 * <src>,
5405 i16 <len>, i32 <align>)
5406 declare void @llvm.memmove.i32(i8 * <dest>, i8 * <src>,
5407 i32 <len>, i32 <align>)
5408 declare void @llvm.memmove.i64(i8 * <dest>, i8 * <src>,
5409 i64 <len>, i32 <align>)
5415 The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
5416 location to the destination location. It is similar to the
5417 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to overlap.
5421 Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
5422 intrinsics do not return a value, and takes an extra alignment argument.
5428 The first argument is a pointer to the destination, the second is a pointer to
5429 the source. The third argument is an integer argument
5430 specifying the number of bytes to copy, and the fourth argument is the alignment
5431 of the source and destination locations.
5435 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5436 the caller guarantees that the source and destination pointers are aligned to
5443 The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
5444 location to the destination location, which may overlap. It
5445 copies "len" bytes of memory over. If the argument is known to be aligned to
5446 some boundary, this can be specified as the fourth argument, otherwise it should
5452 <!-- _______________________________________________________________________ -->
5453 <div class="doc_subsubsection">
5454 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
5457 <div class="doc_text">
5460 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
5461 width. Not all targets support all bit widths however.</p>
5463 declare void @llvm.memset.i8(i8 * <dest>, i8 <val>,
5464 i8 <len>, i32 <align>)
5465 declare void @llvm.memset.i16(i8 * <dest>, i8 <val>,
5466 i16 <len>, i32 <align>)
5467 declare void @llvm.memset.i32(i8 * <dest>, i8 <val>,
5468 i32 <len>, i32 <align>)
5469 declare void @llvm.memset.i64(i8 * <dest>, i8 <val>,
5470 i64 <len>, i32 <align>)
5476 The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
5481 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
5482 does not return a value, and takes an extra alignment argument.
5488 The first argument is a pointer to the destination to fill, the second is the
5489 byte value to fill it with, the third argument is an integer
5490 argument specifying the number of bytes to fill, and the fourth argument is the
5491 known alignment of destination location.
5495 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5496 the caller guarantees that the destination pointer is aligned to that boundary.
5502 The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
5504 destination location. If the argument is known to be aligned to some boundary,
5505 this can be specified as the fourth argument, otherwise it should be set to 0 or
5511 <!-- _______________________________________________________________________ -->
5512 <div class="doc_subsubsection">
5513 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
5516 <div class="doc_text">
5519 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
5520 floating point or vector of floating point type. Not all targets support all
5523 declare float @llvm.sqrt.f32(float %Val)
5524 declare double @llvm.sqrt.f64(double %Val)
5525 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
5526 declare fp128 @llvm.sqrt.f128(fp128 %Val)
5527 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
5533 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
5534 returning the same value as the libm '<tt>sqrt</tt>' functions would. Unlike
5535 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
5536 negative numbers other than -0.0 (which allows for better optimization, because
5537 there is no need to worry about errno being set). <tt>llvm.sqrt(-0.0)</tt> is
5538 defined to return -0.0 like IEEE sqrt.
5544 The argument and return value are floating point numbers of the same type.
5550 This function returns the sqrt of the specified operand if it is a nonnegative
5551 floating point number.
5555 <!-- _______________________________________________________________________ -->
5556 <div class="doc_subsubsection">
5557 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
5560 <div class="doc_text">
5563 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
5564 floating point or vector of floating point type. Not all targets support all
5567 declare float @llvm.powi.f32(float %Val, i32 %power)
5568 declare double @llvm.powi.f64(double %Val, i32 %power)
5569 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
5570 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
5571 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
5577 The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
5578 specified (positive or negative) power. The order of evaluation of
5579 multiplications is not defined. When a vector of floating point type is
5580 used, the second argument remains a scalar integer value.
5586 The second argument is an integer power, and the first is a value to raise to
5593 This function returns the first value raised to the second power with an
5594 unspecified sequence of rounding operations.</p>
5597 <!-- _______________________________________________________________________ -->
5598 <div class="doc_subsubsection">
5599 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
5602 <div class="doc_text">
5605 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
5606 floating point or vector of floating point type. Not all targets support all
5609 declare float @llvm.sin.f32(float %Val)
5610 declare double @llvm.sin.f64(double %Val)
5611 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
5612 declare fp128 @llvm.sin.f128(fp128 %Val)
5613 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
5619 The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.
5625 The argument and return value are floating point numbers of the same type.
5631 This function returns the sine of the specified operand, returning the
5632 same values as the libm <tt>sin</tt> functions would, and handles error
5633 conditions in the same way.</p>
5636 <!-- _______________________________________________________________________ -->
5637 <div class="doc_subsubsection">
5638 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
5641 <div class="doc_text">
5644 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
5645 floating point or vector of floating point type. Not all targets support all
5648 declare float @llvm.cos.f32(float %Val)
5649 declare double @llvm.cos.f64(double %Val)
5650 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
5651 declare fp128 @llvm.cos.f128(fp128 %Val)
5652 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
5658 The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.
5664 The argument and return value are floating point numbers of the same type.
5670 This function returns the cosine of the specified operand, returning the
5671 same values as the libm <tt>cos</tt> functions would, and handles error
5672 conditions in the same way.</p>
5675 <!-- _______________________________________________________________________ -->
5676 <div class="doc_subsubsection">
5677 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
5680 <div class="doc_text">
5683 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
5684 floating point or vector of floating point type. Not all targets support all
5687 declare float @llvm.pow.f32(float %Val, float %Power)
5688 declare double @llvm.pow.f64(double %Val, double %Power)
5689 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
5690 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
5691 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
5697 The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
5698 specified (positive or negative) power.
5704 The second argument is a floating point power, and the first is a value to
5705 raise to that power.
5711 This function returns the first value raised to the second power,
5713 same values as the libm <tt>pow</tt> functions would, and handles error
5714 conditions in the same way.</p>
5718 <!-- ======================================================================= -->
5719 <div class="doc_subsection">
5720 <a name="int_manip">Bit Manipulation Intrinsics</a>
5723 <div class="doc_text">
5725 LLVM provides intrinsics for a few important bit manipulation operations.
5726 These allow efficient code generation for some algorithms.
5731 <!-- _______________________________________________________________________ -->
5732 <div class="doc_subsubsection">
5733 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
5736 <div class="doc_text">
5739 <p>This is an overloaded intrinsic function. You can use bswap on any integer
5740 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
5742 declare i16 @llvm.bswap.i16(i16 <id>)
5743 declare i32 @llvm.bswap.i32(i32 <id>)
5744 declare i64 @llvm.bswap.i64(i64 <id>)
5750 The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
5751 values with an even number of bytes (positive multiple of 16 bits). These are
5752 useful for performing operations on data that is not in the target's native
5759 The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
5760 and low byte of the input i16 swapped. Similarly, the <tt>llvm.bswap.i32</tt>
5761 intrinsic returns an i32 value that has the four bytes of the input i32
5762 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the returned
5763 i32 will have its bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i48</tt>,
5764 <tt>llvm.bswap.i64</tt> and other intrinsics extend this concept to
5765 additional even-byte lengths (6 bytes, 8 bytes and more, respectively).
5770 <!-- _______________________________________________________________________ -->
5771 <div class="doc_subsubsection">
5772 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
5775 <div class="doc_text">
5778 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
5779 width. Not all targets support all bit widths however.</p>
5781 declare i8 @llvm.ctpop.i8(i8 <src>)
5782 declare i16 @llvm.ctpop.i16(i16 <src>)
5783 declare i32 @llvm.ctpop.i32(i32 <src>)
5784 declare i64 @llvm.ctpop.i64(i64 <src>)
5785 declare i256 @llvm.ctpop.i256(i256 <src>)
5791 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
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.ctpop</tt>' intrinsic counts the 1's in a variable.
5809 <!-- _______________________________________________________________________ -->
5810 <div class="doc_subsubsection">
5811 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
5814 <div class="doc_text">
5817 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
5818 integer bit width. Not all targets support all bit widths however.</p>
5820 declare i8 @llvm.ctlz.i8 (i8 <src>)
5821 declare i16 @llvm.ctlz.i16(i16 <src>)
5822 declare i32 @llvm.ctlz.i32(i32 <src>)
5823 declare i64 @llvm.ctlz.i64(i64 <src>)
5824 declare i256 @llvm.ctlz.i256(i256 <src>)
5830 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
5831 leading zeros in a variable.
5837 The only argument is the value to be counted. The argument may be of any
5838 integer type. The return type must match the argument type.
5844 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
5845 in a variable. If the src == 0 then the result is the size in bits of the type
5846 of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.
5852 <!-- _______________________________________________________________________ -->
5853 <div class="doc_subsubsection">
5854 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
5857 <div class="doc_text">
5860 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
5861 integer bit width. Not all targets support all bit widths however.</p>
5863 declare i8 @llvm.cttz.i8 (i8 <src>)
5864 declare i16 @llvm.cttz.i16(i16 <src>)
5865 declare i32 @llvm.cttz.i32(i32 <src>)
5866 declare i64 @llvm.cttz.i64(i64 <src>)
5867 declare i256 @llvm.cttz.i256(i256 <src>)
5873 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
5880 The only argument is the value to be counted. The argument may be of any
5881 integer type. The return type must match the argument type.
5887 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
5888 in a variable. If the src == 0 then the result is the size in bits of the type
5889 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
5893 <!-- _______________________________________________________________________ -->
5894 <div class="doc_subsubsection">
5895 <a name="int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic</a>
5898 <div class="doc_text">
5901 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.select</tt>
5902 on any integer bit width.</p>
5904 declare i17 @llvm.part.select.i17 (i17 %val, i32 %loBit, i32 %hiBit)
5905 declare i29 @llvm.part.select.i29 (i29 %val, i32 %loBit, i32 %hiBit)
5909 <p>The '<tt>llvm.part.select</tt>' family of intrinsic functions selects a
5910 range of bits from an integer value and returns them in the same bit width as
5911 the original value.</p>
5914 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5915 any bit width but they must have the same bit width. The second and third
5916 arguments must be <tt>i32</tt> type since they specify only a bit index.</p>
5919 <p>The operation of the '<tt>llvm.part.select</tt>' intrinsic has two modes
5920 of operation: forwards and reverse. If <tt>%loBit</tt> is greater than
5921 <tt>%hiBits</tt> then the intrinsic operates in reverse mode. Otherwise it
5922 operates in forward mode.</p>
5923 <p>In forward mode, this intrinsic is the equivalent of shifting <tt>%val</tt>
5924 right by <tt>%loBit</tt> bits and then ANDing it with a mask with
5925 only the <tt>%hiBit - %loBit</tt> bits set, as follows:</p>
5927 <li>The <tt>%val</tt> is shifted right (LSHR) by the number of bits specified
5928 by <tt>%loBits</tt>. This normalizes the value to the low order bits.</li>
5929 <li>The <tt>%loBits</tt> value is subtracted from the <tt>%hiBits</tt> value
5930 to determine the number of bits to retain.</li>
5931 <li>A mask of the retained bits is created by shifting a -1 value.</li>
5932 <li>The mask is ANDed with <tt>%val</tt> to produce the result.</li>
5934 <p>In reverse mode, a similar computation is made except that the bits are
5935 returned in the reverse order. So, for example, if <tt>X</tt> has the value
5936 <tt>i16 0x0ACF (101011001111)</tt> and we apply
5937 <tt>part.select(i16 X, 8, 3)</tt> to it, we get back the value
5938 <tt>i16 0x0026 (000000100110)</tt>.</p>
5941 <div class="doc_subsubsection">
5942 <a name="int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic</a>
5945 <div class="doc_text">
5948 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.set</tt>
5949 on any integer bit width.</p>
5951 declare i17 @llvm.part.set.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
5952 declare i29 @llvm.part.set.i29.i9 (i29 %val, i9 %repl, i32 %lo, i32 %hi)
5956 <p>The '<tt>llvm.part.set</tt>' family of intrinsic functions replaces a range
5957 of bits in an integer value with another integer value. It returns the integer
5958 with the replaced bits.</p>
5961 <p>The first argument, <tt>%val</tt>, and the result may be integer types of
5962 any bit width, but they must have the same bit width. <tt>%val</tt> is the value
5963 whose bits will be replaced. The second argument, <tt>%repl</tt> may be an
5964 integer of any bit width. The third and fourth arguments must be <tt>i32</tt>
5965 type since they specify only a bit index.</p>
5968 <p>The operation of the '<tt>llvm.part.set</tt>' intrinsic has two modes
5969 of operation: forwards and reverse. If <tt>%lo</tt> is greater than
5970 <tt>%hi</tt> then the intrinsic operates in reverse mode. Otherwise it
5971 operates in forward mode.</p>
5973 <p>For both modes, the <tt>%repl</tt> value is prepared for use by either
5974 truncating it down to the size of the replacement area or zero extending it
5975 up to that size.</p>
5977 <p>In forward mode, the bits between <tt>%lo</tt> and <tt>%hi</tt> (inclusive)
5978 are replaced with corresponding bits from <tt>%repl</tt>. That is the 0th bit
5979 in <tt>%repl</tt> replaces the <tt>%lo</tt>th bit in <tt>%val</tt> and etc. up
5980 to the <tt>%hi</tt>th bit.</p>
5982 <p>In reverse mode, a similar computation is made except that the bits are
5983 reversed. That is, the <tt>0</tt>th bit in <tt>%repl</tt> replaces the
5984 <tt>%hi</tt> bit in <tt>%val</tt> and etc. down to the <tt>%lo</tt>th bit.</p>
5989 llvm.part.set(0xFFFF, 0, 4, 7) -> 0xFF0F
5990 llvm.part.set(0xFFFF, 0, 7, 4) -> 0xFF0F
5991 llvm.part.set(0xFFFF, 1, 7, 4) -> 0xFF8F
5992 llvm.part.set(0xFFFF, F, 8, 3) -> 0xFFE7
5993 llvm.part.set(0xFFFF, 0, 3, 8) -> 0xFE07
5998 <!-- ======================================================================= -->
5999 <div class="doc_subsection">
6000 <a name="int_overflow">Arithmetic with Overflow Intrinsics</a>
6003 <div class="doc_text">
6005 LLVM provides intrinsics for some arithmetic with overflow operations.
6010 <!-- _______________________________________________________________________ -->
6011 <div class="doc_subsubsection">
6012 <a name="int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt>' Intrinsics</a>
6015 <div class="doc_text">
6019 <p>This is an overloaded intrinsic. You can use <tt>llvm.sadd.with.overflow</tt>
6020 on any integer bit width.</p>
6023 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
6024 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6025 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
6030 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
6031 a signed addition of the two arguments, and indicate whether an overflow
6032 occurred during the signed summation.</p>
6036 <p>The arguments (%a and %b) and the first element of the result structure may
6037 be of integer types of any bit width, but they must have the same bit width. The
6038 second element of the result structure must be of type <tt>i1</tt>. <tt>%a</tt>
6039 and <tt>%b</tt> are the two values that will undergo signed addition.</p>
6043 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
6044 a signed addition of the two variables. They return a structure — the
6045 first element of which is the signed summation, and the second element of which
6046 is a bit specifying if the signed summation resulted in an overflow.</p>
6050 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6051 %sum = extractvalue {i32, i1} %res, 0
6052 %obit = extractvalue {i32, i1} %res, 1
6053 br i1 %obit, label %overflow, label %normal
6058 <!-- _______________________________________________________________________ -->
6059 <div class="doc_subsubsection">
6060 <a name="int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt>' Intrinsics</a>
6063 <div class="doc_text">
6067 <p>This is an overloaded intrinsic. You can use <tt>llvm.uadd.with.overflow</tt>
6068 on any integer bit width.</p>
6071 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
6072 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6073 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
6078 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
6079 an unsigned addition of the two arguments, and indicate whether a carry occurred
6080 during the unsigned summation.</p>
6084 <p>The arguments (%a and %b) and the first element of the result structure may
6085 be of integer types of any bit width, but they must have the same bit width. The
6086 second element of the result structure must be of type <tt>i1</tt>. <tt>%a</tt>
6087 and <tt>%b</tt> are the two values that will undergo unsigned addition.</p>
6091 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
6092 an unsigned addition of the two arguments. They return a structure — the
6093 first element of which is the sum, and the second element of which is a bit
6094 specifying if the unsigned summation resulted in a carry.</p>
6098 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6099 %sum = extractvalue {i32, i1} %res, 0
6100 %obit = extractvalue {i32, i1} %res, 1
6101 br i1 %obit, label %carry, label %normal
6106 <!-- _______________________________________________________________________ -->
6107 <div class="doc_subsubsection">
6108 <a name="int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt>' Intrinsics</a>
6111 <div class="doc_text">
6115 <p>This is an overloaded intrinsic. You can use <tt>llvm.ssub.with.overflow</tt>
6116 on any integer bit width.</p>
6119 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
6120 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6121 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
6126 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
6127 a signed subtraction of the two arguments, and indicate whether an overflow
6128 occurred during the signed subtraction.</p>
6132 <p>The arguments (%a and %b) and the first element of the result structure may
6133 be of integer types of any bit width, but they must have the same bit width. The
6134 second element of the result structure must be of type <tt>i1</tt>. <tt>%a</tt>
6135 and <tt>%b</tt> are the two values that will undergo signed subtraction.</p>
6139 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
6140 a signed subtraction of the two arguments. They return a structure — the
6141 first element of which is the subtraction, and the second element of which is a bit
6142 specifying if the signed subtraction resulted in an overflow.</p>
6146 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6147 %sum = extractvalue {i32, i1} %res, 0
6148 %obit = extractvalue {i32, i1} %res, 1
6149 br i1 %obit, label %overflow, label %normal
6154 <!-- _______________________________________________________________________ -->
6155 <div class="doc_subsubsection">
6156 <a name="int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt>' Intrinsics</a>
6159 <div class="doc_text">
6163 <p>This is an overloaded intrinsic. You can use <tt>llvm.usub.with.overflow</tt>
6164 on any integer bit width.</p>
6167 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
6168 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6169 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
6174 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
6175 an unsigned subtraction of the two arguments, and indicate whether an overflow
6176 occurred during the unsigned subtraction.</p>
6180 <p>The arguments (%a and %b) and the first element of the result structure may
6181 be of integer types of any bit width, but they must have the same bit width. The
6182 second element of the result structure must be of type <tt>i1</tt>. <tt>%a</tt>
6183 and <tt>%b</tt> are the two values that will undergo unsigned subtraction.</p>
6187 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
6188 an unsigned subtraction of the two arguments. They return a structure — the
6189 first element of which is the subtraction, and the second element of which is a bit
6190 specifying if the unsigned subtraction resulted in an overflow.</p>
6194 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6195 %sum = extractvalue {i32, i1} %res, 0
6196 %obit = extractvalue {i32, i1} %res, 1
6197 br i1 %obit, label %overflow, label %normal
6202 <!-- _______________________________________________________________________ -->
6203 <div class="doc_subsubsection">
6204 <a name="int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt>' Intrinsics</a>
6207 <div class="doc_text">
6211 <p>This is an overloaded intrinsic. You can use <tt>llvm.smul.with.overflow</tt>
6212 on any integer bit width.</p>
6215 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
6216 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6217 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
6222 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
6223 a signed multiplication of the two arguments, and indicate whether an overflow
6224 occurred during the signed multiplication.</p>
6228 <p>The arguments (%a and %b) and the first element of the result structure may
6229 be of integer types of any bit width, but they must have the same bit width. The
6230 second element of the result structure must be of type <tt>i1</tt>. <tt>%a</tt>
6231 and <tt>%b</tt> are the two values that will undergo signed multiplication.</p>
6235 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
6236 a signed multiplication of the two arguments. They return a structure —
6237 the first element of which is the multiplication, and the second element of
6238 which is a bit specifying if the signed multiplication resulted in an
6243 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6244 %sum = extractvalue {i32, i1} %res, 0
6245 %obit = extractvalue {i32, i1} %res, 1
6246 br i1 %obit, label %overflow, label %normal
6251 <!-- _______________________________________________________________________ -->
6252 <div class="doc_subsubsection">
6253 <a name="int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt>' Intrinsics</a>
6256 <div class="doc_text">
6260 <p>This is an overloaded intrinsic. You can use <tt>llvm.umul.with.overflow</tt>
6261 on any integer bit width.</p>
6264 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
6265 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6266 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
6271 <p><i><b>Warning:</b> '<tt>llvm.umul.with.overflow</tt>' is badly broken. It is
6272 actively being fixed, but it should not currently be used!</i></p>
6274 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
6275 a unsigned multiplication of the two arguments, and indicate whether an overflow
6276 occurred during the unsigned multiplication.</p>
6280 <p>The arguments (%a and %b) and the first element of the result structure may
6281 be of integer types of any bit width, but they must have the same bit width. The
6282 second element of the result structure must be of type <tt>i1</tt>. <tt>%a</tt>
6283 and <tt>%b</tt> are the two values that will undergo unsigned
6288 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
6289 an unsigned multiplication of the two arguments. They return a structure —
6290 the first element of which is the multiplication, and the second element of
6291 which is a bit specifying if the unsigned multiplication resulted in an
6296 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6297 %sum = extractvalue {i32, i1} %res, 0
6298 %obit = extractvalue {i32, i1} %res, 1
6299 br i1 %obit, label %overflow, label %normal
6304 <!-- ======================================================================= -->
6305 <div class="doc_subsection">
6306 <a name="int_debugger">Debugger Intrinsics</a>
6309 <div class="doc_text">
6311 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
6312 are described in the <a
6313 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
6314 Debugging</a> document.
6319 <!-- ======================================================================= -->
6320 <div class="doc_subsection">
6321 <a name="int_eh">Exception Handling Intrinsics</a>
6324 <div class="doc_text">
6325 <p> The LLVM exception handling intrinsics (which all start with
6326 <tt>llvm.eh.</tt> prefix), are described in the <a
6327 href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
6328 Handling</a> document. </p>
6331 <!-- ======================================================================= -->
6332 <div class="doc_subsection">
6333 <a name="int_trampoline">Trampoline Intrinsic</a>
6336 <div class="doc_text">
6338 This intrinsic makes it possible to excise one parameter, marked with
6339 the <tt>nest</tt> attribute, from a function. The result is a callable
6340 function pointer lacking the nest parameter - the caller does not need
6341 to provide a value for it. Instead, the value to use is stored in
6342 advance in a "trampoline", a block of memory usually allocated
6343 on the stack, which also contains code to splice the nest value into the
6344 argument list. This is used to implement the GCC nested function address
6348 For example, if the function is
6349 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
6350 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as follows:</p>
6352 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
6353 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
6354 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
6355 %fp = bitcast i8* %p to i32 (i32, i32)*
6357 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
6358 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
6361 <!-- _______________________________________________________________________ -->
6362 <div class="doc_subsubsection">
6363 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
6365 <div class="doc_text">
6368 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
6372 This fills the memory pointed to by <tt>tramp</tt> with code
6373 and returns a function pointer suitable for executing it.
6377 The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
6378 pointers. The <tt>tramp</tt> argument must point to a sufficiently large
6379 and sufficiently aligned block of memory; this memory is written to by the
6380 intrinsic. Note that the size and the alignment are target-specific - LLVM
6381 currently provides no portable way of determining them, so a front-end that
6382 generates this intrinsic needs to have some target-specific knowledge.
6383 The <tt>func</tt> argument must hold a function bitcast to an <tt>i8*</tt>.
6387 The block of memory pointed to by <tt>tramp</tt> is filled with target
6388 dependent code, turning it into a function. A pointer to this function is
6389 returned, but needs to be bitcast to an
6390 <a href="#int_trampoline">appropriate function pointer type</a>
6391 before being called. The new function's signature is the same as that of
6392 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
6393 removed. At most one such <tt>nest</tt> argument is allowed, and it must be
6394 of pointer type. Calling the new function is equivalent to calling
6395 <tt>func</tt> with the same argument list, but with <tt>nval</tt> used for the
6396 missing <tt>nest</tt> argument. If, after calling
6397 <tt>llvm.init.trampoline</tt>, the memory pointed to by <tt>tramp</tt> is
6398 modified, then the effect of any later call to the returned function pointer is
6403 <!-- ======================================================================= -->
6404 <div class="doc_subsection">
6405 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
6408 <div class="doc_text">
6410 These intrinsic functions expand the "universal IR" of LLVM to represent
6411 hardware constructs for atomic operations and memory synchronization. This
6412 provides an interface to the hardware, not an interface to the programmer. It
6413 is aimed at a low enough level to allow any programming models or APIs
6414 (Application Programming Interfaces) which
6415 need atomic behaviors to map cleanly onto it. It is also modeled primarily on
6416 hardware behavior. Just as hardware provides a "universal IR" for source
6417 languages, it also provides a starting point for developing a "universal"
6418 atomic operation and synchronization IR.
6421 These do <em>not</em> form an API such as high-level threading libraries,
6422 software transaction memory systems, atomic primitives, and intrinsic
6423 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
6424 application libraries. The hardware interface provided by LLVM should allow
6425 a clean implementation of all of these APIs and parallel programming models.
6426 No one model or paradigm should be selected above others unless the hardware
6427 itself ubiquitously does so.
6432 <!-- _______________________________________________________________________ -->
6433 <div class="doc_subsubsection">
6434 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
6436 <div class="doc_text">
6439 declare void @llvm.memory.barrier( i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>,
6445 The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
6446 specific pairs of memory access types.
6450 The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
6451 The first four arguments enables a specific barrier as listed below. The fith
6452 argument specifies that the barrier applies to io or device or uncached memory.
6456 <li><tt>ll</tt>: load-load barrier</li>
6457 <li><tt>ls</tt>: load-store barrier</li>
6458 <li><tt>sl</tt>: store-load barrier</li>
6459 <li><tt>ss</tt>: store-store barrier</li>
6460 <li><tt>device</tt>: barrier applies to device and uncached memory also.</li>
6464 This intrinsic causes the system to enforce some ordering constraints upon
6465 the loads and stores of the program. This barrier does not indicate
6466 <em>when</em> any events will occur, it only enforces an <em>order</em> in
6467 which they occur. For any of the specified pairs of load and store operations
6468 (f.ex. load-load, or store-load), all of the first operations preceding the
6469 barrier will complete before any of the second operations succeeding the
6470 barrier begin. Specifically the semantics for each pairing is as follows:
6473 <li><tt>ll</tt>: All loads before the barrier must complete before any load
6474 after the barrier begins.</li>
6476 <li><tt>ls</tt>: All loads before the barrier must complete before any
6477 store after the barrier begins.</li>
6478 <li><tt>ss</tt>: All stores before the barrier must complete before any
6479 store after the barrier begins.</li>
6480 <li><tt>sl</tt>: All stores before the barrier must complete before any
6481 load after the barrier begins.</li>
6484 These semantics are applied with a logical "and" behavior when more than one
6485 is enabled in a single memory barrier intrinsic.
6488 Backends may implement stronger barriers than those requested when they do not
6489 support as fine grained a barrier as requested. Some architectures do not
6490 need all types of barriers and on such architectures, these become noops.
6497 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
6498 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
6499 <i>; guarantee the above finishes</i>
6500 store i32 8, %ptr <i>; before this begins</i>
6504 <!-- _______________________________________________________________________ -->
6505 <div class="doc_subsubsection">
6506 <a name="int_atomic_cmp_swap">'<tt>llvm.atomic.cmp.swap.*</tt>' Intrinsic</a>
6508 <div class="doc_text">
6511 This is an overloaded intrinsic. You can use <tt>llvm.atomic.cmp.swap</tt> on
6512 any integer bit width and for different address spaces. Not all targets
6513 support all bit widths however.</p>
6516 declare i8 @llvm.atomic.cmp.swap.i8.p0i8( i8* <ptr>, i8 <cmp>, i8 <val> )
6517 declare i16 @llvm.atomic.cmp.swap.i16.p0i16( i16* <ptr>, i16 <cmp>, i16 <val> )
6518 declare i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* <ptr>, i32 <cmp>, i32 <val> )
6519 declare i64 @llvm.atomic.cmp.swap.i64.p0i64( i64* <ptr>, i64 <cmp>, i64 <val> )
6524 This loads a value in memory and compares it to a given value. If they are
6525 equal, it stores a new value into the memory.
6529 The <tt>llvm.atomic.cmp.swap</tt> intrinsic takes three arguments. The result as
6530 well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
6531 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
6532 this integer type. While any bit width integer may be used, targets may only
6533 lower representations they support in hardware.
6538 This entire intrinsic must be executed atomically. It first loads the value
6539 in memory pointed to by <tt>ptr</tt> and compares it with the value
6540 <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the memory. The
6541 loaded value is yielded in all cases. This provides the equivalent of an
6542 atomic compare-and-swap operation within the SSA framework.
6550 %val1 = add i32 4, 4
6551 %result1 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 4, %val1 )
6552 <i>; yields {i32}:result1 = 4</i>
6553 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
6554 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
6556 %val2 = add i32 1, 1
6557 %result2 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 5, %val2 )
6558 <i>; yields {i32}:result2 = 8</i>
6559 %stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
6561 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
6565 <!-- _______________________________________________________________________ -->
6566 <div class="doc_subsubsection">
6567 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
6569 <div class="doc_text">
6573 This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
6574 integer bit width. Not all targets support all bit widths however.</p>
6576 declare i8 @llvm.atomic.swap.i8.p0i8( i8* <ptr>, i8 <val> )
6577 declare i16 @llvm.atomic.swap.i16.p0i16( i16* <ptr>, i16 <val> )
6578 declare i32 @llvm.atomic.swap.i32.p0i32( i32* <ptr>, i32 <val> )
6579 declare i64 @llvm.atomic.swap.i64.p0i64( i64* <ptr>, i64 <val> )
6584 This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
6585 the value from memory. It then stores the value in <tt>val</tt> in the memory
6591 The <tt>llvm.atomic.swap</tt> intrinsic takes two arguments. Both the
6592 <tt>val</tt> argument and the result must be integers of the same bit width.
6593 The first argument, <tt>ptr</tt>, must be a pointer to a value of this
6594 integer type. The targets may only lower integer representations they
6599 This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
6600 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
6601 equivalent of an atomic swap operation within the SSA framework.
6609 %val1 = add i32 4, 4
6610 %result1 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val1 )
6611 <i>; yields {i32}:result1 = 4</i>
6612 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
6613 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
6615 %val2 = add i32 1, 1
6616 %result2 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val2 )
6617 <i>; yields {i32}:result2 = 8</i>
6619 %stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
6620 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
6624 <!-- _______________________________________________________________________ -->
6625 <div class="doc_subsubsection">
6626 <a name="int_atomic_load_add">'<tt>llvm.atomic.load.add.*</tt>' Intrinsic</a>
6629 <div class="doc_text">
6632 This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.add</tt> on any
6633 integer bit width. Not all targets support all bit widths however.</p>
6635 declare i8 @llvm.atomic.load.add.i8..p0i8( i8* <ptr>, i8 <delta> )
6636 declare i16 @llvm.atomic.load.add.i16..p0i16( i16* <ptr>, i16 <delta> )
6637 declare i32 @llvm.atomic.load.add.i32..p0i32( i32* <ptr>, i32 <delta> )
6638 declare i64 @llvm.atomic.load.add.i64..p0i64( i64* <ptr>, i64 <delta> )
6643 This intrinsic adds <tt>delta</tt> to the value stored in memory at
6644 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
6649 The intrinsic takes two arguments, the first a pointer to an integer value
6650 and the second an integer value. The result is also an integer value. These
6651 integer types can have any bit width, but they must all have the same bit
6652 width. The targets may only lower integer representations they support.
6656 This intrinsic does a series of operations atomically. It first loads the
6657 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
6658 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
6665 %result1 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 4 )
6666 <i>; yields {i32}:result1 = 4</i>
6667 %result2 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 2 )
6668 <i>; yields {i32}:result2 = 8</i>
6669 %result3 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 5 )
6670 <i>; yields {i32}:result3 = 10</i>
6671 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
6675 <!-- _______________________________________________________________________ -->
6676 <div class="doc_subsubsection">
6677 <a name="int_atomic_load_sub">'<tt>llvm.atomic.load.sub.*</tt>' Intrinsic</a>
6680 <div class="doc_text">
6683 This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.sub</tt> on
6684 any integer bit width and for different address spaces. Not all targets
6685 support all bit widths however.</p>
6687 declare i8 @llvm.atomic.load.sub.i8.p0i32( i8* <ptr>, i8 <delta> )
6688 declare i16 @llvm.atomic.load.sub.i16.p0i32( i16* <ptr>, i16 <delta> )
6689 declare i32 @llvm.atomic.load.sub.i32.p0i32( i32* <ptr>, i32 <delta> )
6690 declare i64 @llvm.atomic.load.sub.i64.p0i32( i64* <ptr>, i64 <delta> )
6695 This intrinsic subtracts <tt>delta</tt> to the value stored in memory at
6696 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
6701 The intrinsic takes two arguments, the first a pointer to an integer value
6702 and the second an integer value. The result is also an integer value. These
6703 integer types can have any bit width, but they must all have the same bit
6704 width. The targets may only lower integer representations they support.
6708 This intrinsic does a series of operations atomically. It first loads the
6709 value stored at <tt>ptr</tt>. It then subtracts <tt>delta</tt>, stores the
6710 result to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
6717 %result1 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 4 )
6718 <i>; yields {i32}:result1 = 8</i>
6719 %result2 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 2 )
6720 <i>; yields {i32}:result2 = 4</i>
6721 %result3 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 5 )
6722 <i>; yields {i32}:result3 = 2</i>
6723 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = -3</i>
6727 <!-- _______________________________________________________________________ -->
6728 <div class="doc_subsubsection">
6729 <a name="int_atomic_load_and">'<tt>llvm.atomic.load.and.*</tt>' Intrinsic</a><br>
6730 <a name="int_atomic_load_nand">'<tt>llvm.atomic.load.nand.*</tt>' Intrinsic</a><br>
6731 <a name="int_atomic_load_or">'<tt>llvm.atomic.load.or.*</tt>' Intrinsic</a><br>
6732 <a name="int_atomic_load_xor">'<tt>llvm.atomic.load.xor.*</tt>' Intrinsic</a><br>
6735 <div class="doc_text">
6738 These are overloaded intrinsics. You can use <tt>llvm.atomic.load_and</tt>,
6739 <tt>llvm.atomic.load_nand</tt>, <tt>llvm.atomic.load_or</tt>, and
6740 <tt>llvm.atomic.load_xor</tt> on any integer bit width and for different
6741 address spaces. Not all targets support all bit widths however.</p>
6743 declare i8 @llvm.atomic.load.and.i8.p0i8( i8* <ptr>, i8 <delta> )
6744 declare i16 @llvm.atomic.load.and.i16.p0i16( i16* <ptr>, i16 <delta> )
6745 declare i32 @llvm.atomic.load.and.i32.p0i32( i32* <ptr>, i32 <delta> )
6746 declare i64 @llvm.atomic.load.and.i64.p0i64( i64* <ptr>, i64 <delta> )
6751 declare i8 @llvm.atomic.load.or.i8.p0i8( i8* <ptr>, i8 <delta> )
6752 declare i16 @llvm.atomic.load.or.i16.p0i16( i16* <ptr>, i16 <delta> )
6753 declare i32 @llvm.atomic.load.or.i32.p0i32( i32* <ptr>, i32 <delta> )
6754 declare i64 @llvm.atomic.load.or.i64.p0i64( i64* <ptr>, i64 <delta> )
6759 declare i8 @llvm.atomic.load.nand.i8.p0i32( i8* <ptr>, i8 <delta> )
6760 declare i16 @llvm.atomic.load.nand.i16.p0i32( i16* <ptr>, i16 <delta> )
6761 declare i32 @llvm.atomic.load.nand.i32.p0i32( i32* <ptr>, i32 <delta> )
6762 declare i64 @llvm.atomic.load.nand.i64.p0i32( i64* <ptr>, i64 <delta> )
6767 declare i8 @llvm.atomic.load.xor.i8.p0i32( i8* <ptr>, i8 <delta> )
6768 declare i16 @llvm.atomic.load.xor.i16.p0i32( i16* <ptr>, i16 <delta> )
6769 declare i32 @llvm.atomic.load.xor.i32.p0i32( i32* <ptr>, i32 <delta> )
6770 declare i64 @llvm.atomic.load.xor.i64.p0i32( i64* <ptr>, i64 <delta> )
6775 These intrinsics bitwise the operation (and, nand, or, xor) <tt>delta</tt> to
6776 the value stored in memory at <tt>ptr</tt>. It yields the original value
6782 These intrinsics take two arguments, the first a pointer to an integer value
6783 and the second an integer value. The result is also an integer value. These
6784 integer types can have any bit width, but they must all have the same bit
6785 width. The targets may only lower integer representations they support.
6789 These intrinsics does a series of operations atomically. They first load the
6790 value stored at <tt>ptr</tt>. They then do the bitwise operation
6791 <tt>delta</tt>, store the result to <tt>ptr</tt>. They yield the original
6792 value stored at <tt>ptr</tt>.
6798 store i32 0x0F0F, %ptr
6799 %result0 = call i32 @llvm.atomic.load.nand.i32.p0i32( i32* %ptr, i32 0xFF )
6800 <i>; yields {i32}:result0 = 0x0F0F</i>
6801 %result1 = call i32 @llvm.atomic.load.and.i32.p0i32( i32* %ptr, i32 0xFF )
6802 <i>; yields {i32}:result1 = 0xFFFFFFF0</i>
6803 %result2 = call i32 @llvm.atomic.load.or.i32.p0i32( i32* %ptr, i32 0F )
6804 <i>; yields {i32}:result2 = 0xF0</i>
6805 %result3 = call i32 @llvm.atomic.load.xor.i32.p0i32( i32* %ptr, i32 0F )
6806 <i>; yields {i32}:result3 = FF</i>
6807 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = F0</i>
6812 <!-- _______________________________________________________________________ -->
6813 <div class="doc_subsubsection">
6814 <a name="int_atomic_load_max">'<tt>llvm.atomic.load.max.*</tt>' Intrinsic</a><br>
6815 <a name="int_atomic_load_min">'<tt>llvm.atomic.load.min.*</tt>' Intrinsic</a><br>
6816 <a name="int_atomic_load_umax">'<tt>llvm.atomic.load.umax.*</tt>' Intrinsic</a><br>
6817 <a name="int_atomic_load_umin">'<tt>llvm.atomic.load.umin.*</tt>' Intrinsic</a><br>
6820 <div class="doc_text">
6823 These are overloaded intrinsics. You can use <tt>llvm.atomic.load_max</tt>,
6824 <tt>llvm.atomic.load_min</tt>, <tt>llvm.atomic.load_umax</tt>, and
6825 <tt>llvm.atomic.load_umin</tt> on any integer bit width and for different
6826 address spaces. Not all targets
6827 support all bit widths however.</p>
6829 declare i8 @llvm.atomic.load.max.i8.p0i8( i8* <ptr>, i8 <delta> )
6830 declare i16 @llvm.atomic.load.max.i16.p0i16( i16* <ptr>, i16 <delta> )
6831 declare i32 @llvm.atomic.load.max.i32.p0i32( i32* <ptr>, i32 <delta> )
6832 declare i64 @llvm.atomic.load.max.i64.p0i64( i64* <ptr>, i64 <delta> )
6837 declare i8 @llvm.atomic.load.min.i8.p0i8( i8* <ptr>, i8 <delta> )
6838 declare i16 @llvm.atomic.load.min.i16.p0i16( i16* <ptr>, i16 <delta> )
6839 declare i32 @llvm.atomic.load.min.i32..p0i32( i32* <ptr>, i32 <delta> )
6840 declare i64 @llvm.atomic.load.min.i64..p0i64( i64* <ptr>, i64 <delta> )
6845 declare i8 @llvm.atomic.load.umax.i8.p0i8( i8* <ptr>, i8 <delta> )
6846 declare i16 @llvm.atomic.load.umax.i16.p0i16( i16* <ptr>, i16 <delta> )
6847 declare i32 @llvm.atomic.load.umax.i32.p0i32( i32* <ptr>, i32 <delta> )
6848 declare i64 @llvm.atomic.load.umax.i64.p0i64( i64* <ptr>, i64 <delta> )
6853 declare i8 @llvm.atomic.load.umin.i8..p0i8( i8* <ptr>, i8 <delta> )
6854 declare i16 @llvm.atomic.load.umin.i16.p0i16( i16* <ptr>, i16 <delta> )
6855 declare i32 @llvm.atomic.load.umin.i32..p0i32( i32* <ptr>, i32 <delta> )
6856 declare i64 @llvm.atomic.load.umin.i64..p0i64( i64* <ptr>, i64 <delta> )
6861 These intrinsics takes the signed or unsigned minimum or maximum of
6862 <tt>delta</tt> and the value stored in memory at <tt>ptr</tt>. It yields the
6863 original value at <tt>ptr</tt>.
6868 These intrinsics take two arguments, the first a pointer to an integer value
6869 and the second an integer value. The result is also an integer value. These
6870 integer types can have any bit width, but they must all have the same bit
6871 width. The targets may only lower integer representations they support.
6875 These intrinsics does a series of operations atomically. They first load the
6876 value stored at <tt>ptr</tt>. They then do the signed or unsigned min or max
6877 <tt>delta</tt> and the value, store the result to <tt>ptr</tt>. They yield
6878 the original value stored at <tt>ptr</tt>.
6885 %result0 = call i32 @llvm.atomic.load.min.i32.p0i32( i32* %ptr, i32 -2 )
6886 <i>; yields {i32}:result0 = 7</i>
6887 %result1 = call i32 @llvm.atomic.load.max.i32.p0i32( i32* %ptr, i32 8 )
6888 <i>; yields {i32}:result1 = -2</i>
6889 %result2 = call i32 @llvm.atomic.load.umin.i32.p0i32( i32* %ptr, i32 10 )
6890 <i>; yields {i32}:result2 = 8</i>
6891 %result3 = call i32 @llvm.atomic.load.umax.i32.p0i32( i32* %ptr, i32 30 )
6892 <i>; yields {i32}:result3 = 8</i>
6893 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 30</i>
6897 <!-- ======================================================================= -->
6898 <div class="doc_subsection">
6899 <a name="int_general">General Intrinsics</a>
6902 <div class="doc_text">
6903 <p> This class of intrinsics is designed to be generic and has
6904 no specific purpose. </p>
6907 <!-- _______________________________________________________________________ -->
6908 <div class="doc_subsubsection">
6909 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
6912 <div class="doc_text">
6916 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
6922 The '<tt>llvm.var.annotation</tt>' intrinsic
6928 The first argument is a pointer to a value, the second is a pointer to a
6929 global string, the third is a pointer to a global string which is the source
6930 file name, and the last argument is the line number.
6936 This intrinsic allows annotation of local variables with arbitrary strings.
6937 This can be useful for special purpose optimizations that want to look for these
6938 annotations. These have no other defined use, they are ignored by code
6939 generation and optimization.
6943 <!-- _______________________________________________________________________ -->
6944 <div class="doc_subsubsection">
6945 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
6948 <div class="doc_text">
6951 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
6952 any integer bit width.
6955 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
6956 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
6957 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
6958 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
6959 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
6965 The '<tt>llvm.annotation</tt>' intrinsic.
6971 The first argument is an integer value (result of some expression),
6972 the second is a pointer to a global string, the third is a pointer to a global
6973 string which is the source file name, and the last argument is the line number.
6974 It returns the value of the first argument.
6980 This intrinsic allows annotations to be put on arbitrary expressions
6981 with arbitrary strings. This can be useful for special purpose optimizations
6982 that want to look for these annotations. These have no other defined use, they
6983 are ignored by code generation and optimization.
6987 <!-- _______________________________________________________________________ -->
6988 <div class="doc_subsubsection">
6989 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
6992 <div class="doc_text">
6996 declare void @llvm.trap()
7002 The '<tt>llvm.trap</tt>' intrinsic
7014 This intrinsics is lowered to the target dependent trap instruction. If the
7015 target does not have a trap instruction, this intrinsic will be lowered to the
7016 call of the abort() function.
7020 <!-- _______________________________________________________________________ -->
7021 <div class="doc_subsubsection">
7022 <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a>
7024 <div class="doc_text">
7027 declare void @llvm.stackprotector( i8* <guard>, i8** <slot> )
7032 The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and stores
7033 it onto the stack at <tt>slot</tt>. The stack slot is adjusted to ensure that
7034 it is placed on the stack before local variables.
7038 The <tt>llvm.stackprotector</tt> intrinsic requires two pointer arguments. The
7039 first argument is the value loaded from the stack guard
7040 <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt> that
7041 has enough space to hold the value of the guard.
7045 This intrinsic causes the prologue/epilogue inserter to force the position of
7046 the <tt>AllocaInst</tt> stack slot to be before local variables on the
7047 stack. This is to ensure that if a local variable on the stack is overwritten,
7048 it will destroy the value of the guard. When the function exits, the guard on
7049 the stack is checked against the original guard. If they're different, then
7050 the program aborts by calling the <tt>__stack_chk_fail()</tt> function.
7054 <!-- *********************************************************************** -->
7057 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
7058 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
7059 <a href="http://validator.w3.org/check/referer"><img
7060 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
7062 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
7063 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
7064 Last modified: $Date$