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_fadd">'<tt>fadd</tt>' Instruction</a></li>
93 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
94 <li><a href="#i_fsub">'<tt>fsub</tt>' Instruction</a></li>
95 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
96 <li><a href="#i_fmul">'<tt>fmul</tt>' Instruction</a></li>
97 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
98 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
99 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
100 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
101 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
102 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
105 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
107 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
108 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
109 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
110 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
111 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
112 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
115 <li><a href="#vectorops">Vector Operations</a>
117 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
118 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
119 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
122 <li><a href="#aggregateops">Aggregate Operations</a>
124 <li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li>
125 <li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li>
128 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
130 <li><a href="#i_malloc">'<tt>malloc</tt>' Instruction</a></li>
131 <li><a href="#i_free">'<tt>free</tt>' Instruction</a></li>
132 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
133 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
134 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
135 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
138 <li><a href="#convertops">Conversion Operations</a>
140 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
141 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
142 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
143 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
144 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
145 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
146 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
147 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
148 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
149 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
150 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
151 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
154 <li><a href="#otherops">Other Operations</a>
156 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
157 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
158 <li><a href="#i_vicmp">'<tt>vicmp</tt>' Instruction</a></li>
159 <li><a href="#i_vfcmp">'<tt>vfcmp</tt>' Instruction</a></li>
160 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
161 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
162 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
163 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
168 <li><a href="#intrinsics">Intrinsic Functions</a>
170 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
172 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
173 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
174 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
177 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
179 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
180 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
181 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
184 <li><a href="#int_codegen">Code Generator Intrinsics</a>
186 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
187 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
188 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
189 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
190 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
191 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
192 <li><a href="#int_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
195 <li><a href="#int_libc">Standard C Library Intrinsics</a>
197 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
198 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
199 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
200 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
201 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
202 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
203 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
204 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
207 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
209 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
210 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
211 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
212 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
213 <li><a href="#int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic </a></li>
214 <li><a href="#int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic </a></li>
217 <li><a href="#int_overflow">Arithmetic with Overflow Intrinsics</a>
219 <li><a href="#int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt> Intrinsics</a></li>
220 <li><a href="#int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt> Intrinsics</a></li>
221 <li><a href="#int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt> Intrinsics</a></li>
222 <li><a href="#int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt> Intrinsics</a></li>
223 <li><a href="#int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt> Intrinsics</a></li>
224 <li><a href="#int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt> Intrinsics</a></li>
227 <li><a href="#int_debugger">Debugger intrinsics</a></li>
228 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
229 <li><a href="#int_trampoline">Trampoline Intrinsic</a>
231 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
234 <li><a href="#int_atomics">Atomic intrinsics</a>
236 <li><a href="#int_memory_barrier"><tt>llvm.memory_barrier</tt></a></li>
237 <li><a href="#int_atomic_cmp_swap"><tt>llvm.atomic.cmp.swap</tt></a></li>
238 <li><a href="#int_atomic_swap"><tt>llvm.atomic.swap</tt></a></li>
239 <li><a href="#int_atomic_load_add"><tt>llvm.atomic.load.add</tt></a></li>
240 <li><a href="#int_atomic_load_sub"><tt>llvm.atomic.load.sub</tt></a></li>
241 <li><a href="#int_atomic_load_and"><tt>llvm.atomic.load.and</tt></a></li>
242 <li><a href="#int_atomic_load_nand"><tt>llvm.atomic.load.nand</tt></a></li>
243 <li><a href="#int_atomic_load_or"><tt>llvm.atomic.load.or</tt></a></li>
244 <li><a href="#int_atomic_load_xor"><tt>llvm.atomic.load.xor</tt></a></li>
245 <li><a href="#int_atomic_load_max"><tt>llvm.atomic.load.max</tt></a></li>
246 <li><a href="#int_atomic_load_min"><tt>llvm.atomic.load.min</tt></a></li>
247 <li><a href="#int_atomic_load_umax"><tt>llvm.atomic.load.umax</tt></a></li>
248 <li><a href="#int_atomic_load_umin"><tt>llvm.atomic.load.umin</tt></a></li>
251 <li><a href="#int_general">General intrinsics</a>
253 <li><a href="#int_var_annotation">
254 '<tt>llvm.var.annotation</tt>' Intrinsic</a></li>
255 <li><a href="#int_annotation">
256 '<tt>llvm.annotation.*</tt>' Intrinsic</a></li>
257 <li><a href="#int_trap">
258 '<tt>llvm.trap</tt>' Intrinsic</a></li>
259 <li><a href="#int_stackprotector">
260 '<tt>llvm.stackprotector</tt>' Intrinsic</a></li>
267 <div class="doc_author">
268 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
269 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
272 <!-- *********************************************************************** -->
273 <div class="doc_section"> <a name="abstract">Abstract </a></div>
274 <!-- *********************************************************************** -->
276 <div class="doc_text">
277 <p>This document is a reference manual for the LLVM assembly language.
278 LLVM is a Static Single Assignment (SSA) based representation that provides
279 type safety, low-level operations, flexibility, and the capability of
280 representing 'all' high-level languages cleanly. It is the common code
281 representation used throughout all phases of the LLVM compilation
285 <!-- *********************************************************************** -->
286 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
287 <!-- *********************************************************************** -->
289 <div class="doc_text">
291 <p>The LLVM code representation is designed to be used in three
292 different forms: as an in-memory compiler IR, as an on-disk bitcode
293 representation (suitable for fast loading by a Just-In-Time compiler),
294 and as a human readable assembly language representation. This allows
295 LLVM to provide a powerful intermediate representation for efficient
296 compiler transformations and analysis, while providing a natural means
297 to debug and visualize the transformations. The three different forms
298 of LLVM are all equivalent. This document describes the human readable
299 representation and notation.</p>
301 <p>The LLVM representation aims to be light-weight and low-level
302 while being expressive, typed, and extensible at the same time. It
303 aims to be a "universal IR" of sorts, by being at a low enough level
304 that high-level ideas may be cleanly mapped to it (similar to how
305 microprocessors are "universal IR's", allowing many source languages to
306 be mapped to them). By providing type information, LLVM can be used as
307 the target of optimizations: for example, through pointer analysis, it
308 can be proven that a C automatic variable is never accessed outside of
309 the current function... allowing it to be promoted to a simple SSA
310 value instead of a memory location.</p>
314 <!-- _______________________________________________________________________ -->
315 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
317 <div class="doc_text">
319 <p>It is important to note that this document describes 'well formed'
320 LLVM assembly language. There is a difference between what the parser
321 accepts and what is considered 'well formed'. For example, the
322 following instruction is syntactically okay, but not well formed:</p>
324 <div class="doc_code">
326 %x = <a href="#i_add">add</a> i32 1, %x
330 <p>...because the definition of <tt>%x</tt> does not dominate all of
331 its uses. The LLVM infrastructure provides a verification pass that may
332 be used to verify that an LLVM module is well formed. This pass is
333 automatically run by the parser after parsing input assembly and by
334 the optimizer before it outputs bitcode. The violations pointed out
335 by the verifier pass indicate bugs in transformation passes or input to
339 <!-- Describe the typesetting conventions here. -->
341 <!-- *********************************************************************** -->
342 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
343 <!-- *********************************************************************** -->
345 <div class="doc_text">
347 <p>LLVM identifiers come in two basic types: global and local. Global
348 identifiers (functions, global variables) begin with the @ character. Local
349 identifiers (register names, types) begin with the % character. Additionally,
350 there are three different formats for identifiers, for different purposes:</p>
353 <li>Named values are represented as a string of characters with their prefix.
354 For example, %foo, @DivisionByZero, %a.really.long.identifier. The actual
355 regular expression used is '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
356 Identifiers which require other characters in their names can be surrounded
357 with quotes. Special characters may be escaped using "\xx" where xx is the
358 ASCII code for the character in hexadecimal. In this way, any character can
359 be used in a name value, even quotes themselves.
361 <li>Unnamed values are represented as an unsigned numeric value with their
362 prefix. For example, %12, @2, %44.</li>
364 <li>Constants, which are described in a <a href="#constants">section about
365 constants</a>, below.</li>
368 <p>LLVM requires that values start with a prefix for two reasons: Compilers
369 don't need to worry about name clashes with reserved words, and the set of
370 reserved words may be expanded in the future without penalty. Additionally,
371 unnamed identifiers allow a compiler to quickly come up with a temporary
372 variable without having to avoid symbol table conflicts.</p>
374 <p>Reserved words in LLVM are very similar to reserved words in other
375 languages. There are keywords for different opcodes
376 ('<tt><a href="#i_add">add</a></tt>',
377 '<tt><a href="#i_bitcast">bitcast</a></tt>',
378 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
379 href="#t_void">void</a></tt>', '<tt><a href="#t_primitive">i32</a></tt>', etc...),
380 and others. These reserved words cannot conflict with variable names, because
381 none of them start with a prefix character ('%' or '@').</p>
383 <p>Here is an example of LLVM code to multiply the integer variable
384 '<tt>%X</tt>' by 8:</p>
388 <div class="doc_code">
390 %result = <a href="#i_mul">mul</a> i32 %X, 8
394 <p>After strength reduction:</p>
396 <div class="doc_code">
398 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
402 <p>And the hard way:</p>
404 <div class="doc_code">
406 <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
407 <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
408 %result = <a href="#i_add">add</a> i32 %1, %1
412 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
413 important lexical features of LLVM:</p>
417 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
420 <li>Unnamed temporaries are created when the result of a computation is not
421 assigned to a named value.</li>
423 <li>Unnamed temporaries are numbered sequentially</li>
427 <p>...and it also shows a convention that we follow in this document. When
428 demonstrating instructions, we will follow an instruction with a comment that
429 defines the type and name of value produced. Comments are shown in italic
434 <!-- *********************************************************************** -->
435 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
436 <!-- *********************************************************************** -->
438 <!-- ======================================================================= -->
439 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
442 <div class="doc_text">
444 <p>LLVM programs are composed of "Module"s, each of which is a
445 translation unit of the input programs. Each module consists of
446 functions, global variables, and symbol table entries. Modules may be
447 combined together with the LLVM linker, which merges function (and
448 global variable) definitions, resolves forward declarations, and merges
449 symbol table entries. Here is an example of the "hello world" module:</p>
451 <div class="doc_code">
452 <pre><i>; Declare the string constant as a global constant...</i>
453 <a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a
454 href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
456 <i>; External declaration of the puts function</i>
457 <a href="#functionstructure">declare</a> i32 @puts(i8 *) <i>; i32(i8 *)* </i>
459 <i>; Definition of main function</i>
460 define i32 @main() { <i>; i32()* </i>
461 <i>; Convert [13 x i8]* to i8 *...</i>
463 href="#i_getelementptr">getelementptr</a> [13 x i8]* @.LC0, i64 0, i64 0 <i>; i8 *</i>
465 <i>; Call puts function to write out the string to stdout...</i>
467 href="#i_call">call</a> i32 @puts(i8 * %cast210) <i>; i32</i>
469 href="#i_ret">ret</a> i32 0<br>}<br>
473 <p>This example is made up of a <a href="#globalvars">global variable</a>
474 named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
475 function, and a <a href="#functionstructure">function definition</a>
476 for "<tt>main</tt>".</p>
478 <p>In general, a module is made up of a list of global values,
479 where both functions and global variables are global values. Global values are
480 represented by a pointer to a memory location (in this case, a pointer to an
481 array of char, and a pointer to a function), and have one of the following <a
482 href="#linkage">linkage types</a>.</p>
486 <!-- ======================================================================= -->
487 <div class="doc_subsection">
488 <a name="linkage">Linkage Types</a>
491 <div class="doc_text">
494 All Global Variables and Functions have one of the following types of linkage:
499 <dt><tt><b><a name="linkage_private">private</a></b></tt>: </dt>
501 <dd>Global values with private linkage are only directly accessible by
502 objects in the current module. In particular, linking code into a module with
503 an private global value may cause the private to be renamed as necessary to
504 avoid collisions. Because the symbol is private to the module, all
505 references can be updated. This doesn't show up in any symbol table in the
509 <dt><tt><b><a name="linkage_internal">internal</a></b></tt>: </dt>
511 <dd> Similar to private, but the value shows as a local symbol (STB_LOCAL in
512 the case of ELF) in the object file. This corresponds to the notion of the
513 '<tt>static</tt>' keyword in C.
516 <dt><tt><b><a name="available_externally">available_externally</a></b></tt>:
519 <dd>Globals with "<tt>available_externally</tt>" linkage are never emitted
520 into the object file corresponding to the LLVM module. They exist to
521 allow inlining and other optimizations to take place given knowledge of the
522 definition of the global, which is known to be somewhere outside the module.
523 Globals with <tt>available_externally</tt> linkage are allowed to be discarded
524 at will, and are otherwise the same as <tt>linkonce_odr</tt>. This linkage
525 type is only allowed on definitions, not declarations.</dd>
527 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
529 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
530 the same name when linkage occurs. This is typically used to implement
531 inline functions, templates, or other code which must be generated in each
532 translation unit that uses it. Unreferenced <tt>linkonce</tt> globals are
533 allowed to be discarded.
536 <dt><tt><b><a name="linkage_common">common</a></b></tt>: </dt>
538 <dd>"<tt>common</tt>" linkage is exactly the same as <tt>linkonce</tt>
539 linkage, except that unreferenced <tt>common</tt> globals may not be
540 discarded. This is used for globals that may be emitted in multiple
541 translation units, but that are not guaranteed to be emitted into every
542 translation unit that uses them. One example of this is tentative
543 definitions in C, such as "<tt>int X;</tt>" at global scope.
546 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
548 <dd>"<tt>weak</tt>" linkage is the same as <tt>common</tt> linkage, except
549 that some targets may choose to emit different assembly sequences for them
550 for target-dependent reasons. This is used for globals that are declared
551 "weak" in C source code.
554 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
556 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
557 pointer to array type. When two global variables with appending linkage are
558 linked together, the two global arrays are appended together. This is the
559 LLVM, typesafe, equivalent of having the system linker append together
560 "sections" with identical names when .o files are linked.
563 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt>: </dt>
565 <dd>The semantics of this linkage follow the ELF object file model: the
566 symbol is weak until linked, if not linked, the symbol becomes null instead
567 of being an undefined reference.
570 <dt><tt><b><a name="linkage_linkonce">linkonce_odr</a></b></tt>: </dt>
571 <dt><tt><b><a name="linkage_weak">weak_odr</a></b></tt>: </dt>
572 <dd>Some languages allow differing globals to be merged, such as two
573 functions with different semantics. Other languages, such as <tt>C++</tt>,
574 ensure that only equivalent globals are ever merged (the "one definition
575 rule" - "ODR"). Such languages can use the <tt>linkonce_odr</tt>
576 and <tt>weak_odr</tt> linkage types to indicate that the global will only
577 be merged with equivalent globals. These linkage types are otherwise the
578 same as their non-<tt>odr</tt> versions.
581 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
583 <dd>If none of the above identifiers are used, the global is externally
584 visible, meaning that it participates in linkage and can be used to resolve
585 external symbol references.
590 The next two types of linkage are targeted for Microsoft Windows platform
591 only. They are designed to support importing (exporting) symbols from (to)
592 DLLs (Dynamic Link Libraries).
596 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt>: </dt>
598 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
599 or variable via a global pointer to a pointer that is set up by the DLL
600 exporting the symbol. On Microsoft Windows targets, the pointer name is
601 formed by combining <code>__imp_</code> and the function or variable name.
604 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt>: </dt>
606 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
607 pointer to a pointer in a DLL, so that it can be referenced with the
608 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
609 name is formed by combining <code>__imp_</code> and the function or variable
615 <p>For example, since the "<tt>.LC0</tt>"
616 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
617 variable and was linked with this one, one of the two would be renamed,
618 preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
619 external (i.e., lacking any linkage declarations), they are accessible
620 outside of the current module.</p>
621 <p>It is illegal for a function <i>declaration</i>
622 to have any linkage type other than "externally visible", <tt>dllimport</tt>
623 or <tt>extern_weak</tt>.</p>
624 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt>, <tt>weak</tt>
625 or <tt>weak_odr</tt> linkages.</p>
628 <!-- ======================================================================= -->
629 <div class="doc_subsection">
630 <a name="callingconv">Calling Conventions</a>
633 <div class="doc_text">
635 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
636 and <a href="#i_invoke">invokes</a> can all have an optional calling convention
637 specified for the call. The calling convention of any pair of dynamic
638 caller/callee must match, or the behavior of the program is undefined. The
639 following calling conventions are supported by LLVM, and more may be added in
643 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
645 <dd>This calling convention (the default if no other calling convention is
646 specified) matches the target C calling conventions. This calling convention
647 supports varargs function calls and tolerates some mismatch in the declared
648 prototype and implemented declaration of the function (as does normal C).
651 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
653 <dd>This calling convention attempts to make calls as fast as possible
654 (e.g. by passing things in registers). This calling convention allows the
655 target to use whatever tricks it wants to produce fast code for the target,
656 without having to conform to an externally specified ABI (Application Binary
657 Interface). Implementations of this convention should allow arbitrary
658 <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> to be
659 supported. This calling convention does not support varargs and requires the
660 prototype of all callees to exactly match the prototype of the function
664 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
666 <dd>This calling convention attempts to make code in the caller as efficient
667 as possible under the assumption that the call is not commonly executed. As
668 such, these calls often preserve all registers so that the call does not break
669 any live ranges in the caller side. This calling convention does not support
670 varargs and requires the prototype of all callees to exactly match the
671 prototype of the function definition.
674 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
676 <dd>Any calling convention may be specified by number, allowing
677 target-specific calling conventions to be used. Target specific calling
678 conventions start at 64.
682 <p>More calling conventions can be added/defined on an as-needed basis, to
683 support pascal conventions or any other well-known target-independent
688 <!-- ======================================================================= -->
689 <div class="doc_subsection">
690 <a name="visibility">Visibility Styles</a>
693 <div class="doc_text">
696 All Global Variables and Functions have one of the following visibility styles:
700 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
702 <dd>On targets that use the ELF object file format, default visibility means
703 that the declaration is visible to other
704 modules and, in shared libraries, means that the declared entity may be
705 overridden. On Darwin, default visibility means that the declaration is
706 visible to other modules. Default visibility corresponds to "external
707 linkage" in the language.
710 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
712 <dd>Two declarations of an object with hidden visibility refer to the same
713 object if they are in the same shared object. Usually, hidden visibility
714 indicates that the symbol will not be placed into the dynamic symbol table,
715 so no other module (executable or shared library) can reference it
719 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
721 <dd>On ELF, protected visibility indicates that the symbol will be placed in
722 the dynamic symbol table, but that references within the defining module will
723 bind to the local symbol. That is, the symbol cannot be overridden by another
730 <!-- ======================================================================= -->
731 <div class="doc_subsection">
732 <a name="namedtypes">Named Types</a>
735 <div class="doc_text">
737 <p>LLVM IR allows you to specify name aliases for certain types. This can make
738 it easier to read the IR and make the IR more condensed (particularly when
739 recursive types are involved). An example of a name specification is:
742 <div class="doc_code">
744 %mytype = type { %mytype*, i32 }
748 <p>You may give a name to any <a href="#typesystem">type</a> except "<a
749 href="t_void">void</a>". Type name aliases may be used anywhere a type is
750 expected with the syntax "%mytype".</p>
752 <p>Note that type names are aliases for the structural type that they indicate,
753 and that you can therefore specify multiple names for the same type. This often
754 leads to confusing behavior when dumping out a .ll file. Since LLVM IR uses
755 structural typing, the name is not part of the type. When printing out LLVM IR,
756 the printer will pick <em>one name</em> to render all types of a particular
757 shape. This means that if you have code where two different source types end up
758 having the same LLVM type, that the dumper will sometimes print the "wrong" or
759 unexpected type. This is an important design point and isn't going to
764 <!-- ======================================================================= -->
765 <div class="doc_subsection">
766 <a name="globalvars">Global Variables</a>
769 <div class="doc_text">
771 <p>Global variables define regions of memory allocated at compilation time
772 instead of run-time. Global variables may optionally be initialized, may have
773 an explicit section to be placed in, and may have an optional explicit alignment
774 specified. A variable may be defined as "thread_local", which means that it
775 will not be shared by threads (each thread will have a separated copy of the
776 variable). A variable may be defined as a global "constant," which indicates
777 that the contents of the variable will <b>never</b> be modified (enabling better
778 optimization, allowing the global data to be placed in the read-only section of
779 an executable, etc). Note that variables that need runtime initialization
780 cannot be marked "constant" as there is a store to the variable.</p>
783 LLVM explicitly allows <em>declarations</em> of global variables to be marked
784 constant, even if the final definition of the global is not. This capability
785 can be used to enable slightly better optimization of the program, but requires
786 the language definition to guarantee that optimizations based on the
787 'constantness' are valid for the translation units that do not include the
791 <p>As SSA values, global variables define pointer values that are in
792 scope (i.e. they dominate) all basic blocks in the program. Global
793 variables always define a pointer to their "content" type because they
794 describe a region of memory, and all memory objects in LLVM are
795 accessed through pointers.</p>
797 <p>A global variable may be declared to reside in a target-specifc numbered
798 address space. For targets that support them, address spaces may affect how
799 optimizations are performed and/or what target instructions are used to access
800 the variable. The default address space is zero. The address space qualifier
801 must precede any other attributes.</p>
803 <p>LLVM allows an explicit section to be specified for globals. If the target
804 supports it, it will emit globals to the section specified.</p>
806 <p>An explicit alignment may be specified for a global. If not present, or if
807 the alignment is set to zero, the alignment of the global is set by the target
808 to whatever it feels convenient. If an explicit alignment is specified, the
809 global is forced to have at least that much alignment. All alignments must be
812 <p>For example, the following defines a global in a numbered address space with
813 an initializer, section, and alignment:</p>
815 <div class="doc_code">
817 @G = addrspace(5) constant float 1.0, section "foo", align 4
824 <!-- ======================================================================= -->
825 <div class="doc_subsection">
826 <a name="functionstructure">Functions</a>
829 <div class="doc_text">
831 <p>LLVM function definitions consist of the "<tt>define</tt>" keyord,
832 an optional <a href="#linkage">linkage type</a>, an optional
833 <a href="#visibility">visibility style</a>, an optional
834 <a href="#callingconv">calling convention</a>, a return type, an optional
835 <a href="#paramattrs">parameter attribute</a> for the return type, a function
836 name, a (possibly empty) argument list (each with optional
837 <a href="#paramattrs">parameter attributes</a>), optional
838 <a href="#fnattrs">function attributes</a>, an optional section,
839 an optional alignment, an optional <a href="#gc">garbage collector name</a>,
840 an opening curly brace, a list of basic blocks, and a closing curly brace.
842 LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
843 optional <a href="#linkage">linkage type</a>, an optional
844 <a href="#visibility">visibility style</a>, an optional
845 <a href="#callingconv">calling convention</a>, a return type, an optional
846 <a href="#paramattrs">parameter attribute</a> for the return type, a function
847 name, a possibly empty list of arguments, an optional alignment, and an optional
848 <a href="#gc">garbage collector name</a>.</p>
850 <p>A function definition contains a list of basic blocks, forming the CFG
851 (Control Flow Graph) for
852 the function. Each basic block may optionally start with a label (giving the
853 basic block a symbol table entry), contains a list of instructions, and ends
854 with a <a href="#terminators">terminator</a> instruction (such as a branch or
855 function return).</p>
857 <p>The first basic block in a function is special in two ways: it is immediately
858 executed on entrance to the function, and it is not allowed to have predecessor
859 basic blocks (i.e. there can not be any branches to the entry block of a
860 function). Because the block can have no predecessors, it also cannot have any
861 <a href="#i_phi">PHI nodes</a>.</p>
863 <p>LLVM allows an explicit section to be specified for functions. If the target
864 supports it, it will emit functions to the section specified.</p>
866 <p>An explicit alignment may be specified for a function. If not present, or if
867 the alignment is set to zero, the alignment of the function is set by the target
868 to whatever it feels convenient. If an explicit alignment is specified, the
869 function is forced to have at least that much alignment. All alignments must be
874 <div class="doc_code">
876 define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
877 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
878 <ResultType> @<FunctionName> ([argument list])
879 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
880 [<a href="#gc">gc</a>] { ... }
887 <!-- ======================================================================= -->
888 <div class="doc_subsection">
889 <a name="aliasstructure">Aliases</a>
891 <div class="doc_text">
892 <p>Aliases act as "second name" for the aliasee value (which can be either
893 function, global variable, another alias or bitcast of global value). Aliases
894 may have an optional <a href="#linkage">linkage type</a>, and an
895 optional <a href="#visibility">visibility style</a>.</p>
899 <div class="doc_code">
901 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
909 <!-- ======================================================================= -->
910 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
911 <div class="doc_text">
912 <p>The return type and each parameter of a function type may have a set of
913 <i>parameter attributes</i> associated with them. Parameter attributes are
914 used to communicate additional information about the result or parameters of
915 a function. Parameter attributes are considered to be part of the function,
916 not of the function type, so functions with different parameter attributes
917 can have the same function type.</p>
919 <p>Parameter attributes are simple keywords that follow the type specified. If
920 multiple parameter attributes are needed, they are space separated. For
923 <div class="doc_code">
925 declare i32 @printf(i8* noalias nocapture, ...)
926 declare i32 @atoi(i8 zeroext)
927 declare signext i8 @returns_signed_char()
931 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
932 <tt>readonly</tt>) come immediately after the argument list.</p>
934 <p>Currently, only the following parameter attributes are defined:</p>
936 <dt><tt>zeroext</tt></dt>
937 <dd>This indicates to the code generator that the parameter or return value
938 should be zero-extended to a 32-bit value by the caller (for a parameter)
939 or the callee (for a return value).</dd>
941 <dt><tt>signext</tt></dt>
942 <dd>This indicates to the code generator that the parameter or return value
943 should be sign-extended to a 32-bit value by the caller (for a parameter)
944 or the callee (for a return value).</dd>
946 <dt><tt>inreg</tt></dt>
947 <dd>This indicates that this parameter or return value should be treated
948 in a special target-dependent fashion during while emitting code for a
949 function call or return (usually, by putting it in a register as opposed
950 to memory, though some targets use it to distinguish between two different
951 kinds of registers). Use of this attribute is target-specific.</dd>
953 <dt><tt><a name="byval">byval</a></tt></dt>
954 <dd>This indicates that the pointer parameter should really be passed by
955 value to the function. The attribute implies that a hidden copy of the
956 pointee is made between the caller and the callee, so the callee is unable
957 to modify the value in the callee. This attribute is only valid on LLVM
958 pointer arguments. It is generally used to pass structs and arrays by
959 value, but is also valid on pointers to scalars. The copy is considered to
960 belong to the caller not the callee (for example,
961 <tt><a href="#readonly">readonly</a></tt> functions should not write to
962 <tt>byval</tt> parameters). This is not a valid attribute for return
963 values. The byval attribute also supports specifying an alignment with the
964 align attribute. This has a target-specific effect on the code generator
965 that usually indicates a desired alignment for the synthesized stack
968 <dt><tt>sret</tt></dt>
969 <dd>This indicates that the pointer parameter specifies the address of a
970 structure that is the return value of the function in the source program.
971 This pointer must be guaranteed by the caller to be valid: loads and stores
972 to the structure may be assumed by the callee to not to trap. This may only
973 be applied to the first parameter. This is not a valid attribute for
976 <dt><tt>noalias</tt></dt>
977 <dd>This indicates that the pointer does not alias any global or any other
978 parameter. The caller is responsible for ensuring that this is the
979 case. On a function return value, <tt>noalias</tt> additionally indicates
980 that the pointer does not alias any other pointers visible to the
981 caller. For further details, please see the discussion of the NoAlias
983 <a href="http://llvm.org/docs/AliasAnalysis.html#MustMayNo">alias
986 <dt><tt>nocapture</tt></dt>
987 <dd>This indicates that the callee does not make any copies of the pointer
988 that outlive the callee itself. This is not a valid attribute for return
991 <dt><tt>nest</tt></dt>
992 <dd>This indicates that the pointer parameter can be excised using the
993 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
994 attribute for return values.</dd>
999 <!-- ======================================================================= -->
1000 <div class="doc_subsection">
1001 <a name="gc">Garbage Collector Names</a>
1004 <div class="doc_text">
1005 <p>Each function may specify a garbage collector name, which is simply a
1008 <div class="doc_code"><pre
1009 >define void @f() gc "name" { ...</pre></div>
1011 <p>The compiler declares the supported values of <i>name</i>. Specifying a
1012 collector which will cause the compiler to alter its output in order to support
1013 the named garbage collection algorithm.</p>
1016 <!-- ======================================================================= -->
1017 <div class="doc_subsection">
1018 <a name="fnattrs">Function Attributes</a>
1021 <div class="doc_text">
1023 <p>Function attributes are set to communicate additional information about
1024 a function. Function attributes are considered to be part of the function,
1025 not of the function type, so functions with different parameter attributes
1026 can have the same function type.</p>
1028 <p>Function attributes are simple keywords that follow the type specified. If
1029 multiple attributes are needed, they are space separated. For
1032 <div class="doc_code">
1034 define void @f() noinline { ... }
1035 define void @f() alwaysinline { ... }
1036 define void @f() alwaysinline optsize { ... }
1037 define void @f() optsize
1042 <dt><tt>alwaysinline</tt></dt>
1043 <dd>This attribute indicates that the inliner should attempt to inline this
1044 function into callers whenever possible, ignoring any active inlining size
1045 threshold for this caller.</dd>
1047 <dt><tt>noinline</tt></dt>
1048 <dd>This attribute indicates that the inliner should never inline this function
1049 in any situation. This attribute may not be used together with the
1050 <tt>alwaysinline</tt> attribute.</dd>
1052 <dt><tt>optsize</tt></dt>
1053 <dd>This attribute suggests that optimization passes and code generator passes
1054 make choices that keep the code size of this function low, and otherwise do
1055 optimizations specifically to reduce code size.</dd>
1057 <dt><tt>noreturn</tt></dt>
1058 <dd>This function attribute indicates that the function never returns normally.
1059 This produces undefined behavior at runtime if the function ever does
1060 dynamically return.</dd>
1062 <dt><tt>nounwind</tt></dt>
1063 <dd>This function attribute indicates that the function never returns with an
1064 unwind or exceptional control flow. If the function does unwind, its runtime
1065 behavior is undefined.</dd>
1067 <dt><tt>readnone</tt></dt>
1068 <dd>This attribute indicates that the function computes its result (or decides to
1069 unwind an exception) based strictly on its arguments, without dereferencing any
1070 pointer arguments or otherwise accessing any mutable state (e.g. memory, control
1071 registers, etc) visible to caller functions. It does not write through any
1072 pointer arguments (including <tt><a href="#byval">byval</a></tt> arguments) and
1073 never changes any state visible to callers. This means that it cannot unwind
1074 exceptions by calling the <tt>C++</tt> exception throwing methods, but could
1075 use the <tt>unwind</tt> instruction.</dd>
1077 <dt><tt><a name="readonly">readonly</a></tt></dt>
1078 <dd>This attribute indicates that the function does not write through any
1079 pointer arguments (including <tt><a href="#byval">byval</a></tt> arguments)
1080 or otherwise modify any state (e.g. memory, control registers, etc) visible to
1081 caller functions. It may dereference pointer arguments and read state that may
1082 be set in the caller. A readonly function always returns the same value (or
1083 unwinds an exception identically) when called with the same set of arguments
1084 and global state. It cannot unwind an exception by calling the <tt>C++</tt>
1085 exception throwing methods, but may use the <tt>unwind</tt> instruction.</dd>
1087 <dt><tt><a name="ssp">ssp</a></tt></dt>
1088 <dd>This attribute indicates that the function should emit a stack smashing
1089 protector. It is in the form of a "canary"—a random value placed on the
1090 stack before the local variables that's checked upon return from the function to
1091 see if it has been overwritten. A heuristic is used to determine if a function
1092 needs stack protectors or not.
1094 <br><br>If a function that has an <tt>ssp</tt> attribute is inlined into a function
1095 that doesn't have an <tt>ssp</tt> attribute, then the resulting function will
1096 have an <tt>ssp</tt> attribute.</dd>
1098 <dt><tt>sspreq</tt></dt>
1099 <dd>This attribute indicates that the function should <em>always</em> emit a
1100 stack smashing protector. This overrides the <tt><a href="#ssp">ssp</a></tt>
1103 If a function that has an <tt>sspreq</tt> attribute is inlined into a
1104 function that doesn't have an <tt>sspreq</tt> attribute or which has
1105 an <tt>ssp</tt> attribute, then the resulting function will have
1106 an <tt>sspreq</tt> attribute.</dd>
1108 <dt><tt>noredzone</tt></dt>
1109 <dd>This attribute indicates that the code generator should not use a
1110 red zone, even if the target-specific ABI normally permits it.
1113 <dt><tt>noimplicitfloat</tt></dt>
1114 <dd>This attributes disables implicit floating point instructions.</dd>
1120 <!-- ======================================================================= -->
1121 <div class="doc_subsection">
1122 <a name="moduleasm">Module-Level Inline Assembly</a>
1125 <div class="doc_text">
1127 Modules may contain "module-level inline asm" blocks, which corresponds to the
1128 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
1129 LLVM and treated as a single unit, but may be separated in the .ll file if
1130 desired. The syntax is very simple:
1133 <div class="doc_code">
1135 module asm "inline asm code goes here"
1136 module asm "more can go here"
1140 <p>The strings can contain any character by escaping non-printable characters.
1141 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1146 The inline asm code is simply printed to the machine code .s file when
1147 assembly code is generated.
1151 <!-- ======================================================================= -->
1152 <div class="doc_subsection">
1153 <a name="datalayout">Data Layout</a>
1156 <div class="doc_text">
1157 <p>A module may specify a target specific data layout string that specifies how
1158 data is to be laid out in memory. The syntax for the data layout is simply:</p>
1159 <pre> target datalayout = "<i>layout specification</i>"</pre>
1160 <p>The <i>layout specification</i> consists of a list of specifications
1161 separated by the minus sign character ('-'). Each specification starts with a
1162 letter and may include other information after the letter to define some
1163 aspect of the data layout. The specifications accepted are as follows: </p>
1166 <dd>Specifies that the target lays out data in big-endian form. That is, the
1167 bits with the most significance have the lowest address location.</dd>
1169 <dd>Specifies that the target lays out data in little-endian form. That is,
1170 the bits with the least significance have the lowest address location.</dd>
1171 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1172 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1173 <i>preferred</i> alignments. All sizes are in bits. Specifying the <i>pref</i>
1174 alignment is optional. If omitted, the preceding <tt>:</tt> should be omitted
1176 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1177 <dd>This specifies the alignment for an integer type of a given bit
1178 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1179 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1180 <dd>This specifies the alignment for a vector type of a given bit
1182 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1183 <dd>This specifies the alignment for a floating point type of a given bit
1184 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
1186 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1187 <dd>This specifies the alignment for an aggregate type of a given bit
1189 <dt><tt>s<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1190 <dd>This specifies the alignment for a stack object of a given bit
1193 <p>When constructing the data layout for a given target, LLVM starts with a
1194 default set of specifications which are then (possibly) overriden by the
1195 specifications in the <tt>datalayout</tt> keyword. The default specifications
1196 are given in this list:</p>
1198 <li><tt>E</tt> - big endian</li>
1199 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
1200 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1201 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1202 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1203 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1204 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1205 alignment of 64-bits</li>
1206 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1207 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1208 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1209 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1210 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1211 <li><tt>s0:64:64</tt> - stack objects are 64-bit aligned</li>
1213 <p>When LLVM is determining the alignment for a given type, it uses the
1214 following rules:</p>
1216 <li>If the type sought is an exact match for one of the specifications, that
1217 specification is used.</li>
1218 <li>If no match is found, and the type sought is an integer type, then the
1219 smallest integer type that is larger than the bitwidth of the sought type is
1220 used. If none of the specifications are larger than the bitwidth then the the
1221 largest integer type is used. For example, given the default specifications
1222 above, the i7 type will use the alignment of i8 (next largest) while both
1223 i65 and i256 will use the alignment of i64 (largest specified).</li>
1224 <li>If no match is found, and the type sought is a vector type, then the
1225 largest vector type that is smaller than the sought vector type will be used
1226 as a fall back. This happens because <128 x double> can be implemented
1227 in terms of 64 <2 x double>, for example.</li>
1231 <!-- *********************************************************************** -->
1232 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
1233 <!-- *********************************************************************** -->
1235 <div class="doc_text">
1237 <p>The LLVM type system is one of the most important features of the
1238 intermediate representation. Being typed enables a number of
1239 optimizations to be performed on the intermediate representation directly,
1240 without having to do
1241 extra analyses on the side before the transformation. A strong type
1242 system makes it easier to read the generated code and enables novel
1243 analyses and transformations that are not feasible to perform on normal
1244 three address code representations.</p>
1248 <!-- ======================================================================= -->
1249 <div class="doc_subsection"> <a name="t_classifications">Type
1250 Classifications</a> </div>
1251 <div class="doc_text">
1252 <p>The types fall into a few useful
1253 classifications:</p>
1255 <table border="1" cellspacing="0" cellpadding="4">
1257 <tr><th>Classification</th><th>Types</th></tr>
1259 <td><a href="#t_integer">integer</a></td>
1260 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1263 <td><a href="#t_floating">floating point</a></td>
1264 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1267 <td><a name="t_firstclass">first class</a></td>
1268 <td><a href="#t_integer">integer</a>,
1269 <a href="#t_floating">floating point</a>,
1270 <a href="#t_pointer">pointer</a>,
1271 <a href="#t_vector">vector</a>,
1272 <a href="#t_struct">structure</a>,
1273 <a href="#t_array">array</a>,
1274 <a href="#t_label">label</a>,
1275 <a href="#t_metadata">metadata</a>.
1279 <td><a href="#t_primitive">primitive</a></td>
1280 <td><a href="#t_label">label</a>,
1281 <a href="#t_void">void</a>,
1282 <a href="#t_floating">floating point</a>,
1283 <a href="#t_metadata">metadata</a>.</td>
1286 <td><a href="#t_derived">derived</a></td>
1287 <td><a href="#t_integer">integer</a>,
1288 <a href="#t_array">array</a>,
1289 <a href="#t_function">function</a>,
1290 <a href="#t_pointer">pointer</a>,
1291 <a href="#t_struct">structure</a>,
1292 <a href="#t_pstruct">packed structure</a>,
1293 <a href="#t_vector">vector</a>,
1294 <a href="#t_opaque">opaque</a>.
1300 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
1301 most important. Values of these types are the only ones which can be
1302 produced by instructions, passed as arguments, or used as operands to
1306 <!-- ======================================================================= -->
1307 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
1309 <div class="doc_text">
1310 <p>The primitive types are the fundamental building blocks of the LLVM
1315 <!-- _______________________________________________________________________ -->
1316 <div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1318 <div class="doc_text">
1321 <tr><th>Type</th><th>Description</th></tr>
1322 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1323 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1324 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1325 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1326 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1331 <!-- _______________________________________________________________________ -->
1332 <div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1334 <div class="doc_text">
1336 <p>The void type does not represent any value and has no size.</p>
1345 <!-- _______________________________________________________________________ -->
1346 <div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1348 <div class="doc_text">
1350 <p>The label type represents code labels.</p>
1359 <!-- _______________________________________________________________________ -->
1360 <div class="doc_subsubsection"> <a name="t_metadata">Metadata Type</a> </div>
1362 <div class="doc_text">
1364 <p>The metadata type represents embedded metadata. The only derived type that
1365 may contain metadata is <tt>metadata*</tt> or a function type that returns or
1366 takes metadata typed parameters, but not pointer to metadata types.</p>
1376 <!-- ======================================================================= -->
1377 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1379 <div class="doc_text">
1381 <p>The real power in LLVM comes from the derived types in the system.
1382 This is what allows a programmer to represent arrays, functions,
1383 pointers, and other useful types. Note that these derived types may be
1384 recursive: For example, it is possible to have a two dimensional array.</p>
1388 <!-- _______________________________________________________________________ -->
1389 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1391 <div class="doc_text">
1394 <p>The integer type is a very simple derived type that simply specifies an
1395 arbitrary bit width for the integer type desired. Any bit width from 1 bit to
1396 2^23-1 (about 8 million) can be specified.</p>
1404 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1408 <table class="layout">
1410 <td class="left"><tt>i1</tt></td>
1411 <td class="left">a single-bit integer.</td>
1414 <td class="left"><tt>i32</tt></td>
1415 <td class="left">a 32-bit integer.</td>
1418 <td class="left"><tt>i1942652</tt></td>
1419 <td class="left">a really big integer of over 1 million bits.</td>
1423 <p>Note that the code generator does not yet support large integer types
1424 to be used as function return types. The specific limit on how large a
1425 return type the code generator can currently handle is target-dependent;
1426 currently it's often 64 bits for 32-bit targets and 128 bits for 64-bit
1431 <!-- _______________________________________________________________________ -->
1432 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1434 <div class="doc_text">
1438 <p>The array type is a very simple derived type that arranges elements
1439 sequentially in memory. The array type requires a size (number of
1440 elements) and an underlying data type.</p>
1445 [<# elements> x <elementtype>]
1448 <p>The number of elements is a constant integer value; elementtype may
1449 be any type with a size.</p>
1452 <table class="layout">
1454 <td class="left"><tt>[40 x i32]</tt></td>
1455 <td class="left">Array of 40 32-bit integer values.</td>
1458 <td class="left"><tt>[41 x i32]</tt></td>
1459 <td class="left">Array of 41 32-bit integer values.</td>
1462 <td class="left"><tt>[4 x i8]</tt></td>
1463 <td class="left">Array of 4 8-bit integer values.</td>
1466 <p>Here are some examples of multidimensional arrays:</p>
1467 <table class="layout">
1469 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1470 <td class="left">3x4 array of 32-bit integer values.</td>
1473 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1474 <td class="left">12x10 array of single precision floating point values.</td>
1477 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1478 <td class="left">2x3x4 array of 16-bit integer values.</td>
1482 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1483 length array. Normally, accesses past the end of an array are undefined in
1484 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
1485 As a special case, however, zero length arrays are recognized to be variable
1486 length. This allows implementation of 'pascal style arrays' with the LLVM
1487 type "{ i32, [0 x float]}", for example.</p>
1489 <p>Note that the code generator does not yet support large aggregate types
1490 to be used as function return types. The specific limit on how large an
1491 aggregate return type the code generator can currently handle is
1492 target-dependent, and also dependent on the aggregate element types.</p>
1496 <!-- _______________________________________________________________________ -->
1497 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1498 <div class="doc_text">
1502 <p>The function type can be thought of as a function signature. It
1503 consists of a return type and a list of formal parameter types. The
1504 return type of a function type is a scalar type, a void type, or a struct type.
1505 If the return type is a struct type then all struct elements must be of first
1506 class types, and the struct must have at least one element.</p>
1511 <returntype list> (<parameter list>)
1514 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1515 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1516 which indicates that the function takes a variable number of arguments.
1517 Variable argument functions can access their arguments with the <a
1518 href="#int_varargs">variable argument handling intrinsic</a> functions.
1519 '<tt><returntype list></tt>' is a comma-separated list of
1520 <a href="#t_firstclass">first class</a> type specifiers.</p>
1523 <table class="layout">
1525 <td class="left"><tt>i32 (i32)</tt></td>
1526 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1528 </tr><tr class="layout">
1529 <td class="left"><tt>float (i16 signext, i32 *) *
1531 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1532 an <tt>i16</tt> that should be sign extended and a
1533 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1536 </tr><tr class="layout">
1537 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1538 <td class="left">A vararg function that takes at least one
1539 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1540 which returns an integer. This is the signature for <tt>printf</tt> in
1543 </tr><tr class="layout">
1544 <td class="left"><tt>{i32, i32} (i32)</tt></td>
1545 <td class="left">A function taking an <tt>i32</tt>, returning two
1546 <tt>i32</tt> values as an aggregate of type <tt>{ i32, i32 }</tt>
1552 <!-- _______________________________________________________________________ -->
1553 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1554 <div class="doc_text">
1556 <p>The structure type is used to represent a collection of data members
1557 together in memory. The packing of the field types is defined to match
1558 the ABI of the underlying processor. The elements of a structure may
1559 be any type that has a size.</p>
1560 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1561 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1562 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1565 <pre> { <type list> }<br></pre>
1567 <table class="layout">
1569 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1570 <td class="left">A triple of three <tt>i32</tt> values</td>
1571 </tr><tr class="layout">
1572 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1573 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1574 second element is a <a href="#t_pointer">pointer</a> to a
1575 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1576 an <tt>i32</tt>.</td>
1580 <p>Note that the code generator does not yet support large aggregate types
1581 to be used as function return types. The specific limit on how large an
1582 aggregate return type the code generator can currently handle is
1583 target-dependent, and also dependent on the aggregate element types.</p>
1587 <!-- _______________________________________________________________________ -->
1588 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1590 <div class="doc_text">
1592 <p>The packed structure type is used to represent a collection of data members
1593 together in memory. There is no padding between fields. Further, the alignment
1594 of a packed structure is 1 byte. The elements of a packed structure may
1595 be any type that has a size.</p>
1596 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1597 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1598 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1601 <pre> < { <type list> } > <br></pre>
1603 <table class="layout">
1605 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1606 <td class="left">A triple of three <tt>i32</tt> values</td>
1607 </tr><tr class="layout">
1609 <tt>< { float, i32 (i32)* } ></tt></td>
1610 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1611 second element is a <a href="#t_pointer">pointer</a> to a
1612 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1613 an <tt>i32</tt>.</td>
1618 <!-- _______________________________________________________________________ -->
1619 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1620 <div class="doc_text">
1622 <p>As in many languages, the pointer type represents a pointer or
1623 reference to another object, which must live in memory. Pointer types may have
1624 an optional address space attribute defining the target-specific numbered
1625 address space where the pointed-to object resides. The default address space is
1628 <p>Note that LLVM does not permit pointers to void (<tt>void*</tt>) nor does
1629 it permit pointers to labels (<tt>label*</tt>). Use <tt>i8*</tt> instead.</p>
1632 <pre> <type> *<br></pre>
1634 <table class="layout">
1636 <td class="left"><tt>[4 x i32]*</tt></td>
1637 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1638 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1641 <td class="left"><tt>i32 (i32 *) *</tt></td>
1642 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
1643 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1647 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1648 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1649 that resides in address space #5.</td>
1654 <!-- _______________________________________________________________________ -->
1655 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1656 <div class="doc_text">
1660 <p>A vector type is a simple derived type that represents a vector
1661 of elements. Vector types are used when multiple primitive data
1662 are operated in parallel using a single instruction (SIMD).
1663 A vector type requires a size (number of
1664 elements) and an underlying primitive data type. Vectors must have a power
1665 of two length (1, 2, 4, 8, 16 ...). Vector types are
1666 considered <a href="#t_firstclass">first class</a>.</p>
1671 < <# elements> x <elementtype> >
1674 <p>The number of elements is a constant integer value; elementtype may
1675 be any integer or floating point type.</p>
1679 <table class="layout">
1681 <td class="left"><tt><4 x i32></tt></td>
1682 <td class="left">Vector of 4 32-bit integer values.</td>
1685 <td class="left"><tt><8 x float></tt></td>
1686 <td class="left">Vector of 8 32-bit floating-point values.</td>
1689 <td class="left"><tt><2 x i64></tt></td>
1690 <td class="left">Vector of 2 64-bit integer values.</td>
1694 <p>Note that the code generator does not yet support large vector types
1695 to be used as function return types. The specific limit on how large a
1696 vector return type codegen can currently handle is target-dependent;
1697 currently it's often a few times longer than a hardware vector register.</p>
1701 <!-- _______________________________________________________________________ -->
1702 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1703 <div class="doc_text">
1707 <p>Opaque types are used to represent unknown types in the system. This
1708 corresponds (for example) to the C notion of a forward declared structure type.
1709 In LLVM, opaque types can eventually be resolved to any type (not just a
1710 structure type).</p>
1720 <table class="layout">
1722 <td class="left"><tt>opaque</tt></td>
1723 <td class="left">An opaque type.</td>
1728 <!-- ======================================================================= -->
1729 <div class="doc_subsection">
1730 <a name="t_uprefs">Type Up-references</a>
1733 <div class="doc_text">
1736 An "up reference" allows you to refer to a lexically enclosing type without
1737 requiring it to have a name. For instance, a structure declaration may contain a
1738 pointer to any of the types it is lexically a member of. Example of up
1739 references (with their equivalent as named type declarations) include:</p>
1742 { \2 * } %x = type { %x* }
1743 { \2 }* %y = type { %y }*
1748 An up reference is needed by the asmprinter for printing out cyclic types when
1749 there is no declared name for a type in the cycle. Because the asmprinter does
1750 not want to print out an infinite type string, it needs a syntax to handle
1751 recursive types that have no names (all names are optional in llvm IR).
1760 The level is the count of the lexical type that is being referred to.
1765 <table class="layout">
1767 <td class="left"><tt>\1*</tt></td>
1768 <td class="left">Self-referential pointer.</td>
1771 <td class="left"><tt>{ { \3*, i8 }, i32 }</tt></td>
1772 <td class="left">Recursive structure where the upref refers to the out-most
1779 <!-- *********************************************************************** -->
1780 <div class="doc_section"> <a name="constants">Constants</a> </div>
1781 <!-- *********************************************************************** -->
1783 <div class="doc_text">
1785 <p>LLVM has several different basic types of constants. This section describes
1786 them all and their syntax.</p>
1790 <!-- ======================================================================= -->
1791 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1793 <div class="doc_text">
1796 <dt><b>Boolean constants</b></dt>
1798 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1799 constants of the <tt><a href="#t_primitive">i1</a></tt> type.
1802 <dt><b>Integer constants</b></dt>
1804 <dd>Standard integers (such as '4') are constants of the <a
1805 href="#t_integer">integer</a> type. Negative numbers may be used with
1809 <dt><b>Floating point constants</b></dt>
1811 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1812 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1813 notation (see below). The assembler requires the exact decimal value of
1814 a floating-point constant. For example, the assembler accepts 1.25 but
1815 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
1816 constants must have a <a href="#t_floating">floating point</a> type. </dd>
1818 <dt><b>Null pointer constants</b></dt>
1820 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1821 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1825 <p>The one non-intuitive notation for constants is the hexadecimal form
1826 of floating point constants. For example, the form '<tt>double
1827 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1828 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1829 (and the only time that they are generated by the disassembler) is when a
1830 floating point constant must be emitted but it cannot be represented as a
1831 decimal floating point number in a reasonable number of digits. For example,
1832 NaN's, infinities, and other
1833 special values are represented in their IEEE hexadecimal format so that
1834 assembly and disassembly do not cause any bits to change in the constants.</p>
1835 <p>When using the hexadecimal form, constants of types float and double are
1836 represented using the 16-digit form shown above (which matches the IEEE754
1837 representation for double); float values must, however, be exactly representable
1838 as IEE754 single precision.
1839 Hexadecimal format is always used for long
1840 double, and there are three forms of long double. The 80-bit
1841 format used by x86 is represented as <tt>0xK</tt>
1842 followed by 20 hexadecimal digits.
1843 The 128-bit format used by PowerPC (two adjacent doubles) is represented
1844 by <tt>0xM</tt> followed by 32 hexadecimal digits. The IEEE 128-bit
1845 format is represented
1846 by <tt>0xL</tt> followed by 32 hexadecimal digits; no currently supported
1847 target uses this format. Long doubles will only work if they match
1848 the long double format on your target. All hexadecimal formats are big-endian
1849 (sign bit at the left).</p>
1852 <!-- ======================================================================= -->
1853 <div class="doc_subsection">
1854 <a name="aggregateconstants"> <!-- old anchor -->
1855 <a name="complexconstants">Complex Constants</a></a>
1858 <div class="doc_text">
1859 <p>Complex constants are a (potentially recursive) combination of simple
1860 constants and smaller complex constants.</p>
1863 <dt><b>Structure constants</b></dt>
1865 <dd>Structure constants are represented with notation similar to structure
1866 type definitions (a comma separated list of elements, surrounded by braces
1867 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
1868 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>". Structure constants
1869 must have <a href="#t_struct">structure type</a>, and the number and
1870 types of elements must match those specified by the type.
1873 <dt><b>Array constants</b></dt>
1875 <dd>Array constants are represented with notation similar to array type
1876 definitions (a comma separated list of elements, surrounded by square brackets
1877 (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74 ]</tt>". Array
1878 constants must have <a href="#t_array">array type</a>, and the number and
1879 types of elements must match those specified by the type.
1882 <dt><b>Vector constants</b></dt>
1884 <dd>Vector constants are represented with notation similar to vector type
1885 definitions (a comma separated list of elements, surrounded by
1886 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32 42,
1887 i32 11, i32 74, i32 100 ></tt>". Vector constants must have <a
1888 href="#t_vector">vector type</a>, and the number and types of elements must
1889 match those specified by the type.
1892 <dt><b>Zero initialization</b></dt>
1894 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1895 value to zero of <em>any</em> type, including scalar and aggregate types.
1896 This is often used to avoid having to print large zero initializers (e.g. for
1897 large arrays) and is always exactly equivalent to using explicit zero
1901 <dt><b>Metadata node</b></dt>
1903 <dd>A metadata node is a structure-like constant with
1904 <a href="#t_metadata">metadata type</a>. For example:
1905 "<tt>metadata !{ i32 0, metadata !"test" }</tt>". Unlike other constants
1906 that are meant to be interpreted as part of the instruction stream, metadata
1907 is a place to attach additional information such as debug info.
1913 <!-- ======================================================================= -->
1914 <div class="doc_subsection">
1915 <a name="globalconstants">Global Variable and Function Addresses</a>
1918 <div class="doc_text">
1920 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1921 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1922 constants. These constants are explicitly referenced when the <a
1923 href="#identifiers">identifier for the global</a> is used and always have <a
1924 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1927 <div class="doc_code">
1931 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1937 <!-- ======================================================================= -->
1938 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1939 <div class="doc_text">
1940 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1941 no specific value. Undefined values may be of any type and be used anywhere
1942 a constant is permitted.</p>
1944 <p>Undefined values indicate to the compiler that the program is well defined
1945 no matter what value is used, giving the compiler more freedom to optimize.
1949 <!-- ======================================================================= -->
1950 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1953 <div class="doc_text">
1955 <p>Constant expressions are used to allow expressions involving other constants
1956 to be used as constants. Constant expressions may be of any <a
1957 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1958 that does not have side effects (e.g. load and call are not supported). The
1959 following is the syntax for constant expressions:</p>
1962 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1963 <dd>Truncate a constant to another type. The bit size of CST must be larger
1964 than the bit size of TYPE. Both types must be integers.</dd>
1966 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1967 <dd>Zero extend a constant to another type. The bit size of CST must be
1968 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1970 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1971 <dd>Sign extend a constant to another type. The bit size of CST must be
1972 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1974 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1975 <dd>Truncate a floating point constant to another floating point type. The
1976 size of CST must be larger than the size of TYPE. Both types must be
1977 floating point.</dd>
1979 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1980 <dd>Floating point extend a constant to another type. The size of CST must be
1981 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1983 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
1984 <dd>Convert a floating point constant to the corresponding unsigned integer
1985 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1986 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1987 of the same number of elements. If the value won't fit in the integer type,
1988 the results are undefined.</dd>
1990 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
1991 <dd>Convert a floating point constant to the corresponding signed integer
1992 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1993 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1994 of the same number of elements. If the value won't fit in the integer type,
1995 the results are undefined.</dd>
1997 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
1998 <dd>Convert an unsigned integer constant to the corresponding floating point
1999 constant. TYPE must be a scalar or vector floating point type. CST must be of
2000 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
2001 of the same number of elements. If the value won't fit in the floating point
2002 type, the results are undefined.</dd>
2004 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
2005 <dd>Convert a signed integer constant to the corresponding floating point
2006 constant. TYPE must be a scalar or vector floating point type. CST must be of
2007 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
2008 of the same number of elements. If the value won't fit in the floating point
2009 type, the results are undefined.</dd>
2011 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
2012 <dd>Convert a pointer typed constant to the corresponding integer constant
2013 TYPE must be an integer type. CST must be of pointer type. The CST value is
2014 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
2016 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
2017 <dd>Convert a integer constant to a pointer constant. TYPE must be a
2018 pointer type. CST must be of integer type. The CST value is zero extended,
2019 truncated, or unchanged to make it fit in a pointer size. This one is
2020 <i>really</i> dangerous!</dd>
2022 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
2023 <dd>Convert a constant, CST, to another TYPE. The constraints of the operands
2024 are the same as those for the <a href="#i_bitcast">bitcast
2025 instruction</a>.</dd>
2027 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
2029 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
2030 constants. As with the <a href="#i_getelementptr">getelementptr</a>
2031 instruction, the index list may have zero or more indexes, which are required
2032 to make sense for the type of "CSTPTR".</dd>
2034 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
2036 <dd>Perform the <a href="#i_select">select operation</a> on
2039 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
2040 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
2042 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
2043 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
2045 <dt><b><tt>vicmp COND ( VAL1, VAL2 )</tt></b></dt>
2046 <dd>Performs the <a href="#i_vicmp">vicmp operation</a> on constants.</dd>
2048 <dt><b><tt>vfcmp COND ( VAL1, VAL2 )</tt></b></dt>
2049 <dd>Performs the <a href="#i_vfcmp">vfcmp operation</a> on constants.</dd>
2051 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
2053 <dd>Perform the <a href="#i_extractelement">extractelement
2054 operation</a> on constants.</dd>
2056 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
2058 <dd>Perform the <a href="#i_insertelement">insertelement
2059 operation</a> on constants.</dd>
2062 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
2064 <dd>Perform the <a href="#i_shufflevector">shufflevector
2065 operation</a> on constants.</dd>
2067 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
2069 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
2070 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
2071 binary</a> operations. The constraints on operands are the same as those for
2072 the corresponding instruction (e.g. no bitwise operations on floating point
2073 values are allowed).</dd>
2077 <!-- ======================================================================= -->
2078 <div class="doc_subsection"><a name="metadata">Embedded Metadata</a>
2081 <div class="doc_text">
2083 <p>Embedded metadata provides a way to attach arbitrary data to the
2084 instruction stream without affecting the behaviour of the program. There are
2085 two metadata primitives, strings and nodes. All metadata has the
2086 <tt>metadata</tt> type and is identified in syntax by a preceding exclamation
2087 point ('<tt>!</tt>').
2090 <p>A metadata string is a string surrounded by double quotes. It can contain
2091 any character by escaping non-printable characters with "\xx" where "xx" is
2092 the two digit hex code. For example: "<tt>!"test\00"</tt>".
2095 <p>Metadata nodes are represented with notation similar to structure constants
2096 (a comma separated list of elements, surrounded by braces and preceeded by an
2097 exclamation point). For example: "<tt>!{ metadata !"test\00", i32 10}</tt>".
2100 <p>A metadata node will attempt to track changes to the values it holds. In
2101 the event that a value is deleted, it will be replaced with a typeless
2102 "<tt>null</tt>", such as "<tt>metadata !{null, i32 10}</tt>".</p>
2104 <p>Optimizations may rely on metadata to provide additional information about
2105 the program that isn't available in the instructions, or that isn't easily
2106 computable. Similarly, the code generator may expect a certain metadata format
2107 to be used to express debugging information.</p>
2110 <!-- *********************************************************************** -->
2111 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
2112 <!-- *********************************************************************** -->
2114 <!-- ======================================================================= -->
2115 <div class="doc_subsection">
2116 <a name="inlineasm">Inline Assembler Expressions</a>
2119 <div class="doc_text">
2122 LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
2123 Module-Level Inline Assembly</a>) through the use of a special value. This
2124 value represents the inline assembler as a string (containing the instructions
2125 to emit), a list of operand constraints (stored as a string), and a flag that
2126 indicates whether or not the inline asm expression has side effects. An example
2127 inline assembler expression is:
2130 <div class="doc_code">
2132 i32 (i32) asm "bswap $0", "=r,r"
2137 Inline assembler expressions may <b>only</b> be used as the callee operand of
2138 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
2141 <div class="doc_code">
2143 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
2148 Inline asms with side effects not visible in the constraint list must be marked
2149 as having side effects. This is done through the use of the
2150 '<tt>sideeffect</tt>' keyword, like so:
2153 <div class="doc_code">
2155 call void asm sideeffect "eieio", ""()
2159 <p>TODO: The format of the asm and constraints string still need to be
2160 documented here. Constraints on what can be done (e.g. duplication, moving, etc
2161 need to be documented). This is probably best done by reference to another
2162 document that covers inline asm from a holistic perspective.
2167 <!-- *********************************************************************** -->
2168 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
2169 <!-- *********************************************************************** -->
2171 <div class="doc_text">
2173 <p>The LLVM instruction set consists of several different
2174 classifications of instructions: <a href="#terminators">terminator
2175 instructions</a>, <a href="#binaryops">binary instructions</a>,
2176 <a href="#bitwiseops">bitwise binary instructions</a>, <a
2177 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
2178 instructions</a>.</p>
2182 <!-- ======================================================================= -->
2183 <div class="doc_subsection"> <a name="terminators">Terminator
2184 Instructions</a> </div>
2186 <div class="doc_text">
2188 <p>As mentioned <a href="#functionstructure">previously</a>, every
2189 basic block in a program ends with a "Terminator" instruction, which
2190 indicates which block should be executed after the current block is
2191 finished. These terminator instructions typically yield a '<tt>void</tt>'
2192 value: they produce control flow, not values (the one exception being
2193 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
2194 <p>There are six different terminator instructions: the '<a
2195 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
2196 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
2197 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
2198 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
2199 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
2203 <!-- _______________________________________________________________________ -->
2204 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
2205 Instruction</a> </div>
2206 <div class="doc_text">
2209 ret <type> <value> <i>; Return a value from a non-void function</i>
2210 ret void <i>; Return from void function</i>
2215 <p>The '<tt>ret</tt>' instruction is used to return control flow (and
2216 optionally a value) from a function back to the caller.</p>
2217 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
2218 returns a value and then causes control flow, and one that just causes
2219 control flow to occur.</p>
2223 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument,
2224 the return value. The type of the return value must be a
2225 '<a href="#t_firstclass">first class</a>' type.</p>
2227 <p>A function is not <a href="#wellformed">well formed</a> if
2228 it it has a non-void return type and contains a '<tt>ret</tt>'
2229 instruction with no return value or a return value with a type that
2230 does not match its type, or if it has a void return type and contains
2231 a '<tt>ret</tt>' instruction with a return value.</p>
2235 <p>When the '<tt>ret</tt>' instruction is executed, control flow
2236 returns back to the calling function's context. If the caller is a "<a
2237 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
2238 the instruction after the call. If the caller was an "<a
2239 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
2240 at the beginning of the "normal" destination block. If the instruction
2241 returns a value, that value shall set the call or invoke instruction's
2247 ret i32 5 <i>; Return an integer value of 5</i>
2248 ret void <i>; Return from a void function</i>
2249 ret { i32, i8 } { i32 4, i8 2 } <i>; Return a struct of values 4 and 2</i>
2252 <p>Note that the code generator does not yet fully support large
2253 return values. The specific sizes that are currently supported are
2254 dependent on the target. For integers, on 32-bit targets the limit
2255 is often 64 bits, and on 64-bit targets the limit is often 128 bits.
2256 For aggregate types, the current limits are dependent on the element
2257 types; for example targets are often limited to 2 total integer
2258 elements and 2 total floating-point elements.</p>
2261 <!-- _______________________________________________________________________ -->
2262 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
2263 <div class="doc_text">
2265 <pre> br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
2268 <p>The '<tt>br</tt>' instruction is used to cause control flow to
2269 transfer to a different basic block in the current function. There are
2270 two forms of this instruction, corresponding to a conditional branch
2271 and an unconditional branch.</p>
2273 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
2274 single '<tt>i1</tt>' value and two '<tt>label</tt>' values. The
2275 unconditional form of the '<tt>br</tt>' instruction takes a single
2276 '<tt>label</tt>' value as a target.</p>
2278 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
2279 argument is evaluated. If the value is <tt>true</tt>, control flows
2280 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
2281 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
2283 <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
2284 href="#i_ret">ret</a> i32 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> i32 0<br></pre>
2286 <!-- _______________________________________________________________________ -->
2287 <div class="doc_subsubsection">
2288 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
2291 <div class="doc_text">
2295 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
2300 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
2301 several different places. It is a generalization of the '<tt>br</tt>'
2302 instruction, allowing a branch to occur to one of many possible
2308 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
2309 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
2310 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
2311 table is not allowed to contain duplicate constant entries.</p>
2315 <p>The <tt>switch</tt> instruction specifies a table of values and
2316 destinations. When the '<tt>switch</tt>' instruction is executed, this
2317 table is searched for the given value. If the value is found, control flow is
2318 transfered to the corresponding destination; otherwise, control flow is
2319 transfered to the default destination.</p>
2321 <h5>Implementation:</h5>
2323 <p>Depending on properties of the target machine and the particular
2324 <tt>switch</tt> instruction, this instruction may be code generated in different
2325 ways. For example, it could be generated as a series of chained conditional
2326 branches or with a lookup table.</p>
2331 <i>; Emulate a conditional br instruction</i>
2332 %Val = <a href="#i_zext">zext</a> i1 %value to i32
2333 switch i32 %Val, label %truedest [ i32 0, label %falsedest ]
2335 <i>; Emulate an unconditional br instruction</i>
2336 switch i32 0, label %dest [ ]
2338 <i>; Implement a jump table:</i>
2339 switch i32 %val, label %otherwise [ i32 0, label %onzero
2341 i32 2, label %ontwo ]
2345 <!-- _______________________________________________________________________ -->
2346 <div class="doc_subsubsection">
2347 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
2350 <div class="doc_text">
2355 <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>]
2356 to label <normal label> unwind label <exception label>
2361 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
2362 function, with the possibility of control flow transfer to either the
2363 '<tt>normal</tt>' label or the
2364 '<tt>exception</tt>' label. If the callee function returns with the
2365 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
2366 "normal" label. If the callee (or any indirect callees) returns with the "<a
2367 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
2368 continued at the dynamically nearest "exception" label.</p>
2372 <p>This instruction requires several arguments:</p>
2376 The optional "cconv" marker indicates which <a href="#callingconv">calling
2377 convention</a> the call should use. If none is specified, the call defaults
2378 to using C calling conventions.
2381 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
2382 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>',
2383 and '<tt>inreg</tt>' attributes are valid here.</li>
2385 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
2386 function value being invoked. In most cases, this is a direct function
2387 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
2388 an arbitrary pointer to function value.
2391 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
2392 function to be invoked. </li>
2394 <li>'<tt>function args</tt>': argument list whose types match the function
2395 signature argument types. If the function signature indicates the function
2396 accepts a variable number of arguments, the extra arguments can be
2399 <li>'<tt>normal label</tt>': the label reached when the called function
2400 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
2402 <li>'<tt>exception label</tt>': the label reached when a callee returns with
2403 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
2405 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
2406 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
2407 '<tt>readnone</tt>' attributes are valid here.</li>
2412 <p>This instruction is designed to operate as a standard '<tt><a
2413 href="#i_call">call</a></tt>' instruction in most regards. The primary
2414 difference is that it establishes an association with a label, which is used by
2415 the runtime library to unwind the stack.</p>
2417 <p>This instruction is used in languages with destructors to ensure that proper
2418 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
2419 exception. Additionally, this is important for implementation of
2420 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
2422 <p>For the purposes of the SSA form, the definition of the value
2423 returned by the '<tt>invoke</tt>' instruction is deemed to occur on
2424 the edge from the current block to the "normal" label. If the callee
2425 unwinds then no return value is available.</p>
2429 %retval = invoke i32 @Test(i32 15) to label %Continue
2430 unwind label %TestCleanup <i>; {i32}:retval set</i>
2431 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
2432 unwind label %TestCleanup <i>; {i32}:retval set</i>
2437 <!-- _______________________________________________________________________ -->
2439 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
2440 Instruction</a> </div>
2442 <div class="doc_text">
2451 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
2452 at the first callee in the dynamic call stack which used an <a
2453 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
2454 primarily used to implement exception handling.</p>
2458 <p>The '<tt>unwind</tt>' instruction causes execution of the current function to
2459 immediately halt. The dynamic call stack is then searched for the first <a
2460 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
2461 execution continues at the "exceptional" destination block specified by the
2462 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
2463 dynamic call chain, undefined behavior results.</p>
2466 <!-- _______________________________________________________________________ -->
2468 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
2469 Instruction</a> </div>
2471 <div class="doc_text">
2480 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
2481 instruction is used to inform the optimizer that a particular portion of the
2482 code is not reachable. This can be used to indicate that the code after a
2483 no-return function cannot be reached, and other facts.</p>
2487 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
2492 <!-- ======================================================================= -->
2493 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
2494 <div class="doc_text">
2495 <p>Binary operators are used to do most of the computation in a
2496 program. They require two operands of the same type, execute an operation on them, and
2497 produce a single value. The operands might represent
2498 multiple data, as is the case with the <a href="#t_vector">vector</a> data type.
2499 The result value has the same type as its operands.</p>
2500 <p>There are several different binary operators:</p>
2502 <!-- _______________________________________________________________________ -->
2503 <div class="doc_subsubsection">
2504 <a name="i_add">'<tt>add</tt>' Instruction</a>
2507 <div class="doc_text">
2512 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2517 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
2521 <p>The two arguments to the '<tt>add</tt>' instruction must be <a
2522 href="#t_integer">integer</a> or
2523 <a href="#t_vector">vector</a> of integer values. Both arguments must
2524 have identical types.</p>
2528 <p>The value produced is the integer sum of the two operands.</p>
2530 <p>If the sum has unsigned overflow, the result returned is the
2531 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2534 <p>Because LLVM integers use a two's complement representation, this
2535 instruction is appropriate for both signed and unsigned integers.</p>
2540 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
2543 <!-- _______________________________________________________________________ -->
2544 <div class="doc_subsubsection">
2545 <a name="i_fadd">'<tt>fadd</tt>' Instruction</a>
2548 <div class="doc_text">
2553 <result> = fadd <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2558 <p>The '<tt>fadd</tt>' instruction returns the sum of its two operands.</p>
2562 <p>The two arguments to the '<tt>fadd</tt>' instruction must be
2563 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a> of
2564 floating point values. Both arguments must have identical types.</p>
2568 <p>The value produced is the floating point sum of the two operands.</p>
2573 <result> = fadd float 4.0, %var <i>; yields {float}:result = 4.0 + %var</i>
2576 <!-- _______________________________________________________________________ -->
2577 <div class="doc_subsubsection">
2578 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
2581 <div class="doc_text">
2586 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2591 <p>The '<tt>sub</tt>' instruction returns the difference of its two
2594 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
2595 '<tt>neg</tt>' instruction present in most other intermediate
2596 representations.</p>
2600 <p>The two arguments to the '<tt>sub</tt>' instruction must be <a
2601 href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of
2602 integer values. Both arguments must have identical types.</p>
2606 <p>The value produced is the integer difference of the two operands.</p>
2608 <p>If the difference has unsigned overflow, the result returned is the
2609 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2612 <p>Because LLVM integers use a two's complement representation, this
2613 instruction is appropriate for both signed and unsigned integers.</p>
2617 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
2618 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
2622 <!-- _______________________________________________________________________ -->
2623 <div class="doc_subsubsection">
2624 <a name="i_fsub">'<tt>fsub</tt>' Instruction</a>
2627 <div class="doc_text">
2632 <result> = fsub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2637 <p>The '<tt>fsub</tt>' instruction returns the difference of its two
2640 <p>Note that the '<tt>fsub</tt>' instruction is used to represent the
2641 '<tt>fneg</tt>' instruction present in most other intermediate
2642 representations.</p>
2646 <p>The two arguments to the '<tt>fsub</tt>' instruction must be <a
2647 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a>
2648 of floating point values. Both arguments must have identical types.</p>
2652 <p>The value produced is the floating point difference of the two operands.</p>
2656 <result> = fsub float 4.0, %var <i>; yields {float}:result = 4.0 - %var</i>
2657 <result> = fsub float -0.0, %val <i>; yields {float}:result = -%var</i>
2661 <!-- _______________________________________________________________________ -->
2662 <div class="doc_subsubsection">
2663 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
2666 <div class="doc_text">
2669 <pre> <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2672 <p>The '<tt>mul</tt>' instruction returns the product of its two
2677 <p>The two arguments to the '<tt>mul</tt>' instruction must be <a
2678 href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2679 values. Both arguments must have identical types.</p>
2683 <p>The value produced is the integer product of the two operands.</p>
2685 <p>If the result of the multiplication has unsigned overflow,
2686 the result returned is the mathematical result modulo
2687 2<sup>n</sup>, where n is the bit width of the result.</p>
2688 <p>Because LLVM integers use a two's complement representation, and the
2689 result is the same width as the operands, this instruction returns the
2690 correct result for both signed and unsigned integers. If a full product
2691 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands
2692 should be sign-extended or zero-extended as appropriate to the
2693 width of the full product.</p>
2695 <pre> <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
2699 <!-- _______________________________________________________________________ -->
2700 <div class="doc_subsubsection">
2701 <a name="i_fmul">'<tt>fmul</tt>' Instruction</a>
2704 <div class="doc_text">
2707 <pre> <result> = fmul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2710 <p>The '<tt>fmul</tt>' instruction returns the product of its two
2715 <p>The two arguments to the '<tt>fmul</tt>' instruction must be
2716 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a>
2717 of floating point values. Both arguments must have identical types.</p>
2721 <p>The value produced is the floating point product of the two operands.</p>
2724 <pre> <result> = fmul float 4.0, %var <i>; yields {float}:result = 4.0 * %var</i>
2728 <!-- _______________________________________________________________________ -->
2729 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
2731 <div class="doc_text">
2733 <pre> <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2736 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two
2741 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
2742 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2743 values. Both arguments must have identical types.</p>
2747 <p>The value produced is the unsigned integer quotient of the two operands.</p>
2748 <p>Note that unsigned integer division and signed integer division are distinct
2749 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
2750 <p>Division by zero leads to undefined behavior.</p>
2752 <pre> <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2755 <!-- _______________________________________________________________________ -->
2756 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
2758 <div class="doc_text">
2761 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2766 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
2771 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
2772 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2773 values. Both arguments must have identical types.</p>
2776 <p>The value produced is the signed integer quotient of the two operands rounded towards zero.</p>
2777 <p>Note that signed integer division and unsigned integer division are distinct
2778 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
2779 <p>Division by zero leads to undefined behavior. Overflow also leads to
2780 undefined behavior; this is a rare case, but can occur, for example,
2781 by doing a 32-bit division of -2147483648 by -1.</p>
2783 <pre> <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2786 <!-- _______________________________________________________________________ -->
2787 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
2788 Instruction</a> </div>
2789 <div class="doc_text">
2792 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2796 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
2801 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
2802 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a>
2803 of floating point values. Both arguments must have identical types.</p>
2807 <p>The value produced is the floating point quotient of the two operands.</p>
2812 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
2816 <!-- _______________________________________________________________________ -->
2817 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
2819 <div class="doc_text">
2821 <pre> <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2824 <p>The '<tt>urem</tt>' instruction returns the remainder from the
2825 unsigned division of its two arguments.</p>
2827 <p>The two arguments to the '<tt>urem</tt>' instruction must be
2828 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2829 values. Both arguments must have identical types.</p>
2831 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
2832 This instruction always performs an unsigned division to get the remainder.</p>
2833 <p>Note that unsigned integer remainder and signed integer remainder are
2834 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
2835 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
2837 <pre> <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2841 <!-- _______________________________________________________________________ -->
2842 <div class="doc_subsubsection">
2843 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
2846 <div class="doc_text">
2851 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2856 <p>The '<tt>srem</tt>' instruction returns the remainder from the
2857 signed division of its two operands. This instruction can also take
2858 <a href="#t_vector">vector</a> versions of the values in which case
2859 the elements must be integers.</p>
2863 <p>The two arguments to the '<tt>srem</tt>' instruction must be
2864 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2865 values. Both arguments must have identical types.</p>
2869 <p>This instruction returns the <i>remainder</i> of a division (where the result
2870 has the same sign as the dividend, <tt>op1</tt>), not the <i>modulo</i>
2871 operator (where the result has the same sign as the divisor, <tt>op2</tt>) of
2872 a value. For more information about the difference, see <a
2873 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
2874 Math Forum</a>. For a table of how this is implemented in various languages,
2875 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
2876 Wikipedia: modulo operation</a>.</p>
2877 <p>Note that signed integer remainder and unsigned integer remainder are
2878 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
2879 <p>Taking the remainder of a division by zero leads to undefined behavior.
2880 Overflow also leads to undefined behavior; this is a rare case, but can occur,
2881 for example, by taking the remainder of a 32-bit division of -2147483648 by -1.
2882 (The remainder doesn't actually overflow, but this rule lets srem be
2883 implemented using instructions that return both the result of the division
2884 and the remainder.)</p>
2886 <pre> <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2890 <!-- _______________________________________________________________________ -->
2891 <div class="doc_subsubsection">
2892 <a name="i_frem">'<tt>frem</tt>' Instruction</a> </div>
2894 <div class="doc_text">
2897 <pre> <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2900 <p>The '<tt>frem</tt>' instruction returns the remainder from the
2901 division of its two operands.</p>
2903 <p>The two arguments to the '<tt>frem</tt>' instruction must be
2904 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a>
2905 of floating point values. Both arguments must have identical types.</p>
2909 <p>This instruction returns the <i>remainder</i> of a division.
2910 The remainder has the same sign as the dividend.</p>
2915 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
2919 <!-- ======================================================================= -->
2920 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
2921 Operations</a> </div>
2922 <div class="doc_text">
2923 <p>Bitwise binary operators are used to do various forms of
2924 bit-twiddling in a program. They are generally very efficient
2925 instructions and can commonly be strength reduced from other
2926 instructions. They require two operands of the same type, execute an operation on them,
2927 and produce a single value. The resulting value is the same type as its operands.</p>
2930 <!-- _______________________________________________________________________ -->
2931 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
2932 Instruction</a> </div>
2933 <div class="doc_text">
2935 <pre> <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2940 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2941 the left a specified number of bits.</p>
2945 <p>Both arguments to the '<tt>shl</tt>' instruction must be the same <a
2946 href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2947 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
2951 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod 2<sup>n</sup>,
2952 where n is the width of the result. If <tt>op2</tt> is (statically or dynamically) negative or
2953 equal to or larger than the number of bits in <tt>op1</tt>, the result is undefined.
2954 If the arguments are vectors, each vector element of <tt>op1</tt> is shifted by the
2955 corresponding shift amount in <tt>op2</tt>.</p>
2957 <h5>Example:</h5><pre>
2958 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
2959 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
2960 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
2961 <result> = shl i32 1, 32 <i>; undefined</i>
2962 <result> = shl <2 x i32> < i32 1, i32 1>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 2, i32 4></i>
2965 <!-- _______________________________________________________________________ -->
2966 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
2967 Instruction</a> </div>
2968 <div class="doc_text">
2970 <pre> <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2974 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
2975 operand shifted to the right a specified number of bits with zero fill.</p>
2978 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
2979 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2980 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
2984 <p>This instruction always performs a logical shift right operation. The most
2985 significant bits of the result will be filled with zero bits after the
2986 shift. If <tt>op2</tt> is (statically or dynamically) equal to or larger than
2987 the number of bits in <tt>op1</tt>, the result is undefined. If the arguments are
2988 vectors, each vector element of <tt>op1</tt> is shifted by the corresponding shift
2989 amount in <tt>op2</tt>.</p>
2993 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
2994 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
2995 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
2996 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
2997 <result> = lshr i32 1, 32 <i>; undefined</i>
2998 <result> = lshr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 2> <i>; yields: result=<2 x i32> < i32 0x7FFFFFFF, i32 1></i>
3002 <!-- _______________________________________________________________________ -->
3003 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
3004 Instruction</a> </div>
3005 <div class="doc_text">
3008 <pre> <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3012 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
3013 operand shifted to the right a specified number of bits with sign extension.</p>
3016 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
3017 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3018 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
3021 <p>This instruction always performs an arithmetic shift right operation,
3022 The most significant bits of the result will be filled with the sign bit
3023 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
3024 larger than the number of bits in <tt>op1</tt>, the result is undefined. If the
3025 arguments are vectors, each vector element of <tt>op1</tt> is shifted by the
3026 corresponding shift amount in <tt>op2</tt>.</p>
3030 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
3031 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
3032 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
3033 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
3034 <result> = ashr i32 1, 32 <i>; undefined</i>
3035 <result> = ashr <2 x i32> < i32 -2, i32 4>, < i32 1, i32 3> <i>; yields: result=<2 x i32> < i32 -1, i32 0></i>
3039 <!-- _______________________________________________________________________ -->
3040 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
3041 Instruction</a> </div>
3043 <div class="doc_text">
3048 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3053 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
3054 its two operands.</p>
3058 <p>The two arguments to the '<tt>and</tt>' instruction must be
3059 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3060 values. Both arguments must have identical types.</p>
3063 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
3066 <table border="1" cellspacing="0" cellpadding="4">
3098 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
3099 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
3100 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
3103 <!-- _______________________________________________________________________ -->
3104 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
3105 <div class="doc_text">
3107 <pre> <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3110 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
3111 or of its two operands.</p>
3114 <p>The two arguments to the '<tt>or</tt>' instruction must be
3115 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3116 values. Both arguments must have identical types.</p>
3118 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
3121 <table border="1" cellspacing="0" cellpadding="4">
3152 <pre> <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
3153 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
3154 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
3157 <!-- _______________________________________________________________________ -->
3158 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
3159 Instruction</a> </div>
3160 <div class="doc_text">
3162 <pre> <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
3165 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
3166 or of its two operands. The <tt>xor</tt> is used to implement the
3167 "one's complement" operation, which is the "~" operator in C.</p>
3169 <p>The two arguments to the '<tt>xor</tt>' instruction must be
3170 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
3171 values. Both arguments must have identical types.</p>
3175 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
3178 <table border="1" cellspacing="0" cellpadding="4">
3210 <pre> <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
3211 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
3212 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
3213 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
3217 <!-- ======================================================================= -->
3218 <div class="doc_subsection">
3219 <a name="vectorops">Vector Operations</a>
3222 <div class="doc_text">
3224 <p>LLVM supports several instructions to represent vector operations in a
3225 target-independent manner. These instructions cover the element-access and
3226 vector-specific operations needed to process vectors effectively. While LLVM
3227 does directly support these vector operations, many sophisticated algorithms
3228 will want to use target-specific intrinsics to take full advantage of a specific
3233 <!-- _______________________________________________________________________ -->
3234 <div class="doc_subsubsection">
3235 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
3238 <div class="doc_text">
3243 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
3249 The '<tt>extractelement</tt>' instruction extracts a single scalar
3250 element from a vector at a specified index.
3257 The first operand of an '<tt>extractelement</tt>' instruction is a
3258 value of <a href="#t_vector">vector</a> type. The second operand is
3259 an index indicating the position from which to extract the element.
3260 The index may be a variable.</p>
3265 The result is a scalar of the same type as the element type of
3266 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
3267 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
3268 results are undefined.
3274 %result = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
3279 <!-- _______________________________________________________________________ -->
3280 <div class="doc_subsubsection">
3281 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
3284 <div class="doc_text">
3289 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
3295 The '<tt>insertelement</tt>' instruction inserts a scalar
3296 element into a vector at a specified index.
3303 The first operand of an '<tt>insertelement</tt>' instruction is a
3304 value of <a href="#t_vector">vector</a> type. The second operand is a
3305 scalar value whose type must equal the element type of the first
3306 operand. The third operand is an index indicating the position at
3307 which to insert the value. The index may be a variable.</p>
3312 The result is a vector of the same type as <tt>val</tt>. Its
3313 element values are those of <tt>val</tt> except at position
3314 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
3315 exceeds the length of <tt>val</tt>, the results are undefined.
3321 %result = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
3325 <!-- _______________________________________________________________________ -->
3326 <div class="doc_subsubsection">
3327 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
3330 <div class="doc_text">
3335 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
3341 The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
3342 from two input vectors, returning a vector with the same element type as
3343 the input and length that is the same as the shuffle mask.
3349 The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
3350 with types that match each other. The third argument is a shuffle mask whose
3351 element type is always 'i32'. The result of the instruction is a vector whose
3352 length is the same as the shuffle mask and whose element type is the same as
3353 the element type of the first two operands.
3357 The shuffle mask operand is required to be a constant vector with either
3358 constant integer or undef values.
3364 The elements of the two input vectors are numbered from left to right across
3365 both of the vectors. The shuffle mask operand specifies, for each element of
3366 the result vector, which element of the two input vectors the result element
3367 gets. The element selector may be undef (meaning "don't care") and the second
3368 operand may be undef if performing a shuffle from only one vector.
3374 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
3375 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
3376 %result = shufflevector <4 x i32> %v1, <4 x i32> undef,
3377 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
3378 %result = shufflevector <8 x i32> %v1, <8 x i32> undef,
3379 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i>
3380 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
3381 <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>
3386 <!-- ======================================================================= -->
3387 <div class="doc_subsection">
3388 <a name="aggregateops">Aggregate Operations</a>
3391 <div class="doc_text">
3393 <p>LLVM supports several instructions for working with aggregate values.
3398 <!-- _______________________________________________________________________ -->
3399 <div class="doc_subsubsection">
3400 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
3403 <div class="doc_text">
3408 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
3414 The '<tt>extractvalue</tt>' instruction extracts the value of a struct field
3415 or array element from an aggregate value.
3422 The first operand of an '<tt>extractvalue</tt>' instruction is a
3423 value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a>
3424 type. The operands are constant indices to specify which value to extract
3425 in a similar manner as indices in a
3426 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
3432 The result is the value at the position in the aggregate specified by
3439 %result = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
3444 <!-- _______________________________________________________________________ -->
3445 <div class="doc_subsubsection">
3446 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
3449 <div class="doc_text">
3454 <result> = insertvalue <aggregate type> <val>, <ty> <val>, <idx> <i>; yields <n x <ty>></i>
3460 The '<tt>insertvalue</tt>' instruction inserts a value
3461 into a struct field or array element in an aggregate.
3468 The first operand of an '<tt>insertvalue</tt>' instruction is a
3469 value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a> type.
3470 The second operand is a first-class value to insert.
3471 The following operands are constant indices
3472 indicating the position at which to insert the value in a similar manner as
3474 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
3475 The value to insert must have the same type as the value identified
3482 The result is an aggregate of the same type as <tt>val</tt>. Its
3483 value is that of <tt>val</tt> except that the value at the position
3484 specified by the indices is that of <tt>elt</tt>.
3490 %result = insertvalue {i32, float} %agg, i32 1, 0 <i>; yields {i32, float}</i>
3495 <!-- ======================================================================= -->
3496 <div class="doc_subsection">
3497 <a name="memoryops">Memory Access and Addressing Operations</a>
3500 <div class="doc_text">
3502 <p>A key design point of an SSA-based representation is how it
3503 represents memory. In LLVM, no memory locations are in SSA form, which
3504 makes things very simple. This section describes how to read, write,
3505 allocate, and free memory in LLVM.</p>
3509 <!-- _______________________________________________________________________ -->
3510 <div class="doc_subsubsection">
3511 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
3514 <div class="doc_text">
3519 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3524 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
3525 heap and returns a pointer to it. The object is always allocated in the generic
3526 address space (address space zero).</p>
3530 <p>The '<tt>malloc</tt>' instruction allocates
3531 <tt>sizeof(<type>)*NumElements</tt>
3532 bytes of memory from the operating system and returns a pointer of the
3533 appropriate type to the program. If "NumElements" is specified, it is the
3534 number of elements allocated, otherwise "NumElements" is defaulted to be one.
3535 If a constant alignment is specified, the value result of the allocation is
3536 guaranteed to be aligned to at least that boundary. If not specified, or if
3537 zero, the target can choose to align the allocation on any convenient boundary
3538 compatible with the type.</p>
3540 <p>'<tt>type</tt>' must be a sized type.</p>
3544 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
3545 a pointer is returned. The result of a zero byte allocation is undefined. The
3546 result is null if there is insufficient memory available.</p>
3551 %array = malloc [4 x i8] <i>; yields {[%4 x i8]*}:array</i>
3553 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
3554 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
3555 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
3556 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
3557 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
3560 <p>Note that the code generator does not yet respect the
3561 alignment value.</p>
3565 <!-- _______________________________________________________________________ -->
3566 <div class="doc_subsubsection">
3567 <a name="i_free">'<tt>free</tt>' Instruction</a>
3570 <div class="doc_text">
3575 free <type> <value> <i>; yields {void}</i>
3580 <p>The '<tt>free</tt>' instruction returns memory back to the unused
3581 memory heap to be reallocated in the future.</p>
3585 <p>'<tt>value</tt>' shall be a pointer value that points to a value
3586 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
3591 <p>Access to the memory pointed to by the pointer is no longer defined
3592 after this instruction executes. If the pointer is null, the operation
3598 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
3599 free [4 x i8]* %array
3603 <!-- _______________________________________________________________________ -->
3604 <div class="doc_subsubsection">
3605 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
3608 <div class="doc_text">
3613 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3618 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
3619 currently executing function, to be automatically released when this function
3620 returns to its caller. The object is always allocated in the generic address
3621 space (address space zero).</p>
3625 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
3626 bytes of memory on the runtime stack, returning a pointer of the
3627 appropriate type to the program. If "NumElements" is specified, it is the
3628 number of elements allocated, otherwise "NumElements" is defaulted to be one.
3629 If a constant alignment is specified, the value result of the allocation is
3630 guaranteed to be aligned to at least that boundary. If not specified, or if
3631 zero, the target can choose to align the allocation on any convenient boundary
3632 compatible with the type.</p>
3634 <p>'<tt>type</tt>' may be any sized type.</p>
3638 <p>Memory is allocated; a pointer is returned. The operation is undefined if
3639 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
3640 memory is automatically released when the function returns. The '<tt>alloca</tt>'
3641 instruction is commonly used to represent automatic variables that must
3642 have an address available. When the function returns (either with the <tt><a
3643 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
3644 instructions), the memory is reclaimed. Allocating zero bytes
3645 is legal, but the result is undefined.</p>
3650 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
3651 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
3652 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
3653 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
3657 <!-- _______________________________________________________________________ -->
3658 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
3659 Instruction</a> </div>
3660 <div class="doc_text">
3662 <pre> <result> = load <ty>* <pointer>[, align <alignment>]<br> <result> = volatile load <ty>* <pointer>[, align <alignment>]<br></pre>
3664 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
3666 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
3667 address from which to load. The pointer must point to a <a
3668 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
3669 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
3670 the number or order of execution of this <tt>load</tt> with other
3671 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
3674 The optional constant "align" argument specifies the alignment of the operation
3675 (that is, the alignment of the memory address). A value of 0 or an
3676 omitted "align" argument means that the operation has the preferential
3677 alignment for the target. It is the responsibility of the code emitter
3678 to ensure that the alignment information is correct. Overestimating
3679 the alignment results in an undefined behavior. Underestimating the
3680 alignment may produce less efficient code. An alignment of 1 is always
3684 <p>The location of memory pointed to is loaded. If the value being loaded
3685 is of scalar type then the number of bytes read does not exceed the minimum
3686 number of bytes needed to hold all bits of the type. For example, loading an
3687 <tt>i24</tt> reads at most three bytes. When loading a value of a type like
3688 <tt>i20</tt> with a size that is not an integral number of bytes, the result
3689 is undefined if the value was not originally written using a store of the
3692 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3694 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
3695 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
3698 <!-- _______________________________________________________________________ -->
3699 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
3700 Instruction</a> </div>
3701 <div class="doc_text">
3703 <pre> store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3704 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3707 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
3709 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
3710 to store and an address at which to store it. The type of the '<tt><pointer></tt>'
3711 operand must be a pointer to the <a href="#t_firstclass">first class</a> type
3712 of the '<tt><value></tt>'
3713 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
3714 optimizer is not allowed to modify the number or order of execution of
3715 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
3716 href="#i_store">store</a></tt> instructions.</p>
3718 The optional constant "align" argument specifies the alignment of the operation
3719 (that is, the alignment of the memory address). A value of 0 or an
3720 omitted "align" argument means that the operation has the preferential
3721 alignment for the target. It is the responsibility of the code emitter
3722 to ensure that the alignment information is correct. Overestimating
3723 the alignment results in an undefined behavior. Underestimating the
3724 alignment may produce less efficient code. An alignment of 1 is always
3728 <p>The contents of memory are updated to contain '<tt><value></tt>'
3729 at the location specified by the '<tt><pointer></tt>' operand.
3730 If '<tt><value></tt>' is of scalar type then the number of bytes
3731 written does not exceed the minimum number of bytes needed to hold all
3732 bits of the type. For example, storing an <tt>i24</tt> writes at most
3733 three bytes. When writing a value of a type like <tt>i20</tt> with a
3734 size that is not an integral number of bytes, it is unspecified what
3735 happens to the extra bits that do not belong to the type, but they will
3736 typically be overwritten.</p>
3738 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3739 store i32 3, i32* %ptr <i>; yields {void}</i>
3740 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
3744 <!-- _______________________________________________________________________ -->
3745 <div class="doc_subsubsection">
3746 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
3749 <div class="doc_text">
3752 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
3758 The '<tt>getelementptr</tt>' instruction is used to get the address of a
3759 subelement of an aggregate data structure. It performs address calculation only
3760 and does not access memory.</p>
3764 <p>The first argument is always a pointer, and forms the basis of the
3765 calculation. The remaining arguments are indices, that indicate which of the
3766 elements of the aggregate object are indexed. The interpretation of each index
3767 is dependent on the type being indexed into. The first index always indexes the
3768 pointer value given as the first argument, the second index indexes a value of
3769 the type pointed to (not necessarily the value directly pointed to, since the
3770 first index can be non-zero), etc. The first type indexed into must be a pointer
3771 value, subsequent types can be arrays, vectors and structs. Note that subsequent
3772 types being indexed into can never be pointers, since that would require loading
3773 the pointer before continuing calculation.</p>
3775 <p>The type of each index argument depends on the type it is indexing into.
3776 When indexing into a (packed) structure, only <tt>i32</tt> integer
3777 <b>constants</b> are allowed. When indexing into an array, pointer or vector,
3778 integers of any width are allowed (also non-constants).</p>
3780 <p>For example, let's consider a C code fragment and how it gets
3781 compiled to LLVM:</p>
3783 <div class="doc_code">
3796 int *foo(struct ST *s) {
3797 return &s[1].Z.B[5][13];
3802 <p>The LLVM code generated by the GCC frontend is:</p>
3804 <div class="doc_code">
3806 %RT = <a href="#namedtypes">type</a> { i8 , [10 x [20 x i32]], i8 }
3807 %ST = <a href="#namedtypes">type</a> { i32, double, %RT }
3809 define i32* %foo(%ST* %s) {
3811 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
3819 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
3820 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
3821 }</tt>' type, a structure. The second index indexes into the third element of
3822 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
3823 i8 }</tt>' type, another structure. The third index indexes into the second
3824 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
3825 array. The two dimensions of the array are subscripted into, yielding an
3826 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
3827 to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
3829 <p>Note that it is perfectly legal to index partially through a
3830 structure, returning a pointer to an inner element. Because of this,
3831 the LLVM code for the given testcase is equivalent to:</p>
3834 define i32* %foo(%ST* %s) {
3835 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
3836 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
3837 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
3838 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
3839 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
3844 <p>Note that it is undefined to access an array out of bounds: array
3845 and pointer indexes must always be within the defined bounds of the
3846 array type when accessed with an instruction that dereferences the
3847 pointer (e.g. a load or store instruction). The one exception for
3848 this rule is zero length arrays. These arrays are defined to be
3849 accessible as variable length arrays, which requires access beyond the
3850 zero'th element.</p>
3852 <p>The getelementptr instruction is often confusing. For some more insight
3853 into how it works, see <a href="GetElementPtr.html">the getelementptr
3859 <i>; yields [12 x i8]*:aptr</i>
3860 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
3861 <i>; yields i8*:vptr</i>
3862 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
3863 <i>; yields i8*:eptr</i>
3864 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
3865 <i>; yields i32*:iptr</i>
3866 %iptr = getelementptr [10 x i32]* @arr, i16 0, i16 0
3870 <!-- ======================================================================= -->
3871 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
3873 <div class="doc_text">
3874 <p>The instructions in this category are the conversion instructions (casting)
3875 which all take a single operand and a type. They perform various bit conversions
3879 <!-- _______________________________________________________________________ -->
3880 <div class="doc_subsubsection">
3881 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
3883 <div class="doc_text">
3887 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
3892 The '<tt>trunc</tt>' instruction truncates its operand to the type <tt>ty2</tt>.
3897 The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
3898 be an <a href="#t_integer">integer</a> type, and a type that specifies the size
3899 and type of the result, which must be an <a href="#t_integer">integer</a>
3900 type. The bit size of <tt>value</tt> must be larger than the bit size of
3901 <tt>ty2</tt>. Equal sized types are not allowed.</p>
3905 The '<tt>trunc</tt>' instruction truncates the high order bits in <tt>value</tt>
3906 and converts the remaining bits to <tt>ty2</tt>. Since the source size must be
3907 larger than the destination size, <tt>trunc</tt> cannot be a <i>no-op cast</i>.
3908 It will always truncate bits.</p>
3912 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
3913 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
3914 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
3918 <!-- _______________________________________________________________________ -->
3919 <div class="doc_subsubsection">
3920 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
3922 <div class="doc_text">
3926 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
3930 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
3935 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
3936 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3937 also be of <a href="#t_integer">integer</a> type. The bit size of the
3938 <tt>value</tt> must be smaller than the bit size of the destination type,
3942 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
3943 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
3945 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
3949 %X = zext i32 257 to i64 <i>; yields i64:257</i>
3950 %Y = zext i1 true to i32 <i>; yields i32:1</i>
3954 <!-- _______________________________________________________________________ -->
3955 <div class="doc_subsubsection">
3956 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
3958 <div class="doc_text">
3962 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
3966 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
3970 The '<tt>sext</tt>' instruction takes a value to cast, which must be of
3971 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3972 also be of <a href="#t_integer">integer</a> type. The bit size of the
3973 <tt>value</tt> must be smaller than the bit size of the destination type,
3978 The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
3979 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size of
3980 the type <tt>ty2</tt>.</p>
3982 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
3986 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
3987 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
3991 <!-- _______________________________________________________________________ -->
3992 <div class="doc_subsubsection">
3993 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
3996 <div class="doc_text">
4001 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
4005 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
4010 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
4011 point</a> value to cast and a <a href="#t_floating">floating point</a> type to
4012 cast it to. The size of <tt>value</tt> must be larger than the size of
4013 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
4014 <i>no-op cast</i>.</p>
4017 <p> The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
4018 <a href="#t_floating">floating point</a> type to a smaller
4019 <a href="#t_floating">floating point</a> type. If the value cannot fit within
4020 the destination type, <tt>ty2</tt>, then the results are undefined.</p>
4024 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
4025 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
4029 <!-- _______________________________________________________________________ -->
4030 <div class="doc_subsubsection">
4031 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
4033 <div class="doc_text">
4037 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
4041 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
4042 floating point value.</p>
4045 <p>The '<tt>fpext</tt>' instruction takes a
4046 <a href="#t_floating">floating point</a> <tt>value</tt> to cast,
4047 and a <a href="#t_floating">floating point</a> type to cast it to. The source
4048 type must be smaller than the destination type.</p>
4051 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
4052 <a href="#t_floating">floating point</a> type to a larger
4053 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
4054 used to make a <i>no-op cast</i> because it always changes bits. Use
4055 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
4059 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
4060 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
4064 <!-- _______________________________________________________________________ -->
4065 <div class="doc_subsubsection">
4066 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
4068 <div class="doc_text">
4072 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
4076 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
4077 unsigned integer equivalent of type <tt>ty2</tt>.
4081 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
4082 scalar or vector <a href="#t_floating">floating point</a> value, and a type
4083 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
4084 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
4085 vector integer type with the same number of elements as <tt>ty</tt></p>
4088 <p> The '<tt>fptoui</tt>' instruction converts its
4089 <a href="#t_floating">floating point</a> operand into the nearest (rounding
4090 towards zero) unsigned integer value. If the value cannot fit in <tt>ty2</tt>,
4091 the results are undefined.</p>
4095 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
4096 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
4097 %X = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
4101 <!-- _______________________________________________________________________ -->
4102 <div class="doc_subsubsection">
4103 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
4105 <div class="doc_text">
4109 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
4113 <p>The '<tt>fptosi</tt>' instruction converts
4114 <a href="#t_floating">floating point</a> <tt>value</tt> to type <tt>ty2</tt>.
4118 <p> The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
4119 scalar or vector <a href="#t_floating">floating point</a> value, and a type
4120 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
4121 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
4122 vector integer type with the same number of elements as <tt>ty</tt></p>
4125 <p>The '<tt>fptosi</tt>' instruction converts its
4126 <a href="#t_floating">floating point</a> operand into the nearest (rounding
4127 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
4128 the results are undefined.</p>
4132 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
4133 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
4134 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
4138 <!-- _______________________________________________________________________ -->
4139 <div class="doc_subsubsection">
4140 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
4142 <div class="doc_text">
4146 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
4150 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
4151 integer and converts that value to the <tt>ty2</tt> type.</p>
4154 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
4155 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
4156 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
4157 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
4158 floating point type with the same number of elements as <tt>ty</tt></p>
4161 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
4162 integer quantity and converts it to the corresponding floating point value. If
4163 the value cannot fit in the floating point value, the results are undefined.</p>
4167 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
4168 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
4172 <!-- _______________________________________________________________________ -->
4173 <div class="doc_subsubsection">
4174 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
4176 <div class="doc_text">
4180 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
4184 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed
4185 integer and converts that value to the <tt>ty2</tt> type.</p>
4188 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
4189 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
4190 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
4191 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
4192 floating point type with the same number of elements as <tt>ty</tt></p>
4195 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed
4196 integer quantity and converts it to the corresponding floating point value. If
4197 the value cannot fit in the floating point value, the results are undefined.</p>
4201 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
4202 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
4206 <!-- _______________________________________________________________________ -->
4207 <div class="doc_subsubsection">
4208 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
4210 <div class="doc_text">
4214 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
4218 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
4219 the integer type <tt>ty2</tt>.</p>
4222 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
4223 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
4224 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.</p>
4227 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
4228 <tt>ty2</tt> by interpreting the pointer value as an integer and either
4229 truncating or zero extending that value to the size of the integer type. If
4230 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
4231 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
4232 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
4237 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
4238 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
4242 <!-- _______________________________________________________________________ -->
4243 <div class="doc_subsubsection">
4244 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
4246 <div class="doc_text">
4250 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
4254 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to
4255 a pointer type, <tt>ty2</tt>.</p>
4258 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
4259 value to cast, and a type to cast it to, which must be a
4260 <a href="#t_pointer">pointer</a> type.</p>
4263 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
4264 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
4265 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
4266 size of a pointer then a truncation is done. If <tt>value</tt> is smaller than
4267 the size of a pointer then a zero extension is done. If they are the same size,
4268 nothing is done (<i>no-op cast</i>).</p>
4272 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
4273 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
4274 %Y = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
4278 <!-- _______________________________________________________________________ -->
4279 <div class="doc_subsubsection">
4280 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
4282 <div class="doc_text">
4286 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
4291 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
4292 <tt>ty2</tt> without changing any bits.</p>
4296 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be
4297 a non-aggregate first class value, and a type to cast it to, which must also be
4298 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes of
4300 and the destination type, <tt>ty2</tt>, must be identical. If the source
4301 type is a pointer, the destination type must also be a pointer. This
4302 instruction supports bitwise conversion of vectors to integers and to vectors
4303 of other types (as long as they have the same size).</p>
4306 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
4307 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
4308 this conversion. The conversion is done as if the <tt>value</tt> had been
4309 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only be
4310 converted to other pointer types with this instruction. To convert pointers to
4311 other types, use the <a href="#i_inttoptr">inttoptr</a> or
4312 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
4316 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
4317 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
4318 %Z = bitcast <2 x int> %V to i64; <i>; yields i64: %V</i>
4322 <!-- ======================================================================= -->
4323 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
4324 <div class="doc_text">
4325 <p>The instructions in this category are the "miscellaneous"
4326 instructions, which defy better classification.</p>
4329 <!-- _______________________________________________________________________ -->
4330 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
4332 <div class="doc_text">
4334 <pre> <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4337 <p>The '<tt>icmp</tt>' instruction returns a boolean value or
4338 a vector of boolean values based on comparison
4339 of its two integer, integer vector, or pointer operands.</p>
4341 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
4342 the condition code indicating the kind of comparison to perform. It is not
4343 a value, just a keyword. The possible condition code are:
4346 <li><tt>eq</tt>: equal</li>
4347 <li><tt>ne</tt>: not equal </li>
4348 <li><tt>ugt</tt>: unsigned greater than</li>
4349 <li><tt>uge</tt>: unsigned greater or equal</li>
4350 <li><tt>ult</tt>: unsigned less than</li>
4351 <li><tt>ule</tt>: unsigned less or equal</li>
4352 <li><tt>sgt</tt>: signed greater than</li>
4353 <li><tt>sge</tt>: signed greater or equal</li>
4354 <li><tt>slt</tt>: signed less than</li>
4355 <li><tt>sle</tt>: signed less or equal</li>
4357 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
4358 <a href="#t_pointer">pointer</a>
4359 or integer <a href="#t_vector">vector</a> typed.
4360 They must also be identical types.</p>
4362 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to
4363 the condition code given as <tt>cond</tt>. The comparison performed always
4364 yields either an <a href="#t_primitive"><tt>i1</tt></a> or vector of <tt>i1</tt> result, as follows:
4367 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
4368 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
4370 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
4371 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.</li>
4372 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
4373 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4374 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
4375 <tt>true</tt> if <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4376 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
4377 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
4378 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
4379 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4380 <li><tt>sgt</tt>: interprets the operands as signed values and yields
4381 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4382 <li><tt>sge</tt>: interprets the operands as signed values and yields
4383 <tt>true</tt> if <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4384 <li><tt>slt</tt>: interprets the operands as signed values and yields
4385 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
4386 <li><tt>sle</tt>: interprets the operands as signed values and yields
4387 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4389 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
4390 values are compared as if they were integers.</p>
4391 <p>If the operands are integer vectors, then they are compared
4392 element by element. The result is an <tt>i1</tt> vector with
4393 the same number of elements as the values being compared.
4394 Otherwise, the result is an <tt>i1</tt>.
4398 <pre> <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
4399 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
4400 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
4401 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
4402 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
4403 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
4406 <p>Note that the code generator does not yet support vector types with
4407 the <tt>icmp</tt> instruction.</p>
4411 <!-- _______________________________________________________________________ -->
4412 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
4414 <div class="doc_text">
4416 <pre> <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4419 <p>The '<tt>fcmp</tt>' instruction returns a boolean value
4420 or vector of boolean values based on comparison
4421 of its operands.</p>
4423 If the operands are floating point scalars, then the result
4424 type is a boolean (<a href="#t_primitive"><tt>i1</tt></a>).
4426 <p>If the operands are floating point vectors, then the result type
4427 is a vector of boolean with the same number of elements as the
4428 operands being compared.</p>
4430 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
4431 the condition code indicating the kind of comparison to perform. It is not
4432 a value, just a keyword. The possible condition code are:</p>
4434 <li><tt>false</tt>: no comparison, always returns false</li>
4435 <li><tt>oeq</tt>: ordered and equal</li>
4436 <li><tt>ogt</tt>: ordered and greater than </li>
4437 <li><tt>oge</tt>: ordered and greater than or equal</li>
4438 <li><tt>olt</tt>: ordered and less than </li>
4439 <li><tt>ole</tt>: ordered and less than or equal</li>
4440 <li><tt>one</tt>: ordered and not equal</li>
4441 <li><tt>ord</tt>: ordered (no nans)</li>
4442 <li><tt>ueq</tt>: unordered or equal</li>
4443 <li><tt>ugt</tt>: unordered or greater than </li>
4444 <li><tt>uge</tt>: unordered or greater than or equal</li>
4445 <li><tt>ult</tt>: unordered or less than </li>
4446 <li><tt>ule</tt>: unordered or less than or equal</li>
4447 <li><tt>une</tt>: unordered or not equal</li>
4448 <li><tt>uno</tt>: unordered (either nans)</li>
4449 <li><tt>true</tt>: no comparison, always returns true</li>
4451 <p><i>Ordered</i> means that neither operand is a QNAN while
4452 <i>unordered</i> means that either operand may be a QNAN.</p>
4453 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be
4454 either a <a href="#t_floating">floating point</a> type
4455 or a <a href="#t_vector">vector</a> of floating point type.
4456 They must have identical types.</p>
4458 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4459 according to the condition code given as <tt>cond</tt>.
4460 If the operands are vectors, then the vectors are compared
4462 Each comparison performed
4463 always yields an <a href="#t_primitive">i1</a> result, as follows:</p>
4465 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
4466 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4467 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
4468 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4469 <tt>op1</tt> is greather than <tt>op2</tt>.</li>
4470 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4471 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4472 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4473 <tt>op1</tt> is less than <tt>op2</tt>.</li>
4474 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4475 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4476 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4477 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
4478 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
4479 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
4480 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
4481 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
4482 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4483 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
4484 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4485 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
4486 <tt>op1</tt> is less than <tt>op2</tt>.</li>
4487 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
4488 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4489 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
4490 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
4491 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
4492 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
4496 <pre> <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
4497 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
4498 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
4499 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
4502 <p>Note that the code generator does not yet support vector types with
4503 the <tt>fcmp</tt> instruction.</p>
4507 <!-- _______________________________________________________________________ -->
4508 <div class="doc_subsubsection">
4509 <a name="i_vicmp">'<tt>vicmp</tt>' Instruction</a>
4511 <div class="doc_text">
4513 <pre> <result> = vicmp <cond> <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4516 <p>The '<tt>vicmp</tt>' instruction returns an integer vector value based on
4517 element-wise comparison of its two integer vector operands.</p>
4519 <p>The '<tt>vicmp</tt>' instruction takes three operands. The first operand is
4520 the condition code indicating the kind of comparison to perform. It is not
4521 a value, just a keyword. The possible condition code are:</p>
4523 <li><tt>eq</tt>: equal</li>
4524 <li><tt>ne</tt>: not equal </li>
4525 <li><tt>ugt</tt>: unsigned greater than</li>
4526 <li><tt>uge</tt>: unsigned greater or equal</li>
4527 <li><tt>ult</tt>: unsigned less than</li>
4528 <li><tt>ule</tt>: unsigned less or equal</li>
4529 <li><tt>sgt</tt>: signed greater than</li>
4530 <li><tt>sge</tt>: signed greater or equal</li>
4531 <li><tt>slt</tt>: signed less than</li>
4532 <li><tt>sle</tt>: signed less or equal</li>
4534 <p>The remaining two arguments must be <a href="#t_vector">vector</a> or
4535 <a href="#t_integer">integer</a> typed. They must also be identical types.</p>
4537 <p>The '<tt>vicmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4538 according to the condition code given as <tt>cond</tt>. The comparison yields a
4539 <a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, of
4540 identical type as the values being compared. The most significant bit in each
4541 element is 1 if the element-wise comparison evaluates to true, and is 0
4542 otherwise. All other bits of the result are undefined. The condition codes
4543 are evaluated identically to the <a href="#i_icmp">'<tt>icmp</tt>'
4544 instruction</a>.</p>
4548 <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>
4549 <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>
4553 <!-- _______________________________________________________________________ -->
4554 <div class="doc_subsubsection">
4555 <a name="i_vfcmp">'<tt>vfcmp</tt>' Instruction</a>
4557 <div class="doc_text">
4559 <pre> <result> = vfcmp <cond> <ty> <op1>, <op2></pre>
4561 <p>The '<tt>vfcmp</tt>' instruction returns an integer vector value based on
4562 element-wise comparison of its two floating point vector operands. The output
4563 elements have the same width as the input elements.</p>
4565 <p>The '<tt>vfcmp</tt>' instruction takes three operands. The first operand is
4566 the condition code indicating the kind of comparison to perform. It is not
4567 a value, just a keyword. The possible condition code are:</p>
4569 <li><tt>false</tt>: no comparison, always returns false</li>
4570 <li><tt>oeq</tt>: ordered and equal</li>
4571 <li><tt>ogt</tt>: ordered and greater than </li>
4572 <li><tt>oge</tt>: ordered and greater than or equal</li>
4573 <li><tt>olt</tt>: ordered and less than </li>
4574 <li><tt>ole</tt>: ordered and less than or equal</li>
4575 <li><tt>one</tt>: ordered and not equal</li>
4576 <li><tt>ord</tt>: ordered (no nans)</li>
4577 <li><tt>ueq</tt>: unordered or equal</li>
4578 <li><tt>ugt</tt>: unordered or greater than </li>
4579 <li><tt>uge</tt>: unordered or greater than or equal</li>
4580 <li><tt>ult</tt>: unordered or less than </li>
4581 <li><tt>ule</tt>: unordered or less than or equal</li>
4582 <li><tt>une</tt>: unordered or not equal</li>
4583 <li><tt>uno</tt>: unordered (either nans)</li>
4584 <li><tt>true</tt>: no comparison, always returns true</li>
4586 <p>The remaining two arguments must be <a href="#t_vector">vector</a> of
4587 <a href="#t_floating">floating point</a> typed. They must also be identical
4590 <p>The '<tt>vfcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4591 according to the condition code given as <tt>cond</tt>. The comparison yields a
4592 <a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, with
4593 an identical number of elements as the values being compared, and each element
4594 having identical with to the width of the floating point elements. The most
4595 significant bit in each element is 1 if the element-wise comparison evaluates to
4596 true, and is 0 otherwise. All other bits of the result are undefined. The
4597 condition codes are evaluated identically to the
4598 <a href="#i_fcmp">'<tt>fcmp</tt>' instruction</a>.</p>
4602 <i>; yields: result=<2 x i32> < i32 0, i32 -1 ></i>
4603 <result> = vfcmp oeq <2 x float> < float 4, float 0 >, < float 5, float 0 >
4605 <i>; yields: result=<2 x i64> < i64 -1, i64 0 ></i>
4606 <result> = vfcmp ult <2 x double> < double 1, double 2 >, < double 2, double 2>
4610 <!-- _______________________________________________________________________ -->
4611 <div class="doc_subsubsection">
4612 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
4615 <div class="doc_text">
4619 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
4621 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
4622 the SSA graph representing the function.</p>
4625 <p>The type of the incoming values is specified with the first type
4626 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
4627 as arguments, with one pair for each predecessor basic block of the
4628 current block. Only values of <a href="#t_firstclass">first class</a>
4629 type may be used as the value arguments to the PHI node. Only labels
4630 may be used as the label arguments.</p>
4632 <p>There must be no non-phi instructions between the start of a basic
4633 block and the PHI instructions: i.e. PHI instructions must be first in
4636 <p>For the purposes of the SSA form, the use of each incoming value is
4637 deemed to occur on the edge from the corresponding predecessor block
4638 to the current block (but after any definition of an '<tt>invoke</tt>'
4639 instruction's return value on the same edge).</p>
4643 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
4644 specified by the pair corresponding to the predecessor basic block that executed
4645 just prior to the current block.</p>
4649 Loop: ; Infinite loop that counts from 0 on up...
4650 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
4651 %nextindvar = add i32 %indvar, 1
4656 <!-- _______________________________________________________________________ -->
4657 <div class="doc_subsubsection">
4658 <a name="i_select">'<tt>select</tt>' Instruction</a>
4661 <div class="doc_text">
4666 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
4668 <i>selty</i> is either i1 or {<N x i1>}
4674 The '<tt>select</tt>' instruction is used to choose one value based on a
4675 condition, without branching.
4682 The '<tt>select</tt>' instruction requires an 'i1' value or
4683 a vector of 'i1' values indicating the
4684 condition, and two values of the same <a href="#t_firstclass">first class</a>
4685 type. If the val1/val2 are vectors and
4686 the condition is a scalar, then entire vectors are selected, not
4687 individual elements.
4693 If the condition is an i1 and it evaluates to 1, the instruction returns the first
4694 value argument; otherwise, it returns the second value argument.
4697 If the condition is a vector of i1, then the value arguments must
4698 be vectors of the same size, and the selection is done element
4705 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
4708 <p>Note that the code generator does not yet support conditions
4709 with vector type.</p>
4714 <!-- _______________________________________________________________________ -->
4715 <div class="doc_subsubsection">
4716 <a name="i_call">'<tt>call</tt>' Instruction</a>
4719 <div class="doc_text">
4723 <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>]
4728 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
4732 <p>This instruction requires several arguments:</p>
4736 <p>The optional "tail" marker indicates whether the callee function accesses
4737 any allocas or varargs in the caller. If the "tail" marker is present, the
4738 function call is eligible for tail call optimization. Note that calls may
4739 be marked "tail" even if they do not occur before a <a
4740 href="#i_ret"><tt>ret</tt></a> instruction.</p>
4743 <p>The optional "cconv" marker indicates which <a href="#callingconv">calling
4744 convention</a> the call should use. If none is specified, the call defaults
4745 to using C calling conventions.</p>
4749 <p>The optional <a href="#paramattrs">Parameter Attributes</a> list for
4750 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>',
4751 and '<tt>inreg</tt>' attributes are valid here.</p>
4755 <p>'<tt>ty</tt>': the type of the call instruction itself which is also
4756 the type of the return value. Functions that return no value are marked
4757 <tt><a href="#t_void">void</a></tt>.</p>
4760 <p>'<tt>fnty</tt>': shall be the signature of the pointer to function
4761 value being invoked. The argument types must match the types implied by
4762 this signature. This type can be omitted if the function is not varargs
4763 and if the function type does not return a pointer to a function.</p>
4766 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
4767 be invoked. In most cases, this is a direct function invocation, but
4768 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
4769 to function value.</p>
4772 <p>'<tt>function args</tt>': argument list whose types match the
4773 function signature argument types. All arguments must be of
4774 <a href="#t_firstclass">first class</a> type. If the function signature
4775 indicates the function accepts a variable number of arguments, the extra
4776 arguments can be specified.</p>
4779 <p>The optional <a href="#fnattrs">function attributes</a> list. Only
4780 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
4781 '<tt>readnone</tt>' attributes are valid here.</p>
4787 <p>The '<tt>call</tt>' instruction is used to cause control flow to
4788 transfer to a specified function, with its incoming arguments bound to
4789 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
4790 instruction in the called function, control flow continues with the
4791 instruction after the function call, and the return value of the
4792 function is bound to the result argument.</p>
4797 %retval = call i32 @test(i32 %argc)
4798 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42) <i>; yields i32</i>
4799 %X = tail call i32 @foo() <i>; yields i32</i>
4800 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
4801 call void %foo(i8 97 signext)
4803 %struct.A = type { i32, i8 }
4804 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
4805 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
4806 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
4807 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
4808 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
4813 <!-- _______________________________________________________________________ -->
4814 <div class="doc_subsubsection">
4815 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
4818 <div class="doc_text">
4823 <resultval> = va_arg <va_list*> <arglist>, <argty>
4828 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
4829 the "variable argument" area of a function call. It is used to implement the
4830 <tt>va_arg</tt> macro in C.</p>
4834 <p>This instruction takes a <tt>va_list*</tt> value and the type of
4835 the argument. It returns a value of the specified argument type and
4836 increments the <tt>va_list</tt> to point to the next argument. The
4837 actual type of <tt>va_list</tt> is target specific.</p>
4841 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
4842 type from the specified <tt>va_list</tt> and causes the
4843 <tt>va_list</tt> to point to the next argument. For more information,
4844 see the variable argument handling <a href="#int_varargs">Intrinsic
4847 <p>It is legal for this instruction to be called in a function which does not
4848 take a variable number of arguments, for example, the <tt>vfprintf</tt>
4851 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
4852 href="#intrinsics">intrinsic function</a> because it takes a type as an
4857 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
4859 <p>Note that the code generator does not yet fully support va_arg
4860 on many targets. Also, it does not currently support va_arg with
4861 aggregate types on any target.</p>
4865 <!-- *********************************************************************** -->
4866 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
4867 <!-- *********************************************************************** -->
4869 <div class="doc_text">
4871 <p>LLVM supports the notion of an "intrinsic function". These functions have
4872 well known names and semantics and are required to follow certain restrictions.
4873 Overall, these intrinsics represent an extension mechanism for the LLVM
4874 language that does not require changing all of the transformations in LLVM when
4875 adding to the language (or the bitcode reader/writer, the parser, etc...).</p>
4877 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
4878 prefix is reserved in LLVM for intrinsic names; thus, function names may not
4879 begin with this prefix. Intrinsic functions must always be external functions:
4880 you cannot define the body of intrinsic functions. Intrinsic functions may
4881 only be used in call or invoke instructions: it is illegal to take the address
4882 of an intrinsic function. Additionally, because intrinsic functions are part
4883 of the LLVM language, it is required if any are added that they be documented
4886 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
4887 a family of functions that perform the same operation but on different data
4888 types. Because LLVM can represent over 8 million different integer types,
4889 overloading is used commonly to allow an intrinsic function to operate on any
4890 integer type. One or more of the argument types or the result type can be
4891 overloaded to accept any integer type. Argument types may also be defined as
4892 exactly matching a previous argument's type or the result type. This allows an
4893 intrinsic function which accepts multiple arguments, but needs all of them to
4894 be of the same type, to only be overloaded with respect to a single argument or
4897 <p>Overloaded intrinsics will have the names of its overloaded argument types
4898 encoded into its function name, each preceded by a period. Only those types
4899 which are overloaded result in a name suffix. Arguments whose type is matched
4900 against another type do not. For example, the <tt>llvm.ctpop</tt> function can
4901 take an integer of any width and returns an integer of exactly the same integer
4902 width. This leads to a family of functions such as
4903 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29 %val)</tt>.
4904 Only one type, the return type, is overloaded, and only one type suffix is
4905 required. Because the argument's type is matched against the return type, it
4906 does not require its own name suffix.</p>
4908 <p>To learn how to add an intrinsic function, please see the
4909 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.
4914 <!-- ======================================================================= -->
4915 <div class="doc_subsection">
4916 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
4919 <div class="doc_text">
4921 <p>Variable argument support is defined in LLVM with the <a
4922 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
4923 intrinsic functions. These functions are related to the similarly
4924 named macros defined in the <tt><stdarg.h></tt> header file.</p>
4926 <p>All of these functions operate on arguments that use a
4927 target-specific value type "<tt>va_list</tt>". The LLVM assembly
4928 language reference manual does not define what this type is, so all
4929 transformations should be prepared to handle these functions regardless of
4932 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
4933 instruction and the variable argument handling intrinsic functions are
4936 <div class="doc_code">
4938 define i32 @test(i32 %X, ...) {
4939 ; Initialize variable argument processing
4941 %ap2 = bitcast i8** %ap to i8*
4942 call void @llvm.va_start(i8* %ap2)
4944 ; Read a single integer argument
4945 %tmp = va_arg i8** %ap, i32
4947 ; Demonstrate usage of llvm.va_copy and llvm.va_end
4949 %aq2 = bitcast i8** %aq to i8*
4950 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
4951 call void @llvm.va_end(i8* %aq2)
4953 ; Stop processing of arguments.
4954 call void @llvm.va_end(i8* %ap2)
4958 declare void @llvm.va_start(i8*)
4959 declare void @llvm.va_copy(i8*, i8*)
4960 declare void @llvm.va_end(i8*)
4966 <!-- _______________________________________________________________________ -->
4967 <div class="doc_subsubsection">
4968 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
4972 <div class="doc_text">
4974 <pre> declare void %llvm.va_start(i8* <arglist>)<br></pre>
4976 <p>The '<tt>llvm.va_start</tt>' intrinsic initializes
4977 <tt>*<arglist></tt> for subsequent use by <tt><a
4978 href="#i_va_arg">va_arg</a></tt>.</p>
4982 <p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
4986 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
4987 macro available in C. In a target-dependent way, it initializes the
4988 <tt>va_list</tt> element to which the argument points, so that the next call to
4989 <tt>va_arg</tt> will produce the first variable argument passed to the function.
4990 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
4991 last argument of the function as the compiler can figure that out.</p>
4995 <!-- _______________________________________________________________________ -->
4996 <div class="doc_subsubsection">
4997 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
5000 <div class="doc_text">
5002 <pre> declare void @llvm.va_end(i8* <arglist>)<br></pre>
5005 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
5006 which has been initialized previously with <tt><a href="#int_va_start">llvm.va_start</a></tt>
5007 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
5011 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
5015 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
5016 macro available in C. In a target-dependent way, it destroys the
5017 <tt>va_list</tt> element to which the argument points. Calls to <a
5018 href="#int_va_start"><tt>llvm.va_start</tt></a> and <a href="#int_va_copy">
5019 <tt>llvm.va_copy</tt></a> must be matched exactly with calls to
5020 <tt>llvm.va_end</tt>.</p>
5024 <!-- _______________________________________________________________________ -->
5025 <div class="doc_subsubsection">
5026 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
5029 <div class="doc_text">
5034 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
5039 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
5040 from the source argument list to the destination argument list.</p>
5044 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
5045 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
5050 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
5051 macro available in C. In a target-dependent way, it copies the source
5052 <tt>va_list</tt> element into the destination <tt>va_list</tt> element. This
5053 intrinsic is necessary because the <tt><a href="#int_va_start">
5054 llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
5055 example, memory allocation.</p>
5059 <!-- ======================================================================= -->
5060 <div class="doc_subsection">
5061 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
5064 <div class="doc_text">
5067 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
5068 Collection</a> (GC) requires the implementation and generation of these
5070 These intrinsics allow identification of <a href="#int_gcroot">GC roots on the
5071 stack</a>, as well as garbage collector implementations that require <a
5072 href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a> barriers.
5073 Front-ends for type-safe garbage collected languages should generate these
5074 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
5075 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
5078 <p>The garbage collection intrinsics only operate on objects in the generic
5079 address space (address space zero).</p>
5083 <!-- _______________________________________________________________________ -->
5084 <div class="doc_subsubsection">
5085 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
5088 <div class="doc_text">
5093 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
5098 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
5099 the code generator, and allows some metadata to be associated with it.</p>
5103 <p>The first argument specifies the address of a stack object that contains the
5104 root pointer. The second pointer (which must be either a constant or a global
5105 value address) contains the meta-data to be associated with the root.</p>
5109 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
5110 location. At compile-time, the code generator generates information to allow
5111 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
5112 intrinsic may only be used in a function which <a href="#gc">specifies a GC
5118 <!-- _______________________________________________________________________ -->
5119 <div class="doc_subsubsection">
5120 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
5123 <div class="doc_text">
5128 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
5133 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
5134 locations, allowing garbage collector implementations that require read
5139 <p>The second argument is the address to read from, which should be an address
5140 allocated from the garbage collector. The first object is a pointer to the
5141 start of the referenced object, if needed by the language runtime (otherwise
5146 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
5147 instruction, but may be replaced with substantially more complex code by the
5148 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
5149 may only be used in a function which <a href="#gc">specifies a GC
5155 <!-- _______________________________________________________________________ -->
5156 <div class="doc_subsubsection">
5157 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
5160 <div class="doc_text">
5165 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
5170 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
5171 locations, allowing garbage collector implementations that require write
5172 barriers (such as generational or reference counting collectors).</p>
5176 <p>The first argument is the reference to store, the second is the start of the
5177 object to store it to, and the third is the address of the field of Obj to
5178 store to. If the runtime does not require a pointer to the object, Obj may be
5183 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
5184 instruction, but may be replaced with substantially more complex code by the
5185 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
5186 may only be used in a function which <a href="#gc">specifies a GC
5193 <!-- ======================================================================= -->
5194 <div class="doc_subsection">
5195 <a name="int_codegen">Code Generator Intrinsics</a>
5198 <div class="doc_text">
5200 These intrinsics are provided by LLVM to expose special features that may only
5201 be implemented with code generator support.
5206 <!-- _______________________________________________________________________ -->
5207 <div class="doc_subsubsection">
5208 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
5211 <div class="doc_text">
5215 declare i8 *@llvm.returnaddress(i32 <level>)
5221 The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
5222 target-specific value indicating the return address of the current function
5223 or one of its callers.
5229 The argument to this intrinsic indicates which function to return the address
5230 for. Zero indicates the calling function, one indicates its caller, etc. The
5231 argument is <b>required</b> to be a constant integer value.
5237 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
5238 the return address of the specified call frame, or zero if it cannot be
5239 identified. The value returned by this intrinsic is likely to be incorrect or 0
5240 for arguments other than zero, so it should only be used for debugging purposes.
5244 Note that calling this intrinsic does not prevent function inlining or other
5245 aggressive transformations, so the value returned may not be that of the obvious
5246 source-language caller.
5251 <!-- _______________________________________________________________________ -->
5252 <div class="doc_subsubsection">
5253 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
5256 <div class="doc_text">
5260 declare i8 *@llvm.frameaddress(i32 <level>)
5266 The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
5267 target-specific frame pointer value for the specified stack frame.
5273 The argument to this intrinsic indicates which function to return the frame
5274 pointer for. Zero indicates the calling function, one indicates its caller,
5275 etc. The argument is <b>required</b> to be a constant integer value.
5281 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
5282 the frame address of the specified call frame, or zero if it cannot be
5283 identified. The value returned by this intrinsic is likely to be incorrect or 0
5284 for arguments other than zero, so it should only be used for debugging purposes.
5288 Note that calling this intrinsic does not prevent function inlining or other
5289 aggressive transformations, so the value returned may not be that of the obvious
5290 source-language caller.
5294 <!-- _______________________________________________________________________ -->
5295 <div class="doc_subsubsection">
5296 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
5299 <div class="doc_text">
5303 declare i8 *@llvm.stacksave()
5309 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
5310 the function stack, for use with <a href="#int_stackrestore">
5311 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
5312 features like scoped automatic variable sized arrays in C99.
5318 This intrinsic returns a opaque pointer value that can be passed to <a
5319 href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
5320 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
5321 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
5322 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
5323 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
5324 that were allocated after the <tt>llvm.stacksave</tt> was executed.
5329 <!-- _______________________________________________________________________ -->
5330 <div class="doc_subsubsection">
5331 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
5334 <div class="doc_text">
5338 declare void @llvm.stackrestore(i8 * %ptr)
5344 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
5345 the function stack to the state it was in when the corresponding <a
5346 href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
5347 useful for implementing language features like scoped automatic variable sized
5354 See the description for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.
5360 <!-- _______________________________________________________________________ -->
5361 <div class="doc_subsubsection">
5362 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
5365 <div class="doc_text">
5369 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>)
5376 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
5377 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
5379 effect on the behavior of the program but can change its performance
5386 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
5387 determining if the fetch should be for a read (0) or write (1), and
5388 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
5389 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
5390 <tt>locality</tt> arguments must be constant integers.
5396 This intrinsic does not modify the behavior of the program. In particular,
5397 prefetches cannot trap and do not produce a value. On targets that support this
5398 intrinsic, the prefetch can provide hints to the processor cache for better
5404 <!-- _______________________________________________________________________ -->
5405 <div class="doc_subsubsection">
5406 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
5409 <div class="doc_text">
5413 declare void @llvm.pcmarker(i32 <id>)
5420 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
5422 code to simulators and other tools. The method is target specific, but it is
5423 expected that the marker will use exported symbols to transmit the PC of the
5425 The marker makes no guarantees that it will remain with any specific instruction
5426 after optimizations. It is possible that the presence of a marker will inhibit
5427 optimizations. The intended use is to be inserted after optimizations to allow
5428 correlations of simulation runs.
5434 <tt>id</tt> is a numerical id identifying the marker.
5440 This intrinsic does not modify the behavior of the program. Backends that do not
5441 support this intrinisic may ignore it.
5446 <!-- _______________________________________________________________________ -->
5447 <div class="doc_subsubsection">
5448 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
5451 <div class="doc_text">
5455 declare i64 @llvm.readcyclecounter( )
5462 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
5463 counter register (or similar low latency, high accuracy clocks) on those targets
5464 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
5465 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
5466 should only be used for small timings.
5472 When directly supported, reading the cycle counter should not modify any memory.
5473 Implementations are allowed to either return a application specific value or a
5474 system wide value. On backends without support, this is lowered to a constant 0.
5479 <!-- ======================================================================= -->
5480 <div class="doc_subsection">
5481 <a name="int_libc">Standard C Library Intrinsics</a>
5484 <div class="doc_text">
5486 LLVM provides intrinsics for a few important standard C library functions.
5487 These intrinsics allow source-language front-ends to pass information about the
5488 alignment of the pointer arguments to the code generator, providing opportunity
5489 for more efficient code generation.
5494 <!-- _______________________________________________________________________ -->
5495 <div class="doc_subsubsection">
5496 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
5499 <div class="doc_text">
5502 <p>This is an overloaded intrinsic. You can use llvm.memcpy on any integer bit
5503 width. Not all targets support all bit widths however.</p>
5505 declare void @llvm.memcpy.i8(i8 * <dest>, i8 * <src>,
5506 i8 <len>, i32 <align>)
5507 declare void @llvm.memcpy.i16(i8 * <dest>, i8 * <src>,
5508 i16 <len>, i32 <align>)
5509 declare void @llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
5510 i32 <len>, i32 <align>)
5511 declare void @llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
5512 i64 <len>, i32 <align>)
5518 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
5519 location to the destination location.
5523 Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
5524 intrinsics do not return a value, and takes an extra alignment argument.
5530 The first argument is a pointer to the destination, the second is a pointer to
5531 the source. The third argument is an integer argument
5532 specifying the number of bytes to copy, and the fourth argument is the alignment
5533 of the source and destination locations.
5537 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5538 the caller guarantees that both the source and destination pointers are aligned
5545 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
5546 location to the destination location, which are not allowed to overlap. It
5547 copies "len" bytes of memory over. If the argument is known to be aligned to
5548 some boundary, this can be specified as the fourth argument, otherwise it should
5554 <!-- _______________________________________________________________________ -->
5555 <div class="doc_subsubsection">
5556 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
5559 <div class="doc_text">
5562 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
5563 width. Not all targets support all bit widths however.</p>
5565 declare void @llvm.memmove.i8(i8 * <dest>, i8 * <src>,
5566 i8 <len>, i32 <align>)
5567 declare void @llvm.memmove.i16(i8 * <dest>, i8 * <src>,
5568 i16 <len>, i32 <align>)
5569 declare void @llvm.memmove.i32(i8 * <dest>, i8 * <src>,
5570 i32 <len>, i32 <align>)
5571 declare void @llvm.memmove.i64(i8 * <dest>, i8 * <src>,
5572 i64 <len>, i32 <align>)
5578 The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
5579 location to the destination location. It is similar to the
5580 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to overlap.
5584 Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
5585 intrinsics do not return a value, and takes an extra alignment argument.
5591 The first argument is a pointer to the destination, the second is a pointer to
5592 the source. The third argument is an integer argument
5593 specifying the number of bytes to copy, and the fourth argument is the alignment
5594 of the source and destination locations.
5598 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5599 the caller guarantees that the source and destination pointers are aligned to
5606 The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
5607 location to the destination location, which may overlap. It
5608 copies "len" bytes of memory over. If the argument is known to be aligned to
5609 some boundary, this can be specified as the fourth argument, otherwise it should
5615 <!-- _______________________________________________________________________ -->
5616 <div class="doc_subsubsection">
5617 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
5620 <div class="doc_text">
5623 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
5624 width. Not all targets support all bit widths however.</p>
5626 declare void @llvm.memset.i8(i8 * <dest>, i8 <val>,
5627 i8 <len>, i32 <align>)
5628 declare void @llvm.memset.i16(i8 * <dest>, i8 <val>,
5629 i16 <len>, i32 <align>)
5630 declare void @llvm.memset.i32(i8 * <dest>, i8 <val>,
5631 i32 <len>, i32 <align>)
5632 declare void @llvm.memset.i64(i8 * <dest>, i8 <val>,
5633 i64 <len>, i32 <align>)
5639 The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
5644 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
5645 does not return a value, and takes an extra alignment argument.
5651 The first argument is a pointer to the destination to fill, the second is the
5652 byte value to fill it with, the third argument is an integer
5653 argument specifying the number of bytes to fill, and the fourth argument is the
5654 known alignment of destination location.
5658 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5659 the caller guarantees that the destination pointer is aligned to that boundary.
5665 The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
5667 destination location. If the argument is known to be aligned to some boundary,
5668 this can be specified as the fourth argument, otherwise it should be set to 0 or
5674 <!-- _______________________________________________________________________ -->
5675 <div class="doc_subsubsection">
5676 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
5679 <div class="doc_text">
5682 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
5683 floating point or vector of floating point type. Not all targets support all
5686 declare float @llvm.sqrt.f32(float %Val)
5687 declare double @llvm.sqrt.f64(double %Val)
5688 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
5689 declare fp128 @llvm.sqrt.f128(fp128 %Val)
5690 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
5696 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
5697 returning the same value as the libm '<tt>sqrt</tt>' functions would. Unlike
5698 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
5699 negative numbers other than -0.0 (which allows for better optimization, because
5700 there is no need to worry about errno being set). <tt>llvm.sqrt(-0.0)</tt> is
5701 defined to return -0.0 like IEEE sqrt.
5707 The argument and return value are floating point numbers of the same type.
5713 This function returns the sqrt of the specified operand if it is a nonnegative
5714 floating point number.
5718 <!-- _______________________________________________________________________ -->
5719 <div class="doc_subsubsection">
5720 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
5723 <div class="doc_text">
5726 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
5727 floating point or vector of floating point type. Not all targets support all
5730 declare float @llvm.powi.f32(float %Val, i32 %power)
5731 declare double @llvm.powi.f64(double %Val, i32 %power)
5732 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
5733 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
5734 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
5740 The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
5741 specified (positive or negative) power. The order of evaluation of
5742 multiplications is not defined. When a vector of floating point type is
5743 used, the second argument remains a scalar integer value.
5749 The second argument is an integer power, and the first is a value to raise to
5756 This function returns the first value raised to the second power with an
5757 unspecified sequence of rounding operations.</p>
5760 <!-- _______________________________________________________________________ -->
5761 <div class="doc_subsubsection">
5762 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
5765 <div class="doc_text">
5768 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
5769 floating point or vector of floating point type. Not all targets support all
5772 declare float @llvm.sin.f32(float %Val)
5773 declare double @llvm.sin.f64(double %Val)
5774 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
5775 declare fp128 @llvm.sin.f128(fp128 %Val)
5776 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
5782 The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.
5788 The argument and return value are floating point numbers of the same type.
5794 This function returns the sine of the specified operand, returning the
5795 same values as the libm <tt>sin</tt> functions would, and handles error
5796 conditions in the same way.</p>
5799 <!-- _______________________________________________________________________ -->
5800 <div class="doc_subsubsection">
5801 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
5804 <div class="doc_text">
5807 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
5808 floating point or vector of floating point type. Not all targets support all
5811 declare float @llvm.cos.f32(float %Val)
5812 declare double @llvm.cos.f64(double %Val)
5813 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
5814 declare fp128 @llvm.cos.f128(fp128 %Val)
5815 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
5821 The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.
5827 The argument and return value are floating point numbers of the same type.
5833 This function returns the cosine of the specified operand, returning the
5834 same values as the libm <tt>cos</tt> functions would, and handles error
5835 conditions in the same way.</p>
5838 <!-- _______________________________________________________________________ -->
5839 <div class="doc_subsubsection">
5840 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
5843 <div class="doc_text">
5846 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
5847 floating point or vector of floating point type. Not all targets support all
5850 declare float @llvm.pow.f32(float %Val, float %Power)
5851 declare double @llvm.pow.f64(double %Val, double %Power)
5852 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
5853 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
5854 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
5860 The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
5861 specified (positive or negative) power.
5867 The second argument is a floating point power, and the first is a value to
5868 raise to that power.
5874 This function returns the first value raised to the second power,
5876 same values as the libm <tt>pow</tt> functions would, and handles error
5877 conditions in the same way.</p>
5881 <!-- ======================================================================= -->
5882 <div class="doc_subsection">
5883 <a name="int_manip">Bit Manipulation Intrinsics</a>
5886 <div class="doc_text">
5888 LLVM provides intrinsics for a few important bit manipulation operations.
5889 These allow efficient code generation for some algorithms.
5894 <!-- _______________________________________________________________________ -->
5895 <div class="doc_subsubsection">
5896 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
5899 <div class="doc_text">
5902 <p>This is an overloaded intrinsic function. You can use bswap on any integer
5903 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
5905 declare i16 @llvm.bswap.i16(i16 <id>)
5906 declare i32 @llvm.bswap.i32(i32 <id>)
5907 declare i64 @llvm.bswap.i64(i64 <id>)
5913 The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
5914 values with an even number of bytes (positive multiple of 16 bits). These are
5915 useful for performing operations on data that is not in the target's native
5922 The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
5923 and low byte of the input i16 swapped. Similarly, the <tt>llvm.bswap.i32</tt>
5924 intrinsic returns an i32 value that has the four bytes of the input i32
5925 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the returned
5926 i32 will have its bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i48</tt>,
5927 <tt>llvm.bswap.i64</tt> and other intrinsics extend this concept to
5928 additional even-byte lengths (6 bytes, 8 bytes and more, respectively).
5933 <!-- _______________________________________________________________________ -->
5934 <div class="doc_subsubsection">
5935 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
5938 <div class="doc_text">
5941 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
5942 width. Not all targets support all bit widths however.</p>
5944 declare i8 @llvm.ctpop.i8(i8 <src>)
5945 declare i16 @llvm.ctpop.i16(i16 <src>)
5946 declare i32 @llvm.ctpop.i32(i32 <src>)
5947 declare i64 @llvm.ctpop.i64(i64 <src>)
5948 declare i256 @llvm.ctpop.i256(i256 <src>)
5954 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
5961 The only argument is the value to be counted. The argument may be of any
5962 integer type. The return type must match the argument type.
5968 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
5972 <!-- _______________________________________________________________________ -->
5973 <div class="doc_subsubsection">
5974 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
5977 <div class="doc_text">
5980 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
5981 integer bit width. Not all targets support all bit widths however.</p>
5983 declare i8 @llvm.ctlz.i8 (i8 <src>)
5984 declare i16 @llvm.ctlz.i16(i16 <src>)
5985 declare i32 @llvm.ctlz.i32(i32 <src>)
5986 declare i64 @llvm.ctlz.i64(i64 <src>)
5987 declare i256 @llvm.ctlz.i256(i256 <src>)
5993 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
5994 leading zeros in a variable.
6000 The only argument is the value to be counted. The argument may be of any
6001 integer type. The return type must match the argument type.
6007 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
6008 in a variable. If the src == 0 then the result is the size in bits of the type
6009 of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.
6015 <!-- _______________________________________________________________________ -->
6016 <div class="doc_subsubsection">
6017 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
6020 <div class="doc_text">
6023 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
6024 integer bit width. Not all targets support all bit widths however.</p>
6026 declare i8 @llvm.cttz.i8 (i8 <src>)
6027 declare i16 @llvm.cttz.i16(i16 <src>)
6028 declare i32 @llvm.cttz.i32(i32 <src>)
6029 declare i64 @llvm.cttz.i64(i64 <src>)
6030 declare i256 @llvm.cttz.i256(i256 <src>)
6036 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
6043 The only argument is the value to be counted. The argument may be of any
6044 integer type. The return type must match the argument type.
6050 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
6051 in a variable. If the src == 0 then the result is the size in bits of the type
6052 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
6056 <!-- _______________________________________________________________________ -->
6057 <div class="doc_subsubsection">
6058 <a name="int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic</a>
6061 <div class="doc_text">
6064 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.select</tt>
6065 on any integer bit width.</p>
6067 declare i17 @llvm.part.select.i17 (i17 %val, i32 %loBit, i32 %hiBit)
6068 declare i29 @llvm.part.select.i29 (i29 %val, i32 %loBit, i32 %hiBit)
6072 <p>The '<tt>llvm.part.select</tt>' family of intrinsic functions selects a
6073 range of bits from an integer value and returns them in the same bit width as
6074 the original value.</p>
6077 <p>The first argument, <tt>%val</tt> and the result may be integer types of
6078 any bit width but they must have the same bit width. The second and third
6079 arguments must be <tt>i32</tt> type since they specify only a bit index.</p>
6082 <p>The operation of the '<tt>llvm.part.select</tt>' intrinsic has two modes
6083 of operation: forwards and reverse. If <tt>%loBit</tt> is greater than
6084 <tt>%hiBits</tt> then the intrinsic operates in reverse mode. Otherwise it
6085 operates in forward mode.</p>
6086 <p>In forward mode, this intrinsic is the equivalent of shifting <tt>%val</tt>
6087 right by <tt>%loBit</tt> bits and then ANDing it with a mask with
6088 only the <tt>%hiBit - %loBit</tt> bits set, as follows:</p>
6090 <li>The <tt>%val</tt> is shifted right (LSHR) by the number of bits specified
6091 by <tt>%loBits</tt>. This normalizes the value to the low order bits.</li>
6092 <li>The <tt>%loBits</tt> value is subtracted from the <tt>%hiBits</tt> value
6093 to determine the number of bits to retain.</li>
6094 <li>A mask of the retained bits is created by shifting a -1 value.</li>
6095 <li>The mask is ANDed with <tt>%val</tt> to produce the result.</li>
6097 <p>In reverse mode, a similar computation is made except that the bits are
6098 returned in the reverse order. So, for example, if <tt>X</tt> has the value
6099 <tt>i16 0x0ACF (101011001111)</tt> and we apply
6100 <tt>part.select(i16 X, 8, 3)</tt> to it, we get back the value
6101 <tt>i16 0x0026 (000000100110)</tt>.</p>
6104 <div class="doc_subsubsection">
6105 <a name="int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic</a>
6108 <div class="doc_text">
6111 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.set</tt>
6112 on any integer bit width.</p>
6114 declare i17 @llvm.part.set.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
6115 declare i29 @llvm.part.set.i29.i9 (i29 %val, i9 %repl, i32 %lo, i32 %hi)
6119 <p>The '<tt>llvm.part.set</tt>' family of intrinsic functions replaces a range
6120 of bits in an integer value with another integer value. It returns the integer
6121 with the replaced bits.</p>
6124 <p>The first argument, <tt>%val</tt>, and the result may be integer types of
6125 any bit width, but they must have the same bit width. <tt>%val</tt> is the value
6126 whose bits will be replaced. The second argument, <tt>%repl</tt> may be an
6127 integer of any bit width. The third and fourth arguments must be <tt>i32</tt>
6128 type since they specify only a bit index.</p>
6131 <p>The operation of the '<tt>llvm.part.set</tt>' intrinsic has two modes
6132 of operation: forwards and reverse. If <tt>%lo</tt> is greater than
6133 <tt>%hi</tt> then the intrinsic operates in reverse mode. Otherwise it
6134 operates in forward mode.</p>
6136 <p>For both modes, the <tt>%repl</tt> value is prepared for use by either
6137 truncating it down to the size of the replacement area or zero extending it
6138 up to that size.</p>
6140 <p>In forward mode, the bits between <tt>%lo</tt> and <tt>%hi</tt> (inclusive)
6141 are replaced with corresponding bits from <tt>%repl</tt>. That is the 0th bit
6142 in <tt>%repl</tt> replaces the <tt>%lo</tt>th bit in <tt>%val</tt> and etc. up
6143 to the <tt>%hi</tt>th bit.</p>
6145 <p>In reverse mode, a similar computation is made except that the bits are
6146 reversed. That is, the <tt>0</tt>th bit in <tt>%repl</tt> replaces the
6147 <tt>%hi</tt> bit in <tt>%val</tt> and etc. down to the <tt>%lo</tt>th bit.</p>
6152 llvm.part.set(0xFFFF, 0, 4, 7) -> 0xFF0F
6153 llvm.part.set(0xFFFF, 0, 7, 4) -> 0xFF0F
6154 llvm.part.set(0xFFFF, 1, 7, 4) -> 0xFF8F
6155 llvm.part.set(0xFFFF, F, 8, 3) -> 0xFFE7
6156 llvm.part.set(0xFFFF, 0, 3, 8) -> 0xFE07
6161 <!-- ======================================================================= -->
6162 <div class="doc_subsection">
6163 <a name="int_overflow">Arithmetic with Overflow Intrinsics</a>
6166 <div class="doc_text">
6168 LLVM provides intrinsics for some arithmetic with overflow operations.
6173 <!-- _______________________________________________________________________ -->
6174 <div class="doc_subsubsection">
6175 <a name="int_sadd_overflow">'<tt>llvm.sadd.with.overflow.*</tt>' Intrinsics</a>
6178 <div class="doc_text">
6182 <p>This is an overloaded intrinsic. You can use <tt>llvm.sadd.with.overflow</tt>
6183 on any integer bit width.</p>
6186 declare {i16, i1} @llvm.sadd.with.overflow.i16(i16 %a, i16 %b)
6187 declare {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6188 declare {i64, i1} @llvm.sadd.with.overflow.i64(i64 %a, i64 %b)
6193 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
6194 a signed addition of the two arguments, and indicate whether an overflow
6195 occurred during the signed summation.</p>
6199 <p>The arguments (%a and %b) and the first element of the result structure may
6200 be of integer types of any bit width, but they must have the same bit width. The
6201 second element of the result structure must be of type <tt>i1</tt>. <tt>%a</tt>
6202 and <tt>%b</tt> are the two values that will undergo signed addition.</p>
6206 <p>The '<tt>llvm.sadd.with.overflow</tt>' family of intrinsic functions perform
6207 a signed addition of the two variables. They return a structure — the
6208 first element of which is the signed summation, and the second element of which
6209 is a bit specifying if the signed summation resulted in an overflow.</p>
6213 %res = call {i32, i1} @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
6214 %sum = extractvalue {i32, i1} %res, 0
6215 %obit = extractvalue {i32, i1} %res, 1
6216 br i1 %obit, label %overflow, label %normal
6221 <!-- _______________________________________________________________________ -->
6222 <div class="doc_subsubsection">
6223 <a name="int_uadd_overflow">'<tt>llvm.uadd.with.overflow.*</tt>' Intrinsics</a>
6226 <div class="doc_text">
6230 <p>This is an overloaded intrinsic. You can use <tt>llvm.uadd.with.overflow</tt>
6231 on any integer bit width.</p>
6234 declare {i16, i1} @llvm.uadd.with.overflow.i16(i16 %a, i16 %b)
6235 declare {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6236 declare {i64, i1} @llvm.uadd.with.overflow.i64(i64 %a, i64 %b)
6241 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
6242 an unsigned addition of the two arguments, and indicate whether a carry occurred
6243 during the unsigned summation.</p>
6247 <p>The arguments (%a and %b) and the first element of the result structure may
6248 be of integer types of any bit width, but they must have the same bit width. The
6249 second element of the result structure must be of type <tt>i1</tt>. <tt>%a</tt>
6250 and <tt>%b</tt> are the two values that will undergo unsigned addition.</p>
6254 <p>The '<tt>llvm.uadd.with.overflow</tt>' family of intrinsic functions perform
6255 an unsigned addition of the two arguments. They return a structure — the
6256 first element of which is the sum, and the second element of which is a bit
6257 specifying if the unsigned summation resulted in a carry.</p>
6261 %res = call {i32, i1} @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
6262 %sum = extractvalue {i32, i1} %res, 0
6263 %obit = extractvalue {i32, i1} %res, 1
6264 br i1 %obit, label %carry, label %normal
6269 <!-- _______________________________________________________________________ -->
6270 <div class="doc_subsubsection">
6271 <a name="int_ssub_overflow">'<tt>llvm.ssub.with.overflow.*</tt>' Intrinsics</a>
6274 <div class="doc_text">
6278 <p>This is an overloaded intrinsic. You can use <tt>llvm.ssub.with.overflow</tt>
6279 on any integer bit width.</p>
6282 declare {i16, i1} @llvm.ssub.with.overflow.i16(i16 %a, i16 %b)
6283 declare {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6284 declare {i64, i1} @llvm.ssub.with.overflow.i64(i64 %a, i64 %b)
6289 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
6290 a signed subtraction of the two arguments, and indicate whether an overflow
6291 occurred during the signed subtraction.</p>
6295 <p>The arguments (%a and %b) and the first element of the result structure may
6296 be of integer types of any bit width, but they must have the same bit width. The
6297 second element of the result structure must be of type <tt>i1</tt>. <tt>%a</tt>
6298 and <tt>%b</tt> are the two values that will undergo signed subtraction.</p>
6302 <p>The '<tt>llvm.ssub.with.overflow</tt>' family of intrinsic functions perform
6303 a signed subtraction of the two arguments. They return a structure — the
6304 first element of which is the subtraction, and the second element of which is a bit
6305 specifying if the signed subtraction resulted in an overflow.</p>
6309 %res = call {i32, i1} @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
6310 %sum = extractvalue {i32, i1} %res, 0
6311 %obit = extractvalue {i32, i1} %res, 1
6312 br i1 %obit, label %overflow, label %normal
6317 <!-- _______________________________________________________________________ -->
6318 <div class="doc_subsubsection">
6319 <a name="int_usub_overflow">'<tt>llvm.usub.with.overflow.*</tt>' Intrinsics</a>
6322 <div class="doc_text">
6326 <p>This is an overloaded intrinsic. You can use <tt>llvm.usub.with.overflow</tt>
6327 on any integer bit width.</p>
6330 declare {i16, i1} @llvm.usub.with.overflow.i16(i16 %a, i16 %b)
6331 declare {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6332 declare {i64, i1} @llvm.usub.with.overflow.i64(i64 %a, i64 %b)
6337 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
6338 an unsigned subtraction of the two arguments, and indicate whether an overflow
6339 occurred during the unsigned subtraction.</p>
6343 <p>The arguments (%a and %b) and the first element of the result structure may
6344 be of integer types of any bit width, but they must have the same bit width. The
6345 second element of the result structure must be of type <tt>i1</tt>. <tt>%a</tt>
6346 and <tt>%b</tt> are the two values that will undergo unsigned subtraction.</p>
6350 <p>The '<tt>llvm.usub.with.overflow</tt>' family of intrinsic functions perform
6351 an unsigned subtraction of the two arguments. They return a structure — the
6352 first element of which is the subtraction, and the second element of which is a bit
6353 specifying if the unsigned subtraction resulted in an overflow.</p>
6357 %res = call {i32, i1} @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
6358 %sum = extractvalue {i32, i1} %res, 0
6359 %obit = extractvalue {i32, i1} %res, 1
6360 br i1 %obit, label %overflow, label %normal
6365 <!-- _______________________________________________________________________ -->
6366 <div class="doc_subsubsection">
6367 <a name="int_smul_overflow">'<tt>llvm.smul.with.overflow.*</tt>' Intrinsics</a>
6370 <div class="doc_text">
6374 <p>This is an overloaded intrinsic. You can use <tt>llvm.smul.with.overflow</tt>
6375 on any integer bit width.</p>
6378 declare {i16, i1} @llvm.smul.with.overflow.i16(i16 %a, i16 %b)
6379 declare {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6380 declare {i64, i1} @llvm.smul.with.overflow.i64(i64 %a, i64 %b)
6385 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
6386 a signed multiplication of the two arguments, and indicate whether an overflow
6387 occurred during the signed multiplication.</p>
6391 <p>The arguments (%a and %b) and the first element of the result structure may
6392 be of integer types of any bit width, but they must have the same bit width. The
6393 second element of the result structure must be of type <tt>i1</tt>. <tt>%a</tt>
6394 and <tt>%b</tt> are the two values that will undergo signed multiplication.</p>
6398 <p>The '<tt>llvm.smul.with.overflow</tt>' family of intrinsic functions perform
6399 a signed multiplication of the two arguments. They return a structure —
6400 the first element of which is the multiplication, and the second element of
6401 which is a bit specifying if the signed multiplication resulted in an
6406 %res = call {i32, i1} @llvm.smul.with.overflow.i32(i32 %a, i32 %b)
6407 %sum = extractvalue {i32, i1} %res, 0
6408 %obit = extractvalue {i32, i1} %res, 1
6409 br i1 %obit, label %overflow, label %normal
6414 <!-- _______________________________________________________________________ -->
6415 <div class="doc_subsubsection">
6416 <a name="int_umul_overflow">'<tt>llvm.umul.with.overflow.*</tt>' Intrinsics</a>
6419 <div class="doc_text">
6423 <p>This is an overloaded intrinsic. You can use <tt>llvm.umul.with.overflow</tt>
6424 on any integer bit width.</p>
6427 declare {i16, i1} @llvm.umul.with.overflow.i16(i16 %a, i16 %b)
6428 declare {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6429 declare {i64, i1} @llvm.umul.with.overflow.i64(i64 %a, i64 %b)
6434 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
6435 a unsigned multiplication of the two arguments, and indicate whether an overflow
6436 occurred during the unsigned multiplication.</p>
6440 <p>The arguments (%a and %b) and the first element of the result structure may
6441 be of integer types of any bit width, but they must have the same bit width. The
6442 second element of the result structure must be of type <tt>i1</tt>. <tt>%a</tt>
6443 and <tt>%b</tt> are the two values that will undergo unsigned
6448 <p>The '<tt>llvm.umul.with.overflow</tt>' family of intrinsic functions perform
6449 an unsigned multiplication of the two arguments. They return a structure —
6450 the first element of which is the multiplication, and the second element of
6451 which is a bit specifying if the unsigned multiplication resulted in an
6456 %res = call {i32, i1} @llvm.umul.with.overflow.i32(i32 %a, i32 %b)
6457 %sum = extractvalue {i32, i1} %res, 0
6458 %obit = extractvalue {i32, i1} %res, 1
6459 br i1 %obit, label %overflow, label %normal
6464 <!-- ======================================================================= -->
6465 <div class="doc_subsection">
6466 <a name="int_debugger">Debugger Intrinsics</a>
6469 <div class="doc_text">
6471 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
6472 are described in the <a
6473 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
6474 Debugging</a> document.
6479 <!-- ======================================================================= -->
6480 <div class="doc_subsection">
6481 <a name="int_eh">Exception Handling Intrinsics</a>
6484 <div class="doc_text">
6485 <p> The LLVM exception handling intrinsics (which all start with
6486 <tt>llvm.eh.</tt> prefix), are described in the <a
6487 href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
6488 Handling</a> document. </p>
6491 <!-- ======================================================================= -->
6492 <div class="doc_subsection">
6493 <a name="int_trampoline">Trampoline Intrinsic</a>
6496 <div class="doc_text">
6498 This intrinsic makes it possible to excise one parameter, marked with
6499 the <tt>nest</tt> attribute, from a function. The result is a callable
6500 function pointer lacking the nest parameter - the caller does not need
6501 to provide a value for it. Instead, the value to use is stored in
6502 advance in a "trampoline", a block of memory usually allocated
6503 on the stack, which also contains code to splice the nest value into the
6504 argument list. This is used to implement the GCC nested function address
6508 For example, if the function is
6509 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
6510 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as follows:</p>
6512 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
6513 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
6514 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
6515 %fp = bitcast i8* %p to i32 (i32, i32)*
6517 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
6518 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
6521 <!-- _______________________________________________________________________ -->
6522 <div class="doc_subsubsection">
6523 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
6525 <div class="doc_text">
6528 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
6532 This fills the memory pointed to by <tt>tramp</tt> with code
6533 and returns a function pointer suitable for executing it.
6537 The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
6538 pointers. The <tt>tramp</tt> argument must point to a sufficiently large
6539 and sufficiently aligned block of memory; this memory is written to by the
6540 intrinsic. Note that the size and the alignment are target-specific - LLVM
6541 currently provides no portable way of determining them, so a front-end that
6542 generates this intrinsic needs to have some target-specific knowledge.
6543 The <tt>func</tt> argument must hold a function bitcast to an <tt>i8*</tt>.
6547 The block of memory pointed to by <tt>tramp</tt> is filled with target
6548 dependent code, turning it into a function. A pointer to this function is
6549 returned, but needs to be bitcast to an
6550 <a href="#int_trampoline">appropriate function pointer type</a>
6551 before being called. The new function's signature is the same as that of
6552 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
6553 removed. At most one such <tt>nest</tt> argument is allowed, and it must be
6554 of pointer type. Calling the new function is equivalent to calling
6555 <tt>func</tt> with the same argument list, but with <tt>nval</tt> used for the
6556 missing <tt>nest</tt> argument. If, after calling
6557 <tt>llvm.init.trampoline</tt>, the memory pointed to by <tt>tramp</tt> is
6558 modified, then the effect of any later call to the returned function pointer is
6563 <!-- ======================================================================= -->
6564 <div class="doc_subsection">
6565 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
6568 <div class="doc_text">
6570 These intrinsic functions expand the "universal IR" of LLVM to represent
6571 hardware constructs for atomic operations and memory synchronization. This
6572 provides an interface to the hardware, not an interface to the programmer. It
6573 is aimed at a low enough level to allow any programming models or APIs
6574 (Application Programming Interfaces) which
6575 need atomic behaviors to map cleanly onto it. It is also modeled primarily on
6576 hardware behavior. Just as hardware provides a "universal IR" for source
6577 languages, it also provides a starting point for developing a "universal"
6578 atomic operation and synchronization IR.
6581 These do <em>not</em> form an API such as high-level threading libraries,
6582 software transaction memory systems, atomic primitives, and intrinsic
6583 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
6584 application libraries. The hardware interface provided by LLVM should allow
6585 a clean implementation of all of these APIs and parallel programming models.
6586 No one model or paradigm should be selected above others unless the hardware
6587 itself ubiquitously does so.
6592 <!-- _______________________________________________________________________ -->
6593 <div class="doc_subsubsection">
6594 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
6596 <div class="doc_text">
6599 declare void @llvm.memory.barrier( i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>,
6605 The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
6606 specific pairs of memory access types.
6610 The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
6611 The first four arguments enables a specific barrier as listed below. The fith
6612 argument specifies that the barrier applies to io or device or uncached memory.
6616 <li><tt>ll</tt>: load-load barrier</li>
6617 <li><tt>ls</tt>: load-store barrier</li>
6618 <li><tt>sl</tt>: store-load barrier</li>
6619 <li><tt>ss</tt>: store-store barrier</li>
6620 <li><tt>device</tt>: barrier applies to device and uncached memory also.</li>
6624 This intrinsic causes the system to enforce some ordering constraints upon
6625 the loads and stores of the program. This barrier does not indicate
6626 <em>when</em> any events will occur, it only enforces an <em>order</em> in
6627 which they occur. For any of the specified pairs of load and store operations
6628 (f.ex. load-load, or store-load), all of the first operations preceding the
6629 barrier will complete before any of the second operations succeeding the
6630 barrier begin. Specifically the semantics for each pairing is as follows:
6633 <li><tt>ll</tt>: All loads before the barrier must complete before any load
6634 after the barrier begins.</li>
6636 <li><tt>ls</tt>: All loads before the barrier must complete before any
6637 store after the barrier begins.</li>
6638 <li><tt>ss</tt>: All stores before the barrier must complete before any
6639 store after the barrier begins.</li>
6640 <li><tt>sl</tt>: All stores before the barrier must complete before any
6641 load after the barrier begins.</li>
6644 These semantics are applied with a logical "and" behavior when more than one
6645 is enabled in a single memory barrier intrinsic.
6648 Backends may implement stronger barriers than those requested when they do not
6649 support as fine grained a barrier as requested. Some architectures do not
6650 need all types of barriers and on such architectures, these become noops.
6657 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
6658 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
6659 <i>; guarantee the above finishes</i>
6660 store i32 8, %ptr <i>; before this begins</i>
6664 <!-- _______________________________________________________________________ -->
6665 <div class="doc_subsubsection">
6666 <a name="int_atomic_cmp_swap">'<tt>llvm.atomic.cmp.swap.*</tt>' Intrinsic</a>
6668 <div class="doc_text">
6671 This is an overloaded intrinsic. You can use <tt>llvm.atomic.cmp.swap</tt> on
6672 any integer bit width and for different address spaces. Not all targets
6673 support all bit widths however.</p>
6676 declare i8 @llvm.atomic.cmp.swap.i8.p0i8( i8* <ptr>, i8 <cmp>, i8 <val> )
6677 declare i16 @llvm.atomic.cmp.swap.i16.p0i16( i16* <ptr>, i16 <cmp>, i16 <val> )
6678 declare i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* <ptr>, i32 <cmp>, i32 <val> )
6679 declare i64 @llvm.atomic.cmp.swap.i64.p0i64( i64* <ptr>, i64 <cmp>, i64 <val> )
6684 This loads a value in memory and compares it to a given value. If they are
6685 equal, it stores a new value into the memory.
6689 The <tt>llvm.atomic.cmp.swap</tt> intrinsic takes three arguments. The result as
6690 well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
6691 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
6692 this integer type. While any bit width integer may be used, targets may only
6693 lower representations they support in hardware.
6698 This entire intrinsic must be executed atomically. It first loads the value
6699 in memory pointed to by <tt>ptr</tt> and compares it with the value
6700 <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the memory. The
6701 loaded value is yielded in all cases. This provides the equivalent of an
6702 atomic compare-and-swap operation within the SSA framework.
6710 %val1 = add i32 4, 4
6711 %result1 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 4, %val1 )
6712 <i>; yields {i32}:result1 = 4</i>
6713 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
6714 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
6716 %val2 = add i32 1, 1
6717 %result2 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 5, %val2 )
6718 <i>; yields {i32}:result2 = 8</i>
6719 %stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
6721 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
6725 <!-- _______________________________________________________________________ -->
6726 <div class="doc_subsubsection">
6727 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
6729 <div class="doc_text">
6733 This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
6734 integer bit width. Not all targets support all bit widths however.</p>
6736 declare i8 @llvm.atomic.swap.i8.p0i8( i8* <ptr>, i8 <val> )
6737 declare i16 @llvm.atomic.swap.i16.p0i16( i16* <ptr>, i16 <val> )
6738 declare i32 @llvm.atomic.swap.i32.p0i32( i32* <ptr>, i32 <val> )
6739 declare i64 @llvm.atomic.swap.i64.p0i64( i64* <ptr>, i64 <val> )
6744 This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
6745 the value from memory. It then stores the value in <tt>val</tt> in the memory
6751 The <tt>llvm.atomic.swap</tt> intrinsic takes two arguments. Both the
6752 <tt>val</tt> argument and the result must be integers of the same bit width.
6753 The first argument, <tt>ptr</tt>, must be a pointer to a value of this
6754 integer type. The targets may only lower integer representations they
6759 This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
6760 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
6761 equivalent of an atomic swap operation within the SSA framework.
6769 %val1 = add i32 4, 4
6770 %result1 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val1 )
6771 <i>; yields {i32}:result1 = 4</i>
6772 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
6773 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
6775 %val2 = add i32 1, 1
6776 %result2 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val2 )
6777 <i>; yields {i32}:result2 = 8</i>
6779 %stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
6780 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
6784 <!-- _______________________________________________________________________ -->
6785 <div class="doc_subsubsection">
6786 <a name="int_atomic_load_add">'<tt>llvm.atomic.load.add.*</tt>' Intrinsic</a>
6789 <div class="doc_text">
6792 This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.add</tt> on any
6793 integer bit width. Not all targets support all bit widths however.</p>
6795 declare i8 @llvm.atomic.load.add.i8..p0i8( i8* <ptr>, i8 <delta> )
6796 declare i16 @llvm.atomic.load.add.i16..p0i16( i16* <ptr>, i16 <delta> )
6797 declare i32 @llvm.atomic.load.add.i32..p0i32( i32* <ptr>, i32 <delta> )
6798 declare i64 @llvm.atomic.load.add.i64..p0i64( i64* <ptr>, i64 <delta> )
6803 This intrinsic adds <tt>delta</tt> to the value stored in memory at
6804 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
6809 The intrinsic takes two arguments, the first a pointer to an integer value
6810 and the second an integer value. The result is also an integer value. These
6811 integer types can have any bit width, but they must all have the same bit
6812 width. The targets may only lower integer representations they support.
6816 This intrinsic does a series of operations atomically. It first loads the
6817 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
6818 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
6825 %result1 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 4 )
6826 <i>; yields {i32}:result1 = 4</i>
6827 %result2 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 2 )
6828 <i>; yields {i32}:result2 = 8</i>
6829 %result3 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 5 )
6830 <i>; yields {i32}:result3 = 10</i>
6831 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
6835 <!-- _______________________________________________________________________ -->
6836 <div class="doc_subsubsection">
6837 <a name="int_atomic_load_sub">'<tt>llvm.atomic.load.sub.*</tt>' Intrinsic</a>
6840 <div class="doc_text">
6843 This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.sub</tt> on
6844 any integer bit width and for different address spaces. Not all targets
6845 support all bit widths however.</p>
6847 declare i8 @llvm.atomic.load.sub.i8.p0i32( i8* <ptr>, i8 <delta> )
6848 declare i16 @llvm.atomic.load.sub.i16.p0i32( i16* <ptr>, i16 <delta> )
6849 declare i32 @llvm.atomic.load.sub.i32.p0i32( i32* <ptr>, i32 <delta> )
6850 declare i64 @llvm.atomic.load.sub.i64.p0i32( i64* <ptr>, i64 <delta> )
6855 This intrinsic subtracts <tt>delta</tt> to the value stored in memory at
6856 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
6861 The intrinsic takes two arguments, the first a pointer to an integer value
6862 and the second an integer value. The result is also an integer value. These
6863 integer types can have any bit width, but they must all have the same bit
6864 width. The targets may only lower integer representations they support.
6868 This intrinsic does a series of operations atomically. It first loads the
6869 value stored at <tt>ptr</tt>. It then subtracts <tt>delta</tt>, stores the
6870 result to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
6877 %result1 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 4 )
6878 <i>; yields {i32}:result1 = 8</i>
6879 %result2 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 2 )
6880 <i>; yields {i32}:result2 = 4</i>
6881 %result3 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 5 )
6882 <i>; yields {i32}:result3 = 2</i>
6883 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = -3</i>
6887 <!-- _______________________________________________________________________ -->
6888 <div class="doc_subsubsection">
6889 <a name="int_atomic_load_and">'<tt>llvm.atomic.load.and.*</tt>' Intrinsic</a><br>
6890 <a name="int_atomic_load_nand">'<tt>llvm.atomic.load.nand.*</tt>' Intrinsic</a><br>
6891 <a name="int_atomic_load_or">'<tt>llvm.atomic.load.or.*</tt>' Intrinsic</a><br>
6892 <a name="int_atomic_load_xor">'<tt>llvm.atomic.load.xor.*</tt>' Intrinsic</a><br>
6895 <div class="doc_text">
6898 These are overloaded intrinsics. You can use <tt>llvm.atomic.load_and</tt>,
6899 <tt>llvm.atomic.load_nand</tt>, <tt>llvm.atomic.load_or</tt>, and
6900 <tt>llvm.atomic.load_xor</tt> on any integer bit width and for different
6901 address spaces. Not all targets support all bit widths however.</p>
6903 declare i8 @llvm.atomic.load.and.i8.p0i8( i8* <ptr>, i8 <delta> )
6904 declare i16 @llvm.atomic.load.and.i16.p0i16( i16* <ptr>, i16 <delta> )
6905 declare i32 @llvm.atomic.load.and.i32.p0i32( i32* <ptr>, i32 <delta> )
6906 declare i64 @llvm.atomic.load.and.i64.p0i64( i64* <ptr>, i64 <delta> )
6911 declare i8 @llvm.atomic.load.or.i8.p0i8( i8* <ptr>, i8 <delta> )
6912 declare i16 @llvm.atomic.load.or.i16.p0i16( i16* <ptr>, i16 <delta> )
6913 declare i32 @llvm.atomic.load.or.i32.p0i32( i32* <ptr>, i32 <delta> )
6914 declare i64 @llvm.atomic.load.or.i64.p0i64( i64* <ptr>, i64 <delta> )
6919 declare i8 @llvm.atomic.load.nand.i8.p0i32( i8* <ptr>, i8 <delta> )
6920 declare i16 @llvm.atomic.load.nand.i16.p0i32( i16* <ptr>, i16 <delta> )
6921 declare i32 @llvm.atomic.load.nand.i32.p0i32( i32* <ptr>, i32 <delta> )
6922 declare i64 @llvm.atomic.load.nand.i64.p0i32( i64* <ptr>, i64 <delta> )
6927 declare i8 @llvm.atomic.load.xor.i8.p0i32( i8* <ptr>, i8 <delta> )
6928 declare i16 @llvm.atomic.load.xor.i16.p0i32( i16* <ptr>, i16 <delta> )
6929 declare i32 @llvm.atomic.load.xor.i32.p0i32( i32* <ptr>, i32 <delta> )
6930 declare i64 @llvm.atomic.load.xor.i64.p0i32( i64* <ptr>, i64 <delta> )
6935 These intrinsics bitwise the operation (and, nand, or, xor) <tt>delta</tt> to
6936 the value stored in memory at <tt>ptr</tt>. It yields the original value
6942 These intrinsics take two arguments, the first a pointer to an integer value
6943 and the second an integer value. The result is also an integer value. These
6944 integer types can have any bit width, but they must all have the same bit
6945 width. The targets may only lower integer representations they support.
6949 These intrinsics does a series of operations atomically. They first load the
6950 value stored at <tt>ptr</tt>. They then do the bitwise operation
6951 <tt>delta</tt>, store the result to <tt>ptr</tt>. They yield the original
6952 value stored at <tt>ptr</tt>.
6958 store i32 0x0F0F, %ptr
6959 %result0 = call i32 @llvm.atomic.load.nand.i32.p0i32( i32* %ptr, i32 0xFF )
6960 <i>; yields {i32}:result0 = 0x0F0F</i>
6961 %result1 = call i32 @llvm.atomic.load.and.i32.p0i32( i32* %ptr, i32 0xFF )
6962 <i>; yields {i32}:result1 = 0xFFFFFFF0</i>
6963 %result2 = call i32 @llvm.atomic.load.or.i32.p0i32( i32* %ptr, i32 0F )
6964 <i>; yields {i32}:result2 = 0xF0</i>
6965 %result3 = call i32 @llvm.atomic.load.xor.i32.p0i32( i32* %ptr, i32 0F )
6966 <i>; yields {i32}:result3 = FF</i>
6967 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = F0</i>
6972 <!-- _______________________________________________________________________ -->
6973 <div class="doc_subsubsection">
6974 <a name="int_atomic_load_max">'<tt>llvm.atomic.load.max.*</tt>' Intrinsic</a><br>
6975 <a name="int_atomic_load_min">'<tt>llvm.atomic.load.min.*</tt>' Intrinsic</a><br>
6976 <a name="int_atomic_load_umax">'<tt>llvm.atomic.load.umax.*</tt>' Intrinsic</a><br>
6977 <a name="int_atomic_load_umin">'<tt>llvm.atomic.load.umin.*</tt>' Intrinsic</a><br>
6980 <div class="doc_text">
6983 These are overloaded intrinsics. You can use <tt>llvm.atomic.load_max</tt>,
6984 <tt>llvm.atomic.load_min</tt>, <tt>llvm.atomic.load_umax</tt>, and
6985 <tt>llvm.atomic.load_umin</tt> on any integer bit width and for different
6986 address spaces. Not all targets
6987 support all bit widths however.</p>
6989 declare i8 @llvm.atomic.load.max.i8.p0i8( i8* <ptr>, i8 <delta> )
6990 declare i16 @llvm.atomic.load.max.i16.p0i16( i16* <ptr>, i16 <delta> )
6991 declare i32 @llvm.atomic.load.max.i32.p0i32( i32* <ptr>, i32 <delta> )
6992 declare i64 @llvm.atomic.load.max.i64.p0i64( i64* <ptr>, i64 <delta> )
6997 declare i8 @llvm.atomic.load.min.i8.p0i8( i8* <ptr>, i8 <delta> )
6998 declare i16 @llvm.atomic.load.min.i16.p0i16( i16* <ptr>, i16 <delta> )
6999 declare i32 @llvm.atomic.load.min.i32..p0i32( i32* <ptr>, i32 <delta> )
7000 declare i64 @llvm.atomic.load.min.i64..p0i64( i64* <ptr>, i64 <delta> )
7005 declare i8 @llvm.atomic.load.umax.i8.p0i8( i8* <ptr>, i8 <delta> )
7006 declare i16 @llvm.atomic.load.umax.i16.p0i16( i16* <ptr>, i16 <delta> )
7007 declare i32 @llvm.atomic.load.umax.i32.p0i32( i32* <ptr>, i32 <delta> )
7008 declare i64 @llvm.atomic.load.umax.i64.p0i64( i64* <ptr>, i64 <delta> )
7013 declare i8 @llvm.atomic.load.umin.i8..p0i8( i8* <ptr>, i8 <delta> )
7014 declare i16 @llvm.atomic.load.umin.i16.p0i16( i16* <ptr>, i16 <delta> )
7015 declare i32 @llvm.atomic.load.umin.i32..p0i32( i32* <ptr>, i32 <delta> )
7016 declare i64 @llvm.atomic.load.umin.i64..p0i64( i64* <ptr>, i64 <delta> )
7021 These intrinsics takes the signed or unsigned minimum or maximum of
7022 <tt>delta</tt> and the value stored in memory at <tt>ptr</tt>. It yields the
7023 original value at <tt>ptr</tt>.
7028 These intrinsics take two arguments, the first a pointer to an integer value
7029 and the second an integer value. The result is also an integer value. These
7030 integer types can have any bit width, but they must all have the same bit
7031 width. The targets may only lower integer representations they support.
7035 These intrinsics does a series of operations atomically. They first load the
7036 value stored at <tt>ptr</tt>. They then do the signed or unsigned min or max
7037 <tt>delta</tt> and the value, store the result to <tt>ptr</tt>. They yield
7038 the original value stored at <tt>ptr</tt>.
7045 %result0 = call i32 @llvm.atomic.load.min.i32.p0i32( i32* %ptr, i32 -2 )
7046 <i>; yields {i32}:result0 = 7</i>
7047 %result1 = call i32 @llvm.atomic.load.max.i32.p0i32( i32* %ptr, i32 8 )
7048 <i>; yields {i32}:result1 = -2</i>
7049 %result2 = call i32 @llvm.atomic.load.umin.i32.p0i32( i32* %ptr, i32 10 )
7050 <i>; yields {i32}:result2 = 8</i>
7051 %result3 = call i32 @llvm.atomic.load.umax.i32.p0i32( i32* %ptr, i32 30 )
7052 <i>; yields {i32}:result3 = 8</i>
7053 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 30</i>
7057 <!-- ======================================================================= -->
7058 <div class="doc_subsection">
7059 <a name="int_general">General Intrinsics</a>
7062 <div class="doc_text">
7063 <p> This class of intrinsics is designed to be generic and has
7064 no specific purpose. </p>
7067 <!-- _______________________________________________________________________ -->
7068 <div class="doc_subsubsection">
7069 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
7072 <div class="doc_text">
7076 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
7082 The '<tt>llvm.var.annotation</tt>' intrinsic
7088 The first argument is a pointer to a value, the second is a pointer to a
7089 global string, the third is a pointer to a global string which is the source
7090 file name, and the last argument is the line number.
7096 This intrinsic allows annotation of local variables with arbitrary strings.
7097 This can be useful for special purpose optimizations that want to look for these
7098 annotations. These have no other defined use, they are ignored by code
7099 generation and optimization.
7103 <!-- _______________________________________________________________________ -->
7104 <div class="doc_subsubsection">
7105 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
7108 <div class="doc_text">
7111 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
7112 any integer bit width.
7115 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
7116 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
7117 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
7118 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
7119 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
7125 The '<tt>llvm.annotation</tt>' intrinsic.
7131 The first argument is an integer value (result of some expression),
7132 the second is a pointer to a global string, the third is a pointer to a global
7133 string which is the source file name, and the last argument is the line number.
7134 It returns the value of the first argument.
7140 This intrinsic allows annotations to be put on arbitrary expressions
7141 with arbitrary strings. This can be useful for special purpose optimizations
7142 that want to look for these annotations. These have no other defined use, they
7143 are ignored by code generation and optimization.
7147 <!-- _______________________________________________________________________ -->
7148 <div class="doc_subsubsection">
7149 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
7152 <div class="doc_text">
7156 declare void @llvm.trap()
7162 The '<tt>llvm.trap</tt>' intrinsic
7174 This intrinsics is lowered to the target dependent trap instruction. If the
7175 target does not have a trap instruction, this intrinsic will be lowered to the
7176 call of the abort() function.
7180 <!-- _______________________________________________________________________ -->
7181 <div class="doc_subsubsection">
7182 <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a>
7184 <div class="doc_text">
7187 declare void @llvm.stackprotector( i8* <guard>, i8** <slot> )
7192 The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and stores
7193 it onto the stack at <tt>slot</tt>. The stack slot is adjusted to ensure that
7194 it is placed on the stack before local variables.
7198 The <tt>llvm.stackprotector</tt> intrinsic requires two pointer arguments. The
7199 first argument is the value loaded from the stack guard
7200 <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt> that
7201 has enough space to hold the value of the guard.
7205 This intrinsic causes the prologue/epilogue inserter to force the position of
7206 the <tt>AllocaInst</tt> stack slot to be before local variables on the
7207 stack. This is to ensure that if a local variable on the stack is overwritten,
7208 it will destroy the value of the guard. When the function exits, the guard on
7209 the stack is checked against the original guard. If they're different, then
7210 the program aborts by calling the <tt>__stack_chk_fail()</tt> function.
7214 <!-- *********************************************************************** -->
7217 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
7218 src="http://jigsaw.w3.org/css-validator/images/vcss-blue" alt="Valid CSS"></a>
7219 <a href="http://validator.w3.org/check/referer"><img
7220 src="http://www.w3.org/Icons/valid-html401-blue" alt="Valid HTML 4.01"></a>
7222 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
7223 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
7224 Last modified: $Date$