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>
44 <li><a href="#t_metadata">Metadata Type</a></li>
47 <li><a href="#t_derived">Derived Types</a>
49 <li><a href="#t_integer">Integer Type</a></li>
50 <li><a href="#t_array">Array Type</a></li>
51 <li><a href="#t_function">Function Type</a></li>
52 <li><a href="#t_pointer">Pointer Type</a></li>
53 <li><a href="#t_struct">Structure Type</a></li>
54 <li><a href="#t_pstruct">Packed Structure Type</a></li>
55 <li><a href="#t_vector">Vector Type</a></li>
56 <li><a href="#t_opaque">Opaque Type</a></li>
59 <li><a href="#t_uprefs">Type Up-references</a></li>
62 <li><a href="#constants">Constants</a>
64 <li><a href="#simpleconstants">Simple Constants</a></li>
65 <li><a href="#complexconstants">Complex Constants</a></li>
66 <li><a href="#globalconstants">Global Variable and Function Addresses</a></li>
67 <li><a href="#undefvalues">Undefined Values</a></li>
68 <li><a href="#constantexprs">Constant Expressions</a></li>
69 <li><a href="#metadata">Embedded Metadata</a></li>
72 <li><a href="#othervalues">Other Values</a>
74 <li><a href="#inlineasm">Inline Assembler Expressions</a></li>
77 <li><a href="#instref">Instruction Reference</a>
79 <li><a href="#terminators">Terminator Instructions</a>
81 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
82 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
83 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
84 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
85 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
86 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
89 <li><a href="#binaryops">Binary Operations</a>
91 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
92 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
93 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
94 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
95 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
96 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
97 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
98 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
99 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
102 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
104 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
105 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
106 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
107 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
108 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
109 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
112 <li><a href="#vectorops">Vector Operations</a>
114 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
115 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
116 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
119 <li><a href="#aggregateops">Aggregate Operations</a>
121 <li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li>
122 <li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li>
125 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
127 <li><a href="#i_malloc">'<tt>malloc</tt>' Instruction</a></li>
128 <li><a href="#i_free">'<tt>free</tt>' Instruction</a></li>
129 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
130 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
131 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
132 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
135 <li><a href="#convertops">Conversion Operations</a>
137 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
138 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
139 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
140 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
141 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
142 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
143 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
144 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
145 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
146 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
147 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
148 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
151 <li><a href="#otherops">Other Operations</a>
153 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
154 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
155 <li><a href="#i_vicmp">'<tt>vicmp</tt>' Instruction</a></li>
156 <li><a href="#i_vfcmp">'<tt>vfcmp</tt>' Instruction</a></li>
157 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
158 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
159 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
160 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
165 <li><a href="#intrinsics">Intrinsic Functions</a>
167 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
169 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
170 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
171 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
174 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
176 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
177 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
178 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
181 <li><a href="#int_codegen">Code Generator Intrinsics</a>
183 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
184 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
185 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
186 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
187 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
188 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
189 <li><a href="#int_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
192 <li><a href="#int_libc">Standard C Library Intrinsics</a>
194 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
195 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
196 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
197 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
198 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
199 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
200 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
201 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
204 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
206 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
207 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
208 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
209 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
210 <li><a href="#int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic </a></li>
211 <li><a href="#int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic </a></li>
214 <li><a href="#int_overflow">Arithmetic with Overflow Intrinsics</a>
216 <li><a href="#int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt> Intrinsics</a></li>
217 <li><a href="#int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt> Intrinsics</a></li>
218 <li><a href="#int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt> Intrinsics</a></li>
219 <li><a href="#int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt> Intrinsics</a></li>
220 <li><a href="#int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt> Intrinsics</a></li>
221 <li><a href="#int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt> Intrinsics</a></li>
224 <li><a href="#int_debugger">Debugger intrinsics</a></li>
225 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
226 <li><a href="#int_trampoline">Trampoline Intrinsic</a>
228 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
231 <li><a href="#int_atomics">Atomic intrinsics</a>
233 <li><a href="#int_memory_barrier"><tt>llvm.memory_barrier</tt></a></li>
234 <li><a href="#int_atomic_cmp_swap"><tt>llvm.atomic.cmp.swap</tt></a></li>
235 <li><a href="#int_atomic_swap"><tt>llvm.atomic.swap</tt></a></li>
236 <li><a href="#int_atomic_load_add"><tt>llvm.atomic.load.add</tt></a></li>
237 <li><a href="#int_atomic_load_sub"><tt>llvm.atomic.load.sub</tt></a></li>
238 <li><a href="#int_atomic_load_and"><tt>llvm.atomic.load.and</tt></a></li>
239 <li><a href="#int_atomic_load_nand"><tt>llvm.atomic.load.nand</tt></a></li>
240 <li><a href="#int_atomic_load_or"><tt>llvm.atomic.load.or</tt></a></li>
241 <li><a href="#int_atomic_load_xor"><tt>llvm.atomic.load.xor</tt></a></li>
242 <li><a href="#int_atomic_load_max"><tt>llvm.atomic.load.max</tt></a></li>
243 <li><a href="#int_atomic_load_min"><tt>llvm.atomic.load.min</tt></a></li>
244 <li><a href="#int_atomic_load_umax"><tt>llvm.atomic.load.umax</tt></a></li>
245 <li><a href="#int_atomic_load_umin"><tt>llvm.atomic.load.umin</tt></a></li>
248 <li><a href="#int_general">General intrinsics</a>
250 <li><a href="#int_var_annotation">
251 '<tt>llvm.var.annotation</tt>' Intrinsic</a></li>
252 <li><a href="#int_annotation">
253 '<tt>llvm.annotation.*</tt>' Intrinsic</a></li>
254 <li><a href="#int_trap">
255 '<tt>llvm.trap</tt>' Intrinsic</a></li>
256 <li><a href="#int_stackprotector">
257 '<tt>llvm.stackprotector</tt>' Intrinsic</a></li>
264 <div class="doc_author">
265 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
266 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
269 <!-- *********************************************************************** -->
270 <div class="doc_section"> <a name="abstract">Abstract </a></div>
271 <!-- *********************************************************************** -->
273 <div class="doc_text">
274 <p>This document is a reference manual for the LLVM assembly language.
275 LLVM is a Static Single Assignment (SSA) based representation that provides
276 type safety, low-level operations, flexibility, and the capability of
277 representing 'all' high-level languages cleanly. It is the common code
278 representation used throughout all phases of the LLVM compilation
282 <!-- *********************************************************************** -->
283 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
284 <!-- *********************************************************************** -->
286 <div class="doc_text">
288 <p>The LLVM code representation is designed to be used in three
289 different forms: as an in-memory compiler IR, as an on-disk bitcode
290 representation (suitable for fast loading by a Just-In-Time compiler),
291 and as a human readable assembly language representation. This allows
292 LLVM to provide a powerful intermediate representation for efficient
293 compiler transformations and analysis, while providing a natural means
294 to debug and visualize the transformations. The three different forms
295 of LLVM are all equivalent. This document describes the human readable
296 representation and notation.</p>
298 <p>The LLVM representation aims to be light-weight and low-level
299 while being expressive, typed, and extensible at the same time. It
300 aims to be a "universal IR" of sorts, by being at a low enough level
301 that high-level ideas may be cleanly mapped to it (similar to how
302 microprocessors are "universal IR's", allowing many source languages to
303 be mapped to them). By providing type information, LLVM can be used as
304 the target of optimizations: for example, through pointer analysis, it
305 can be proven that a C automatic variable is never accessed outside of
306 the current function... allowing it to be promoted to a simple SSA
307 value instead of a memory location.</p>
311 <!-- _______________________________________________________________________ -->
312 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
314 <div class="doc_text">
316 <p>It is important to note that this document describes 'well formed'
317 LLVM assembly language. There is a difference between what the parser
318 accepts and what is considered 'well formed'. For example, the
319 following instruction is syntactically okay, but not well formed:</p>
321 <div class="doc_code">
323 %x = <a href="#i_add">add</a> i32 1, %x
327 <p>...because the definition of <tt>%x</tt> does not dominate all of
328 its uses. The LLVM infrastructure provides a verification pass that may
329 be used to verify that an LLVM module is well formed. This pass is
330 automatically run by the parser after parsing input assembly and by
331 the optimizer before it outputs bitcode. The violations pointed out
332 by the verifier pass indicate bugs in transformation passes or input to
336 <!-- Describe the typesetting conventions here. -->
338 <!-- *********************************************************************** -->
339 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
340 <!-- *********************************************************************** -->
342 <div class="doc_text">
344 <p>LLVM identifiers come in two basic types: global and local. Global
345 identifiers (functions, global variables) begin with the @ character. Local
346 identifiers (register names, types) begin with the % character. Additionally,
347 there are three different formats for identifiers, for different purposes:</p>
350 <li>Named values are represented as a string of characters with their prefix.
351 For example, %foo, @DivisionByZero, %a.really.long.identifier. The actual
352 regular expression used is '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
353 Identifiers which require other characters in their names can be surrounded
354 with quotes. Special characters may be escaped using "\xx" where xx is the
355 ASCII code for the character in hexadecimal. In this way, any character can
356 be used in a name value, even quotes themselves.
358 <li>Unnamed values are represented as an unsigned numeric value with their
359 prefix. For example, %12, @2, %44.</li>
361 <li>Constants, which are described in a <a href="#constants">section about
362 constants</a>, below.</li>
365 <p>LLVM requires that values start with a prefix for two reasons: Compilers
366 don't need to worry about name clashes with reserved words, and the set of
367 reserved words may be expanded in the future without penalty. Additionally,
368 unnamed identifiers allow a compiler to quickly come up with a temporary
369 variable without having to avoid symbol table conflicts.</p>
371 <p>Reserved words in LLVM are very similar to reserved words in other
372 languages. There are keywords for different opcodes
373 ('<tt><a href="#i_add">add</a></tt>',
374 '<tt><a href="#i_bitcast">bitcast</a></tt>',
375 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
376 href="#t_void">void</a></tt>', '<tt><a href="#t_primitive">i32</a></tt>', etc...),
377 and others. These reserved words cannot conflict with variable names, because
378 none of them start with a prefix character ('%' or '@').</p>
380 <p>Here is an example of LLVM code to multiply the integer variable
381 '<tt>%X</tt>' by 8:</p>
385 <div class="doc_code">
387 %result = <a href="#i_mul">mul</a> i32 %X, 8
391 <p>After strength reduction:</p>
393 <div class="doc_code">
395 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
399 <p>And the hard way:</p>
401 <div class="doc_code">
403 <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
404 <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
405 %result = <a href="#i_add">add</a> i32 %1, %1
409 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
410 important lexical features of LLVM:</p>
414 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
417 <li>Unnamed temporaries are created when the result of a computation is not
418 assigned to a named value.</li>
420 <li>Unnamed temporaries are numbered sequentially</li>
424 <p>...and it also shows a convention that we follow in this document. When
425 demonstrating instructions, we will follow an instruction with a comment that
426 defines the type and name of value produced. Comments are shown in italic
431 <!-- *********************************************************************** -->
432 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
433 <!-- *********************************************************************** -->
435 <!-- ======================================================================= -->
436 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
439 <div class="doc_text">
441 <p>LLVM programs are composed of "Module"s, each of which is a
442 translation unit of the input programs. Each module consists of
443 functions, global variables, and symbol table entries. Modules may be
444 combined together with the LLVM linker, which merges function (and
445 global variable) definitions, resolves forward declarations, and merges
446 symbol table entries. Here is an example of the "hello world" module:</p>
448 <div class="doc_code">
449 <pre><i>; Declare the string constant as a global constant...</i>
450 <a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a
451 href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
453 <i>; External declaration of the puts function</i>
454 <a href="#functionstructure">declare</a> i32 @puts(i8 *) <i>; i32(i8 *)* </i>
456 <i>; Definition of main function</i>
457 define i32 @main() { <i>; i32()* </i>
458 <i>; Convert [13 x i8]* to i8 *...</i>
460 href="#i_getelementptr">getelementptr</a> [13 x i8]* @.LC0, i64 0, i64 0 <i>; i8 *</i>
462 <i>; Call puts function to write out the string to stdout...</i>
464 href="#i_call">call</a> i32 @puts(i8 * %cast210) <i>; i32</i>
466 href="#i_ret">ret</a> i32 0<br>}<br>
470 <p>This example is made up of a <a href="#globalvars">global variable</a>
471 named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
472 function, and a <a href="#functionstructure">function definition</a>
473 for "<tt>main</tt>".</p>
475 <p>In general, a module is made up of a list of global values,
476 where both functions and global variables are global values. Global values are
477 represented by a pointer to a memory location (in this case, a pointer to an
478 array of char, and a pointer to a function), and have one of the following <a
479 href="#linkage">linkage types</a>.</p>
483 <!-- ======================================================================= -->
484 <div class="doc_subsection">
485 <a name="linkage">Linkage Types</a>
488 <div class="doc_text">
491 All Global Variables and Functions have one of the following types of linkage:
496 <dt><tt><b><a name="linkage_private">private</a></b></tt>: </dt>
498 <dd>Global values with private linkage are only directly accessible by
499 objects in the current module. In particular, linking code into a module with
500 an private global value may cause the private to be renamed as necessary to
501 avoid collisions. Because the symbol is private to the module, all
502 references can be updated. This doesn't show up in any symbol table in the
506 <dt><tt><b><a name="linkage_internal">internal</a></b></tt>: </dt>
508 <dd> Similar to private, but the value shows as a local symbol (STB_LOCAL in
509 the case of ELF) in the object file. This corresponds to the notion of the
510 '<tt>static</tt>' keyword in C.
513 <dt><tt><b><a name="available_externally">available_externally</a></b></tt>:
516 <dd>Globals with "<tt>available_externally</tt>" linkage are never emitted
517 into the object file corresponding to the LLVM module. They exist to
518 allow inlining and other optimizations to take place given knowledge of the
519 definition of the global, which is known to be somewhere outside the module.
520 Globals with <tt>available_externally</tt> linkage are allowed to be discarded
521 at will, and are otherwise the same as <tt>linkonce_odr</tt>. This linkage
522 type is only allowed on definitions, not declarations.</dd>
524 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
526 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
527 the same name when linkage occurs. This is typically used to implement
528 inline functions, templates, or other code which must be generated in each
529 translation unit that uses it. Unreferenced <tt>linkonce</tt> globals are
530 allowed to be discarded.
533 <dt><tt><b><a name="linkage_common">common</a></b></tt>: </dt>
535 <dd>"<tt>common</tt>" linkage is exactly the same as <tt>linkonce</tt>
536 linkage, except that unreferenced <tt>common</tt> globals may not be
537 discarded. This is used for globals that may be emitted in multiple
538 translation units, but that are not guaranteed to be emitted into every
539 translation unit that uses them. One example of this is tentative
540 definitions in C, such as "<tt>int X;</tt>" at global scope.
543 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
545 <dd>"<tt>weak</tt>" linkage is the same as <tt>common</tt> linkage, except
546 that some targets may choose to emit different assembly sequences for them
547 for target-dependent reasons. This is used for globals that are declared
548 "weak" in C source code.
551 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
553 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
554 pointer to array type. When two global variables with appending linkage are
555 linked together, the two global arrays are appended together. This is the
556 LLVM, typesafe, equivalent of having the system linker append together
557 "sections" with identical names when .o files are linked.
560 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt>: </dt>
562 <dd>The semantics of this linkage follow the ELF object file model: the
563 symbol is weak until linked, if not linked, the symbol becomes null instead
564 of being an undefined reference.
567 <dt><tt><b><a name="linkage_linkonce">linkonce_odr</a></b></tt>: </dt>
568 <dt><tt><b><a name="linkage_weak">weak_odr</a></b></tt>: </dt>
569 <dd>Some languages allow differing globals to be merged, such as two
570 functions with different semantics. Other languages, such as <tt>C++</tt>,
571 ensure that only equivalent globals are ever merged (the "one definition
572 rule" - "ODR"). Such languages can use the <tt>linkonce_odr</tt>
573 and <tt>weak_odr</tt> linkage types to indicate that the global will only
574 be merged with equivalent globals. These linkage types are otherwise the
575 same as their non-<tt>odr</tt> versions.
578 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
580 <dd>If none of the above identifiers are used, the global is externally
581 visible, meaning that it participates in linkage and can be used to resolve
582 external symbol references.
587 The next two types of linkage are targeted for Microsoft Windows platform
588 only. They are designed to support importing (exporting) symbols from (to)
589 DLLs (Dynamic Link Libraries).
593 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt>: </dt>
595 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
596 or variable via a global pointer to a pointer that is set up by the DLL
597 exporting the symbol. On Microsoft Windows targets, the pointer name is
598 formed by combining <code>__imp_</code> and the function or variable name.
601 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt>: </dt>
603 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
604 pointer to a pointer in a DLL, so that it can be referenced with the
605 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
606 name is formed by combining <code>__imp_</code> and the function or variable
612 <p>For example, since the "<tt>.LC0</tt>"
613 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
614 variable and was linked with this one, one of the two would be renamed,
615 preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
616 external (i.e., lacking any linkage declarations), they are accessible
617 outside of the current module.</p>
618 <p>It is illegal for a function <i>declaration</i>
619 to have any linkage type other than "externally visible", <tt>dllimport</tt>
620 or <tt>extern_weak</tt>.</p>
621 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt>, <tt>weak</tt>
622 or <tt>weak_odr</tt> linkages.</p>
625 <!-- ======================================================================= -->
626 <div class="doc_subsection">
627 <a name="callingconv">Calling Conventions</a>
630 <div class="doc_text">
632 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
633 and <a href="#i_invoke">invokes</a> can all have an optional calling convention
634 specified for the call. The calling convention of any pair of dynamic
635 caller/callee must match, or the behavior of the program is undefined. The
636 following calling conventions are supported by LLVM, and more may be added in
640 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
642 <dd>This calling convention (the default if no other calling convention is
643 specified) matches the target C calling conventions. This calling convention
644 supports varargs function calls and tolerates some mismatch in the declared
645 prototype and implemented declaration of the function (as does normal C).
648 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
650 <dd>This calling convention attempts to make calls as fast as possible
651 (e.g. by passing things in registers). This calling convention allows the
652 target to use whatever tricks it wants to produce fast code for the target,
653 without having to conform to an externally specified ABI (Application Binary
654 Interface). Implementations of this convention should allow arbitrary
655 <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> to be
656 supported. This calling convention does not support varargs and requires the
657 prototype of all callees to exactly match the prototype of the function
661 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
663 <dd>This calling convention attempts to make code in the caller as efficient
664 as possible under the assumption that the call is not commonly executed. As
665 such, these calls often preserve all registers so that the call does not break
666 any live ranges in the caller side. This calling convention does not support
667 varargs and requires the prototype of all callees to exactly match the
668 prototype of the function definition.
671 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
673 <dd>Any calling convention may be specified by number, allowing
674 target-specific calling conventions to be used. Target specific calling
675 conventions start at 64.
679 <p>More calling conventions can be added/defined on an as-needed basis, to
680 support pascal conventions or any other well-known target-independent
685 <!-- ======================================================================= -->
686 <div class="doc_subsection">
687 <a name="visibility">Visibility Styles</a>
690 <div class="doc_text">
693 All Global Variables and Functions have one of the following visibility styles:
697 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
699 <dd>On targets that use the ELF object file format, default visibility means
700 that the declaration is visible to other
701 modules and, in shared libraries, means that the declared entity may be
702 overridden. On Darwin, default visibility means that the declaration is
703 visible to other modules. Default visibility corresponds to "external
704 linkage" in the language.
707 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
709 <dd>Two declarations of an object with hidden visibility refer to the same
710 object if they are in the same shared object. Usually, hidden visibility
711 indicates that the symbol will not be placed into the dynamic symbol table,
712 so no other module (executable or shared library) can reference it
716 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
718 <dd>On ELF, protected visibility indicates that the symbol will be placed in
719 the dynamic symbol table, but that references within the defining module will
720 bind to the local symbol. That is, the symbol cannot be overridden by another
727 <!-- ======================================================================= -->
728 <div class="doc_subsection">
729 <a name="namedtypes">Named Types</a>
732 <div class="doc_text">
734 <p>LLVM IR allows you to specify name aliases for certain types. This can make
735 it easier to read the IR and make the IR more condensed (particularly when
736 recursive types are involved). An example of a name specification is:
739 <div class="doc_code">
741 %mytype = type { %mytype*, i32 }
745 <p>You may give a name to any <a href="#typesystem">type</a> except "<a
746 href="t_void">void</a>". Type name aliases may be used anywhere a type is
747 expected with the syntax "%mytype".</p>
749 <p>Note that type names are aliases for the structural type that they indicate,
750 and that you can therefore specify multiple names for the same type. This often
751 leads to confusing behavior when dumping out a .ll file. Since LLVM IR uses
752 structural typing, the name is not part of the type. When printing out LLVM IR,
753 the printer will pick <em>one name</em> to render all types of a particular
754 shape. This means that if you have code where two different source types end up
755 having the same LLVM type, that the dumper will sometimes print the "wrong" or
756 unexpected type. This is an important design point and isn't going to
761 <!-- ======================================================================= -->
762 <div class="doc_subsection">
763 <a name="globalvars">Global Variables</a>
766 <div class="doc_text">
768 <p>Global variables define regions of memory allocated at compilation time
769 instead of run-time. Global variables may optionally be initialized, may have
770 an explicit section to be placed in, and may have an optional explicit alignment
771 specified. A variable may be defined as "thread_local", which means that it
772 will not be shared by threads (each thread will have a separated copy of the
773 variable). A variable may be defined as a global "constant," which indicates
774 that the contents of the variable will <b>never</b> be modified (enabling better
775 optimization, allowing the global data to be placed in the read-only section of
776 an executable, etc). Note that variables that need runtime initialization
777 cannot be marked "constant" as there is a store to the variable.</p>
780 LLVM explicitly allows <em>declarations</em> of global variables to be marked
781 constant, even if the final definition of the global is not. This capability
782 can be used to enable slightly better optimization of the program, but requires
783 the language definition to guarantee that optimizations based on the
784 'constantness' are valid for the translation units that do not include the
788 <p>As SSA values, global variables define pointer values that are in
789 scope (i.e. they dominate) all basic blocks in the program. Global
790 variables always define a pointer to their "content" type because they
791 describe a region of memory, and all memory objects in LLVM are
792 accessed through pointers.</p>
794 <p>A global variable may be declared to reside in a target-specifc numbered
795 address space. For targets that support them, address spaces may affect how
796 optimizations are performed and/or what target instructions are used to access
797 the variable. The default address space is zero. The address space qualifier
798 must precede any other attributes.</p>
800 <p>LLVM allows an explicit section to be specified for globals. If the target
801 supports it, it will emit globals to the section specified.</p>
803 <p>An explicit alignment may be specified for a global. If not present, or if
804 the alignment is set to zero, the alignment of the global is set by the target
805 to whatever it feels convenient. If an explicit alignment is specified, the
806 global is forced to have at least that much alignment. All alignments must be
809 <p>For example, the following defines a global in a numbered address space with
810 an initializer, section, and alignment:</p>
812 <div class="doc_code">
814 @G = addrspace(5) constant float 1.0, section "foo", align 4
821 <!-- ======================================================================= -->
822 <div class="doc_subsection">
823 <a name="functionstructure">Functions</a>
826 <div class="doc_text">
828 <p>LLVM function definitions consist of the "<tt>define</tt>" keyord,
829 an optional <a href="#linkage">linkage type</a>, an optional
830 <a href="#visibility">visibility style</a>, an optional
831 <a href="#callingconv">calling convention</a>, a return type, an optional
832 <a href="#paramattrs">parameter attribute</a> for the return type, a function
833 name, a (possibly empty) argument list (each with optional
834 <a href="#paramattrs">parameter attributes</a>), optional
835 <a href="#fnattrs">function attributes</a>, an optional section,
836 an optional alignment, an optional <a href="#gc">garbage collector name</a>,
837 an opening curly brace, a list of basic blocks, and a closing curly brace.
839 LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
840 optional <a href="#linkage">linkage type</a>, an optional
841 <a href="#visibility">visibility style</a>, an optional
842 <a href="#callingconv">calling convention</a>, a return type, an optional
843 <a href="#paramattrs">parameter attribute</a> for the return type, a function
844 name, a possibly empty list of arguments, an optional alignment, and an optional
845 <a href="#gc">garbage collector name</a>.</p>
847 <p>A function definition contains a list of basic blocks, forming the CFG
848 (Control Flow Graph) for
849 the function. Each basic block may optionally start with a label (giving the
850 basic block a symbol table entry), contains a list of instructions, and ends
851 with a <a href="#terminators">terminator</a> instruction (such as a branch or
852 function return).</p>
854 <p>The first basic block in a function is special in two ways: it is immediately
855 executed on entrance to the function, and it is not allowed to have predecessor
856 basic blocks (i.e. there can not be any branches to the entry block of a
857 function). Because the block can have no predecessors, it also cannot have any
858 <a href="#i_phi">PHI nodes</a>.</p>
860 <p>LLVM allows an explicit section to be specified for functions. If the target
861 supports it, it will emit functions to the section specified.</p>
863 <p>An explicit alignment may be specified for a function. If not present, or if
864 the alignment is set to zero, the alignment of the function is set by the target
865 to whatever it feels convenient. If an explicit alignment is specified, the
866 function is forced to have at least that much alignment. All alignments must be
871 <div class="doc_code">
873 define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
874 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
875 <ResultType> @<FunctionName> ([argument list])
876 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
877 [<a href="#gc">gc</a>] { ... }
884 <!-- ======================================================================= -->
885 <div class="doc_subsection">
886 <a name="aliasstructure">Aliases</a>
888 <div class="doc_text">
889 <p>Aliases act as "second name" for the aliasee value (which can be either
890 function, global variable, another alias or bitcast of global value). Aliases
891 may have an optional <a href="#linkage">linkage type</a>, and an
892 optional <a href="#visibility">visibility style</a>.</p>
896 <div class="doc_code">
898 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
906 <!-- ======================================================================= -->
907 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
908 <div class="doc_text">
909 <p>The return type and each parameter of a function type may have a set of
910 <i>parameter attributes</i> associated with them. Parameter attributes are
911 used to communicate additional information about the result or parameters of
912 a function. Parameter attributes are considered to be part of the function,
913 not of the function type, so functions with different parameter attributes
914 can have the same function type.</p>
916 <p>Parameter attributes are simple keywords that follow the type specified. If
917 multiple parameter attributes are needed, they are space separated. For
920 <div class="doc_code">
922 declare i32 @printf(i8* noalias nocapture, ...)
923 declare i32 @atoi(i8 zeroext)
924 declare signext i8 @returns_signed_char()
928 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
929 <tt>readonly</tt>) come immediately after the argument list.</p>
931 <p>Currently, only the following parameter attributes are defined:</p>
933 <dt><tt>zeroext</tt></dt>
934 <dd>This indicates to the code generator that the parameter or return value
935 should be zero-extended to a 32-bit value by the caller (for a parameter)
936 or the callee (for a return value).</dd>
938 <dt><tt>signext</tt></dt>
939 <dd>This indicates to the code generator that the parameter or return value
940 should be sign-extended to a 32-bit value by the caller (for a parameter)
941 or the callee (for a return value).</dd>
943 <dt><tt>inreg</tt></dt>
944 <dd>This indicates that this parameter or return value should be treated
945 in a special target-dependent fashion during while emitting code for a
946 function call or return (usually, by putting it in a register as opposed
947 to memory, though some targets use it to distinguish between two different
948 kinds of registers). Use of this attribute is target-specific.</dd>
950 <dt><tt><a name="byval">byval</a></tt></dt>
951 <dd>This indicates that the pointer parameter should really be passed by
952 value to the function. The attribute implies that a hidden copy of the
953 pointee is made between the caller and the callee, so the callee is unable
954 to modify the value in the callee. This attribute is only valid on LLVM
955 pointer arguments. It is generally used to pass structs and arrays by
956 value, but is also valid on pointers to scalars. The copy is considered to
957 belong to the caller not the callee (for example,
958 <tt><a href="#readonly">readonly</a></tt> functions should not write to
959 <tt>byval</tt> parameters). This is not a valid attribute for return
960 values. The byval attribute also supports specifying an alignment with the
961 align attribute. This has a target-specific effect on the code generator
962 that usually indicates a desired alignment for the synthesized stack
965 <dt><tt>sret</tt></dt>
966 <dd>This indicates that the pointer parameter specifies the address of a
967 structure that is the return value of the function in the source program.
968 This pointer must be guaranteed by the caller to be valid: loads and stores
969 to the structure may be assumed by the callee to not to trap. This may only
970 be applied to the first parameter. This is not a valid attribute for
973 <dt><tt>noalias</tt></dt>
974 <dd>This indicates that the pointer does not alias any global or any other
975 parameter. The caller is responsible for ensuring that this is the
976 case. On a function return value, <tt>noalias</tt> additionally indicates
977 that the pointer does not alias any other pointers visible to the
978 caller. For further details, please see the discussion of the NoAlias
980 <a href="http://llvm.org/docs/AliasAnalysis.html#MustMayNo">alias
983 <dt><tt>nocapture</tt></dt>
984 <dd>This indicates that the callee does not make any copies of the pointer
985 that outlive the callee itself. This is not a valid attribute for return
988 <dt><tt>nest</tt></dt>
989 <dd>This indicates that the pointer parameter can be excised using the
990 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
991 attribute for return values.</dd>
996 <!-- ======================================================================= -->
997 <div class="doc_subsection">
998 <a name="gc">Garbage Collector Names</a>
1001 <div class="doc_text">
1002 <p>Each function may specify a garbage collector name, which is simply a
1005 <div class="doc_code"><pre
1006 >define void @f() gc "name" { ...</pre></div>
1008 <p>The compiler declares the supported values of <i>name</i>. Specifying a
1009 collector which will cause the compiler to alter its output in order to support
1010 the named garbage collection algorithm.</p>
1013 <!-- ======================================================================= -->
1014 <div class="doc_subsection">
1015 <a name="fnattrs">Function Attributes</a>
1018 <div class="doc_text">
1020 <p>Function attributes are set to communicate additional information about
1021 a function. Function attributes are considered to be part of the function,
1022 not of the function type, so functions with different parameter attributes
1023 can have the same function type.</p>
1025 <p>Function attributes are simple keywords that follow the type specified. If
1026 multiple attributes are needed, they are space separated. For
1029 <div class="doc_code">
1031 define void @f() noinline { ... }
1032 define void @f() alwaysinline { ... }
1033 define void @f() alwaysinline optsize { ... }
1034 define void @f() optsize
1039 <dt><tt>alwaysinline</tt></dt>
1040 <dd>This attribute indicates that the inliner should attempt to inline this
1041 function into callers whenever possible, ignoring any active inlining size
1042 threshold for this caller.</dd>
1044 <dt><tt>noinline</tt></dt>
1045 <dd>This attribute indicates that the inliner should never inline this function
1046 in any situation. This attribute may not be used together with the
1047 <tt>alwaysinline</tt> attribute.</dd>
1049 <dt><tt>optsize</tt></dt>
1050 <dd>This attribute suggests that optimization passes and code generator passes
1051 make choices that keep the code size of this function low, and otherwise do
1052 optimizations specifically to reduce code size.</dd>
1054 <dt><tt>noreturn</tt></dt>
1055 <dd>This function attribute indicates that the function never returns normally.
1056 This produces undefined behavior at runtime if the function ever does
1057 dynamically return.</dd>
1059 <dt><tt>nounwind</tt></dt>
1060 <dd>This function attribute indicates that the function never returns with an
1061 unwind or exceptional control flow. If the function does unwind, its runtime
1062 behavior is undefined.</dd>
1064 <dt><tt>readnone</tt></dt>
1065 <dd>This attribute indicates that the function computes its result (or decides to
1066 unwind an exception) based strictly on its arguments, without dereferencing any
1067 pointer arguments or otherwise accessing any mutable state (e.g. memory, control
1068 registers, etc) visible to caller functions. It does not write through any
1069 pointer arguments (including <tt><a href="#byval">byval</a></tt> arguments) and
1070 never changes any state visible to callers. This means that it cannot unwind
1071 exceptions by calling the <tt>C++</tt> exception throwing methods, but could
1072 use the <tt>unwind</tt> instruction.</dd>
1074 <dt><tt><a name="readonly">readonly</a></tt></dt>
1075 <dd>This attribute indicates that the function does not write through any
1076 pointer arguments (including <tt><a href="#byval">byval</a></tt> arguments)
1077 or otherwise modify any state (e.g. memory, control registers, etc) visible to
1078 caller functions. It may dereference pointer arguments and read state that may
1079 be set in the caller. A readonly function always returns the same value (or
1080 unwinds an exception identically) when called with the same set of arguments
1081 and global state. It cannot unwind an exception by calling the <tt>C++</tt>
1082 exception throwing methods, but may use the <tt>unwind</tt> instruction.</dd>
1084 <dt><tt><a name="ssp">ssp</a></tt></dt>
1085 <dd>This attribute indicates that the function should emit a stack smashing
1086 protector. It is in the form of a "canary"—a random value placed on the
1087 stack before the local variables that's checked upon return from the function to
1088 see if it has been overwritten. A heuristic is used to determine if a function
1089 needs stack protectors or not.
1091 <p>If a function that has an <tt>ssp</tt> attribute is inlined into a function
1092 that doesn't have an <tt>ssp</tt> attribute, then the resulting function will
1093 have an <tt>ssp</tt> attribute.</p></dd>
1095 <dt><tt>sspreq</tt></dt>
1096 <dd>This attribute indicates that the function should <em>always</em> emit a
1097 stack smashing protector. This overrides the <tt><a href="#ssp">ssp</a></tt>
1100 <p>If a function that has an <tt>sspreq</tt> attribute is inlined into a
1101 function that doesn't have an <tt>sspreq</tt> attribute or which has
1102 an <tt>ssp</tt> attribute, then the resulting function will have
1103 an <tt>sspreq</tt> attribute.</p></dd>
1108 <!-- ======================================================================= -->
1109 <div class="doc_subsection">
1110 <a name="moduleasm">Module-Level Inline Assembly</a>
1113 <div class="doc_text">
1115 Modules may contain "module-level inline asm" blocks, which corresponds to the
1116 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
1117 LLVM and treated as a single unit, but may be separated in the .ll file if
1118 desired. The syntax is very simple:
1121 <div class="doc_code">
1123 module asm "inline asm code goes here"
1124 module asm "more can go here"
1128 <p>The strings can contain any character by escaping non-printable characters.
1129 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1134 The inline asm code is simply printed to the machine code .s file when
1135 assembly code is generated.
1139 <!-- ======================================================================= -->
1140 <div class="doc_subsection">
1141 <a name="datalayout">Data Layout</a>
1144 <div class="doc_text">
1145 <p>A module may specify a target specific data layout string that specifies how
1146 data is to be laid out in memory. The syntax for the data layout is simply:</p>
1147 <pre> target datalayout = "<i>layout specification</i>"</pre>
1148 <p>The <i>layout specification</i> consists of a list of specifications
1149 separated by the minus sign character ('-'). Each specification starts with a
1150 letter and may include other information after the letter to define some
1151 aspect of the data layout. The specifications accepted are as follows: </p>
1154 <dd>Specifies that the target lays out data in big-endian form. That is, the
1155 bits with the most significance have the lowest address location.</dd>
1157 <dd>Specifies that the target lays out data in little-endian form. That is,
1158 the bits with the least significance have the lowest address location.</dd>
1159 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1160 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1161 <i>preferred</i> alignments. All sizes are in bits. Specifying the <i>pref</i>
1162 alignment is optional. If omitted, the preceding <tt>:</tt> should be omitted
1164 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1165 <dd>This specifies the alignment for an integer type of a given bit
1166 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1167 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1168 <dd>This specifies the alignment for a vector type of a given bit
1170 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1171 <dd>This specifies the alignment for a floating point type of a given bit
1172 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
1174 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1175 <dd>This specifies the alignment for an aggregate type of a given bit
1178 <p>When constructing the data layout for a given target, LLVM starts with a
1179 default set of specifications which are then (possibly) overriden by the
1180 specifications in the <tt>datalayout</tt> keyword. The default specifications
1181 are given in this list:</p>
1183 <li><tt>E</tt> - big endian</li>
1184 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
1185 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1186 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1187 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1188 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1189 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1190 alignment of 64-bits</li>
1191 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1192 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1193 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1194 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1195 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1197 <p>When LLVM is determining the alignment for a given type, it uses the
1198 following rules:</p>
1200 <li>If the type sought is an exact match for one of the specifications, that
1201 specification is used.</li>
1202 <li>If no match is found, and the type sought is an integer type, then the
1203 smallest integer type that is larger than the bitwidth of the sought type is
1204 used. If none of the specifications are larger than the bitwidth then the the
1205 largest integer type is used. For example, given the default specifications
1206 above, the i7 type will use the alignment of i8 (next largest) while both
1207 i65 and i256 will use the alignment of i64 (largest specified).</li>
1208 <li>If no match is found, and the type sought is a vector type, then the
1209 largest vector type that is smaller than the sought vector type will be used
1210 as a fall back. This happens because <128 x double> can be implemented
1211 in terms of 64 <2 x double>, for example.</li>
1215 <!-- *********************************************************************** -->
1216 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
1217 <!-- *********************************************************************** -->
1219 <div class="doc_text">
1221 <p>The LLVM type system is one of the most important features of the
1222 intermediate representation. Being typed enables a number of
1223 optimizations to be performed on the intermediate representation directly,
1224 without having to do
1225 extra analyses on the side before the transformation. A strong type
1226 system makes it easier to read the generated code and enables novel
1227 analyses and transformations that are not feasible to perform on normal
1228 three address code representations.</p>
1232 <!-- ======================================================================= -->
1233 <div class="doc_subsection"> <a name="t_classifications">Type
1234 Classifications</a> </div>
1235 <div class="doc_text">
1236 <p>The types fall into a few useful
1237 classifications:</p>
1239 <table border="1" cellspacing="0" cellpadding="4">
1241 <tr><th>Classification</th><th>Types</th></tr>
1243 <td><a href="#t_integer">integer</a></td>
1244 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1247 <td><a href="#t_floating">floating point</a></td>
1248 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1251 <td><a name="t_firstclass">first class</a></td>
1252 <td><a href="#t_integer">integer</a>,
1253 <a href="#t_floating">floating point</a>,
1254 <a href="#t_pointer">pointer</a>,
1255 <a href="#t_vector">vector</a>,
1256 <a href="#t_struct">structure</a>,
1257 <a href="#t_array">array</a>,
1258 <a href="#t_label">label</a>,
1259 <a href="#t_metadata">metadata</a>.
1263 <td><a href="#t_primitive">primitive</a></td>
1264 <td><a href="#t_label">label</a>,
1265 <a href="#t_void">void</a>,
1266 <a href="#t_floating">floating point</a>,
1267 <a href="#t_metadata">metadata</a>.</td>
1270 <td><a href="#t_derived">derived</a></td>
1271 <td><a href="#t_integer">integer</a>,
1272 <a href="#t_array">array</a>,
1273 <a href="#t_function">function</a>,
1274 <a href="#t_pointer">pointer</a>,
1275 <a href="#t_struct">structure</a>,
1276 <a href="#t_pstruct">packed structure</a>,
1277 <a href="#t_vector">vector</a>,
1278 <a href="#t_opaque">opaque</a>.
1284 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
1285 most important. Values of these types are the only ones which can be
1286 produced by instructions, passed as arguments, or used as operands to
1290 <!-- ======================================================================= -->
1291 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
1293 <div class="doc_text">
1294 <p>The primitive types are the fundamental building blocks of the LLVM
1299 <!-- _______________________________________________________________________ -->
1300 <div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1302 <div class="doc_text">
1305 <tr><th>Type</th><th>Description</th></tr>
1306 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1307 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1308 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1309 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1310 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1315 <!-- _______________________________________________________________________ -->
1316 <div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1318 <div class="doc_text">
1320 <p>The void type does not represent any value and has no size.</p>
1329 <!-- _______________________________________________________________________ -->
1330 <div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1332 <div class="doc_text">
1334 <p>The label type represents code labels.</p>
1343 <!-- _______________________________________________________________________ -->
1344 <div class="doc_subsubsection"> <a name="t_metadata">Metadata Type</a> </div>
1346 <div class="doc_text">
1348 <p>The metadata type represents embedded metadata. The only derived type that
1349 may contain metadata is <tt>metadata*</tt> or a function type that returns or
1350 takes metadata typed parameters, but not pointer to metadata types.</p>
1360 <!-- ======================================================================= -->
1361 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1363 <div class="doc_text">
1365 <p>The real power in LLVM comes from the derived types in the system.
1366 This is what allows a programmer to represent arrays, functions,
1367 pointers, and other useful types. Note that these derived types may be
1368 recursive: For example, it is possible to have a two dimensional array.</p>
1372 <!-- _______________________________________________________________________ -->
1373 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1375 <div class="doc_text">
1378 <p>The integer type is a very simple derived type that simply specifies an
1379 arbitrary bit width for the integer type desired. Any bit width from 1 bit to
1380 2^23-1 (about 8 million) can be specified.</p>
1388 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1392 <table class="layout">
1394 <td class="left"><tt>i1</tt></td>
1395 <td class="left">a single-bit integer.</td>
1398 <td class="left"><tt>i32</tt></td>
1399 <td class="left">a 32-bit integer.</td>
1402 <td class="left"><tt>i1942652</tt></td>
1403 <td class="left">a really big integer of over 1 million bits.</td>
1407 <p>Note that the code generator does not yet support large integer types
1408 to be used as function return types. The specific limit on how large a
1409 return type the code generator can currently handle is target-dependent;
1410 currently it's often 64 bits for 32-bit targets and 128 bits for 64-bit
1415 <!-- _______________________________________________________________________ -->
1416 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1418 <div class="doc_text">
1422 <p>The array type is a very simple derived type that arranges elements
1423 sequentially in memory. The array type requires a size (number of
1424 elements) and an underlying data type.</p>
1429 [<# elements> x <elementtype>]
1432 <p>The number of elements is a constant integer value; elementtype may
1433 be any type with a size.</p>
1436 <table class="layout">
1438 <td class="left"><tt>[40 x i32]</tt></td>
1439 <td class="left">Array of 40 32-bit integer values.</td>
1442 <td class="left"><tt>[41 x i32]</tt></td>
1443 <td class="left">Array of 41 32-bit integer values.</td>
1446 <td class="left"><tt>[4 x i8]</tt></td>
1447 <td class="left">Array of 4 8-bit integer values.</td>
1450 <p>Here are some examples of multidimensional arrays:</p>
1451 <table class="layout">
1453 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1454 <td class="left">3x4 array of 32-bit integer values.</td>
1457 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1458 <td class="left">12x10 array of single precision floating point values.</td>
1461 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1462 <td class="left">2x3x4 array of 16-bit integer values.</td>
1466 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1467 length array. Normally, accesses past the end of an array are undefined in
1468 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
1469 As a special case, however, zero length arrays are recognized to be variable
1470 length. This allows implementation of 'pascal style arrays' with the LLVM
1471 type "{ i32, [0 x float]}", for example.</p>
1473 <p>Note that the code generator does not yet support large aggregate types
1474 to be used as function return types. The specific limit on how large an
1475 aggregate return type the code generator can currently handle is
1476 target-dependent, and also dependent on the aggregate element types.</p>
1480 <!-- _______________________________________________________________________ -->
1481 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1482 <div class="doc_text">
1486 <p>The function type can be thought of as a function signature. It
1487 consists of a return type and a list of formal parameter types. The
1488 return type of a function type is a scalar type, a void type, or a struct type.
1489 If the return type is a struct type then all struct elements must be of first
1490 class types, and the struct must have at least one element.</p>
1495 <returntype list> (<parameter list>)
1498 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1499 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1500 which indicates that the function takes a variable number of arguments.
1501 Variable argument functions can access their arguments with the <a
1502 href="#int_varargs">variable argument handling intrinsic</a> functions.
1503 '<tt><returntype list></tt>' is a comma-separated list of
1504 <a href="#t_firstclass">first class</a> type specifiers.</p>
1507 <table class="layout">
1509 <td class="left"><tt>i32 (i32)</tt></td>
1510 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1512 </tr><tr class="layout">
1513 <td class="left"><tt>float (i16 signext, i32 *) *
1515 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1516 an <tt>i16</tt> that should be sign extended and a
1517 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1520 </tr><tr class="layout">
1521 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1522 <td class="left">A vararg function that takes at least one
1523 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1524 which returns an integer. This is the signature for <tt>printf</tt> in
1527 </tr><tr class="layout">
1528 <td class="left"><tt>{i32, i32} (i32)</tt></td>
1529 <td class="left">A function taking an <tt>i32</tt>, returning two
1530 <tt>i32</tt> values as an aggregate of type <tt>{ i32, i32 }</tt>
1536 <!-- _______________________________________________________________________ -->
1537 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1538 <div class="doc_text">
1540 <p>The structure type is used to represent a collection of data members
1541 together in memory. The packing of the field types is defined to match
1542 the ABI of the underlying processor. The elements of a structure may
1543 be any type that has a size.</p>
1544 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1545 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1546 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1549 <pre> { <type list> }<br></pre>
1551 <table class="layout">
1553 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1554 <td class="left">A triple of three <tt>i32</tt> values</td>
1555 </tr><tr class="layout">
1556 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1557 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1558 second element is a <a href="#t_pointer">pointer</a> to a
1559 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1560 an <tt>i32</tt>.</td>
1564 <p>Note that the code generator does not yet support large aggregate types
1565 to be used as function return types. The specific limit on how large an
1566 aggregate return type the code generator can currently handle is
1567 target-dependent, and also dependent on the aggregate element types.</p>
1571 <!-- _______________________________________________________________________ -->
1572 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1574 <div class="doc_text">
1576 <p>The packed structure type is used to represent a collection of data members
1577 together in memory. There is no padding between fields. Further, the alignment
1578 of a packed structure is 1 byte. The elements of a packed structure may
1579 be any type that has a size.</p>
1580 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1581 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1582 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1585 <pre> < { <type list> } > <br></pre>
1587 <table class="layout">
1589 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1590 <td class="left">A triple of three <tt>i32</tt> values</td>
1591 </tr><tr class="layout">
1593 <tt>< { float, i32 (i32)* } ></tt></td>
1594 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1595 second element is a <a href="#t_pointer">pointer</a> to a
1596 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1597 an <tt>i32</tt>.</td>
1602 <!-- _______________________________________________________________________ -->
1603 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1604 <div class="doc_text">
1606 <p>As in many languages, the pointer type represents a pointer or
1607 reference to another object, which must live in memory. Pointer types may have
1608 an optional address space attribute defining the target-specific numbered
1609 address space where the pointed-to object resides. The default address space is
1612 <p>Note that LLVM does not permit pointers to void (<tt>void*</tt>) nor does
1613 it permit pointers to labels (<tt>label*</tt>). Use <tt>i8*</tt> instead.</p>
1616 <pre> <type> *<br></pre>
1618 <table class="layout">
1620 <td class="left"><tt>[4 x i32]*</tt></td>
1621 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1622 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1625 <td class="left"><tt>i32 (i32 *) *</tt></td>
1626 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
1627 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1631 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1632 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1633 that resides in address space #5.</td>
1638 <!-- _______________________________________________________________________ -->
1639 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1640 <div class="doc_text">
1644 <p>A vector type is a simple derived type that represents a vector
1645 of elements. Vector types are used when multiple primitive data
1646 are operated in parallel using a single instruction (SIMD).
1647 A vector type requires a size (number of
1648 elements) and an underlying primitive data type. Vectors must have a power
1649 of two length (1, 2, 4, 8, 16 ...). Vector types are
1650 considered <a href="#t_firstclass">first class</a>.</p>
1655 < <# elements> x <elementtype> >
1658 <p>The number of elements is a constant integer value; elementtype may
1659 be any integer or floating point type.</p>
1663 <table class="layout">
1665 <td class="left"><tt><4 x i32></tt></td>
1666 <td class="left">Vector of 4 32-bit integer values.</td>
1669 <td class="left"><tt><8 x float></tt></td>
1670 <td class="left">Vector of 8 32-bit floating-point values.</td>
1673 <td class="left"><tt><2 x i64></tt></td>
1674 <td class="left">Vector of 2 64-bit integer values.</td>
1678 <p>Note that the code generator does not yet support large vector types
1679 to be used as function return types. The specific limit on how large a
1680 vector return type codegen can currently handle is target-dependent;
1681 currently it's often a few times longer than a hardware vector register.</p>
1685 <!-- _______________________________________________________________________ -->
1686 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1687 <div class="doc_text">
1691 <p>Opaque types are used to represent unknown types in the system. This
1692 corresponds (for example) to the C notion of a forward declared structure type.
1693 In LLVM, opaque types can eventually be resolved to any type (not just a
1694 structure type).</p>
1704 <table class="layout">
1706 <td class="left"><tt>opaque</tt></td>
1707 <td class="left">An opaque type.</td>
1712 <!-- ======================================================================= -->
1713 <div class="doc_subsection">
1714 <a name="t_uprefs">Type Up-references</a>
1717 <div class="doc_text">
1720 An "up reference" allows you to refer to a lexically enclosing type without
1721 requiring it to have a name. For instance, a structure declaration may contain a
1722 pointer to any of the types it is lexically a member of. Example of up
1723 references (with their equivalent as named type declarations) include:</p>
1726 { \2 * } %x = type { %x* }
1727 { \2 }* %y = type { %y }*
1732 An up reference is needed by the asmprinter for printing out cyclic types when
1733 there is no declared name for a type in the cycle. Because the asmprinter does
1734 not want to print out an infinite type string, it needs a syntax to handle
1735 recursive types that have no names (all names are optional in llvm IR).
1744 The level is the count of the lexical type that is being referred to.
1749 <table class="layout">
1751 <td class="left"><tt>\1*</tt></td>
1752 <td class="left">Self-referential pointer.</td>
1755 <td class="left"><tt>{ { \3*, i8 }, i32 }</tt></td>
1756 <td class="left">Recursive structure where the upref refers to the out-most
1763 <!-- *********************************************************************** -->
1764 <div class="doc_section"> <a name="constants">Constants</a> </div>
1765 <!-- *********************************************************************** -->
1767 <div class="doc_text">
1769 <p>LLVM has several different basic types of constants. This section describes
1770 them all and their syntax.</p>
1774 <!-- ======================================================================= -->
1775 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1777 <div class="doc_text">
1780 <dt><b>Boolean constants</b></dt>
1782 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1783 constants of the <tt><a href="#t_primitive">i1</a></tt> type.
1786 <dt><b>Integer constants</b></dt>
1788 <dd>Standard integers (such as '4') are constants of the <a
1789 href="#t_integer">integer</a> type. Negative numbers may be used with
1793 <dt><b>Floating point constants</b></dt>
1795 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1796 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1797 notation (see below). The assembler requires the exact decimal value of
1798 a floating-point constant. For example, the assembler accepts 1.25 but
1799 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
1800 constants must have a <a href="#t_floating">floating point</a> type. </dd>
1802 <dt><b>Null pointer constants</b></dt>
1804 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1805 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1809 <p>The one non-intuitive notation for constants is the hexadecimal form
1810 of floating point constants. For example, the form '<tt>double
1811 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1812 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1813 (and the only time that they are generated by the disassembler) is when a
1814 floating point constant must be emitted but it cannot be represented as a
1815 decimal floating point number in a reasonable number of digits. For example,
1816 NaN's, infinities, and other
1817 special values are represented in their IEEE hexadecimal format so that
1818 assembly and disassembly do not cause any bits to change in the constants.</p>
1819 <p>When using the hexadecimal form, constants of types float and double are
1820 represented using the 16-digit form shown above (which matches the IEEE754
1821 representation for double); float values must, however, be exactly representable
1822 as IEE754 single precision.
1823 Hexadecimal format is always used for long
1824 double, and there are three forms of long double. The 80-bit
1825 format used by x86 is represented as <tt>0xK</tt>
1826 followed by 20 hexadecimal digits.
1827 The 128-bit format used by PowerPC (two adjacent doubles) is represented
1828 by <tt>0xM</tt> followed by 32 hexadecimal digits. The IEEE 128-bit
1829 format is represented
1830 by <tt>0xL</tt> followed by 32 hexadecimal digits; no currently supported
1831 target uses this format. Long doubles will only work if they match
1832 the long double format on your target. All hexadecimal formats are big-endian
1833 (sign bit at the left).</p>
1836 <!-- ======================================================================= -->
1837 <div class="doc_subsection">
1838 <a name="aggregateconstants"> <!-- old anchor -->
1839 <a name="complexconstants">Complex Constants</a></a>
1842 <div class="doc_text">
1843 <p>Complex constants are a (potentially recursive) combination of simple
1844 constants and smaller complex constants.</p>
1847 <dt><b>Structure constants</b></dt>
1849 <dd>Structure constants are represented with notation similar to structure
1850 type definitions (a comma separated list of elements, surrounded by braces
1851 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
1852 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>". Structure constants
1853 must have <a href="#t_struct">structure type</a>, and the number and
1854 types of elements must match those specified by the type.
1857 <dt><b>Array constants</b></dt>
1859 <dd>Array constants are represented with notation similar to array type
1860 definitions (a comma separated list of elements, surrounded by square brackets
1861 (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74 ]</tt>". Array
1862 constants must have <a href="#t_array">array type</a>, and the number and
1863 types of elements must match those specified by the type.
1866 <dt><b>Vector constants</b></dt>
1868 <dd>Vector constants are represented with notation similar to vector type
1869 definitions (a comma separated list of elements, surrounded by
1870 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32 42,
1871 i32 11, i32 74, i32 100 ></tt>". Vector constants must have <a
1872 href="#t_vector">vector type</a>, and the number and types of elements must
1873 match those specified by the type.
1876 <dt><b>Zero initialization</b></dt>
1878 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1879 value to zero of <em>any</em> type, including scalar and aggregate types.
1880 This is often used to avoid having to print large zero initializers (e.g. for
1881 large arrays) and is always exactly equivalent to using explicit zero
1885 <dt><b>Metadata node</b></dt>
1887 <dd>A metadata node is a structure-like constant with
1888 <a href="#t_metadata">metadata type</a>. For example:
1889 "<tt>metadata !{ i32 0, metadata !"test" }</tt>". Unlike other constants
1890 that are meant to be interpreted as part of the instruction stream, metadata
1891 is a place to attach additional information such as debug info.
1897 <!-- ======================================================================= -->
1898 <div class="doc_subsection">
1899 <a name="globalconstants">Global Variable and Function Addresses</a>
1902 <div class="doc_text">
1904 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1905 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1906 constants. These constants are explicitly referenced when the <a
1907 href="#identifiers">identifier for the global</a> is used and always have <a
1908 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1911 <div class="doc_code">
1915 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1921 <!-- ======================================================================= -->
1922 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1923 <div class="doc_text">
1924 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1925 no specific value. Undefined values may be of any type and be used anywhere
1926 a constant is permitted.</p>
1928 <p>Undefined values indicate to the compiler that the program is well defined
1929 no matter what value is used, giving the compiler more freedom to optimize.
1933 <!-- ======================================================================= -->
1934 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1937 <div class="doc_text">
1939 <p>Constant expressions are used to allow expressions involving other constants
1940 to be used as constants. Constant expressions may be of any <a
1941 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1942 that does not have side effects (e.g. load and call are not supported). The
1943 following is the syntax for constant expressions:</p>
1946 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1947 <dd>Truncate a constant to another type. The bit size of CST must be larger
1948 than the bit size of TYPE. Both types must be integers.</dd>
1950 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1951 <dd>Zero extend a constant to another type. The bit size of CST must be
1952 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1954 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1955 <dd>Sign extend a constant to another type. The bit size of CST must be
1956 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1958 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1959 <dd>Truncate a floating point constant to another floating point type. The
1960 size of CST must be larger than the size of TYPE. Both types must be
1961 floating point.</dd>
1963 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1964 <dd>Floating point extend a constant to another type. The size of CST must be
1965 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1967 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
1968 <dd>Convert a floating point constant to the corresponding unsigned integer
1969 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1970 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1971 of the same number of elements. If the value won't fit in the integer type,
1972 the results are undefined.</dd>
1974 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
1975 <dd>Convert a floating point constant to the corresponding signed integer
1976 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1977 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1978 of the same number of elements. If the value won't fit in the integer type,
1979 the results are undefined.</dd>
1981 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
1982 <dd>Convert an unsigned integer constant to the corresponding floating point
1983 constant. TYPE must be a scalar or vector floating point type. CST must be of
1984 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1985 of the same number of elements. If the value won't fit in the floating point
1986 type, the results are undefined.</dd>
1988 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
1989 <dd>Convert a signed integer constant to the corresponding floating point
1990 constant. TYPE must be a scalar or vector floating point type. CST must be of
1991 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1992 of the same number of elements. If the value won't fit in the floating point
1993 type, the results are undefined.</dd>
1995 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
1996 <dd>Convert a pointer typed constant to the corresponding integer constant
1997 TYPE must be an integer type. CST must be of pointer type. The CST value is
1998 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
2000 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
2001 <dd>Convert a integer constant to a pointer constant. TYPE must be a
2002 pointer type. CST must be of integer type. The CST value is zero extended,
2003 truncated, or unchanged to make it fit in a pointer size. This one is
2004 <i>really</i> dangerous!</dd>
2006 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
2007 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
2008 are the same as those for the <a href="#i_bitcast">bitcast
2009 instruction</a>.</dd>
2011 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
2013 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
2014 constants. As with the <a href="#i_getelementptr">getelementptr</a>
2015 instruction, the index list may have zero or more indexes, which are required
2016 to make sense for the type of "CSTPTR".</dd>
2018 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
2020 <dd>Perform the <a href="#i_select">select operation</a> on
2023 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
2024 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
2026 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
2027 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
2029 <dt><b><tt>vicmp COND ( VAL1, VAL2 )</tt></b></dt>
2030 <dd>Performs the <a href="#i_vicmp">vicmp operation</a> on constants.</dd>
2032 <dt><b><tt>vfcmp COND ( VAL1, VAL2 )</tt></b></dt>
2033 <dd>Performs the <a href="#i_vfcmp">vfcmp operation</a> on constants.</dd>
2035 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
2037 <dd>Perform the <a href="#i_extractelement">extractelement
2038 operation</a> on constants.</dd>
2040 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
2042 <dd>Perform the <a href="#i_insertelement">insertelement
2043 operation</a> on constants.</dd>
2046 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
2048 <dd>Perform the <a href="#i_shufflevector">shufflevector
2049 operation</a> on constants.</dd>
2051 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
2053 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2054 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
2055 binary</a> operations. The constraints on operands are the same as those for
2056 the corresponding instruction (e.g. no bitwise operations on floating point
2057 values are allowed).</dd>
2061 <!-- ======================================================================= -->
2062 <div class="doc_subsection"><a name="metadata">Embedded Metadata</a>
2065 <div class="doc_text">
2067 <p>Embedded metadata provides a way to attach arbitrary data to the
2068 instruction stream without affecting the behaviour of the program. There are
2069 two metadata primitives, strings and nodes. All metadata has the
2070 <tt>metadata</tt> type and is identified in syntax by a preceding exclamation
2071 point ('<tt>!</tt>').
2074 <p>A metadata string is a string surrounded by double quotes. It can contain
2075 any character by escaping non-printable characters with "\xx" where "xx" is
2076 the two digit hex code. For example: "<tt>!"test\00"</tt>".
2079 <p>Metadata nodes are represented with notation similar to structure constants
2080 (a comma separated list of elements, surrounded by braces and preceeded by an
2081 exclamation point). For example: "<tt>!{ metadata !"test\00", i32 10}</tt>".
2084 <p>A metadata node will attempt to track changes to the values it holds. In
2085 the event that a value is deleted, it will be replaced with a typeless
2086 "<tt>null</tt>", such as "<tt>metadata !{null, i32 10}</tt>".</p>
2088 <p>Optimizations may rely on metadata to provide additional information about
2089 the program that isn't available in the instructions, or that isn't easily
2090 computable. Similarly, the code generator may expect a certain metadata format
2091 to be used to express debugging information.</p>
2094 <!-- *********************************************************************** -->
2095 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
2096 <!-- *********************************************************************** -->
2098 <!-- ======================================================================= -->
2099 <div class="doc_subsection">
2100 <a name="inlineasm">Inline Assembler Expressions</a>
2103 <div class="doc_text">
2106 LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
2107 Module-Level Inline Assembly</a>) through the use of a special value. This
2108 value represents the inline assembler as a string (containing the instructions
2109 to emit), a list of operand constraints (stored as a string), and a flag that
2110 indicates whether or not the inline asm expression has side effects. An example
2111 inline assembler expression is:
2114 <div class="doc_code">
2116 i32 (i32) asm "bswap $0", "=r,r"
2121 Inline assembler expressions may <b>only</b> be used as the callee operand of
2122 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
2125 <div class="doc_code">
2127 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
2132 Inline asms with side effects not visible in the constraint list must be marked
2133 as having side effects. This is done through the use of the
2134 '<tt>sideeffect</tt>' keyword, like so:
2137 <div class="doc_code">
2139 call void asm sideeffect "eieio", ""()
2143 <p>TODO: The format of the asm and constraints string still need to be
2144 documented here. Constraints on what can be done (e.g. duplication, moving, etc
2145 need to be documented). This is probably best done by reference to another
2146 document that covers inline asm from a holistic perspective.
2151 <!-- *********************************************************************** -->
2152 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
2153 <!-- *********************************************************************** -->
2155 <div class="doc_text">
2157 <p>The LLVM instruction set consists of several different
2158 classifications of instructions: <a href="#terminators">terminator
2159 instructions</a>, <a href="#binaryops">binary instructions</a>,
2160 <a href="#bitwiseops">bitwise binary instructions</a>, <a
2161 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
2162 instructions</a>.</p>
2166 <!-- ======================================================================= -->
2167 <div class="doc_subsection"> <a name="terminators">Terminator
2168 Instructions</a> </div>
2170 <div class="doc_text">
2172 <p>As mentioned <a href="#functionstructure">previously</a>, every
2173 basic block in a program ends with a "Terminator" instruction, which
2174 indicates which block should be executed after the current block is
2175 finished. These terminator instructions typically yield a '<tt>void</tt>'
2176 value: they produce control flow, not values (the one exception being
2177 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
2178 <p>There are six different terminator instructions: the '<a
2179 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
2180 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
2181 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
2182 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
2183 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
2187 <!-- _______________________________________________________________________ -->
2188 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
2189 Instruction</a> </div>
2190 <div class="doc_text">
2193 ret <type> <value> <i>; Return a value from a non-void function</i>
2194 ret void <i>; Return from void function</i>
2199 <p>The '<tt>ret</tt>' instruction is used to return control flow (and
2200 optionally a value) from a function back to the caller.</p>
2201 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
2202 returns a value and then causes control flow, and one that just causes
2203 control flow to occur.</p>
2207 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument,
2208 the return value. The type of the return value must be a
2209 '<a href="#t_firstclass">first class</a>' type.</p>
2211 <p>A function is not <a href="#wellformed">well formed</a> if
2212 it it has a non-void return type and contains a '<tt>ret</tt>'
2213 instruction with no return value or a return value with a type that
2214 does not match its type, or if it has a void return type and contains
2215 a '<tt>ret</tt>' instruction with a return value.</p>
2219 <p>When the '<tt>ret</tt>' instruction is executed, control flow
2220 returns back to the calling function's context. If the caller is a "<a
2221 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
2222 the instruction after the call. If the caller was an "<a
2223 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
2224 at the beginning of the "normal" destination block. If the instruction
2225 returns a value, that value shall set the call or invoke instruction's
2231 ret i32 5 <i>; Return an integer value of 5</i>
2232 ret void <i>; Return from a void function</i>
2233 ret { i32, i8 } { i32 4, i8 2 } <i>; Return a struct of values 4 and 2</i>
2236 <p>Note that the code generator does not yet fully support large
2237 return values. The specific sizes that are currently supported are
2238 dependent on the target. For integers, on 32-bit targets the limit
2239 is often 64 bits, and on 64-bit targets the limit is often 128 bits.
2240 For aggregate types, the current limits are dependent on the element
2241 types; for example targets are often limited to 2 total integer
2242 elements and 2 total floating-point elements.</p>
2245 <!-- _______________________________________________________________________ -->
2246 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
2247 <div class="doc_text">
2249 <pre> br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
2252 <p>The '<tt>br</tt>' instruction is used to cause control flow to
2253 transfer to a different basic block in the current function. There are
2254 two forms of this instruction, corresponding to a conditional branch
2255 and an unconditional branch.</p>
2257 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
2258 single '<tt>i1</tt>' value and two '<tt>label</tt>' values. The
2259 unconditional form of the '<tt>br</tt>' instruction takes a single
2260 '<tt>label</tt>' value as a target.</p>
2262 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
2263 argument is evaluated. If the value is <tt>true</tt>, control flows
2264 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
2265 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
2267 <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
2268 href="#i_ret">ret</a> i32 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> i32 0<br></pre>
2270 <!-- _______________________________________________________________________ -->
2271 <div class="doc_subsubsection">
2272 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
2275 <div class="doc_text">
2279 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
2284 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
2285 several different places. It is a generalization of the '<tt>br</tt>'
2286 instruction, allowing a branch to occur to one of many possible
2292 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
2293 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
2294 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
2295 table is not allowed to contain duplicate constant entries.</p>
2299 <p>The <tt>switch</tt> instruction specifies a table of values and
2300 destinations. When the '<tt>switch</tt>' instruction is executed, this
2301 table is searched for the given value. If the value is found, control flow is
2302 transfered to the corresponding destination; otherwise, control flow is
2303 transfered to the default destination.</p>
2305 <h5>Implementation:</h5>
2307 <p>Depending on properties of the target machine and the particular
2308 <tt>switch</tt> instruction, this instruction may be code generated in different
2309 ways. For example, it could be generated as a series of chained conditional
2310 branches or with a lookup table.</p>
2315 <i>; Emulate a conditional br instruction</i>
2316 %Val = <a href="#i_zext">zext</a> i1 %value to i32
2317 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
2319 <i>; Emulate an unconditional br instruction</i>
2320 switch i32 0, label %dest [ ]
2322 <i>; Implement a jump table:</i>
2323 switch i32 %val, label %otherwise [ i32 0, label %onzero
2325 i32 2, label %ontwo ]
2329 <!-- _______________________________________________________________________ -->
2330 <div class="doc_subsubsection">
2331 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
2334 <div class="doc_text">
2339 <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>]
2340 to label <normal label> unwind label <exception label>
2345 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
2346 function, with the possibility of control flow transfer to either the
2347 '<tt>normal</tt>' label or the
2348 '<tt>exception</tt>' label. If the callee function returns with the
2349 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
2350 "normal" label. If the callee (or any indirect callees) returns with the "<a
2351 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
2352 continued at the dynamically nearest "exception" label.</p>
2356 <p>This instruction requires several arguments:</p>
2360 The optional "cconv" marker indicates which <a href="#callingconv">calling
2361 convention</a> the call should use. If none is specified, the call defaults
2362 to using C calling conventions.
2365 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
2366 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>',
2367 and '<tt>inreg</tt>' attributes are valid here.</li>
2369 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
2370 function value being invoked. In most cases, this is a direct function
2371 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
2372 an arbitrary pointer to function value.
2375 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
2376 function to be invoked. </li>
2378 <li>'<tt>function args</tt>': argument list whose types match the function
2379 signature argument types. If the function signature indicates the function
2380 accepts a variable number of arguments, the extra arguments can be
2383 <li>'<tt>normal label</tt>': the label reached when the called function
2384 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
2386 <li>'<tt>exception label</tt>': the label reached when a callee returns with
2387 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
2389 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
2390 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
2391 '<tt>readnone</tt>' attributes are valid here.</li>
2396 <p>This instruction is designed to operate as a standard '<tt><a
2397 href="#i_call">call</a></tt>' instruction in most regards. The primary
2398 difference is that it establishes an association with a label, which is used by
2399 the runtime library to unwind the stack.</p>
2401 <p>This instruction is used in languages with destructors to ensure that proper
2402 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
2403 exception. Additionally, this is important for implementation of
2404 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
2406 <p>For the purposes of the SSA form, the definition of the value
2407 returned by the '<tt>invoke</tt>' instruction is deemed to occur on
2408 the edge from the current block to the "normal" label. If the callee
2409 unwinds then no return value is available.</p>
2413 %retval = invoke i32 @Test(i32 15) to label %Continue
2414 unwind label %TestCleanup <i>; {i32}:retval set</i>
2415 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
2416 unwind label %TestCleanup <i>; {i32}:retval set</i>
2421 <!-- _______________________________________________________________________ -->
2423 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
2424 Instruction</a> </div>
2426 <div class="doc_text">
2435 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
2436 at the first callee in the dynamic call stack which used an <a
2437 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
2438 primarily used to implement exception handling.</p>
2442 <p>The '<tt>unwind</tt>' instruction causes execution of the current function to
2443 immediately halt. The dynamic call stack is then searched for the first <a
2444 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
2445 execution continues at the "exceptional" destination block specified by the
2446 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
2447 dynamic call chain, undefined behavior results.</p>
2450 <!-- _______________________________________________________________________ -->
2452 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
2453 Instruction</a> </div>
2455 <div class="doc_text">
2464 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
2465 instruction is used to inform the optimizer that a particular portion of the
2466 code is not reachable. This can be used to indicate that the code after a
2467 no-return function cannot be reached, and other facts.</p>
2471 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
2476 <!-- ======================================================================= -->
2477 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
2478 <div class="doc_text">
2479 <p>Binary operators are used to do most of the computation in a
2480 program. They require two operands of the same type, execute an operation on them, and
2481 produce a single value. The operands might represent
2482 multiple data, as is the case with the <a href="#t_vector">vector</a> data type.
2483 The result value has the same type as its operands.</p>
2484 <p>There are several different binary operators:</p>
2486 <!-- _______________________________________________________________________ -->
2487 <div class="doc_subsubsection">
2488 <a name="i_add">'<tt>add</tt>' Instruction</a>
2491 <div class="doc_text">
2496 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2501 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
2505 <p>The two arguments to the '<tt>add</tt>' instruction must be <a
2506 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>, or
2507 <a href="#t_vector">vector</a> values. Both arguments must have identical
2512 <p>The value produced is the integer or floating point sum of the two
2515 <p>If an integer sum has unsigned overflow, the result returned is the
2516 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2519 <p>Because LLVM integers use a two's complement representation, this
2520 instruction is appropriate for both signed and unsigned integers.</p>
2525 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
2528 <!-- _______________________________________________________________________ -->
2529 <div class="doc_subsubsection">
2530 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
2533 <div class="doc_text">
2538 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2543 <p>The '<tt>sub</tt>' instruction returns the difference of its two
2546 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
2547 '<tt>neg</tt>' instruction present in most other intermediate
2548 representations.</p>
2552 <p>The two arguments to the '<tt>sub</tt>' instruction must be <a
2553 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>,
2554 or <a href="#t_vector">vector</a> values. Both arguments must have identical
2559 <p>The value produced is the integer or floating point difference of
2560 the two operands.</p>
2562 <p>If an integer difference has unsigned overflow, the result returned is the
2563 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2566 <p>Because LLVM integers use a two's complement representation, this
2567 instruction is appropriate for both signed and unsigned integers.</p>
2571 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
2572 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
2576 <!-- _______________________________________________________________________ -->
2577 <div class="doc_subsubsection">
2578 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
2581 <div class="doc_text">
2584 <pre> <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2587 <p>The '<tt>mul</tt>' instruction returns the product of its two
2592 <p>The two arguments to the '<tt>mul</tt>' instruction must be <a
2593 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>,
2594 or <a href="#t_vector">vector</a> values. Both arguments must have identical
2599 <p>The value produced is the integer or floating point product of the
2602 <p>If the result of an integer multiplication has unsigned overflow,
2603 the result returned is the mathematical result modulo
2604 2<sup>n</sup>, where n is the bit width of the result.</p>
2605 <p>Because LLVM integers use a two's complement representation, and the
2606 result is the same width as the operands, this instruction returns the
2607 correct result for both signed and unsigned integers. If a full product
2608 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands
2609 should be sign-extended or zero-extended as appropriate to the
2610 width of the full product.</p>
2612 <pre> <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
2616 <!-- _______________________________________________________________________ -->
2617 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
2619 <div class="doc_text">
2621 <pre> <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2624 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two
2629 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
2630 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2631 values. Both arguments must have identical types.</p>
2635 <p>The value produced is the unsigned integer quotient of the two operands.</p>
2636 <p>Note that unsigned integer division and signed integer division are distinct
2637 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
2638 <p>Division by zero leads to undefined behavior.</p>
2640 <pre> <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2643 <!-- _______________________________________________________________________ -->
2644 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
2646 <div class="doc_text">
2649 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2654 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
2659 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
2660 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2661 values. Both arguments must have identical types.</p>
2664 <p>The value produced is the signed integer quotient of the two operands rounded towards zero.</p>
2665 <p>Note that signed integer division and unsigned integer division are distinct
2666 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
2667 <p>Division by zero leads to undefined behavior. Overflow also leads to
2668 undefined behavior; this is a rare case, but can occur, for example,
2669 by doing a 32-bit division of -2147483648 by -1.</p>
2671 <pre> <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2674 <!-- _______________________________________________________________________ -->
2675 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
2676 Instruction</a> </div>
2677 <div class="doc_text">
2680 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2684 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
2689 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
2690 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a>
2691 of floating point values. Both arguments must have identical types.</p>
2695 <p>The value produced is the floating point quotient of the two operands.</p>
2700 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
2704 <!-- _______________________________________________________________________ -->
2705 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
2707 <div class="doc_text">
2709 <pre> <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2712 <p>The '<tt>urem</tt>' instruction returns the remainder from the
2713 unsigned division of its two arguments.</p>
2715 <p>The two arguments to the '<tt>urem</tt>' instruction must be
2716 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2717 values. Both arguments must have identical types.</p>
2719 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
2720 This instruction always performs an unsigned division to get the remainder.</p>
2721 <p>Note that unsigned integer remainder and signed integer remainder are
2722 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
2723 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
2725 <pre> <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2729 <!-- _______________________________________________________________________ -->
2730 <div class="doc_subsubsection">
2731 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
2734 <div class="doc_text">
2739 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2744 <p>The '<tt>srem</tt>' instruction returns the remainder from the
2745 signed division of its two operands. This instruction can also take
2746 <a href="#t_vector">vector</a> versions of the values in which case
2747 the elements must be integers.</p>
2751 <p>The two arguments to the '<tt>srem</tt>' instruction must be
2752 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2753 values. Both arguments must have identical types.</p>
2757 <p>This instruction returns the <i>remainder</i> of a division (where the result
2758 has the same sign as the dividend, <tt>op1</tt>), not the <i>modulo</i>
2759 operator (where the result has the same sign as the divisor, <tt>op2</tt>) of
2760 a value. For more information about the difference, see <a
2761 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
2762 Math Forum</a>. For a table of how this is implemented in various languages,
2763 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
2764 Wikipedia: modulo operation</a>.</p>
2765 <p>Note that signed integer remainder and unsigned integer remainder are
2766 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
2767 <p>Taking the remainder of a division by zero leads to undefined behavior.
2768 Overflow also leads to undefined behavior; this is a rare case, but can occur,
2769 for example, by taking the remainder of a 32-bit division of -2147483648 by -1.
2770 (The remainder doesn't actually overflow, but this rule lets srem be
2771 implemented using instructions that return both the result of the division
2772 and the remainder.)</p>
2774 <pre> <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2778 <!-- _______________________________________________________________________ -->
2779 <div class="doc_subsubsection">
2780 <a name="i_frem">'<tt>frem</tt>' Instruction</a> </div>
2782 <div class="doc_text">
2785 <pre> <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2788 <p>The '<tt>frem</tt>' instruction returns the remainder from the
2789 division of its two operands.</p>
2791 <p>The two arguments to the '<tt>frem</tt>' instruction must be
2792 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a>
2793 of floating point values. Both arguments must have identical types.</p>
2797 <p>This instruction returns the <i>remainder</i> of a division.
2798 The remainder has the same sign as the dividend.</p>
2803 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
2807 <!-- ======================================================================= -->
2808 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
2809 Operations</a> </div>
2810 <div class="doc_text">
2811 <p>Bitwise binary operators are used to do various forms of
2812 bit-twiddling in a program. They are generally very efficient
2813 instructions and can commonly be strength reduced from other
2814 instructions. They require two operands of the same type, execute an operation on them,
2815 and produce a single value. The resulting value is the same type as its operands.</p>
2818 <!-- _______________________________________________________________________ -->
2819 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
2820 Instruction</a> </div>
2821 <div class="doc_text">
2823 <pre> <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2828 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2829 the left a specified number of bits.</p>
2833 <p>Both arguments to the '<tt>shl</tt>' instruction must be the same <a
2834 href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2835 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
2839 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod 2<sup>n</sup>,
2840 where n is the width of the result. If <tt>op2</tt> is (statically or dynamically) negative or
2841 equal to or larger than the number of bits in <tt>op1</tt>, the result is undefined.
2842 If the arguments are vectors, each vector element of <tt>op1</tt> is shifted by the
2843 corresponding shift amount in <tt>op2</tt>.</p>
2845 <h5>Example:</h5><pre>
2846 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
2847 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
2848 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
2849 <result> = shl i32 1, 32 <i>; undefined</i>
2850 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
2853 <!-- _______________________________________________________________________ -->
2854 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
2855 Instruction</a> </div>
2856 <div class="doc_text">
2858 <pre> <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2862 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
2863 operand shifted to the right a specified number of bits with zero fill.</p>
2866 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
2867 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2868 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
2872 <p>This instruction always performs a logical shift right operation. The most
2873 significant bits of the result will be filled with zero bits after the
2874 shift. If <tt>op2</tt> is (statically or dynamically) equal to or larger than
2875 the number of bits in <tt>op1</tt>, the result is undefined. If the arguments are
2876 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding shift
2877 amount in <tt>op2</tt>.</p>
2881 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
2882 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
2883 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
2884 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
2885 <result> = lshr i32 1, 32 <i>; undefined</i>
2886 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
2890 <!-- _______________________________________________________________________ -->
2891 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
2892 Instruction</a> </div>
2893 <div class="doc_text">
2896 <pre> <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2900 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
2901 operand shifted to the right a specified number of bits with sign extension.</p>
2904 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
2905 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2906 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
2909 <p>This instruction always performs an arithmetic shift right operation,
2910 The most significant bits of the result will be filled with the sign bit
2911 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
2912 larger than the number of bits in <tt>op1</tt>, the result is undefined. If the
2913 arguments are vectors, each vector element of <tt>op1</tt> is shifted by the
2914 corresponding shift amount in <tt>op2</tt>.</p>
2918 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
2919 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
2920 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
2921 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
2922 <result> = ashr i32 1, 32 <i>; undefined</i>
2923 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
2927 <!-- _______________________________________________________________________ -->
2928 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
2929 Instruction</a> </div>
2931 <div class="doc_text">
2936 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2941 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
2942 its two operands.</p>
2946 <p>The two arguments to the '<tt>and</tt>' instruction must be
2947 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2948 values. Both arguments must have identical types.</p>
2951 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
2954 <table border="1" cellspacing="0" cellpadding="4">
2986 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
2987 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
2988 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
2991 <!-- _______________________________________________________________________ -->
2992 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
2993 <div class="doc_text">
2995 <pre> <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2998 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
2999 or of its two operands.</p>
3002 <p>The two arguments to the '<tt>or</tt>' instruction must be
3003 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3004 values. Both arguments must have identical types.</p>
3006 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
3009 <table border="1" cellspacing="0" cellpadding="4">
3040 <pre> <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
3041 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
3042 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
3045 <!-- _______________________________________________________________________ -->
3046 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
3047 Instruction</a> </div>
3048 <div class="doc_text">
3050 <pre> <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3053 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
3054 or of its two operands. The <tt>xor</tt> is used to implement the
3055 "one's complement" operation, which is the "~" operator in C.</p>
3057 <p>The two arguments to the '<tt>xor</tt>' instruction must be
3058 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3059 values. Both arguments must have identical types.</p>
3063 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
3066 <table border="1" cellspacing="0" cellpadding="4">
3098 <pre> <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
3099 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
3100 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
3101 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
3105 <!-- ======================================================================= -->
3106 <div class="doc_subsection">
3107 <a name="vectorops">Vector Operations</a>
3110 <div class="doc_text">
3112 <p>LLVM supports several instructions to represent vector operations in a
3113 target-independent manner. These instructions cover the element-access and
3114 vector-specific operations needed to process vectors effectively. While LLVM
3115 does directly support these vector operations, many sophisticated algorithms
3116 will want to use target-specific intrinsics to take full advantage of a specific
3121 <!-- _______________________________________________________________________ -->
3122 <div class="doc_subsubsection">
3123 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
3126 <div class="doc_text">
3131 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
3137 The '<tt>extractelement</tt>' instruction extracts a single scalar
3138 element from a vector at a specified index.
3145 The first operand of an '<tt>extractelement</tt>' instruction is a
3146 value of <a href="#t_vector">vector</a> type. The second operand is
3147 an index indicating the position from which to extract the element.
3148 The index may be a variable.</p>
3153 The result is a scalar of the same type as the element type of
3154 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
3155 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
3156 results are undefined.
3162 %result = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
3167 <!-- _______________________________________________________________________ -->
3168 <div class="doc_subsubsection">
3169 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
3172 <div class="doc_text">
3177 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
3183 The '<tt>insertelement</tt>' instruction inserts a scalar
3184 element into a vector at a specified index.
3191 The first operand of an '<tt>insertelement</tt>' instruction is a
3192 value of <a href="#t_vector">vector</a> type. The second operand is a
3193 scalar value whose type must equal the element type of the first
3194 operand. The third operand is an index indicating the position at
3195 which to insert the value. The index may be a variable.</p>
3200 The result is a vector of the same type as <tt>val</tt>. Its
3201 element values are those of <tt>val</tt> except at position
3202 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
3203 exceeds the length of <tt>val</tt>, the results are undefined.
3209 %result = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
3213 <!-- _______________________________________________________________________ -->
3214 <div class="doc_subsubsection">
3215 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
3218 <div class="doc_text">
3223 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
3229 The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
3230 from two input vectors, returning a vector with the same element type as
3231 the input and length that is the same as the shuffle mask.
3237 The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
3238 with types that match each other. The third argument is a shuffle mask whose
3239 element type is always 'i32'. The result of the instruction is a vector whose
3240 length is the same as the shuffle mask and whose element type is the same as
3241 the element type of the first two operands.
3245 The shuffle mask operand is required to be a constant vector with either
3246 constant integer or undef values.
3252 The elements of the two input vectors are numbered from left to right across
3253 both of the vectors. The shuffle mask operand specifies, for each element of
3254 the result vector, which element of the two input vectors the result element
3255 gets. The element selector may be undef (meaning "don't care") and the second
3256 operand may be undef if performing a shuffle from only one vector.
3262 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
3263 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
3264 %result = shufflevector <4 x i32> %v1, <4 x i32> undef,
3265 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
3266 %result = shufflevector <8 x i32> %v1, <8 x i32> undef,
3267 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i>
3268 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
3269 <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>
3274 <!-- ======================================================================= -->
3275 <div class="doc_subsection">
3276 <a name="aggregateops">Aggregate Operations</a>
3279 <div class="doc_text">
3281 <p>LLVM supports several instructions for working with aggregate values.
3286 <!-- _______________________________________________________________________ -->
3287 <div class="doc_subsubsection">
3288 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
3291 <div class="doc_text">
3296 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
3302 The '<tt>extractvalue</tt>' instruction extracts the value of a struct field
3303 or array element from an aggregate value.
3310 The first operand of an '<tt>extractvalue</tt>' instruction is a
3311 value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a>
3312 type. The operands are constant indices to specify which value to extract
3313 in a similar manner as indices in a
3314 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
3320 The result is the value at the position in the aggregate specified by
3327 %result = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
3332 <!-- _______________________________________________________________________ -->
3333 <div class="doc_subsubsection">
3334 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
3337 <div class="doc_text">
3342 <result> = insertvalue <aggregate type> <val>, <ty> <val>, <idx> <i>; yields <n x <ty>></i>
3348 The '<tt>insertvalue</tt>' instruction inserts a value
3349 into a struct field or array element in an aggregate.
3356 The first operand of an '<tt>insertvalue</tt>' instruction is a
3357 value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a> type.
3358 The second operand is a first-class value to insert.
3359 The following operands are constant indices
3360 indicating the position at which to insert the value in a similar manner as
3362 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
3363 The value to insert must have the same type as the value identified
3370 The result is an aggregate of the same type as <tt>val</tt>. Its
3371 value is that of <tt>val</tt> except that the value at the position
3372 specified by the indices is that of <tt>elt</tt>.
3378 %result = insertvalue {i32, float} %agg, i32 1, 0 <i>; yields {i32, float}</i>
3383 <!-- ======================================================================= -->
3384 <div class="doc_subsection">
3385 <a name="memoryops">Memory Access and Addressing Operations</a>
3388 <div class="doc_text">
3390 <p>A key design point of an SSA-based representation is how it
3391 represents memory. In LLVM, no memory locations are in SSA form, which
3392 makes things very simple. This section describes how to read, write,
3393 allocate, and free memory in LLVM.</p>
3397 <!-- _______________________________________________________________________ -->
3398 <div class="doc_subsubsection">
3399 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
3402 <div class="doc_text">
3407 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3412 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
3413 heap and returns a pointer to it. The object is always allocated in the generic
3414 address space (address space zero).</p>
3418 <p>The '<tt>malloc</tt>' instruction allocates
3419 <tt>sizeof(<type>)*NumElements</tt>
3420 bytes of memory from the operating system and returns a pointer of the
3421 appropriate type to the program. If "NumElements" is specified, it is the
3422 number of elements allocated, otherwise "NumElements" is defaulted to be one.
3423 If a constant alignment is specified, the value result of the allocation is guaranteed to
3424 be aligned to at least that boundary. If not specified, or if zero, the target can
3425 choose to align the allocation on any convenient boundary.</p>
3427 <p>'<tt>type</tt>' must be a sized type.</p>
3431 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
3432 a pointer is returned. The result of a zero byte allocation is undefined. The
3433 result is null if there is insufficient memory available.</p>
3438 %array = malloc [4 x i8] <i>; yields {[%4 x i8]*}:array</i>
3440 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
3441 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
3442 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
3443 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
3444 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
3447 <p>Note that the code generator does not yet respect the
3448 alignment value.</p>
3452 <!-- _______________________________________________________________________ -->
3453 <div class="doc_subsubsection">
3454 <a name="i_free">'<tt>free</tt>' Instruction</a>
3457 <div class="doc_text">
3462 free <type> <value> <i>; yields {void}</i>
3467 <p>The '<tt>free</tt>' instruction returns memory back to the unused
3468 memory heap to be reallocated in the future.</p>
3472 <p>'<tt>value</tt>' shall be a pointer value that points to a value
3473 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
3478 <p>Access to the memory pointed to by the pointer is no longer defined
3479 after this instruction executes. If the pointer is null, the operation
3485 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
3486 free [4 x i8]* %array
3490 <!-- _______________________________________________________________________ -->
3491 <div class="doc_subsubsection">
3492 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
3495 <div class="doc_text">
3500 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3505 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
3506 currently executing function, to be automatically released when this function
3507 returns to its caller. The object is always allocated in the generic address
3508 space (address space zero).</p>
3512 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
3513 bytes of memory on the runtime stack, returning a pointer of the
3514 appropriate type to the program. If "NumElements" is specified, it is the
3515 number of elements allocated, otherwise "NumElements" is defaulted to be one.
3516 If a constant alignment is specified, the value result of the allocation is guaranteed
3517 to be aligned to at least that boundary. If not specified, or if zero, the target
3518 can choose to align the allocation on any convenient boundary.</p>
3520 <p>'<tt>type</tt>' may be any sized type.</p>
3524 <p>Memory is allocated; a pointer is returned. The operation is undefined if
3525 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
3526 memory is automatically released when the function returns. The '<tt>alloca</tt>'
3527 instruction is commonly used to represent automatic variables that must
3528 have an address available. When the function returns (either with the <tt><a
3529 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
3530 instructions), the memory is reclaimed. Allocating zero bytes
3531 is legal, but the result is undefined.</p>
3536 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
3537 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
3538 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
3539 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
3543 <!-- _______________________________________________________________________ -->
3544 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
3545 Instruction</a> </div>
3546 <div class="doc_text">
3548 <pre> <result> = load <ty>* <pointer>[, align <alignment>]<br> <result> = volatile load <ty>* <pointer>[, align <alignment>]<br></pre>
3550 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
3552 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
3553 address from which to load. The pointer must point to a <a
3554 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
3555 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
3556 the number or order of execution of this <tt>load</tt> with other
3557 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
3560 The optional constant "align" argument specifies the alignment of the operation
3561 (that is, the alignment of the memory address). A value of 0 or an
3562 omitted "align" argument means that the operation has the preferential
3563 alignment for the target. It is the responsibility of the code emitter
3564 to ensure that the alignment information is correct. Overestimating
3565 the alignment results in an undefined behavior. Underestimating the
3566 alignment may produce less efficient code. An alignment of 1 is always
3570 <p>The location of memory pointed to is loaded. If the value being loaded
3571 is of scalar type then the number of bytes read does not exceed the minimum
3572 number of bytes needed to hold all bits of the type. For example, loading an
3573 <tt>i24</tt> reads at most three bytes. When loading a value of a type like
3574 <tt>i20</tt> with a size that is not an integral number of bytes, the result
3575 is undefined if the value was not originally written using a store of the
3578 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3580 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
3581 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
3584 <!-- _______________________________________________________________________ -->
3585 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
3586 Instruction</a> </div>
3587 <div class="doc_text">
3589 <pre> store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3590 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3593 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
3595 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
3596 to store and an address at which to store it. The type of the '<tt><pointer></tt>'
3597 operand must be a pointer to the <a href="#t_firstclass">first class</a> type
3598 of the '<tt><value></tt>'
3599 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
3600 optimizer is not allowed to modify the number or order of execution of
3601 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
3602 href="#i_store">store</a></tt> instructions.</p>
3604 The optional constant "align" argument specifies the alignment of the operation
3605 (that is, the alignment of the memory address). A value of 0 or an
3606 omitted "align" argument means that the operation has the preferential
3607 alignment for the target. It is the responsibility of the code emitter
3608 to ensure that the alignment information is correct. Overestimating
3609 the alignment results in an undefined behavior. Underestimating the
3610 alignment may produce less efficient code. An alignment of 1 is always
3614 <p>The contents of memory are updated to contain '<tt><value></tt>'
3615 at the location specified by the '<tt><pointer></tt>' operand.
3616 If '<tt><value></tt>' is of scalar type then the number of bytes
3617 written does not exceed the minimum number of bytes needed to hold all
3618 bits of the type. For example, storing an <tt>i24</tt> writes at most
3619 three bytes. When writing a value of a type like <tt>i20</tt> with a
3620 size that is not an integral number of bytes, it is unspecified what
3621 happens to the extra bits that do not belong to the type, but they will
3622 typically be overwritten.</p>
3624 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3625 store i32 3, i32* %ptr <i>; yields {void}</i>
3626 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
3630 <!-- _______________________________________________________________________ -->
3631 <div class="doc_subsubsection">
3632 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
3635 <div class="doc_text">
3638 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
3644 The '<tt>getelementptr</tt>' instruction is used to get the address of a
3645 subelement of an aggregate data structure. It performs address calculation only
3646 and does not access memory.</p>
3650 <p>The first argument is always a pointer, and forms the basis of the
3651 calculation. The remaining arguments are indices, that indicate which of the
3652 elements of the aggregate object are indexed. The interpretation of each index
3653 is dependent on the type being indexed into. The first index always indexes the
3654 pointer value given as the first argument, the second index indexes a value of
3655 the type pointed to (not necessarily the value directly pointed to, since the
3656 first index can be non-zero), etc. The first type indexed into must be a pointer
3657 value, subsequent types can be arrays, vectors and structs. Note that subsequent
3658 types being indexed into can never be pointers, since that would require loading
3659 the pointer before continuing calculation.</p>
3661 <p>The type of each index argument depends on the type it is indexing into.
3662 When indexing into a (packed) structure, only <tt>i32</tt> integer
3663 <b>constants</b> are allowed. When indexing into an array, pointer or vector,
3664 integers of any width are allowed (also non-constants).</p>
3666 <p>For example, let's consider a C code fragment and how it gets
3667 compiled to LLVM:</p>
3669 <div class="doc_code">
3682 int *foo(struct ST *s) {
3683 return &s[1].Z.B[5][13];
3688 <p>The LLVM code generated by the GCC frontend is:</p>
3690 <div class="doc_code">
3692 %RT = <a href="#namedtypes">type</a> { i8 , [10 x [20 x i32]], i8 }
3693 %ST = <a href="#namedtypes">type</a> { i32, double, %RT }
3695 define i32* %foo(%ST* %s) {
3697 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
3705 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
3706 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
3707 }</tt>' type, a structure. The second index indexes into the third element of
3708 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
3709 i8 }</tt>' type, another structure. The third index indexes into the second
3710 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
3711 array. The two dimensions of the array are subscripted into, yielding an
3712 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
3713 to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
3715 <p>Note that it is perfectly legal to index partially through a
3716 structure, returning a pointer to an inner element. Because of this,
3717 the LLVM code for the given testcase is equivalent to:</p>
3720 define i32* %foo(%ST* %s) {
3721 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
3722 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
3723 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
3724 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
3725 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
3730 <p>Note that it is undefined to access an array out of bounds: array
3731 and pointer indexes must always be within the defined bounds of the
3732 array type when accessed with an instruction that dereferences the
3733 pointer (e.g. a load or store instruction). The one exception for
3734 this rule is zero length arrays. These arrays are defined to be
3735 accessible as variable length arrays, which requires access beyond the
3736 zero'th element.</p>
3738 <p>The getelementptr instruction is often confusing. For some more insight
3739 into how it works, see <a href="GetElementPtr.html">the getelementptr
3745 <i>; yields [12 x i8]*:aptr</i>
3746 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
3747 <i>; yields i8*:vptr</i>
3748 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
3749 <i>; yields i8*:eptr</i>
3750 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
3751 <i>; yields i32*:iptr</i>
3752 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
3756 <!-- ======================================================================= -->
3757 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
3759 <div class="doc_text">
3760 <p>The instructions in this category are the conversion instructions (casting)
3761 which all take a single operand and a type. They perform various bit conversions
3765 <!-- _______________________________________________________________________ -->
3766 <div class="doc_subsubsection">
3767 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
3769 <div class="doc_text">
3773 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
3778 The '<tt>trunc</tt>' instruction truncates its operand to the type <tt>ty2</tt>.
3783 The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
3784 be an <a href="#t_integer">integer</a> type, and a type that specifies the size
3785 and type of the result, which must be an <a href="#t_integer">integer</a>
3786 type. The bit size of <tt>value</tt> must be larger than the bit size of
3787 <tt>ty2</tt>. Equal sized types are not allowed.</p>
3791 The '<tt>trunc</tt>' instruction truncates the high order bits in <tt>value</tt>
3792 and converts the remaining bits to <tt>ty2</tt>. Since the source size must be
3793 larger than the destination size, <tt>trunc</tt> cannot be a <i>no-op cast</i>.
3794 It will always truncate bits.</p>
3798 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
3799 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
3800 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
3804 <!-- _______________________________________________________________________ -->
3805 <div class="doc_subsubsection">
3806 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
3808 <div class="doc_text">
3812 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
3816 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
3821 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
3822 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3823 also be of <a href="#t_integer">integer</a> type. The bit size of the
3824 <tt>value</tt> must be smaller than the bit size of the destination type,
3828 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
3829 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
3831 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
3835 %X = zext i32 257 to i64 <i>; yields i64:257</i>
3836 %Y = zext i1 true to i32 <i>; yields i32:1</i>
3840 <!-- _______________________________________________________________________ -->
3841 <div class="doc_subsubsection">
3842 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
3844 <div class="doc_text">
3848 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
3852 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
3856 The '<tt>sext</tt>' instruction takes a value to cast, which must be of
3857 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3858 also be of <a href="#t_integer">integer</a> type. The bit size of the
3859 <tt>value</tt> must be smaller than the bit size of the destination type,
3864 The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
3865 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size of
3866 the type <tt>ty2</tt>.</p>
3868 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
3872 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
3873 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
3877 <!-- _______________________________________________________________________ -->
3878 <div class="doc_subsubsection">
3879 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
3882 <div class="doc_text">
3887 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
3891 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
3896 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
3897 point</a> value to cast and a <a href="#t_floating">floating point</a> type to
3898 cast it to. The size of <tt>value</tt> must be larger than the size of
3899 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
3900 <i>no-op cast</i>.</p>
3903 <p> The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
3904 <a href="#t_floating">floating point</a> type to a smaller
3905 <a href="#t_floating">floating point</a> type. If the value cannot fit within
3906 the destination type, <tt>ty2</tt>, then the results are undefined.</p>
3910 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
3911 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
3915 <!-- _______________________________________________________________________ -->
3916 <div class="doc_subsubsection">
3917 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
3919 <div class="doc_text">
3923 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
3927 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
3928 floating point value.</p>
3931 <p>The '<tt>fpext</tt>' instruction takes a
3932 <a href="#t_floating">floating point</a> <tt>value</tt> to cast,
3933 and a <a href="#t_floating">floating point</a> type to cast it to. The source
3934 type must be smaller than the destination type.</p>
3937 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
3938 <a href="#t_floating">floating point</a> type to a larger
3939 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
3940 used to make a <i>no-op cast</i> because it always changes bits. Use
3941 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
3945 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
3946 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
3950 <!-- _______________________________________________________________________ -->
3951 <div class="doc_subsubsection">
3952 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
3954 <div class="doc_text">
3958 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
3962 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
3963 unsigned integer equivalent of type <tt>ty2</tt>.
3967 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
3968 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3969 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3970 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3971 vector integer type with the same number of elements as <tt>ty</tt></p>
3974 <p> The '<tt>fptoui</tt>' instruction converts its
3975 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3976 towards zero) unsigned integer value. If the value cannot fit in <tt>ty2</tt>,
3977 the results are undefined.</p>
3981 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
3982 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
3983 %X = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
3987 <!-- _______________________________________________________________________ -->
3988 <div class="doc_subsubsection">
3989 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
3991 <div class="doc_text">
3995 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
3999 <p>The '<tt>fptosi</tt>' instruction converts
4000 <a href="#t_floating">floating point</a> <tt>value</tt> to type <tt>ty2</tt>.
4004 <p> The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
4005 scalar or vector <a href="#t_floating">floating point</a> value, and a type
4006 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
4007 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
4008 vector integer type with the same number of elements as <tt>ty</tt></p>
4011 <p>The '<tt>fptosi</tt>' instruction converts its
4012 <a href="#t_floating">floating point</a> operand into the nearest (rounding
4013 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
4014 the results are undefined.</p>
4018 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
4019 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
4020 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
4024 <!-- _______________________________________________________________________ -->
4025 <div class="doc_subsubsection">
4026 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
4028 <div class="doc_text">
4032 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
4036 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
4037 integer and converts that value to the <tt>ty2</tt> type.</p>
4040 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
4041 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
4042 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
4043 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
4044 floating point type with the same number of elements as <tt>ty</tt></p>
4047 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
4048 integer quantity and converts it to the corresponding floating point value. If
4049 the value cannot fit in the floating point value, the results are undefined.</p>
4053 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
4054 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
4058 <!-- _______________________________________________________________________ -->
4059 <div class="doc_subsubsection">
4060 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
4062 <div class="doc_text">
4066 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
4070 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed
4071 integer and converts that value to the <tt>ty2</tt> type.</p>
4074 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
4075 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
4076 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
4077 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
4078 floating point type with the same number of elements as <tt>ty</tt></p>
4081 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed
4082 integer quantity and converts it to the corresponding floating point value. If
4083 the value cannot fit in the floating point value, the results are undefined.</p>
4087 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
4088 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
4092 <!-- _______________________________________________________________________ -->
4093 <div class="doc_subsubsection">
4094 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
4096 <div class="doc_text">
4100 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
4104 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
4105 the integer type <tt>ty2</tt>.</p>
4108 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
4109 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
4110 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.</p>
4113 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
4114 <tt>ty2</tt> by interpreting the pointer value as an integer and either
4115 truncating or zero extending that value to the size of the integer type. If
4116 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
4117 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
4118 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
4123 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
4124 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
4128 <!-- _______________________________________________________________________ -->
4129 <div class="doc_subsubsection">
4130 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
4132 <div class="doc_text">
4136 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
4140 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to
4141 a pointer type, <tt>ty2</tt>.</p>
4144 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
4145 value to cast, and a type to cast it to, which must be a
4146 <a href="#t_pointer">pointer</a> type.</p>
4149 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
4150 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
4151 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
4152 size of a pointer then a truncation is done. If <tt>value</tt> is smaller than
4153 the size of a pointer then a zero extension is done. If they are the same size,
4154 nothing is done (<i>no-op cast</i>).</p>
4158 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
4159 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
4160 %Y = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
4164 <!-- _______________________________________________________________________ -->
4165 <div class="doc_subsubsection">
4166 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
4168 <div class="doc_text">
4172 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
4177 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
4178 <tt>ty2</tt> without changing any bits.</p>
4182 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be
4183 a non-aggregate first class value, and a type to cast it to, which must also be
4184 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes of
4186 and the destination type, <tt>ty2</tt>, must be identical. If the source
4187 type is a pointer, the destination type must also be a pointer. This
4188 instruction supports bitwise conversion of vectors to integers and to vectors
4189 of other types (as long as they have the same size).</p>
4192 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
4193 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
4194 this conversion. The conversion is done as if the <tt>value</tt> had been
4195 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only be
4196 converted to other pointer types with this instruction. To convert pointers to
4197 other types, use the <a href="#i_inttoptr">inttoptr</a> or
4198 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
4202 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
4203 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
4204 %Z = bitcast <2 x int> %V to i64; <i>; yields i64: %V</i>
4208 <!-- ======================================================================= -->
4209 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
4210 <div class="doc_text">
4211 <p>The instructions in this category are the "miscellaneous"
4212 instructions, which defy better classification.</p>
4215 <!-- _______________________________________________________________________ -->
4216 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
4218 <div class="doc_text">
4220 <pre> <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4223 <p>The '<tt>icmp</tt>' instruction returns a boolean value or
4224 a vector of boolean values based on comparison
4225 of its two integer, integer vector, or pointer operands.</p>
4227 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
4228 the condition code indicating the kind of comparison to perform. It is not
4229 a value, just a keyword. The possible condition code are:
4232 <li><tt>eq</tt>: equal</li>
4233 <li><tt>ne</tt>: not equal </li>
4234 <li><tt>ugt</tt>: unsigned greater than</li>
4235 <li><tt>uge</tt>: unsigned greater or equal</li>
4236 <li><tt>ult</tt>: unsigned less than</li>
4237 <li><tt>ule</tt>: unsigned less or equal</li>
4238 <li><tt>sgt</tt>: signed greater than</li>
4239 <li><tt>sge</tt>: signed greater or equal</li>
4240 <li><tt>slt</tt>: signed less than</li>
4241 <li><tt>sle</tt>: signed less or equal</li>
4243 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
4244 <a href="#t_pointer">pointer</a>
4245 or integer <a href="#t_vector">vector</a> typed.
4246 They must also be identical types.</p>
4248 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to
4249 the condition code given as <tt>cond</tt>. The comparison performed always
4250 yields either an <a href="#t_primitive"><tt>i1</tt></a> or vector of <tt>i1</tt> result, as follows:
4253 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
4254 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
4256 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
4257 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.</li>
4258 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
4259 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4260 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
4261 <tt>true</tt> if <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4262 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
4263 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
4264 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
4265 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4266 <li><tt>sgt</tt>: interprets the operands as signed values and yields
4267 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4268 <li><tt>sge</tt>: interprets the operands as signed values and yields
4269 <tt>true</tt> if <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4270 <li><tt>slt</tt>: interprets the operands as signed values and yields
4271 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
4272 <li><tt>sle</tt>: interprets the operands as signed values and yields
4273 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4275 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
4276 values are compared as if they were integers.</p>
4277 <p>If the operands are integer vectors, then they are compared
4278 element by element. The result is an <tt>i1</tt> vector with
4279 the same number of elements as the values being compared.
4280 Otherwise, the result is an <tt>i1</tt>.
4284 <pre> <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
4285 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
4286 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
4287 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
4288 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
4289 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
4292 <p>Note that the code generator does not yet support vector types with
4293 the <tt>icmp</tt> instruction.</p>
4297 <!-- _______________________________________________________________________ -->
4298 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
4300 <div class="doc_text">
4302 <pre> <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4305 <p>The '<tt>fcmp</tt>' instruction returns a boolean value
4306 or vector of boolean values based on comparison
4307 of its operands.</p>
4309 If the operands are floating point scalars, then the result
4310 type is a boolean (<a href="#t_primitive"><tt>i1</tt></a>).
4312 <p>If the operands are floating point vectors, then the result type
4313 is a vector of boolean with the same number of elements as the
4314 operands being compared.</p>
4316 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
4317 the condition code indicating the kind of comparison to perform. It is not
4318 a value, just a keyword. The possible condition code are:</p>
4320 <li><tt>false</tt>: no comparison, always returns false</li>
4321 <li><tt>oeq</tt>: ordered and equal</li>
4322 <li><tt>ogt</tt>: ordered and greater than </li>
4323 <li><tt>oge</tt>: ordered and greater than or equal</li>
4324 <li><tt>olt</tt>: ordered and less than </li>
4325 <li><tt>ole</tt>: ordered and less than or equal</li>
4326 <li><tt>one</tt>: ordered and not equal</li>
4327 <li><tt>ord</tt>: ordered (no nans)</li>
4328 <li><tt>ueq</tt>: unordered or equal</li>
4329 <li><tt>ugt</tt>: unordered or greater than </li>
4330 <li><tt>uge</tt>: unordered or greater than or equal</li>
4331 <li><tt>ult</tt>: unordered or less than </li>
4332 <li><tt>ule</tt>: unordered or less than or equal</li>
4333 <li><tt>une</tt>: unordered or not equal</li>
4334 <li><tt>uno</tt>: unordered (either nans)</li>
4335 <li><tt>true</tt>: no comparison, always returns true</li>
4337 <p><i>Ordered</i> means that neither operand is a QNAN while
4338 <i>unordered</i> means that either operand may be a QNAN.</p>
4339 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be
4340 either a <a href="#t_floating">floating point</a> type
4341 or a <a href="#t_vector">vector</a> of floating point type.
4342 They must have identical types.</p>
4344 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4345 according to the condition code given as <tt>cond</tt>.
4346 If the operands are vectors, then the vectors are compared
4348 Each comparison performed
4349 always yields an <a href="#t_primitive">i1</a> result, as follows:</p>
4351 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
4352 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4353 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
4354 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4355 <tt>op1</tt> is greather than <tt>op2</tt>.</li>
4356 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4357 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4358 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4359 <tt>op1</tt> is less than <tt>op2</tt>.</li>
4360 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4361 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4362 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4363 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
4364 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
4365 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
4366 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
4367 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
4368 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4369 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
4370 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4371 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
4372 <tt>op1</tt> is less than <tt>op2</tt>.</li>
4373 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
4374 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4375 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
4376 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
4377 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
4378 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
4382 <pre> <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
4383 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
4384 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
4385 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
4388 <p>Note that the code generator does not yet support vector types with
4389 the <tt>fcmp</tt> instruction.</p>
4393 <!-- _______________________________________________________________________ -->
4394 <div class="doc_subsubsection">
4395 <a name="i_vicmp">'<tt>vicmp</tt>' Instruction</a>
4397 <div class="doc_text">
4399 <pre> <result> = vicmp <cond> <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4402 <p>The '<tt>vicmp</tt>' instruction returns an integer vector value based on
4403 element-wise comparison of its two integer vector operands.</p>
4405 <p>The '<tt>vicmp</tt>' instruction takes three operands. The first operand is
4406 the condition code indicating the kind of comparison to perform. It is not
4407 a value, just a keyword. The possible condition code are:</p>
4409 <li><tt>eq</tt>: equal</li>
4410 <li><tt>ne</tt>: not equal </li>
4411 <li><tt>ugt</tt>: unsigned greater than</li>
4412 <li><tt>uge</tt>: unsigned greater or equal</li>
4413 <li><tt>ult</tt>: unsigned less than</li>
4414 <li><tt>ule</tt>: unsigned less or equal</li>
4415 <li><tt>sgt</tt>: signed greater than</li>
4416 <li><tt>sge</tt>: signed greater or equal</li>
4417 <li><tt>slt</tt>: signed less than</li>
4418 <li><tt>sle</tt>: signed less or equal</li>
4420 <p>The remaining two arguments must be <a href="#t_vector">vector</a> or
4421 <a href="#t_integer">integer</a> typed. They must also be identical types.</p>
4423 <p>The '<tt>vicmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4424 according to the condition code given as <tt>cond</tt>. The comparison yields a
4425 <a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, of
4426 identical type as the values being compared. The most significant bit in each
4427 element is 1 if the element-wise comparison evaluates to true, and is 0
4428 otherwise. All other bits of the result are undefined. The condition codes
4429 are evaluated identically to the <a href="#i_icmp">'<tt>icmp</tt>'
4430 instruction</a>.</p>
4434 <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>
4435 <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>
4439 <!-- _______________________________________________________________________ -->
4440 <div class="doc_subsubsection">
4441 <a name="i_vfcmp">'<tt>vfcmp</tt>' Instruction</a>
4443 <div class="doc_text">
4445 <pre> <result> = vfcmp <cond> <ty> <op1>, <op2></pre>
4447 <p>The '<tt>vfcmp</tt>' instruction returns an integer vector value based on
4448 element-wise comparison of its two floating point vector operands. The output
4449 elements have the same width as the input elements.</p>
4451 <p>The '<tt>vfcmp</tt>' instruction takes three operands. The first operand is
4452 the condition code indicating the kind of comparison to perform. It is not
4453 a value, just a keyword. The possible condition code are:</p>
4455 <li><tt>false</tt>: no comparison, always returns false</li>
4456 <li><tt>oeq</tt>: ordered and equal</li>
4457 <li><tt>ogt</tt>: ordered and greater than </li>
4458 <li><tt>oge</tt>: ordered and greater than or equal</li>
4459 <li><tt>olt</tt>: ordered and less than </li>
4460 <li><tt>ole</tt>: ordered and less than or equal</li>
4461 <li><tt>one</tt>: ordered and not equal</li>
4462 <li><tt>ord</tt>: ordered (no nans)</li>
4463 <li><tt>ueq</tt>: unordered or equal</li>
4464 <li><tt>ugt</tt>: unordered or greater than </li>
4465 <li><tt>uge</tt>: unordered or greater than or equal</li>
4466 <li><tt>ult</tt>: unordered or less than </li>
4467 <li><tt>ule</tt>: unordered or less than or equal</li>
4468 <li><tt>une</tt>: unordered or not equal</li>
4469 <li><tt>uno</tt>: unordered (either nans)</li>
4470 <li><tt>true</tt>: no comparison, always returns true</li>
4472 <p>The remaining two arguments must be <a href="#t_vector">vector</a> of
4473 <a href="#t_floating">floating point</a> typed. They must also be identical
4476 <p>The '<tt>vfcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4477 according to the condition code given as <tt>cond</tt>. The comparison yields a
4478 <a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, with
4479 an identical number of elements as the values being compared, and each element
4480 having identical with to the width of the floating point elements. The most
4481 significant bit in each element is 1 if the element-wise comparison evaluates to
4482 true, and is 0 otherwise. All other bits of the result are undefined. The
4483 condition codes are evaluated identically to the
4484 <a href="#i_fcmp">'<tt>fcmp</tt>' instruction</a>.</p>
4488 <i>; yields: result=<2 x i32> < i32 0, i32 -1 ></i>
4489 <result> = vfcmp oeq <2 x float> < float 4, float 0 >, < float 5, float 0 >
4491 <i>; yields: result=<2 x i64> < i64 -1, i64 0 ></i>
4492 <result> = vfcmp ult <2 x double> < double 1, double 2 >, < double 2, double 2>
4496 <!-- _______________________________________________________________________ -->
4497 <div class="doc_subsubsection">
4498 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
4501 <div class="doc_text">
4505 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
4507 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
4508 the SSA graph representing the function.</p>
4511 <p>The type of the incoming values is specified with the first type
4512 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
4513 as arguments, with one pair for each predecessor basic block of the
4514 current block. Only values of <a href="#t_firstclass">first class</a>
4515 type may be used as the value arguments to the PHI node. Only labels
4516 may be used as the label arguments.</p>
4518 <p>There must be no non-phi instructions between the start of a basic
4519 block and the PHI instructions: i.e. PHI instructions must be first in
4522 <p>For the purposes of the SSA form, the use of each incoming value is
4523 deemed to occur on the edge from the corresponding predecessor block
4524 to the current block (but after any definition of an '<tt>invoke</tt>'
4525 instruction's return value on the same edge).</p>
4529 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
4530 specified by the pair corresponding to the predecessor basic block that executed
4531 just prior to the current block.</p>
4535 Loop: ; Infinite loop that counts from 0 on up...
4536 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
4537 %nextindvar = add i32 %indvar, 1
4542 <!-- _______________________________________________________________________ -->
4543 <div class="doc_subsubsection">
4544 <a name="i_select">'<tt>select</tt>' Instruction</a>
4547 <div class="doc_text">
4552 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
4554 <i>selty</i> is either i1 or {<N x i1>}
4560 The '<tt>select</tt>' instruction is used to choose one value based on a
4561 condition, without branching.
4568 The '<tt>select</tt>' instruction requires an 'i1' value or
4569 a vector of 'i1' values indicating the
4570 condition, and two values of the same <a href="#t_firstclass">first class</a>
4571 type. If the val1/val2 are vectors and
4572 the condition is a scalar, then entire vectors are selected, not
4573 individual elements.
4579 If the condition is an i1 and it evaluates to 1, the instruction returns the first
4580 value argument; otherwise, it returns the second value argument.
4583 If the condition is a vector of i1, then the value arguments must
4584 be vectors of the same size, and the selection is done element
4591 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
4594 <p>Note that the code generator does not yet support conditions
4595 with vector type.</p>
4600 <!-- _______________________________________________________________________ -->
4601 <div class="doc_subsubsection">
4602 <a name="i_call">'<tt>call</tt>' Instruction</a>
4605 <div class="doc_text">
4609 <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>]
4614 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
4618 <p>This instruction requires several arguments:</p>
4622 <p>The optional "tail" marker indicates whether the callee function accesses
4623 any allocas or varargs in the caller. If the "tail" marker is present, the
4624 function call is eligible for tail call optimization. Note that calls may
4625 be marked "tail" even if they do not occur before a <a
4626 href="#i_ret"><tt>ret</tt></a> instruction.</p>
4629 <p>The optional "cconv" marker indicates which <a href="#callingconv">calling
4630 convention</a> the call should use. If none is specified, the call defaults
4631 to using C calling conventions.</p>
4635 <p>The optional <a href="#paramattrs">Parameter Attributes</a> list for
4636 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>',
4637 and '<tt>inreg</tt>' attributes are valid here.</p>
4641 <p>'<tt>ty</tt>': the type of the call instruction itself which is also
4642 the type of the return value. Functions that return no value are marked
4643 <tt><a href="#t_void">void</a></tt>.</p>
4646 <p>'<tt>fnty</tt>': shall be the signature of the pointer to function
4647 value being invoked. The argument types must match the types implied by
4648 this signature. This type can be omitted if the function is not varargs
4649 and if the function type does not return a pointer to a function.</p>
4652 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
4653 be invoked. In most cases, this is a direct function invocation, but
4654 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
4655 to function value.</p>
4658 <p>'<tt>function args</tt>': argument list whose types match the
4659 function signature argument types. All arguments must be of
4660 <a href="#t_firstclass">first class</a> type. If the function signature
4661 indicates the function accepts a variable number of arguments, the extra
4662 arguments can be specified.</p>
4665 <p>The optional <a href="#fnattrs">function attributes</a> list. Only
4666 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
4667 '<tt>readnone</tt>' attributes are valid here.</p>
4673 <p>The '<tt>call</tt>' instruction is used to cause control flow to
4674 transfer to a specified function, with its incoming arguments bound to
4675 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
4676 instruction in the called function, control flow continues with the
4677 instruction after the function call, and the return value of the
4678 function is bound to the result argument.</p>
4683 %retval = call i32 @test(i32 %argc)
4684 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42) <i>; yields i32</i>
4685 %X = tail call i32 @foo() <i>; yields i32</i>
4686 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
4687 call void %foo(i8 97 signext)
4689 %struct.A = type { i32, i8 }
4690 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
4691 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
4692 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
4693 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
4694 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
4699 <!-- _______________________________________________________________________ -->
4700 <div class="doc_subsubsection">
4701 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
4704 <div class="doc_text">
4709 <resultval> = va_arg <va_list*> <arglist>, <argty>
4714 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
4715 the "variable argument" area of a function call. It is used to implement the
4716 <tt>va_arg</tt> macro in C.</p>
4720 <p>This instruction takes a <tt>va_list*</tt> value and the type of
4721 the argument. It returns a value of the specified argument type and
4722 increments the <tt>va_list</tt> to point to the next argument. The
4723 actual type of <tt>va_list</tt> is target specific.</p>
4727 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
4728 type from the specified <tt>va_list</tt> and causes the
4729 <tt>va_list</tt> to point to the next argument. For more information,
4730 see the variable argument handling <a href="#int_varargs">Intrinsic
4733 <p>It is legal for this instruction to be called in a function which does not
4734 take a variable number of arguments, for example, the <tt>vfprintf</tt>
4737 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
4738 href="#intrinsics">intrinsic function</a> because it takes a type as an
4743 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
4745 <p>Note that the code generator does not yet fully support va_arg
4746 on many targets. Also, it does not currently support va_arg with
4747 aggregate types on any target.</p>
4751 <!-- *********************************************************************** -->
4752 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
4753 <!-- *********************************************************************** -->
4755 <div class="doc_text">
4757 <p>LLVM supports the notion of an "intrinsic function". These functions have
4758 well known names and semantics and are required to follow certain restrictions.
4759 Overall, these intrinsics represent an extension mechanism for the LLVM
4760 language that does not require changing all of the transformations in LLVM when
4761 adding to the language (or the bitcode reader/writer, the parser, etc...).</p>
4763 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
4764 prefix is reserved in LLVM for intrinsic names; thus, function names may not
4765 begin with this prefix. Intrinsic functions must always be external functions:
4766 you cannot define the body of intrinsic functions. Intrinsic functions may
4767 only be used in call or invoke instructions: it is illegal to take the address
4768 of an intrinsic function. Additionally, because intrinsic functions are part
4769 of the LLVM language, it is required if any are added that they be documented
4772 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
4773 a family of functions that perform the same operation but on different data
4774 types. Because LLVM can represent over 8 million different integer types,
4775 overloading is used commonly to allow an intrinsic function to operate on any
4776 integer type. One or more of the argument types or the result type can be
4777 overloaded to accept any integer type. Argument types may also be defined as
4778 exactly matching a previous argument's type or the result type. This allows an
4779 intrinsic function which accepts multiple arguments, but needs all of them to
4780 be of the same type, to only be overloaded with respect to a single argument or
4783 <p>Overloaded intrinsics will have the names of its overloaded argument types
4784 encoded into its function name, each preceded by a period. Only those types
4785 which are overloaded result in a name suffix. Arguments whose type is matched
4786 against another type do not. For example, the <tt>llvm.ctpop</tt> function can
4787 take an integer of any width and returns an integer of exactly the same integer
4788 width. This leads to a family of functions such as
4789 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29 %val)</tt>.
4790 Only one type, the return type, is overloaded, and only one type suffix is
4791 required. Because the argument's type is matched against the return type, it
4792 does not require its own name suffix.</p>
4794 <p>To learn how to add an intrinsic function, please see the
4795 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.
4800 <!-- ======================================================================= -->
4801 <div class="doc_subsection">
4802 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
4805 <div class="doc_text">
4807 <p>Variable argument support is defined in LLVM with the <a
4808 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
4809 intrinsic functions. These functions are related to the similarly
4810 named macros defined in the <tt><stdarg.h></tt> header file.</p>
4812 <p>All of these functions operate on arguments that use a
4813 target-specific value type "<tt>va_list</tt>". The LLVM assembly
4814 language reference manual does not define what this type is, so all
4815 transformations should be prepared to handle these functions regardless of
4818 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
4819 instruction and the variable argument handling intrinsic functions are
4822 <div class="doc_code">
4824 define i32 @test(i32 %X, ...) {
4825 ; Initialize variable argument processing
4827 %ap2 = bitcast i8** %ap to i8*
4828 call void @llvm.va_start(i8* %ap2)
4830 ; Read a single integer argument
4831 %tmp = va_arg i8** %ap, i32
4833 ; Demonstrate usage of llvm.va_copy and llvm.va_end
4835 %aq2 = bitcast i8** %aq to i8*
4836 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
4837 call void @llvm.va_end(i8* %aq2)
4839 ; Stop processing of arguments.
4840 call void @llvm.va_end(i8* %ap2)
4844 declare void @llvm.va_start(i8*)
4845 declare void @llvm.va_copy(i8*, i8*)
4846 declare void @llvm.va_end(i8*)
4852 <!-- _______________________________________________________________________ -->
4853 <div class="doc_subsubsection">
4854 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
4858 <div class="doc_text">
4860 <pre> declare void %llvm.va_start(i8* <arglist>)<br></pre>
4862 <p>The '<tt>llvm.va_start</tt>' intrinsic initializes
4863 <tt>*<arglist></tt> for subsequent use by <tt><a
4864 href="#i_va_arg">va_arg</a></tt>.</p>
4868 <p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
4872 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
4873 macro available in C. In a target-dependent way, it initializes the
4874 <tt>va_list</tt> element to which the argument points, so that the next call to
4875 <tt>va_arg</tt> will produce the first variable argument passed to the function.
4876 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
4877 last argument of the function as the compiler can figure that out.</p>
4881 <!-- _______________________________________________________________________ -->
4882 <div class="doc_subsubsection">
4883 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
4886 <div class="doc_text">
4888 <pre> declare void @llvm.va_end(i8* <arglist>)<br></pre>
4891 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
4892 which has been initialized previously with <tt><a href="#int_va_start">llvm.va_start</a></tt>
4893 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
4897 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
4901 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
4902 macro available in C. In a target-dependent way, it destroys the
4903 <tt>va_list</tt> element to which the argument points. Calls to <a
4904 href="#int_va_start"><tt>llvm.va_start</tt></a> and <a href="#int_va_copy">
4905 <tt>llvm.va_copy</tt></a> must be matched exactly with calls to
4906 <tt>llvm.va_end</tt>.</p>
4910 <!-- _______________________________________________________________________ -->
4911 <div class="doc_subsubsection">
4912 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
4915 <div class="doc_text">
4920 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
4925 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
4926 from the source argument list to the destination argument list.</p>
4930 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
4931 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
4936 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
4937 macro available in C. In a target-dependent way, it copies the source
4938 <tt>va_list</tt> element into the destination <tt>va_list</tt> element. This
4939 intrinsic is necessary because the <tt><a href="#int_va_start">
4940 llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
4941 example, memory allocation.</p>
4945 <!-- ======================================================================= -->
4946 <div class="doc_subsection">
4947 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
4950 <div class="doc_text">
4953 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
4954 Collection</a> (GC) requires the implementation and generation of these
4956 These intrinsics allow identification of <a href="#int_gcroot">GC roots on the
4957 stack</a>, as well as garbage collector implementations that require <a
4958 href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a> barriers.
4959 Front-ends for type-safe garbage collected languages should generate these
4960 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
4961 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
4964 <p>The garbage collection intrinsics only operate on objects in the generic
4965 address space (address space zero).</p>
4969 <!-- _______________________________________________________________________ -->
4970 <div class="doc_subsubsection">
4971 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
4974 <div class="doc_text">
4979 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
4984 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
4985 the code generator, and allows some metadata to be associated with it.</p>
4989 <p>The first argument specifies the address of a stack object that contains the
4990 root pointer. The second pointer (which must be either a constant or a global
4991 value address) contains the meta-data to be associated with the root.</p>
4995 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
4996 location. At compile-time, the code generator generates information to allow
4997 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
4998 intrinsic may only be used in a function which <a href="#gc">specifies a GC
5004 <!-- _______________________________________________________________________ -->
5005 <div class="doc_subsubsection">
5006 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
5009 <div class="doc_text">
5014 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
5019 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
5020 locations, allowing garbage collector implementations that require read
5025 <p>The second argument is the address to read from, which should be an address
5026 allocated from the garbage collector. The first object is a pointer to the
5027 start of the referenced object, if needed by the language runtime (otherwise
5032 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
5033 instruction, but may be replaced with substantially more complex code by the
5034 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
5035 may only be used in a function which <a href="#gc">specifies a GC
5041 <!-- _______________________________________________________________________ -->
5042 <div class="doc_subsubsection">
5043 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
5046 <div class="doc_text">
5051 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
5056 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
5057 locations, allowing garbage collector implementations that require write
5058 barriers (such as generational or reference counting collectors).</p>
5062 <p>The first argument is the reference to store, the second is the start of the
5063 object to store it to, and the third is the address of the field of Obj to
5064 store to. If the runtime does not require a pointer to the object, Obj may be
5069 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
5070 instruction, but may be replaced with substantially more complex code by the
5071 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
5072 may only be used in a function which <a href="#gc">specifies a GC
5079 <!-- ======================================================================= -->
5080 <div class="doc_subsection">
5081 <a name="int_codegen">Code Generator Intrinsics</a>
5084 <div class="doc_text">
5086 These intrinsics are provided by LLVM to expose special features that may only
5087 be implemented with code generator support.
5092 <!-- _______________________________________________________________________ -->
5093 <div class="doc_subsubsection">
5094 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
5097 <div class="doc_text">
5101 declare i8 *@llvm.returnaddress(i32 <level>)
5107 The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
5108 target-specific value indicating the return address of the current function
5109 or one of its callers.
5115 The argument to this intrinsic indicates which function to return the address
5116 for. Zero indicates the calling function, one indicates its caller, etc. The
5117 argument is <b>required</b> to be a constant integer value.
5123 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
5124 the return address of the specified call frame, or zero if it cannot be
5125 identified. The value returned by this intrinsic is likely to be incorrect or 0
5126 for arguments other than zero, so it should only be used for debugging purposes.
5130 Note that calling this intrinsic does not prevent function inlining or other
5131 aggressive transformations, so the value returned may not be that of the obvious
5132 source-language caller.
5137 <!-- _______________________________________________________________________ -->
5138 <div class="doc_subsubsection">
5139 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
5142 <div class="doc_text">
5146 declare i8 *@llvm.frameaddress(i32 <level>)
5152 The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
5153 target-specific frame pointer value for the specified stack frame.
5159 The argument to this intrinsic indicates which function to return the frame
5160 pointer for. Zero indicates the calling function, one indicates its caller,
5161 etc. The argument is <b>required</b> to be a constant integer value.
5167 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
5168 the frame address of the specified call frame, or zero if it cannot be
5169 identified. The value returned by this intrinsic is likely to be incorrect or 0
5170 for arguments other than zero, so it should only be used for debugging purposes.
5174 Note that calling this intrinsic does not prevent function inlining or other
5175 aggressive transformations, so the value returned may not be that of the obvious
5176 source-language caller.
5180 <!-- _______________________________________________________________________ -->
5181 <div class="doc_subsubsection">
5182 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
5185 <div class="doc_text">
5189 declare i8 *@llvm.stacksave()
5195 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
5196 the function stack, for use with <a href="#int_stackrestore">
5197 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
5198 features like scoped automatic variable sized arrays in C99.
5204 This intrinsic returns a opaque pointer value that can be passed to <a
5205 href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
5206 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
5207 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
5208 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
5209 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
5210 that were allocated after the <tt>llvm.stacksave</tt> was executed.
5215 <!-- _______________________________________________________________________ -->
5216 <div class="doc_subsubsection">
5217 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
5220 <div class="doc_text">
5224 declare void @llvm.stackrestore(i8 * %ptr)
5230 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
5231 the function stack to the state it was in when the corresponding <a
5232 href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
5233 useful for implementing language features like scoped automatic variable sized
5240 See the description for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.
5246 <!-- _______________________________________________________________________ -->
5247 <div class="doc_subsubsection">
5248 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
5251 <div class="doc_text">
5255 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>)
5262 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
5263 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
5265 effect on the behavior of the program but can change its performance
5272 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
5273 determining if the fetch should be for a read (0) or write (1), and
5274 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
5275 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
5276 <tt>locality</tt> arguments must be constant integers.
5282 This intrinsic does not modify the behavior of the program. In particular,
5283 prefetches cannot trap and do not produce a value. On targets that support this
5284 intrinsic, the prefetch can provide hints to the processor cache for better
5290 <!-- _______________________________________________________________________ -->
5291 <div class="doc_subsubsection">
5292 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
5295 <div class="doc_text">
5299 declare void @llvm.pcmarker(i32 <id>)
5306 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
5308 code to simulators and other tools. The method is target specific, but it is
5309 expected that the marker will use exported symbols to transmit the PC of the
5311 The marker makes no guarantees that it will remain with any specific instruction
5312 after optimizations. It is possible that the presence of a marker will inhibit
5313 optimizations. The intended use is to be inserted after optimizations to allow
5314 correlations of simulation runs.
5320 <tt>id</tt> is a numerical id identifying the marker.
5326 This intrinsic does not modify the behavior of the program. Backends that do not
5327 support this intrinisic may ignore it.
5332 <!-- _______________________________________________________________________ -->
5333 <div class="doc_subsubsection">
5334 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
5337 <div class="doc_text">
5341 declare i64 @llvm.readcyclecounter( )
5348 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
5349 counter register (or similar low latency, high accuracy clocks) on those targets
5350 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
5351 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
5352 should only be used for small timings.
5358 When directly supported, reading the cycle counter should not modify any memory.
5359 Implementations are allowed to either return a application specific value or a
5360 system wide value. On backends without support, this is lowered to a constant 0.
5365 <!-- ======================================================================= -->
5366 <div class="doc_subsection">
5367 <a name="int_libc">Standard C Library Intrinsics</a>
5370 <div class="doc_text">
5372 LLVM provides intrinsics for a few important standard C library functions.
5373 These intrinsics allow source-language front-ends to pass information about the
5374 alignment of the pointer arguments to the code generator, providing opportunity
5375 for more efficient code generation.
5380 <!-- _______________________________________________________________________ -->
5381 <div class="doc_subsubsection">
5382 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
5385 <div class="doc_text">
5388 <p>This is an overloaded intrinsic. You can use llvm.memcpy on any integer bit
5389 width. Not all targets support all bit widths however.</p>
5391 declare void @llvm.memcpy.i8(i8 * <dest>, i8 * <src>,
5392 i8 <len>, i32 <align>)
5393 declare void @llvm.memcpy.i16(i8 * <dest>, i8 * <src>,
5394 i16 <len>, i32 <align>)
5395 declare void @llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
5396 i32 <len>, i32 <align>)
5397 declare void @llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
5398 i64 <len>, i32 <align>)
5404 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
5405 location to the destination location.
5409 Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
5410 intrinsics do not return a value, and takes an extra alignment argument.
5416 The first argument is a pointer to the destination, the second is a pointer to
5417 the source. The third argument is an integer argument
5418 specifying the number of bytes to copy, and the fourth argument is the alignment
5419 of the source and destination locations.
5423 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5424 the caller guarantees that both the source and destination pointers are aligned
5431 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
5432 location to the destination location, which are not allowed to overlap. It
5433 copies "len" bytes of memory over. If the argument is known to be aligned to
5434 some boundary, this can be specified as the fourth argument, otherwise it should
5440 <!-- _______________________________________________________________________ -->
5441 <div class="doc_subsubsection">
5442 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
5445 <div class="doc_text">
5448 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
5449 width. Not all targets support all bit widths however.</p>
5451 declare void @llvm.memmove.i8(i8 * <dest>, i8 * <src>,
5452 i8 <len>, i32 <align>)
5453 declare void @llvm.memmove.i16(i8 * <dest>, i8 * <src>,
5454 i16 <len>, i32 <align>)
5455 declare void @llvm.memmove.i32(i8 * <dest>, i8 * <src>,
5456 i32 <len>, i32 <align>)
5457 declare void @llvm.memmove.i64(i8 * <dest>, i8 * <src>,
5458 i64 <len>, i32 <align>)
5464 The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
5465 location to the destination location. It is similar to the
5466 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to overlap.
5470 Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
5471 intrinsics do not return a value, and takes an extra alignment argument.
5477 The first argument is a pointer to the destination, the second is a pointer to
5478 the source. The third argument is an integer argument
5479 specifying the number of bytes to copy, and the fourth argument is the alignment
5480 of the source and destination locations.
5484 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5485 the caller guarantees that the source and destination pointers are aligned to
5492 The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
5493 location to the destination location, which may overlap. It
5494 copies "len" bytes of memory over. If the argument is known to be aligned to
5495 some boundary, this can be specified as the fourth argument, otherwise it should
5501 <!-- _______________________________________________________________________ -->
5502 <div class="doc_subsubsection">
5503 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
5506 <div class="doc_text">
5509 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
5510 width. Not all targets support all bit widths however.</p>
5512 declare void @llvm.memset.i8(i8 * <dest>, i8 <val>,
5513 i8 <len>, i32 <align>)
5514 declare void @llvm.memset.i16(i8 * <dest>, i8 <val>,
5515 i16 <len>, i32 <align>)
5516 declare void @llvm.memset.i32(i8 * <dest>, i8 <val>,
5517 i32 <len>, i32 <align>)
5518 declare void @llvm.memset.i64(i8 * <dest>, i8 <val>,
5519 i64 <len>, i32 <align>)
5525 The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
5530 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
5531 does not return a value, and takes an extra alignment argument.
5537 The first argument is a pointer to the destination to fill, the second is the
5538 byte value to fill it with, the third argument is an integer
5539 argument specifying the number of bytes to fill, and the fourth argument is the
5540 known alignment of destination location.
5544 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5545 the caller guarantees that the destination pointer is aligned to that boundary.
5551 The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
5553 destination location. If the argument is known to be aligned to some boundary,
5554 this can be specified as the fourth argument, otherwise it should be set to 0 or
5560 <!-- _______________________________________________________________________ -->
5561 <div class="doc_subsubsection">
5562 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
5565 <div class="doc_text">
5568 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
5569 floating point or vector of floating point type. Not all targets support all
5572 declare float @llvm.sqrt.f32(float %Val)
5573 declare double @llvm.sqrt.f64(double %Val)
5574 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
5575 declare fp128 @llvm.sqrt.f128(fp128 %Val)
5576 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
5582 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
5583 returning the same value as the libm '<tt>sqrt</tt>' functions would. Unlike
5584 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
5585 negative numbers other than -0.0 (which allows for better optimization, because
5586 there is no need to worry about errno being set). <tt>llvm.sqrt(-0.0)</tt> is
5587 defined to return -0.0 like IEEE sqrt.
5593 The argument and return value are floating point numbers of the same type.
5599 This function returns the sqrt of the specified operand if it is a nonnegative
5600 floating point number.
5604 <!-- _______________________________________________________________________ -->
5605 <div class="doc_subsubsection">
5606 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
5609 <div class="doc_text">
5612 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
5613 floating point or vector of floating point type. Not all targets support all
5616 declare float @llvm.powi.f32(float %Val, i32 %power)
5617 declare double @llvm.powi.f64(double %Val, i32 %power)
5618 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
5619 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
5620 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
5626 The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
5627 specified (positive or negative) power. The order of evaluation of
5628 multiplications is not defined. When a vector of floating point type is
5629 used, the second argument remains a scalar integer value.
5635 The second argument is an integer power, and the first is a value to raise to
5642 This function returns the first value raised to the second power with an
5643 unspecified sequence of rounding operations.</p>
5646 <!-- _______________________________________________________________________ -->
5647 <div class="doc_subsubsection">
5648 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
5651 <div class="doc_text">
5654 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
5655 floating point or vector of floating point type. Not all targets support all
5658 declare float @llvm.sin.f32(float %Val)
5659 declare double @llvm.sin.f64(double %Val)
5660 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
5661 declare fp128 @llvm.sin.f128(fp128 %Val)
5662 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
5668 The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.
5674 The argument and return value are floating point numbers of the same type.
5680 This function returns the sine of the specified operand, returning the
5681 same values as the libm <tt>sin</tt> functions would, and handles error
5682 conditions in the same way.</p>
5685 <!-- _______________________________________________________________________ -->
5686 <div class="doc_subsubsection">
5687 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
5690 <div class="doc_text">
5693 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
5694 floating point or vector of floating point type. Not all targets support all
5697 declare float @llvm.cos.f32(float %Val)
5698 declare double @llvm.cos.f64(double %Val)
5699 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
5700 declare fp128 @llvm.cos.f128(fp128 %Val)
5701 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
5707 The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.
5713 The argument and return value are floating point numbers of the same type.
5719 This function returns the cosine of the specified operand, returning the
5720 same values as the libm <tt>cos</tt> functions would, and handles error
5721 conditions in the same way.</p>
5724 <!-- _______________________________________________________________________ -->
5725 <div class="doc_subsubsection">
5726 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
5729 <div class="doc_text">
5732 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
5733 floating point or vector of floating point type. Not all targets support all
5736 declare float @llvm.pow.f32(float %Val, float %Power)
5737 declare double @llvm.pow.f64(double %Val, double %Power)
5738 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
5739 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
5740 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
5746 The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
5747 specified (positive or negative) power.
5753 The second argument is a floating point power, and the first is a value to
5754 raise to that power.
5760 This function returns the first value raised to the second power,
5762 same values as the libm <tt>pow</tt> functions would, and handles error
5763 conditions in the same way.</p>
5767 <!-- ======================================================================= -->
5768 <div class="doc_subsection">
5769 <a name="int_manip">Bit Manipulation Intrinsics</a>
5772 <div class="doc_text">
5774 LLVM provides intrinsics for a few important bit manipulation operations.
5775 These allow efficient code generation for some algorithms.
5780 <!-- _______________________________________________________________________ -->
5781 <div class="doc_subsubsection">
5782 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
5785 <div class="doc_text">
5788 <p>This is an overloaded intrinsic function. You can use bswap on any integer
5789 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
5791 declare i16 @llvm.bswap.i16(i16 <id>)
5792 declare i32 @llvm.bswap.i32(i32 <id>)
5793 declare i64 @llvm.bswap.i64(i64 <id>)
5799 The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
5800 values with an even number of bytes (positive multiple of 16 bits). These are
5801 useful for performing operations on data that is not in the target's native
5808 The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
5809 and low byte of the input i16 swapped. Similarly, the <tt>llvm.bswap.i32</tt>
5810 intrinsic returns an i32 value that has the four bytes of the input i32
5811 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the returned
5812 i32 will have its bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i48</tt>,
5813 <tt>llvm.bswap.i64</tt> and other intrinsics extend this concept to
5814 additional even-byte lengths (6 bytes, 8 bytes and more, respectively).
5819 <!-- _______________________________________________________________________ -->
5820 <div class="doc_subsubsection">
5821 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
5824 <div class="doc_text">
5827 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
5828 width. Not all targets support all bit widths however.</p>
5830 declare i8 @llvm.ctpop.i8(i8 <src>)
5831 declare i16 @llvm.ctpop.i16(i16 <src>)
5832 declare i32 @llvm.ctpop.i32(i32 <src>)
5833 declare i64 @llvm.ctpop.i64(i64 <src>)
5834 declare i256 @llvm.ctpop.i256(i256 <src>)
5840 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
5847 The only argument is the value to be counted. The argument may be of any
5848 integer type. The return type must match the argument type.
5854 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
5858 <!-- _______________________________________________________________________ -->
5859 <div class="doc_subsubsection">
5860 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
5863 <div class="doc_text">
5866 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
5867 integer bit width. Not all targets support all bit widths however.</p>
5869 declare i8 @llvm.ctlz.i8 (i8 <src>)
5870 declare i16 @llvm.ctlz.i16(i16 <src>)
5871 declare i32 @llvm.ctlz.i32(i32 <src>)
5872 declare i64 @llvm.ctlz.i64(i64 <src>)
5873 declare i256 @llvm.ctlz.i256(i256 <src>)
5879 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
5880 leading zeros in a variable.
5886 The only argument is the value to be counted. The argument may be of any
5887 integer type. The return type must match the argument type.
5893 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
5894 in a variable. If the src == 0 then the result is the size in bits of the type
5895 of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.
5901 <!-- _______________________________________________________________________ -->
5902 <div class="doc_subsubsection">
5903 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
5906 <div class="doc_text">
5909 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
5910 integer bit width. Not all targets support all bit widths however.</p>
5912 declare i8 @llvm.cttz.i8 (i8 <src>)
5913 declare i16 @llvm.cttz.i16(i16 <src>)
5914 declare i32 @llvm.cttz.i32(i32 <src>)
5915 declare i64 @llvm.cttz.i64(i64 <src>)
5916 declare i256 @llvm.cttz.i256(i256 <src>)
5922 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
5929 The only argument is the value to be counted. The argument may be of any
5930 integer type. The return type must match the argument type.
5936 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
5937 in a variable. If the src == 0 then the result is the size in bits of the type
5938 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
5942 <!-- _______________________________________________________________________ -->
5943 <div class="doc_subsubsection">
5944 <a name="int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic</a>
5947 <div class="doc_text">
5950 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.select</tt>
5951 on any integer bit width.</p>
5953 declare i17 @llvm.part.select.i17 (i17 %val, i32 %loBit, i32 %hiBit)
5954 declare i29 @llvm.part.select.i29 (i29 %val, i32 %loBit, i32 %hiBit)
5958 <p>The '<tt>llvm.part.select</tt>' family of intrinsic functions selects a
5959 range of bits from an integer value and returns them in the same bit width as
5960 the original value.</p>
5963 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5964 any bit width but they must have the same bit width. The second and third
5965 arguments must be <tt>i32</tt> type since they specify only a bit index.</p>
5968 <p>The operation of the '<tt>llvm.part.select</tt>' intrinsic has two modes
5969 of operation: forwards and reverse. If <tt>%loBit</tt> is greater than
5970 <tt>%hiBits</tt> then the intrinsic operates in reverse mode. Otherwise it
5971 operates in forward mode.</p>
5972 <p>In forward mode, this intrinsic is the equivalent of shifting <tt>%val</tt>
5973 right by <tt>%loBit</tt> bits and then ANDing it with a mask with
5974 only the <tt>%hiBit - %loBit</tt> bits set, as follows:</p>
5976 <li>The <tt>%val</tt> is shifted right (LSHR) by the number of bits specified
5977 by <tt>%loBits</tt>. This normalizes the value to the low order bits.</li>
5978 <li>The <tt>%loBits</tt> value is subtracted from the <tt>%hiBits</tt> value
5979 to determine the number of bits to retain.</li>
5980 <li>A mask of the retained bits is created by shifting a -1 value.</li>
5981 <li>The mask is ANDed with <tt>%val</tt> to produce the result.</li>
5983 <p>In reverse mode, a similar computation is made except that the bits are
5984 returned in the reverse order. So, for example, if <tt>X</tt> has the value
5985 <tt>i16 0x0ACF (101011001111)</tt> and we apply
5986 <tt>part.select(i16 X, 8, 3)</tt> to it, we get back the value
5987 <tt>i16 0x0026 (000000100110)</tt>.</p>
5990 <div class="doc_subsubsection">
5991 <a name="int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic</a>
5994 <div class="doc_text">
5997 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.set</tt>
5998 on any integer bit width.</p>
6000 declare i17 @llvm.part.set.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
6001 declare i29 @llvm.part.set.i29.i9 (i29 %val, i9 %repl, i32 %lo, i32 %hi)
6005 <p>The '<tt>llvm.part.set</tt>' family of intrinsic functions replaces a range
6006 of bits in an integer value with another integer value. It returns the integer
6007 with the replaced bits.</p>
6010 <p>The first argument, <tt>%val</tt>, and the result may be integer types of
6011 any bit width, but they must have the same bit width. <tt>%val</tt> is the value
6012 whose bits will be replaced. The second argument, <tt>%repl</tt> may be an
6013 integer of any bit width. The third and fourth arguments must be <tt>i32</tt>
6014 type since they specify only a bit index.</p>
6017 <p>The operation of the '<tt>llvm.part.set</tt>' intrinsic has two modes
6018 of operation: forwards and reverse. If <tt>%lo</tt> is greater than
6019 <tt>%hi</tt> then the intrinsic operates in reverse mode. Otherwise it
6020 operates in forward mode.</p>
6022 <p>For both modes, the <tt>%repl</tt> value is prepared for use by either
6023 truncating it down to the size of the replacement area or zero extending it
6024 up to that size.</p>
6026 <p>In forward mode, the bits between <tt>%lo</tt> and <tt>%hi</tt> (inclusive)
6027 are replaced with corresponding bits from <tt>%repl</tt>. That is the 0th bit
6028 in <tt>%repl</tt> replaces the <tt>%lo</tt>th bit in <tt>%val</tt> and etc. up
6029 to the <tt>%hi</tt>th bit.</p>
6031 <p>In reverse mode, a similar computation is made except that the bits are
6032 reversed. That is, the <tt>0</tt>th bit in <tt>%repl</tt> replaces the
6033 <tt>%hi</tt> bit in <tt>%val</tt> and etc. down to the <tt>%lo</tt>th bit.</p>
6038 llvm.part.set(0xFFFF, 0, 4, 7) -> 0xFF0F
6039 llvm.part.set(0xFFFF, 0, 7, 4) -> 0xFF0F
6040 llvm.part.set(0xFFFF, 1, 7, 4) -> 0xFF8F
6041 llvm.part.set(0xFFFF, F, 8, 3) -> 0xFFE7
6042 llvm.part.set(0xFFFF, 0, 3, 8) -> 0xFE07
6047 <!-- ======================================================================= -->
6048 <div class="doc_subsection">
6049 <a name="int_overflow">Arithmetic with Overflow Intrinsics</a>
6052 <div class="doc_text">
6054 LLVM provides intrinsics for some arithmetic with overflow operations.
6059 <!-- _______________________________________________________________________ -->
6060 <div class="doc_subsubsection">
6061 <a name="int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt>' Intrinsics</a>
6064 <div class="doc_text">
6068 <p>This is an overloaded intrinsic. You can use <tt>llvm.sadd.with.overflow</tt>
6069 on any integer bit width.</p>
6072 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
6073 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6074 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
6079 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
6080 a signed addition of the two arguments, and indicate whether an overflow
6081 occurred during the signed summation.</p>
6085 <p>The arguments (%a and %b) and the first element of the result structure may
6086 be of integer types of any bit width, but they must have the same bit width. The
6087 second element of the result structure must be of type <tt>i1</tt>. <tt>%a</tt>
6088 and <tt>%b</tt> are the two values that will undergo signed addition.</p>
6092 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
6093 a signed addition of the two variables. They return a structure — the
6094 first element of which is the signed summation, and the second element of which
6095 is a bit specifying if the signed summation resulted in an overflow.</p>
6099 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6100 %sum = extractvalue {i32, i1} %res, 0
6101 %obit = extractvalue {i32, i1} %res, 1
6102 br i1 %obit, label %overflow, label %normal
6107 <!-- _______________________________________________________________________ -->
6108 <div class="doc_subsubsection">
6109 <a name="int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt>' Intrinsics</a>
6112 <div class="doc_text">
6116 <p>This is an overloaded intrinsic. You can use <tt>llvm.uadd.with.overflow</tt>
6117 on any integer bit width.</p>
6120 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
6121 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6122 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
6127 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
6128 an unsigned addition of the two arguments, and indicate whether a carry occurred
6129 during the unsigned summation.</p>
6133 <p>The arguments (%a and %b) and the first element of the result structure may
6134 be of integer types of any bit width, but they must have the same bit width. The
6135 second element of the result structure must be of type <tt>i1</tt>. <tt>%a</tt>
6136 and <tt>%b</tt> are the two values that will undergo unsigned addition.</p>
6140 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
6141 an unsigned addition of the two arguments. They return a structure — the
6142 first element of which is the sum, and the second element of which is a bit
6143 specifying if the unsigned summation resulted in a carry.</p>
6147 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6148 %sum = extractvalue {i32, i1} %res, 0
6149 %obit = extractvalue {i32, i1} %res, 1
6150 br i1 %obit, label %carry, label %normal
6155 <!-- _______________________________________________________________________ -->
6156 <div class="doc_subsubsection">
6157 <a name="int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt>' Intrinsics</a>
6160 <div class="doc_text">
6164 <p>This is an overloaded intrinsic. You can use <tt>llvm.ssub.with.overflow</tt>
6165 on any integer bit width.</p>
6168 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
6169 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6170 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
6175 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
6176 a signed subtraction of the two arguments, and indicate whether an overflow
6177 occurred during the signed subtraction.</p>
6181 <p>The arguments (%a and %b) and the first element of the result structure may
6182 be of integer types of any bit width, but they must have the same bit width. The
6183 second element of the result structure must be of type <tt>i1</tt>. <tt>%a</tt>
6184 and <tt>%b</tt> are the two values that will undergo signed subtraction.</p>
6188 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
6189 a signed subtraction of the two arguments. They return a structure — the
6190 first element of which is the subtraction, and the second element of which is a bit
6191 specifying if the signed subtraction resulted in an overflow.</p>
6195 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6196 %sum = extractvalue {i32, i1} %res, 0
6197 %obit = extractvalue {i32, i1} %res, 1
6198 br i1 %obit, label %overflow, label %normal
6203 <!-- _______________________________________________________________________ -->
6204 <div class="doc_subsubsection">
6205 <a name="int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt>' Intrinsics</a>
6208 <div class="doc_text">
6212 <p>This is an overloaded intrinsic. You can use <tt>llvm.usub.with.overflow</tt>
6213 on any integer bit width.</p>
6216 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
6217 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6218 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
6223 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
6224 an unsigned subtraction of the two arguments, and indicate whether an overflow
6225 occurred during the unsigned subtraction.</p>
6229 <p>The arguments (%a and %b) and the first element of the result structure may
6230 be of integer types of any bit width, but they must have the same bit width. The
6231 second element of the result structure must be of type <tt>i1</tt>. <tt>%a</tt>
6232 and <tt>%b</tt> are the two values that will undergo unsigned subtraction.</p>
6236 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
6237 an unsigned subtraction of the two arguments. They return a structure — the
6238 first element of which is the subtraction, and the second element of which is a bit
6239 specifying if the unsigned subtraction resulted in an overflow.</p>
6243 %res = call {i32, i1} @llvm.usub.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_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt>' Intrinsics</a>
6256 <div class="doc_text">
6260 <p>This is an overloaded intrinsic. You can use <tt>llvm.smul.with.overflow</tt>
6261 on any integer bit width.</p>
6264 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
6265 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6266 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
6271 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
6272 a signed multiplication of the two arguments, and indicate whether an overflow
6273 occurred during the signed multiplication.</p>
6277 <p>The arguments (%a and %b) and the first element of the result structure may
6278 be of integer types of any bit width, but they must have the same bit width. The
6279 second element of the result structure must be of type <tt>i1</tt>. <tt>%a</tt>
6280 and <tt>%b</tt> are the two values that will undergo signed multiplication.</p>
6284 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
6285 a signed multiplication of the two arguments. They return a structure —
6286 the first element of which is the multiplication, and the second element of
6287 which is a bit specifying if the signed multiplication resulted in an
6292 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6293 %sum = extractvalue {i32, i1} %res, 0
6294 %obit = extractvalue {i32, i1} %res, 1
6295 br i1 %obit, label %overflow, label %normal
6300 <!-- _______________________________________________________________________ -->
6301 <div class="doc_subsubsection">
6302 <a name="int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt>' Intrinsics</a>
6305 <div class="doc_text">
6309 <p>This is an overloaded intrinsic. You can use <tt>llvm.umul.with.overflow</tt>
6310 on any integer bit width.</p>
6313 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
6314 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6315 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
6320 <p><i><b>Warning:</b> '<tt>llvm.umul.with.overflow</tt>' is badly broken. It is
6321 actively being fixed, but it should not currently be used!</i></p>
6323 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
6324 a unsigned multiplication of the two arguments, and indicate whether an overflow
6325 occurred during the unsigned multiplication.</p>
6329 <p>The arguments (%a and %b) and the first element of the result structure may
6330 be of integer types of any bit width, but they must have the same bit width. The
6331 second element of the result structure must be of type <tt>i1</tt>. <tt>%a</tt>
6332 and <tt>%b</tt> are the two values that will undergo unsigned
6337 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
6338 an unsigned multiplication of the two arguments. They return a structure —
6339 the first element of which is the multiplication, and the second element of
6340 which is a bit specifying if the unsigned multiplication resulted in an
6345 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6346 %sum = extractvalue {i32, i1} %res, 0
6347 %obit = extractvalue {i32, i1} %res, 1
6348 br i1 %obit, label %overflow, label %normal
6353 <!-- ======================================================================= -->
6354 <div class="doc_subsection">
6355 <a name="int_debugger">Debugger Intrinsics</a>
6358 <div class="doc_text">
6360 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
6361 are described in the <a
6362 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
6363 Debugging</a> document.
6368 <!-- ======================================================================= -->
6369 <div class="doc_subsection">
6370 <a name="int_eh">Exception Handling Intrinsics</a>
6373 <div class="doc_text">
6374 <p> The LLVM exception handling intrinsics (which all start with
6375 <tt>llvm.eh.</tt> prefix), are described in the <a
6376 href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
6377 Handling</a> document. </p>
6380 <!-- ======================================================================= -->
6381 <div class="doc_subsection">
6382 <a name="int_trampoline">Trampoline Intrinsic</a>
6385 <div class="doc_text">
6387 This intrinsic makes it possible to excise one parameter, marked with
6388 the <tt>nest</tt> attribute, from a function. The result is a callable
6389 function pointer lacking the nest parameter - the caller does not need
6390 to provide a value for it. Instead, the value to use is stored in
6391 advance in a "trampoline", a block of memory usually allocated
6392 on the stack, which also contains code to splice the nest value into the
6393 argument list. This is used to implement the GCC nested function address
6397 For example, if the function is
6398 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
6399 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as follows:</p>
6401 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
6402 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
6403 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
6404 %fp = bitcast i8* %p to i32 (i32, i32)*
6406 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
6407 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
6410 <!-- _______________________________________________________________________ -->
6411 <div class="doc_subsubsection">
6412 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
6414 <div class="doc_text">
6417 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
6421 This fills the memory pointed to by <tt>tramp</tt> with code
6422 and returns a function pointer suitable for executing it.
6426 The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
6427 pointers. The <tt>tramp</tt> argument must point to a sufficiently large
6428 and sufficiently aligned block of memory; this memory is written to by the
6429 intrinsic. Note that the size and the alignment are target-specific - LLVM
6430 currently provides no portable way of determining them, so a front-end that
6431 generates this intrinsic needs to have some target-specific knowledge.
6432 The <tt>func</tt> argument must hold a function bitcast to an <tt>i8*</tt>.
6436 The block of memory pointed to by <tt>tramp</tt> is filled with target
6437 dependent code, turning it into a function. A pointer to this function is
6438 returned, but needs to be bitcast to an
6439 <a href="#int_trampoline">appropriate function pointer type</a>
6440 before being called. The new function's signature is the same as that of
6441 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
6442 removed. At most one such <tt>nest</tt> argument is allowed, and it must be
6443 of pointer type. Calling the new function is equivalent to calling
6444 <tt>func</tt> with the same argument list, but with <tt>nval</tt> used for the
6445 missing <tt>nest</tt> argument. If, after calling
6446 <tt>llvm.init.trampoline</tt>, the memory pointed to by <tt>tramp</tt> is
6447 modified, then the effect of any later call to the returned function pointer is
6452 <!-- ======================================================================= -->
6453 <div class="doc_subsection">
6454 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
6457 <div class="doc_text">
6459 These intrinsic functions expand the "universal IR" of LLVM to represent
6460 hardware constructs for atomic operations and memory synchronization. This
6461 provides an interface to the hardware, not an interface to the programmer. It
6462 is aimed at a low enough level to allow any programming models or APIs
6463 (Application Programming Interfaces) which
6464 need atomic behaviors to map cleanly onto it. It is also modeled primarily on
6465 hardware behavior. Just as hardware provides a "universal IR" for source
6466 languages, it also provides a starting point for developing a "universal"
6467 atomic operation and synchronization IR.
6470 These do <em>not</em> form an API such as high-level threading libraries,
6471 software transaction memory systems, atomic primitives, and intrinsic
6472 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
6473 application libraries. The hardware interface provided by LLVM should allow
6474 a clean implementation of all of these APIs and parallel programming models.
6475 No one model or paradigm should be selected above others unless the hardware
6476 itself ubiquitously does so.
6481 <!-- _______________________________________________________________________ -->
6482 <div class="doc_subsubsection">
6483 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
6485 <div class="doc_text">
6488 declare void @llvm.memory.barrier( i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>,
6494 The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
6495 specific pairs of memory access types.
6499 The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
6500 The first four arguments enables a specific barrier as listed below. The fith
6501 argument specifies that the barrier applies to io or device or uncached memory.
6505 <li><tt>ll</tt>: load-load barrier</li>
6506 <li><tt>ls</tt>: load-store barrier</li>
6507 <li><tt>sl</tt>: store-load barrier</li>
6508 <li><tt>ss</tt>: store-store barrier</li>
6509 <li><tt>device</tt>: barrier applies to device and uncached memory also.</li>
6513 This intrinsic causes the system to enforce some ordering constraints upon
6514 the loads and stores of the program. This barrier does not indicate
6515 <em>when</em> any events will occur, it only enforces an <em>order</em> in
6516 which they occur. For any of the specified pairs of load and store operations
6517 (f.ex. load-load, or store-load), all of the first operations preceding the
6518 barrier will complete before any of the second operations succeeding the
6519 barrier begin. Specifically the semantics for each pairing is as follows:
6522 <li><tt>ll</tt>: All loads before the barrier must complete before any load
6523 after the barrier begins.</li>
6525 <li><tt>ls</tt>: All loads before the barrier must complete before any
6526 store after the barrier begins.</li>
6527 <li><tt>ss</tt>: All stores before the barrier must complete before any
6528 store after the barrier begins.</li>
6529 <li><tt>sl</tt>: All stores before the barrier must complete before any
6530 load after the barrier begins.</li>
6533 These semantics are applied with a logical "and" behavior when more than one
6534 is enabled in a single memory barrier intrinsic.
6537 Backends may implement stronger barriers than those requested when they do not
6538 support as fine grained a barrier as requested. Some architectures do not
6539 need all types of barriers and on such architectures, these become noops.
6546 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
6547 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
6548 <i>; guarantee the above finishes</i>
6549 store i32 8, %ptr <i>; before this begins</i>
6553 <!-- _______________________________________________________________________ -->
6554 <div class="doc_subsubsection">
6555 <a name="int_atomic_cmp_swap">'<tt>llvm.atomic.cmp.swap.*</tt>' Intrinsic</a>
6557 <div class="doc_text">
6560 This is an overloaded intrinsic. You can use <tt>llvm.atomic.cmp.swap</tt> on
6561 any integer bit width and for different address spaces. Not all targets
6562 support all bit widths however.</p>
6565 declare i8 @llvm.atomic.cmp.swap.i8.p0i8( i8* <ptr>, i8 <cmp>, i8 <val> )
6566 declare i16 @llvm.atomic.cmp.swap.i16.p0i16( i16* <ptr>, i16 <cmp>, i16 <val> )
6567 declare i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* <ptr>, i32 <cmp>, i32 <val> )
6568 declare i64 @llvm.atomic.cmp.swap.i64.p0i64( i64* <ptr>, i64 <cmp>, i64 <val> )
6573 This loads a value in memory and compares it to a given value. If they are
6574 equal, it stores a new value into the memory.
6578 The <tt>llvm.atomic.cmp.swap</tt> intrinsic takes three arguments. The result as
6579 well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
6580 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
6581 this integer type. While any bit width integer may be used, targets may only
6582 lower representations they support in hardware.
6587 This entire intrinsic must be executed atomically. It first loads the value
6588 in memory pointed to by <tt>ptr</tt> and compares it with the value
6589 <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the memory. The
6590 loaded value is yielded in all cases. This provides the equivalent of an
6591 atomic compare-and-swap operation within the SSA framework.
6599 %val1 = add i32 4, 4
6600 %result1 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 4, %val1 )
6601 <i>; yields {i32}:result1 = 4</i>
6602 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
6603 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
6605 %val2 = add i32 1, 1
6606 %result2 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 5, %val2 )
6607 <i>; yields {i32}:result2 = 8</i>
6608 %stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
6610 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
6614 <!-- _______________________________________________________________________ -->
6615 <div class="doc_subsubsection">
6616 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
6618 <div class="doc_text">
6622 This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
6623 integer bit width. Not all targets support all bit widths however.</p>
6625 declare i8 @llvm.atomic.swap.i8.p0i8( i8* <ptr>, i8 <val> )
6626 declare i16 @llvm.atomic.swap.i16.p0i16( i16* <ptr>, i16 <val> )
6627 declare i32 @llvm.atomic.swap.i32.p0i32( i32* <ptr>, i32 <val> )
6628 declare i64 @llvm.atomic.swap.i64.p0i64( i64* <ptr>, i64 <val> )
6633 This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
6634 the value from memory. It then stores the value in <tt>val</tt> in the memory
6640 The <tt>llvm.atomic.swap</tt> intrinsic takes two arguments. Both the
6641 <tt>val</tt> argument and the result must be integers of the same bit width.
6642 The first argument, <tt>ptr</tt>, must be a pointer to a value of this
6643 integer type. The targets may only lower integer representations they
6648 This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
6649 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
6650 equivalent of an atomic swap operation within the SSA framework.
6658 %val1 = add i32 4, 4
6659 %result1 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val1 )
6660 <i>; yields {i32}:result1 = 4</i>
6661 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
6662 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
6664 %val2 = add i32 1, 1
6665 %result2 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val2 )
6666 <i>; yields {i32}:result2 = 8</i>
6668 %stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
6669 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
6673 <!-- _______________________________________________________________________ -->
6674 <div class="doc_subsubsection">
6675 <a name="int_atomic_load_add">'<tt>llvm.atomic.load.add.*</tt>' Intrinsic</a>
6678 <div class="doc_text">
6681 This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.add</tt> on any
6682 integer bit width. Not all targets support all bit widths however.</p>
6684 declare i8 @llvm.atomic.load.add.i8..p0i8( i8* <ptr>, i8 <delta> )
6685 declare i16 @llvm.atomic.load.add.i16..p0i16( i16* <ptr>, i16 <delta> )
6686 declare i32 @llvm.atomic.load.add.i32..p0i32( i32* <ptr>, i32 <delta> )
6687 declare i64 @llvm.atomic.load.add.i64..p0i64( i64* <ptr>, i64 <delta> )
6692 This intrinsic adds <tt>delta</tt> to the value stored in memory at
6693 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
6698 The intrinsic takes two arguments, the first a pointer to an integer value
6699 and the second an integer value. The result is also an integer value. These
6700 integer types can have any bit width, but they must all have the same bit
6701 width. The targets may only lower integer representations they support.
6705 This intrinsic does a series of operations atomically. It first loads the
6706 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
6707 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
6714 %result1 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 4 )
6715 <i>; yields {i32}:result1 = 4</i>
6716 %result2 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 2 )
6717 <i>; yields {i32}:result2 = 8</i>
6718 %result3 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 5 )
6719 <i>; yields {i32}:result3 = 10</i>
6720 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
6724 <!-- _______________________________________________________________________ -->
6725 <div class="doc_subsubsection">
6726 <a name="int_atomic_load_sub">'<tt>llvm.atomic.load.sub.*</tt>' Intrinsic</a>
6729 <div class="doc_text">
6732 This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.sub</tt> on
6733 any integer bit width and for different address spaces. Not all targets
6734 support all bit widths however.</p>
6736 declare i8 @llvm.atomic.load.sub.i8.p0i32( i8* <ptr>, i8 <delta> )
6737 declare i16 @llvm.atomic.load.sub.i16.p0i32( i16* <ptr>, i16 <delta> )
6738 declare i32 @llvm.atomic.load.sub.i32.p0i32( i32* <ptr>, i32 <delta> )
6739 declare i64 @llvm.atomic.load.sub.i64.p0i32( i64* <ptr>, i64 <delta> )
6744 This intrinsic subtracts <tt>delta</tt> to the value stored in memory at
6745 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
6750 The intrinsic takes two arguments, the first a pointer to an integer value
6751 and the second an integer value. The result is also an integer value. These
6752 integer types can have any bit width, but they must all have the same bit
6753 width. The targets may only lower integer representations they support.
6757 This intrinsic does a series of operations atomically. It first loads the
6758 value stored at <tt>ptr</tt>. It then subtracts <tt>delta</tt>, stores the
6759 result to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
6766 %result1 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 4 )
6767 <i>; yields {i32}:result1 = 8</i>
6768 %result2 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 2 )
6769 <i>; yields {i32}:result2 = 4</i>
6770 %result3 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 5 )
6771 <i>; yields {i32}:result3 = 2</i>
6772 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = -3</i>
6776 <!-- _______________________________________________________________________ -->
6777 <div class="doc_subsubsection">
6778 <a name="int_atomic_load_and">'<tt>llvm.atomic.load.and.*</tt>' Intrinsic</a><br>
6779 <a name="int_atomic_load_nand">'<tt>llvm.atomic.load.nand.*</tt>' Intrinsic</a><br>
6780 <a name="int_atomic_load_or">'<tt>llvm.atomic.load.or.*</tt>' Intrinsic</a><br>
6781 <a name="int_atomic_load_xor">'<tt>llvm.atomic.load.xor.*</tt>' Intrinsic</a><br>
6784 <div class="doc_text">
6787 These are overloaded intrinsics. You can use <tt>llvm.atomic.load_and</tt>,
6788 <tt>llvm.atomic.load_nand</tt>, <tt>llvm.atomic.load_or</tt>, and
6789 <tt>llvm.atomic.load_xor</tt> on any integer bit width and for different
6790 address spaces. Not all targets support all bit widths however.</p>
6792 declare i8 @llvm.atomic.load.and.i8.p0i8( i8* <ptr>, i8 <delta> )
6793 declare i16 @llvm.atomic.load.and.i16.p0i16( i16* <ptr>, i16 <delta> )
6794 declare i32 @llvm.atomic.load.and.i32.p0i32( i32* <ptr>, i32 <delta> )
6795 declare i64 @llvm.atomic.load.and.i64.p0i64( i64* <ptr>, i64 <delta> )
6800 declare i8 @llvm.atomic.load.or.i8.p0i8( i8* <ptr>, i8 <delta> )
6801 declare i16 @llvm.atomic.load.or.i16.p0i16( i16* <ptr>, i16 <delta> )
6802 declare i32 @llvm.atomic.load.or.i32.p0i32( i32* <ptr>, i32 <delta> )
6803 declare i64 @llvm.atomic.load.or.i64.p0i64( i64* <ptr>, i64 <delta> )
6808 declare i8 @llvm.atomic.load.nand.i8.p0i32( i8* <ptr>, i8 <delta> )
6809 declare i16 @llvm.atomic.load.nand.i16.p0i32( i16* <ptr>, i16 <delta> )
6810 declare i32 @llvm.atomic.load.nand.i32.p0i32( i32* <ptr>, i32 <delta> )
6811 declare i64 @llvm.atomic.load.nand.i64.p0i32( i64* <ptr>, i64 <delta> )
6816 declare i8 @llvm.atomic.load.xor.i8.p0i32( i8* <ptr>, i8 <delta> )
6817 declare i16 @llvm.atomic.load.xor.i16.p0i32( i16* <ptr>, i16 <delta> )
6818 declare i32 @llvm.atomic.load.xor.i32.p0i32( i32* <ptr>, i32 <delta> )
6819 declare i64 @llvm.atomic.load.xor.i64.p0i32( i64* <ptr>, i64 <delta> )
6824 These intrinsics bitwise the operation (and, nand, or, xor) <tt>delta</tt> to
6825 the value stored in memory at <tt>ptr</tt>. It yields the original value
6831 These intrinsics take two arguments, the first a pointer to an integer value
6832 and the second an integer value. The result is also an integer value. These
6833 integer types can have any bit width, but they must all have the same bit
6834 width. The targets may only lower integer representations they support.
6838 These intrinsics does a series of operations atomically. They first load the
6839 value stored at <tt>ptr</tt>. They then do the bitwise operation
6840 <tt>delta</tt>, store the result to <tt>ptr</tt>. They yield the original
6841 value stored at <tt>ptr</tt>.
6847 store i32 0x0F0F, %ptr
6848 %result0 = call i32 @llvm.atomic.load.nand.i32.p0i32( i32* %ptr, i32 0xFF )
6849 <i>; yields {i32}:result0 = 0x0F0F</i>
6850 %result1 = call i32 @llvm.atomic.load.and.i32.p0i32( i32* %ptr, i32 0xFF )
6851 <i>; yields {i32}:result1 = 0xFFFFFFF0</i>
6852 %result2 = call i32 @llvm.atomic.load.or.i32.p0i32( i32* %ptr, i32 0F )
6853 <i>; yields {i32}:result2 = 0xF0</i>
6854 %result3 = call i32 @llvm.atomic.load.xor.i32.p0i32( i32* %ptr, i32 0F )
6855 <i>; yields {i32}:result3 = FF</i>
6856 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = F0</i>
6861 <!-- _______________________________________________________________________ -->
6862 <div class="doc_subsubsection">
6863 <a name="int_atomic_load_max">'<tt>llvm.atomic.load.max.*</tt>' Intrinsic</a><br>
6864 <a name="int_atomic_load_min">'<tt>llvm.atomic.load.min.*</tt>' Intrinsic</a><br>
6865 <a name="int_atomic_load_umax">'<tt>llvm.atomic.load.umax.*</tt>' Intrinsic</a><br>
6866 <a name="int_atomic_load_umin">'<tt>llvm.atomic.load.umin.*</tt>' Intrinsic</a><br>
6869 <div class="doc_text">
6872 These are overloaded intrinsics. You can use <tt>llvm.atomic.load_max</tt>,
6873 <tt>llvm.atomic.load_min</tt>, <tt>llvm.atomic.load_umax</tt>, and
6874 <tt>llvm.atomic.load_umin</tt> on any integer bit width and for different
6875 address spaces. Not all targets
6876 support all bit widths however.</p>
6878 declare i8 @llvm.atomic.load.max.i8.p0i8( i8* <ptr>, i8 <delta> )
6879 declare i16 @llvm.atomic.load.max.i16.p0i16( i16* <ptr>, i16 <delta> )
6880 declare i32 @llvm.atomic.load.max.i32.p0i32( i32* <ptr>, i32 <delta> )
6881 declare i64 @llvm.atomic.load.max.i64.p0i64( i64* <ptr>, i64 <delta> )
6886 declare i8 @llvm.atomic.load.min.i8.p0i8( i8* <ptr>, i8 <delta> )
6887 declare i16 @llvm.atomic.load.min.i16.p0i16( i16* <ptr>, i16 <delta> )
6888 declare i32 @llvm.atomic.load.min.i32..p0i32( i32* <ptr>, i32 <delta> )
6889 declare i64 @llvm.atomic.load.min.i64..p0i64( i64* <ptr>, i64 <delta> )
6894 declare i8 @llvm.atomic.load.umax.i8.p0i8( i8* <ptr>, i8 <delta> )
6895 declare i16 @llvm.atomic.load.umax.i16.p0i16( i16* <ptr>, i16 <delta> )
6896 declare i32 @llvm.atomic.load.umax.i32.p0i32( i32* <ptr>, i32 <delta> )
6897 declare i64 @llvm.atomic.load.umax.i64.p0i64( i64* <ptr>, i64 <delta> )
6902 declare i8 @llvm.atomic.load.umin.i8..p0i8( i8* <ptr>, i8 <delta> )
6903 declare i16 @llvm.atomic.load.umin.i16.p0i16( i16* <ptr>, i16 <delta> )
6904 declare i32 @llvm.atomic.load.umin.i32..p0i32( i32* <ptr>, i32 <delta> )
6905 declare i64 @llvm.atomic.load.umin.i64..p0i64( i64* <ptr>, i64 <delta> )
6910 These intrinsics takes the signed or unsigned minimum or maximum of
6911 <tt>delta</tt> and the value stored in memory at <tt>ptr</tt>. It yields the
6912 original value at <tt>ptr</tt>.
6917 These intrinsics take two arguments, the first a pointer to an integer value
6918 and the second an integer value. The result is also an integer value. These
6919 integer types can have any bit width, but they must all have the same bit
6920 width. The targets may only lower integer representations they support.
6924 These intrinsics does a series of operations atomically. They first load the
6925 value stored at <tt>ptr</tt>. They then do the signed or unsigned min or max
6926 <tt>delta</tt> and the value, store the result to <tt>ptr</tt>. They yield
6927 the original value stored at <tt>ptr</tt>.
6934 %result0 = call i32 @llvm.atomic.load.min.i32.p0i32( i32* %ptr, i32 -2 )
6935 <i>; yields {i32}:result0 = 7</i>
6936 %result1 = call i32 @llvm.atomic.load.max.i32.p0i32( i32* %ptr, i32 8 )
6937 <i>; yields {i32}:result1 = -2</i>
6938 %result2 = call i32 @llvm.atomic.load.umin.i32.p0i32( i32* %ptr, i32 10 )
6939 <i>; yields {i32}:result2 = 8</i>
6940 %result3 = call i32 @llvm.atomic.load.umax.i32.p0i32( i32* %ptr, i32 30 )
6941 <i>; yields {i32}:result3 = 8</i>
6942 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 30</i>
6946 <!-- ======================================================================= -->
6947 <div class="doc_subsection">
6948 <a name="int_general">General Intrinsics</a>
6951 <div class="doc_text">
6952 <p> This class of intrinsics is designed to be generic and has
6953 no specific purpose. </p>
6956 <!-- _______________________________________________________________________ -->
6957 <div class="doc_subsubsection">
6958 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
6961 <div class="doc_text">
6965 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
6971 The '<tt>llvm.var.annotation</tt>' intrinsic
6977 The first argument is a pointer to a value, the second is a pointer to a
6978 global string, the third is a pointer to a global string which is the source
6979 file name, and the last argument is the line number.
6985 This intrinsic allows annotation of local variables with arbitrary strings.
6986 This can be useful for special purpose optimizations that want to look for these
6987 annotations. These have no other defined use, they are ignored by code
6988 generation and optimization.
6992 <!-- _______________________________________________________________________ -->
6993 <div class="doc_subsubsection">
6994 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
6997 <div class="doc_text">
7000 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
7001 any integer bit width.
7004 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
7005 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
7006 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
7007 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
7008 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
7014 The '<tt>llvm.annotation</tt>' intrinsic.
7020 The first argument is an integer value (result of some expression),
7021 the second is a pointer to a global string, the third is a pointer to a global
7022 string which is the source file name, and the last argument is the line number.
7023 It returns the value of the first argument.
7029 This intrinsic allows annotations to be put on arbitrary expressions
7030 with arbitrary strings. This can be useful for special purpose optimizations
7031 that want to look for these annotations. These have no other defined use, they
7032 are ignored by code generation and optimization.
7036 <!-- _______________________________________________________________________ -->
7037 <div class="doc_subsubsection">
7038 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
7041 <div class="doc_text">
7045 declare void @llvm.trap()
7051 The '<tt>llvm.trap</tt>' intrinsic
7063 This intrinsics is lowered to the target dependent trap instruction. If the
7064 target does not have a trap instruction, this intrinsic will be lowered to the
7065 call of the abort() function.
7069 <!-- _______________________________________________________________________ -->
7070 <div class="doc_subsubsection">
7071 <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a>
7073 <div class="doc_text">
7076 declare void @llvm.stackprotector( i8* <guard>, i8** <slot> )
7081 The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and stores
7082 it onto the stack at <tt>slot</tt>. The stack slot is adjusted to ensure that
7083 it is placed on the stack before local variables.
7087 The <tt>llvm.stackprotector</tt> intrinsic requires two pointer arguments. The
7088 first argument is the value loaded from the stack guard
7089 <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt> that
7090 has enough space to hold the value of the guard.
7094 This intrinsic causes the prologue/epilogue inserter to force the position of
7095 the <tt>AllocaInst</tt> stack slot to be before local variables on the
7096 stack. This is to ensure that if a local variable on the stack is overwritten,
7097 it will destroy the value of the guard. When the function exits, the guard on
7098 the stack is checked against the original guard. If they're different, then
7099 the program aborts by calling the <tt>__stack_chk_fail()</tt> function.
7103 <!-- *********************************************************************** -->
7106 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
7107 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
7108 <a href="http://validator.w3.org/check/referer"><img
7109 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
7111 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
7112 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
7113 Last modified: $Date$