1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <title>LLVM Assembly Language Reference Manual</title>
6 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
7 <meta name="author" content="Chris Lattner">
8 <meta name="description"
9 content="LLVM Assembly Language Reference Manual.">
10 <link rel="stylesheet" href="llvm.css" type="text/css">
15 <div class="doc_title"> LLVM Language Reference Manual </div>
17 <li><a href="#abstract">Abstract</a></li>
18 <li><a href="#introduction">Introduction</a></li>
19 <li><a href="#identifiers">Identifiers</a></li>
20 <li><a href="#highlevel">High Level Structure</a>
22 <li><a href="#modulestructure">Module Structure</a></li>
23 <li><a href="#linkage">Linkage Types</a></li>
24 <li><a href="#callingconv">Calling Conventions</a></li>
25 <li><a href="#globalvars">Global Variables</a></li>
26 <li><a href="#functionstructure">Functions</a></li>
27 <li><a href="#aliasstructure">Aliases</a></li>
28 <li><a href="#paramattrs">Parameter Attributes</a></li>
29 <li><a href="#fnattrs">Function Attributes</a></li>
30 <li><a href="#gc">Garbage Collector Names</a></li>
31 <li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
32 <li><a href="#datalayout">Data Layout</a></li>
35 <li><a href="#typesystem">Type System</a>
37 <li><a href="#t_classifications">Type Classifications</a></li>
38 <li><a href="#t_primitive">Primitive Types</a>
40 <li><a href="#t_floating">Floating Point Types</a></li>
41 <li><a href="#t_void">Void Type</a></li>
42 <li><a href="#t_label">Label Type</a></li>
45 <li><a href="#t_derived">Derived Types</a>
47 <li><a href="#t_integer">Integer Type</a></li>
48 <li><a href="#t_array">Array Type</a></li>
49 <li><a href="#t_function">Function Type</a></li>
50 <li><a href="#t_pointer">Pointer Type</a></li>
51 <li><a href="#t_struct">Structure Type</a></li>
52 <li><a href="#t_pstruct">Packed Structure Type</a></li>
53 <li><a href="#t_vector">Vector Type</a></li>
54 <li><a href="#t_opaque">Opaque Type</a></li>
59 <li><a href="#constants">Constants</a>
61 <li><a href="#simpleconstants">Simple Constants</a></li>
62 <li><a href="#aggregateconstants">Aggregate Constants</a></li>
63 <li><a href="#globalconstants">Global Variable and Function Addresses</a></li>
64 <li><a href="#undefvalues">Undefined Values</a></li>
65 <li><a href="#constantexprs">Constant Expressions</a></li>
68 <li><a href="#othervalues">Other Values</a>
70 <li><a href="#inlineasm">Inline Assembler Expressions</a></li>
73 <li><a href="#instref">Instruction Reference</a>
75 <li><a href="#terminators">Terminator Instructions</a>
77 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
78 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
79 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
80 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
81 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
82 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
85 <li><a href="#binaryops">Binary Operations</a>
87 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
88 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
89 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
90 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
91 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
92 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
93 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
94 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
95 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
98 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
100 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
101 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
102 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
103 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
104 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
105 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
108 <li><a href="#vectorops">Vector Operations</a>
110 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
111 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
112 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
115 <li><a href="#aggregateops">Aggregate Operations</a>
117 <li><a href="#i_extractvalue">'<tt>extractvalue</tt>' Instruction</a></li>
118 <li><a href="#i_insertvalue">'<tt>insertvalue</tt>' Instruction</a></li>
121 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
123 <li><a href="#i_malloc">'<tt>malloc</tt>' Instruction</a></li>
124 <li><a href="#i_free">'<tt>free</tt>' Instruction</a></li>
125 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
126 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
127 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
128 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
131 <li><a href="#convertops">Conversion Operations</a>
133 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
134 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
135 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
136 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
137 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
138 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
139 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
140 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
141 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
142 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
143 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
144 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
147 <li><a href="#otherops">Other Operations</a>
149 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
150 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
151 <li><a href="#i_vicmp">'<tt>vicmp</tt>' Instruction</a></li>
152 <li><a href="#i_vfcmp">'<tt>vfcmp</tt>' Instruction</a></li>
153 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
154 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
155 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
156 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
161 <li><a href="#intrinsics">Intrinsic Functions</a>
163 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
165 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
166 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
167 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
170 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
172 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
173 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
174 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
177 <li><a href="#int_codegen">Code Generator Intrinsics</a>
179 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
180 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
181 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
182 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
183 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
184 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
185 <li><a href="#int_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
188 <li><a href="#int_libc">Standard C Library Intrinsics</a>
190 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
191 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
192 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
193 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
194 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
195 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
196 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
197 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
200 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
202 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
203 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
204 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
205 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
206 <li><a href="#int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic </a></li>
207 <li><a href="#int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic </a></li>
210 <li><a href="#int_debugger">Debugger intrinsics</a></li>
211 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
212 <li><a href="#int_trampoline">Trampoline Intrinsic</a>
214 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
217 <li><a href="#int_stackprotect">Stack Protector Intrinsic</a>
219 <li><a href="#int_ssp">'<tt>llvm.stackprotector</tt>' Intrinsic</a></li>
222 <li><a href="#int_atomics">Atomic intrinsics</a>
224 <li><a href="#int_memory_barrier"><tt>llvm.memory_barrier</tt></a></li>
225 <li><a href="#int_atomic_cmp_swap"><tt>llvm.atomic.cmp.swap</tt></a></li>
226 <li><a href="#int_atomic_swap"><tt>llvm.atomic.swap</tt></a></li>
227 <li><a href="#int_atomic_load_add"><tt>llvm.atomic.load.add</tt></a></li>
228 <li><a href="#int_atomic_load_sub"><tt>llvm.atomic.load.sub</tt></a></li>
229 <li><a href="#int_atomic_load_and"><tt>llvm.atomic.load.and</tt></a></li>
230 <li><a href="#int_atomic_load_nand"><tt>llvm.atomic.load.nand</tt></a></li>
231 <li><a href="#int_atomic_load_or"><tt>llvm.atomic.load.or</tt></a></li>
232 <li><a href="#int_atomic_load_xor"><tt>llvm.atomic.load.xor</tt></a></li>
233 <li><a href="#int_atomic_load_max"><tt>llvm.atomic.load.max</tt></a></li>
234 <li><a href="#int_atomic_load_min"><tt>llvm.atomic.load.min</tt></a></li>
235 <li><a href="#int_atomic_load_umax"><tt>llvm.atomic.load.umax</tt></a></li>
236 <li><a href="#int_atomic_load_umin"><tt>llvm.atomic.load.umin</tt></a></li>
239 <li><a href="#int_general">General intrinsics</a>
241 <li><a href="#int_var_annotation">
242 <tt>llvm.var.annotation</tt>' Intrinsic</a></li>
243 <li><a href="#int_annotation">
244 <tt>llvm.annotation.*</tt>' Intrinsic</a></li>
245 <li><a href="#int_trap">
246 <tt>llvm.trap</tt>' Intrinsic</a></li>
253 <div class="doc_author">
254 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
255 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
258 <!-- *********************************************************************** -->
259 <div class="doc_section"> <a name="abstract">Abstract </a></div>
260 <!-- *********************************************************************** -->
262 <div class="doc_text">
263 <p>This document is a reference manual for the LLVM assembly language.
264 LLVM is a Static Single Assignment (SSA) based representation that provides
265 type safety, low-level operations, flexibility, and the capability of
266 representing 'all' high-level languages cleanly. It is the common code
267 representation used throughout all phases of the LLVM compilation
271 <!-- *********************************************************************** -->
272 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
273 <!-- *********************************************************************** -->
275 <div class="doc_text">
277 <p>The LLVM code representation is designed to be used in three
278 different forms: as an in-memory compiler IR, as an on-disk bitcode
279 representation (suitable for fast loading by a Just-In-Time compiler),
280 and as a human readable assembly language representation. This allows
281 LLVM to provide a powerful intermediate representation for efficient
282 compiler transformations and analysis, while providing a natural means
283 to debug and visualize the transformations. The three different forms
284 of LLVM are all equivalent. This document describes the human readable
285 representation and notation.</p>
287 <p>The LLVM representation aims to be light-weight and low-level
288 while being expressive, typed, and extensible at the same time. It
289 aims to be a "universal IR" of sorts, by being at a low enough level
290 that high-level ideas may be cleanly mapped to it (similar to how
291 microprocessors are "universal IR's", allowing many source languages to
292 be mapped to them). By providing type information, LLVM can be used as
293 the target of optimizations: for example, through pointer analysis, it
294 can be proven that a C automatic variable is never accessed outside of
295 the current function... allowing it to be promoted to a simple SSA
296 value instead of a memory location.</p>
300 <!-- _______________________________________________________________________ -->
301 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
303 <div class="doc_text">
305 <p>It is important to note that this document describes 'well formed'
306 LLVM assembly language. There is a difference between what the parser
307 accepts and what is considered 'well formed'. For example, the
308 following instruction is syntactically okay, but not well formed:</p>
310 <div class="doc_code">
312 %x = <a href="#i_add">add</a> i32 1, %x
316 <p>...because the definition of <tt>%x</tt> does not dominate all of
317 its uses. The LLVM infrastructure provides a verification pass that may
318 be used to verify that an LLVM module is well formed. This pass is
319 automatically run by the parser after parsing input assembly and by
320 the optimizer before it outputs bitcode. The violations pointed out
321 by the verifier pass indicate bugs in transformation passes or input to
325 <!-- Describe the typesetting conventions here. -->
327 <!-- *********************************************************************** -->
328 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
329 <!-- *********************************************************************** -->
331 <div class="doc_text">
333 <p>LLVM identifiers come in two basic types: global and local. Global
334 identifiers (functions, global variables) begin with the @ character. Local
335 identifiers (register names, types) begin with the % character. Additionally,
336 there are three different formats for identifiers, for different purposes:</p>
339 <li>Named values are represented as a string of characters with their prefix.
340 For example, %foo, @DivisionByZero, %a.really.long.identifier. The actual
341 regular expression used is '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
342 Identifiers which require other characters in their names can be surrounded
343 with quotes. Special characters may be escaped using "\xx" where xx is the
344 ASCII code for the character in hexadecimal. In this way, any character can
345 be used in a name value, even quotes themselves.
347 <li>Unnamed values are represented as an unsigned numeric value with their
348 prefix. For example, %12, @2, %44.</li>
350 <li>Constants, which are described in a <a href="#constants">section about
351 constants</a>, below.</li>
354 <p>LLVM requires that values start with a prefix for two reasons: Compilers
355 don't need to worry about name clashes with reserved words, and the set of
356 reserved words may be expanded in the future without penalty. Additionally,
357 unnamed identifiers allow a compiler to quickly come up with a temporary
358 variable without having to avoid symbol table conflicts.</p>
360 <p>Reserved words in LLVM are very similar to reserved words in other
361 languages. There are keywords for different opcodes
362 ('<tt><a href="#i_add">add</a></tt>',
363 '<tt><a href="#i_bitcast">bitcast</a></tt>',
364 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
365 href="#t_void">void</a></tt>', '<tt><a href="#t_primitive">i32</a></tt>', etc...),
366 and others. These reserved words cannot conflict with variable names, because
367 none of them start with a prefix character ('%' or '@').</p>
369 <p>Here is an example of LLVM code to multiply the integer variable
370 '<tt>%X</tt>' by 8:</p>
374 <div class="doc_code">
376 %result = <a href="#i_mul">mul</a> i32 %X, 8
380 <p>After strength reduction:</p>
382 <div class="doc_code">
384 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
388 <p>And the hard way:</p>
390 <div class="doc_code">
392 <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
393 <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
394 %result = <a href="#i_add">add</a> i32 %1, %1
398 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
399 important lexical features of LLVM:</p>
403 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
406 <li>Unnamed temporaries are created when the result of a computation is not
407 assigned to a named value.</li>
409 <li>Unnamed temporaries are numbered sequentially</li>
413 <p>...and it also shows a convention that we follow in this document. When
414 demonstrating instructions, we will follow an instruction with a comment that
415 defines the type and name of value produced. Comments are shown in italic
420 <!-- *********************************************************************** -->
421 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
422 <!-- *********************************************************************** -->
424 <!-- ======================================================================= -->
425 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
428 <div class="doc_text">
430 <p>LLVM programs are composed of "Module"s, each of which is a
431 translation unit of the input programs. Each module consists of
432 functions, global variables, and symbol table entries. Modules may be
433 combined together with the LLVM linker, which merges function (and
434 global variable) definitions, resolves forward declarations, and merges
435 symbol table entries. Here is an example of the "hello world" module:</p>
437 <div class="doc_code">
438 <pre><i>; Declare the string constant as a global constant...</i>
439 <a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a
440 href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
442 <i>; External declaration of the puts function</i>
443 <a href="#functionstructure">declare</a> i32 @puts(i8 *) <i>; i32(i8 *)* </i>
445 <i>; Definition of main function</i>
446 define i32 @main() { <i>; i32()* </i>
447 <i>; Convert [13x i8 ]* to i8 *...</i>
449 href="#i_getelementptr">getelementptr</a> [13 x i8 ]* @.LC0, i64 0, i64 0 <i>; i8 *</i>
451 <i>; Call puts function to write out the string to stdout...</i>
453 href="#i_call">call</a> i32 @puts(i8 * %cast210) <i>; i32</i>
455 href="#i_ret">ret</a> i32 0<br>}<br>
459 <p>This example is made up of a <a href="#globalvars">global variable</a>
460 named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
461 function, and a <a href="#functionstructure">function definition</a>
462 for "<tt>main</tt>".</p>
464 <p>In general, a module is made up of a list of global values,
465 where both functions and global variables are global values. Global values are
466 represented by a pointer to a memory location (in this case, a pointer to an
467 array of char, and a pointer to a function), and have one of the following <a
468 href="#linkage">linkage types</a>.</p>
472 <!-- ======================================================================= -->
473 <div class="doc_subsection">
474 <a name="linkage">Linkage Types</a>
477 <div class="doc_text">
480 All Global Variables and Functions have one of the following types of linkage:
485 <dt><tt><b><a name="linkage_internal">internal</a></b></tt>: </dt>
487 <dd>Global values with internal linkage are only directly accessible by
488 objects in the current module. In particular, linking code into a module with
489 an internal global value may cause the internal to be renamed as necessary to
490 avoid collisions. Because the symbol is internal to the module, all
491 references can be updated. This corresponds to the notion of the
492 '<tt>static</tt>' keyword in C.
495 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
497 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
498 the same name when linkage occurs. This is typically used to implement
499 inline functions, templates, or other code which must be generated in each
500 translation unit that uses it. Unreferenced <tt>linkonce</tt> globals are
501 allowed to be discarded.
504 <dt><tt><b><a name="linkage_common">common</a></b></tt>: </dt>
506 <dd>"<tt>common</tt>" linkage is exactly the same as <tt>linkonce</tt>
507 linkage, except that unreferenced <tt>common</tt> globals may not be
508 discarded. This is used for globals that may be emitted in multiple
509 translation units, but that are not guaranteed to be emitted into every
510 translation unit that uses them. One example of this is tentative
511 definitions in C, such as "<tt>int X;</tt>" at global scope.
514 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
516 <dd>"<tt>weak</tt>" linkage is the same as <tt>common</tt> linkage, except
517 that some targets may choose to emit different assembly sequences for them
518 for target-dependent reasons. This is used for globals that are declared
519 "weak" in C source code.
522 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
524 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
525 pointer to array type. When two global variables with appending linkage are
526 linked together, the two global arrays are appended together. This is the
527 LLVM, typesafe, equivalent of having the system linker append together
528 "sections" with identical names when .o files are linked.
531 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt>: </dt>
532 <dd>The semantics of this linkage follow the ELF object file model: the
533 symbol is weak until linked, if not linked, the symbol becomes null instead
534 of being an undefined reference.
537 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
539 <dd>If none of the above identifiers are used, the global is externally
540 visible, meaning that it participates in linkage and can be used to resolve
541 external symbol references.
546 The next two types of linkage are targeted for Microsoft Windows platform
547 only. They are designed to support importing (exporting) symbols from (to)
548 DLLs (Dynamic Link Libraries).
552 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt>: </dt>
554 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
555 or variable via a global pointer to a pointer that is set up by the DLL
556 exporting the symbol. On Microsoft Windows targets, the pointer name is
557 formed by combining <code>_imp__</code> and the function or variable name.
560 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt>: </dt>
562 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
563 pointer to a pointer in a DLL, so that it can be referenced with the
564 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
565 name is formed by combining <code>_imp__</code> and the function or variable
571 <p><a name="linkage_external"></a>For example, since the "<tt>.LC0</tt>"
572 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
573 variable and was linked with this one, one of the two would be renamed,
574 preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
575 external (i.e., lacking any linkage declarations), they are accessible
576 outside of the current module.</p>
577 <p>It is illegal for a function <i>declaration</i>
578 to have any linkage type other than "externally visible", <tt>dllimport</tt>,
579 or <tt>extern_weak</tt>.</p>
580 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt> and <tt>weak</tt>
584 <!-- ======================================================================= -->
585 <div class="doc_subsection">
586 <a name="callingconv">Calling Conventions</a>
589 <div class="doc_text">
591 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
592 and <a href="#i_invoke">invokes</a> can all have an optional calling convention
593 specified for the call. The calling convention of any pair of dynamic
594 caller/callee must match, or the behavior of the program is undefined. The
595 following calling conventions are supported by LLVM, and more may be added in
599 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
601 <dd>This calling convention (the default if no other calling convention is
602 specified) matches the target C calling conventions. This calling convention
603 supports varargs function calls and tolerates some mismatch in the declared
604 prototype and implemented declaration of the function (as does normal C).
607 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
609 <dd>This calling convention attempts to make calls as fast as possible
610 (e.g. by passing things in registers). This calling convention allows the
611 target to use whatever tricks it wants to produce fast code for the target,
612 without having to conform to an externally specified ABI (Application Binary
613 Interface). Implementations of this convention should allow arbitrary
614 <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> to be
615 supported. This calling convention does not support varargs and requires the
616 prototype of all callees to exactly match the prototype of the function
620 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
622 <dd>This calling convention attempts to make code in the caller as efficient
623 as possible under the assumption that the call is not commonly executed. As
624 such, these calls often preserve all registers so that the call does not break
625 any live ranges in the caller side. This calling convention does not support
626 varargs and requires the prototype of all callees to exactly match the
627 prototype of the function definition.
630 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
632 <dd>Any calling convention may be specified by number, allowing
633 target-specific calling conventions to be used. Target specific calling
634 conventions start at 64.
638 <p>More calling conventions can be added/defined on an as-needed basis, to
639 support pascal conventions or any other well-known target-independent
644 <!-- ======================================================================= -->
645 <div class="doc_subsection">
646 <a name="visibility">Visibility Styles</a>
649 <div class="doc_text">
652 All Global Variables and Functions have one of the following visibility styles:
656 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
658 <dd>On targets that use the ELF object file format, default visibility means
659 that the declaration is visible to other
660 modules and, in shared libraries, means that the declared entity may be
661 overridden. On Darwin, default visibility means that the declaration is
662 visible to other modules. Default visibility corresponds to "external
663 linkage" in the language.
666 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
668 <dd>Two declarations of an object with hidden visibility refer to the same
669 object if they are in the same shared object. Usually, hidden visibility
670 indicates that the symbol will not be placed into the dynamic symbol table,
671 so no other module (executable or shared library) can reference it
675 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
677 <dd>On ELF, protected visibility indicates that the symbol will be placed in
678 the dynamic symbol table, but that references within the defining module will
679 bind to the local symbol. That is, the symbol cannot be overridden by another
686 <!-- ======================================================================= -->
687 <div class="doc_subsection">
688 <a name="globalvars">Global Variables</a>
691 <div class="doc_text">
693 <p>Global variables define regions of memory allocated at compilation time
694 instead of run-time. Global variables may optionally be initialized, may have
695 an explicit section to be placed in, and may have an optional explicit alignment
696 specified. A variable may be defined as "thread_local", which means that it
697 will not be shared by threads (each thread will have a separated copy of the
698 variable). A variable may be defined as a global "constant," which indicates
699 that the contents of the variable will <b>never</b> be modified (enabling better
700 optimization, allowing the global data to be placed in the read-only section of
701 an executable, etc). Note that variables that need runtime initialization
702 cannot be marked "constant" as there is a store to the variable.</p>
705 LLVM explicitly allows <em>declarations</em> of global variables to be marked
706 constant, even if the final definition of the global is not. This capability
707 can be used to enable slightly better optimization of the program, but requires
708 the language definition to guarantee that optimizations based on the
709 'constantness' are valid for the translation units that do not include the
713 <p>As SSA values, global variables define pointer values that are in
714 scope (i.e. they dominate) all basic blocks in the program. Global
715 variables always define a pointer to their "content" type because they
716 describe a region of memory, and all memory objects in LLVM are
717 accessed through pointers.</p>
719 <p>A global variable may be declared to reside in a target-specifc numbered
720 address space. For targets that support them, address spaces may affect how
721 optimizations are performed and/or what target instructions are used to access
722 the variable. The default address space is zero. The address space qualifier
723 must precede any other attributes.</p>
725 <p>LLVM allows an explicit section to be specified for globals. If the target
726 supports it, it will emit globals to the section specified.</p>
728 <p>An explicit alignment may be specified for a global. If not present, or if
729 the alignment is set to zero, the alignment of the global is set by the target
730 to whatever it feels convenient. If an explicit alignment is specified, the
731 global is forced to have at least that much alignment. All alignments must be
734 <p>For example, the following defines a global in a numbered address space with
735 an initializer, section, and alignment:</p>
737 <div class="doc_code">
739 @G = constant float 1.0 addrspace(5), section "foo", align 4
746 <!-- ======================================================================= -->
747 <div class="doc_subsection">
748 <a name="functionstructure">Functions</a>
751 <div class="doc_text">
753 <p>LLVM function definitions consist of the "<tt>define</tt>" keyord,
754 an optional <a href="#linkage">linkage type</a>, an optional
755 <a href="#visibility">visibility style</a>, an optional
756 <a href="#callingconv">calling convention</a>, a return type, an optional
757 <a href="#paramattrs">parameter attribute</a> for the return type, a function
758 name, a (possibly empty) argument list (each with optional
759 <a href="#paramattrs">parameter attributes</a>), optional
760 <a href="#fnattrs">function attributes</a>, an optional section,
761 an optional alignment, an optional <a href="#gc">garbage collector name</a>,
762 an opening curly brace, a list of basic blocks, and a closing curly brace.
764 LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
765 optional <a href="#linkage">linkage type</a>, an optional
766 <a href="#visibility">visibility style</a>, an optional
767 <a href="#callingconv">calling convention</a>, a return type, an optional
768 <a href="#paramattrs">parameter attribute</a> for the return type, a function
769 name, a possibly empty list of arguments, an optional alignment, and an optional
770 <a href="#gc">garbage collector name</a>.</p>
772 <p>A function definition contains a list of basic blocks, forming the CFG
773 (Control Flow Graph) for
774 the function. Each basic block may optionally start with a label (giving the
775 basic block a symbol table entry), contains a list of instructions, and ends
776 with a <a href="#terminators">terminator</a> instruction (such as a branch or
777 function return).</p>
779 <p>The first basic block in a function is special in two ways: it is immediately
780 executed on entrance to the function, and it is not allowed to have predecessor
781 basic blocks (i.e. there can not be any branches to the entry block of a
782 function). Because the block can have no predecessors, it also cannot have any
783 <a href="#i_phi">PHI nodes</a>.</p>
785 <p>LLVM allows an explicit section to be specified for functions. If the target
786 supports it, it will emit functions to the section specified.</p>
788 <p>An explicit alignment may be specified for a function. If not present, or if
789 the alignment is set to zero, the alignment of the function is set by the target
790 to whatever it feels convenient. If an explicit alignment is specified, the
791 function is forced to have at least that much alignment. All alignments must be
796 <div class="doc_code">
798 define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
799 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
800 <ResultType> @<FunctionName> ([argument list])
801 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
802 [<a href="#gc">gc</a>] { ... }
809 <!-- ======================================================================= -->
810 <div class="doc_subsection">
811 <a name="aliasstructure">Aliases</a>
813 <div class="doc_text">
814 <p>Aliases act as "second name" for the aliasee value (which can be either
815 function, global variable, another alias or bitcast of global value). Aliases
816 may have an optional <a href="#linkage">linkage type</a>, and an
817 optional <a href="#visibility">visibility style</a>.</p>
821 <div class="doc_code">
823 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
831 <!-- ======================================================================= -->
832 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
833 <div class="doc_text">
834 <p>The return type and each parameter of a function type may have a set of
835 <i>parameter attributes</i> associated with them. Parameter attributes are
836 used to communicate additional information about the result or parameters of
837 a function. Parameter attributes are considered to be part of the function,
838 not of the function type, so functions with different parameter attributes
839 can have the same function type.</p>
841 <p>Parameter attributes are simple keywords that follow the type specified. If
842 multiple parameter attributes are needed, they are space separated. For
845 <div class="doc_code">
847 declare i32 @printf(i8* noalias , ...)
848 declare i32 @atoi(i8 zeroext)
849 declare signext i8 @returns_signed_char()
853 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
854 <tt>readonly</tt>) come immediately after the argument list.</p>
856 <p>Currently, only the following parameter attributes are defined:</p>
858 <dt><tt>zeroext</tt></dt>
859 <dd>This indicates to the code generator that the parameter or return value
860 should be zero-extended to a 32-bit value by the caller (for a parameter)
861 or the callee (for a return value).</dd>
863 <dt><tt>signext</tt></dt>
864 <dd>This indicates to the code generator that the parameter or return value
865 should be sign-extended to a 32-bit value by the caller (for a parameter)
866 or the callee (for a return value).</dd>
868 <dt><tt>inreg</tt></dt>
869 <dd>This indicates that this parameter or return value should be treated
870 in a special target-dependent fashion during while emitting code for a
871 function call or return (usually, by putting it in a register as opposed
872 to memory, though some targets use it to distinguish between two different
873 kinds of registers). Use of this attribute is target-specific.</dd>
875 <dt><tt><a name="byval">byval</a></tt></dt>
876 <dd>This indicates that the pointer parameter should really be passed by
877 value to the function. The attribute implies that a hidden copy of the
878 pointee is made between the caller and the callee, so the callee is unable
879 to modify the value in the callee. This attribute is only valid on LLVM
880 pointer arguments. It is generally used to pass structs and arrays by
881 value, but is also valid on pointers to scalars. The copy is considered to
882 belong to the caller not the callee (for example,
883 <tt><a href="#readonly">readonly</a></tt> functions should not write to
884 <tt>byval</tt> parameters). This is not a valid attribute for return
887 <dt><tt>sret</tt></dt>
888 <dd>This indicates that the pointer parameter specifies the address of a
889 structure that is the return value of the function in the source program.
890 This pointer must be guaranteed by the caller to be valid: loads and stores
891 to the structure may be assumed by the callee to not to trap. This may only
892 be applied to the first parameter. This is not a valid attribute for
895 <dt><tt>noalias</tt></dt>
896 <dd>This indicates that the parameter does not alias any global or any other
897 parameter. The caller is responsible for ensuring that this is the case,
898 usually by placing the value in a stack allocation. This is not a valid
899 attribute for return values.</dd>
901 <dt><tt>nest</tt></dt>
902 <dd>This indicates that the pointer parameter can be excised using the
903 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
904 attribute for return values.</dd>
909 <!-- ======================================================================= -->
910 <div class="doc_subsection">
911 <a name="gc">Garbage Collector Names</a>
914 <div class="doc_text">
915 <p>Each function may specify a garbage collector name, which is simply a
918 <div class="doc_code"><pre
919 >define void @f() gc "name" { ...</pre></div>
921 <p>The compiler declares the supported values of <i>name</i>. Specifying a
922 collector which will cause the compiler to alter its output in order to support
923 the named garbage collection algorithm.</p>
926 <!-- ======================================================================= -->
927 <div class="doc_subsection">
928 <a name="fnattrs">Function Attributes</a>
931 <div class="doc_text">
933 <p>Function attributes are set to communicate additional information about
934 a function. Function attributes are considered to be part of the function,
935 not of the function type, so functions with different parameter attributes
936 can have the same function type.</p>
938 <p>Function attributes are simple keywords that follow the type specified. If
939 multiple attributes are needed, they are space separated. For
942 <div class="doc_code">
944 define void @f() noinline { ... }
945 define void @f() alwaysinline { ... }
946 define void @f() alwaysinline optsize { ... }
947 define void @f() optsize
952 <dt><tt>alwaysinline</tt></dt>
953 <dd>This attribute indicates that the inliner should attempt to inline this
954 function into callers whenever possible, ignoring any active inlining size
955 threshold for this caller.</dd>
957 <dt><tt>noinline</tt></dt>
958 <dd>This attribute indicates that the inliner should never inline this function
959 in any situation. This attribute may not be used together with the
960 <tt>alwaysinline</tt> attribute.</dd>
962 <dt><tt>optsize</tt></dt>
963 <dd>This attribute suggests that optimization passes and code generator passes
964 make choices that keep the code size of this function low, and otherwise do
965 optimizations specifically to reduce code size.</dd>
967 <dt><tt>noreturn</tt></dt>
968 <dd>This function attribute indicates that the function never returns normally.
969 This produces undefined behavior at runtime if the function ever does
970 dynamically return.</dd>
972 <dt><tt>nounwind</tt></dt>
973 <dd>This function attribute indicates that the function never returns with an
974 unwind or exceptional control flow. If the function does unwind, its runtime
975 behavior is undefined.</dd>
977 <dt><tt>readnone</tt></dt>
978 <dd>This attribute indicates that the function computes its result (or the
979 exception it throws) based strictly on its arguments, without dereferencing any
980 pointer arguments or otherwise accessing any mutable state (e.g. memory, control
981 registers, etc) visible to caller functions. It does not write through any
982 pointer arguments (including <tt><a href="#byval">byval</a></tt> arguments) and
983 never changes any state visible to callers.</dd>
985 <dt><tt><a name="readonly">readonly</a></tt></dt>
986 <dd>This attribute indicates that the function does not write through any
987 pointer arguments (including <tt><a href="#byval">byval</a></tt> arguments)
988 or otherwise modify any state (e.g. memory, control registers, etc) visible to
989 caller functions. It may dereference pointer arguments and read state that may
990 be set in the caller. A readonly function always returns the same value (or
991 throws the same exception) when called with the same set of arguments and global
994 <dt><tt><a name="ssp">ssp</a></tt></dt>
995 <dd>This attribute indicates that the function should emit a stack smashing
996 protector. It is in the form of a "canary"—a random value placed on the
997 stack before the local variables that's checked upon return from the function to
998 see if it has been overwritten. A heuristic is used to determine if a function
999 needs stack protectors or not.</dd>
1001 <dt><tt>ssp-req</tt></dt>
1002 <dd>This attribute indicates that the function should <em>always</em> emit a
1003 stack smashing protector. This overrides the <tt><a href="#ssp">ssp</a></tt>
1004 function attribute.</dd>
1009 <!-- ======================================================================= -->
1010 <div class="doc_subsection">
1011 <a name="moduleasm">Module-Level Inline Assembly</a>
1014 <div class="doc_text">
1016 Modules may contain "module-level inline asm" blocks, which corresponds to the
1017 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
1018 LLVM and treated as a single unit, but may be separated in the .ll file if
1019 desired. The syntax is very simple:
1022 <div class="doc_code">
1024 module asm "inline asm code goes here"
1025 module asm "more can go here"
1029 <p>The strings can contain any character by escaping non-printable characters.
1030 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1035 The inline asm code is simply printed to the machine code .s file when
1036 assembly code is generated.
1040 <!-- ======================================================================= -->
1041 <div class="doc_subsection">
1042 <a name="datalayout">Data Layout</a>
1045 <div class="doc_text">
1046 <p>A module may specify a target specific data layout string that specifies how
1047 data is to be laid out in memory. The syntax for the data layout is simply:</p>
1048 <pre> target datalayout = "<i>layout specification</i>"</pre>
1049 <p>The <i>layout specification</i> consists of a list of specifications
1050 separated by the minus sign character ('-'). Each specification starts with a
1051 letter and may include other information after the letter to define some
1052 aspect of the data layout. The specifications accepted are as follows: </p>
1055 <dd>Specifies that the target lays out data in big-endian form. That is, the
1056 bits with the most significance have the lowest address location.</dd>
1058 <dd>Specifies that the target lays out data in little-endian form. That is,
1059 the bits with the least significance have the lowest address location.</dd>
1060 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1061 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1062 <i>preferred</i> alignments. All sizes are in bits. Specifying the <i>pref</i>
1063 alignment is optional. If omitted, the preceding <tt>:</tt> should be omitted
1065 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1066 <dd>This specifies the alignment for an integer type of a given bit
1067 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1068 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1069 <dd>This specifies the alignment for a vector type of a given bit
1071 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1072 <dd>This specifies the alignment for a floating point type of a given bit
1073 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
1075 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1076 <dd>This specifies the alignment for an aggregate type of a given bit
1079 <p>When constructing the data layout for a given target, LLVM starts with a
1080 default set of specifications which are then (possibly) overriden by the
1081 specifications in the <tt>datalayout</tt> keyword. The default specifications
1082 are given in this list:</p>
1084 <li><tt>E</tt> - big endian</li>
1085 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
1086 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1087 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1088 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1089 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1090 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1091 alignment of 64-bits</li>
1092 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1093 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1094 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1095 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1096 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1098 <p>When LLVM is determining the alignment for a given type, it uses the
1099 following rules:</p>
1101 <li>If the type sought is an exact match for one of the specifications, that
1102 specification is used.</li>
1103 <li>If no match is found, and the type sought is an integer type, then the
1104 smallest integer type that is larger than the bitwidth of the sought type is
1105 used. If none of the specifications are larger than the bitwidth then the the
1106 largest integer type is used. For example, given the default specifications
1107 above, the i7 type will use the alignment of i8 (next largest) while both
1108 i65 and i256 will use the alignment of i64 (largest specified).</li>
1109 <li>If no match is found, and the type sought is a vector type, then the
1110 largest vector type that is smaller than the sought vector type will be used
1111 as a fall back. This happens because <128 x double> can be implemented
1112 in terms of 64 <2 x double>, for example.</li>
1116 <!-- *********************************************************************** -->
1117 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
1118 <!-- *********************************************************************** -->
1120 <div class="doc_text">
1122 <p>The LLVM type system is one of the most important features of the
1123 intermediate representation. Being typed enables a number of
1124 optimizations to be performed on the intermediate representation directly,
1125 without having to do
1126 extra analyses on the side before the transformation. A strong type
1127 system makes it easier to read the generated code and enables novel
1128 analyses and transformations that are not feasible to perform on normal
1129 three address code representations.</p>
1133 <!-- ======================================================================= -->
1134 <div class="doc_subsection"> <a name="t_classifications">Type
1135 Classifications</a> </div>
1136 <div class="doc_text">
1137 <p>The types fall into a few useful
1138 classifications:</p>
1140 <table border="1" cellspacing="0" cellpadding="4">
1142 <tr><th>Classification</th><th>Types</th></tr>
1144 <td><a href="#t_integer">integer</a></td>
1145 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1148 <td><a href="#t_floating">floating point</a></td>
1149 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1152 <td><a name="t_firstclass">first class</a></td>
1153 <td><a href="#t_integer">integer</a>,
1154 <a href="#t_floating">floating point</a>,
1155 <a href="#t_pointer">pointer</a>,
1156 <a href="#t_vector">vector</a>,
1157 <a href="#t_struct">structure</a>,
1158 <a href="#t_array">array</a>,
1159 <a href="#t_label">label</a>.
1163 <td><a href="#t_primitive">primitive</a></td>
1164 <td><a href="#t_label">label</a>,
1165 <a href="#t_void">void</a>,
1166 <a href="#t_floating">floating point</a>.</td>
1169 <td><a href="#t_derived">derived</a></td>
1170 <td><a href="#t_integer">integer</a>,
1171 <a href="#t_array">array</a>,
1172 <a href="#t_function">function</a>,
1173 <a href="#t_pointer">pointer</a>,
1174 <a href="#t_struct">structure</a>,
1175 <a href="#t_pstruct">packed structure</a>,
1176 <a href="#t_vector">vector</a>,
1177 <a href="#t_opaque">opaque</a>.
1183 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
1184 most important. Values of these types are the only ones which can be
1185 produced by instructions, passed as arguments, or used as operands to
1189 <!-- ======================================================================= -->
1190 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
1192 <div class="doc_text">
1193 <p>The primitive types are the fundamental building blocks of the LLVM
1198 <!-- _______________________________________________________________________ -->
1199 <div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1201 <div class="doc_text">
1204 <tr><th>Type</th><th>Description</th></tr>
1205 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1206 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1207 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1208 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1209 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1214 <!-- _______________________________________________________________________ -->
1215 <div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1217 <div class="doc_text">
1219 <p>The void type does not represent any value and has no size.</p>
1228 <!-- _______________________________________________________________________ -->
1229 <div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1231 <div class="doc_text">
1233 <p>The label type represents code labels.</p>
1243 <!-- ======================================================================= -->
1244 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1246 <div class="doc_text">
1248 <p>The real power in LLVM comes from the derived types in the system.
1249 This is what allows a programmer to represent arrays, functions,
1250 pointers, and other useful types. Note that these derived types may be
1251 recursive: For example, it is possible to have a two dimensional array.</p>
1255 <!-- _______________________________________________________________________ -->
1256 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1258 <div class="doc_text">
1261 <p>The integer type is a very simple derived type that simply specifies an
1262 arbitrary bit width for the integer type desired. Any bit width from 1 bit to
1263 2^23-1 (about 8 million) can be specified.</p>
1271 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1275 <table class="layout">
1278 <td><tt>i1</tt></td>
1279 <td>a single-bit integer.</td>
1281 <td><tt>i32</tt></td>
1282 <td>a 32-bit integer.</td>
1284 <td><tt>i1942652</tt></td>
1285 <td>a really big integer of over 1 million bits.</td>
1291 <!-- _______________________________________________________________________ -->
1292 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1294 <div class="doc_text">
1298 <p>The array type is a very simple derived type that arranges elements
1299 sequentially in memory. The array type requires a size (number of
1300 elements) and an underlying data type.</p>
1305 [<# elements> x <elementtype>]
1308 <p>The number of elements is a constant integer value; elementtype may
1309 be any type with a size.</p>
1312 <table class="layout">
1314 <td class="left"><tt>[40 x i32]</tt></td>
1315 <td class="left">Array of 40 32-bit integer values.</td>
1318 <td class="left"><tt>[41 x i32]</tt></td>
1319 <td class="left">Array of 41 32-bit integer values.</td>
1322 <td class="left"><tt>[4 x i8]</tt></td>
1323 <td class="left">Array of 4 8-bit integer values.</td>
1326 <p>Here are some examples of multidimensional arrays:</p>
1327 <table class="layout">
1329 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1330 <td class="left">3x4 array of 32-bit integer values.</td>
1333 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1334 <td class="left">12x10 array of single precision floating point values.</td>
1337 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1338 <td class="left">2x3x4 array of 16-bit integer values.</td>
1342 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1343 length array. Normally, accesses past the end of an array are undefined in
1344 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
1345 As a special case, however, zero length arrays are recognized to be variable
1346 length. This allows implementation of 'pascal style arrays' with the LLVM
1347 type "{ i32, [0 x float]}", for example.</p>
1351 <!-- _______________________________________________________________________ -->
1352 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1353 <div class="doc_text">
1357 <p>The function type can be thought of as a function signature. It
1358 consists of a return type and a list of formal parameter types. The
1359 return type of a function type is a scalar type, a void type, or a struct type.
1360 If the return type is a struct type then all struct elements must be of first
1361 class types, and the struct must have at least one element.</p>
1366 <returntype list> (<parameter list>)
1369 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1370 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1371 which indicates that the function takes a variable number of arguments.
1372 Variable argument functions can access their arguments with the <a
1373 href="#int_varargs">variable argument handling intrinsic</a> functions.
1374 '<tt><returntype list></tt>' is a comma-separated list of
1375 <a href="#t_firstclass">first class</a> type specifiers.</p>
1378 <table class="layout">
1380 <td class="left"><tt>i32 (i32)</tt></td>
1381 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1383 </tr><tr class="layout">
1384 <td class="left"><tt>float (i16 signext, i32 *) *
1386 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1387 an <tt>i16</tt> that should be sign extended and a
1388 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1391 </tr><tr class="layout">
1392 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1393 <td class="left">A vararg function that takes at least one
1394 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1395 which returns an integer. This is the signature for <tt>printf</tt> in
1398 </tr><tr class="layout">
1399 <td class="left"><tt>{i32, i32} (i32)</tt></td>
1400 <td class="left">A function taking an <tt>i32></tt>, returning two
1401 <tt> i32 </tt> values as an aggregate of type <tt>{ i32, i32 }</tt>
1407 <!-- _______________________________________________________________________ -->
1408 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1409 <div class="doc_text">
1411 <p>The structure type is used to represent a collection of data members
1412 together in memory. The packing of the field types is defined to match
1413 the ABI of the underlying processor. The elements of a structure may
1414 be any type that has a size.</p>
1415 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1416 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1417 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1420 <pre> { <type list> }<br></pre>
1422 <table class="layout">
1424 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1425 <td class="left">A triple of three <tt>i32</tt> values</td>
1426 </tr><tr class="layout">
1427 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1428 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1429 second element is a <a href="#t_pointer">pointer</a> to a
1430 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1431 an <tt>i32</tt>.</td>
1436 <!-- _______________________________________________________________________ -->
1437 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1439 <div class="doc_text">
1441 <p>The packed structure type is used to represent a collection of data members
1442 together in memory. There is no padding between fields. Further, the alignment
1443 of a packed structure is 1 byte. The elements of a packed structure may
1444 be any type that has a size.</p>
1445 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1446 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1447 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1450 <pre> < { <type list> } > <br></pre>
1452 <table class="layout">
1454 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1455 <td class="left">A triple of three <tt>i32</tt> values</td>
1456 </tr><tr class="layout">
1458 <tt>< { float, i32 (i32)* } ></tt></td>
1459 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1460 second element is a <a href="#t_pointer">pointer</a> to a
1461 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1462 an <tt>i32</tt>.</td>
1467 <!-- _______________________________________________________________________ -->
1468 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1469 <div class="doc_text">
1471 <p>As in many languages, the pointer type represents a pointer or
1472 reference to another object, which must live in memory. Pointer types may have
1473 an optional address space attribute defining the target-specific numbered
1474 address space where the pointed-to object resides. The default address space is
1477 <pre> <type> *<br></pre>
1479 <table class="layout">
1481 <td class="left"><tt>[4x i32]*</tt></td>
1482 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1483 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1486 <td class="left"><tt>i32 (i32 *) *</tt></td>
1487 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
1488 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1492 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1493 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1494 that resides in address space #5.</td>
1499 <!-- _______________________________________________________________________ -->
1500 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1501 <div class="doc_text">
1505 <p>A vector type is a simple derived type that represents a vector
1506 of elements. Vector types are used when multiple primitive data
1507 are operated in parallel using a single instruction (SIMD).
1508 A vector type requires a size (number of
1509 elements) and an underlying primitive data type. Vectors must have a power
1510 of two length (1, 2, 4, 8, 16 ...). Vector types are
1511 considered <a href="#t_firstclass">first class</a>.</p>
1516 < <# elements> x <elementtype> >
1519 <p>The number of elements is a constant integer value; elementtype may
1520 be any integer or floating point type.</p>
1524 <table class="layout">
1526 <td class="left"><tt><4 x i32></tt></td>
1527 <td class="left">Vector of 4 32-bit integer values.</td>
1530 <td class="left"><tt><8 x float></tt></td>
1531 <td class="left">Vector of 8 32-bit floating-point values.</td>
1534 <td class="left"><tt><2 x i64></tt></td>
1535 <td class="left">Vector of 2 64-bit integer values.</td>
1540 <!-- _______________________________________________________________________ -->
1541 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1542 <div class="doc_text">
1546 <p>Opaque types are used to represent unknown types in the system. This
1547 corresponds (for example) to the C notion of a forward declared structure type.
1548 In LLVM, opaque types can eventually be resolved to any type (not just a
1549 structure type).</p>
1559 <table class="layout">
1561 <td class="left"><tt>opaque</tt></td>
1562 <td class="left">An opaque type.</td>
1568 <!-- *********************************************************************** -->
1569 <div class="doc_section"> <a name="constants">Constants</a> </div>
1570 <!-- *********************************************************************** -->
1572 <div class="doc_text">
1574 <p>LLVM has several different basic types of constants. This section describes
1575 them all and their syntax.</p>
1579 <!-- ======================================================================= -->
1580 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1582 <div class="doc_text">
1585 <dt><b>Boolean constants</b></dt>
1587 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1588 constants of the <tt><a href="#t_primitive">i1</a></tt> type.
1591 <dt><b>Integer constants</b></dt>
1593 <dd>Standard integers (such as '4') are constants of the <a
1594 href="#t_integer">integer</a> type. Negative numbers may be used with
1598 <dt><b>Floating point constants</b></dt>
1600 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1601 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1602 notation (see below). The assembler requires the exact decimal value of
1603 a floating-point constant. For example, the assembler accepts 1.25 but
1604 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
1605 constants must have a <a href="#t_floating">floating point</a> type. </dd>
1607 <dt><b>Null pointer constants</b></dt>
1609 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1610 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1614 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1615 of floating point constants. For example, the form '<tt>double
1616 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1617 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1618 (and the only time that they are generated by the disassembler) is when a
1619 floating point constant must be emitted but it cannot be represented as a
1620 decimal floating point number. For example, NaN's, infinities, and other
1621 special values are represented in their IEEE hexadecimal format so that
1622 assembly and disassembly do not cause any bits to change in the constants.</p>
1626 <!-- ======================================================================= -->
1627 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1630 <div class="doc_text">
1631 <p>Aggregate constants arise from aggregation of simple constants
1632 and smaller aggregate constants.</p>
1635 <dt><b>Structure constants</b></dt>
1637 <dd>Structure constants are represented with notation similar to structure
1638 type definitions (a comma separated list of elements, surrounded by braces
1639 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
1640 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>". Structure constants
1641 must have <a href="#t_struct">structure type</a>, and the number and
1642 types of elements must match those specified by the type.
1645 <dt><b>Array constants</b></dt>
1647 <dd>Array constants are represented with notation similar to array type
1648 definitions (a comma separated list of elements, surrounded by square brackets
1649 (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74 ]</tt>". Array
1650 constants must have <a href="#t_array">array type</a>, and the number and
1651 types of elements must match those specified by the type.
1654 <dt><b>Vector constants</b></dt>
1656 <dd>Vector constants are represented with notation similar to vector type
1657 definitions (a comma separated list of elements, surrounded by
1658 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32 42,
1659 i32 11, i32 74, i32 100 ></tt>". Vector constants must have <a
1660 href="#t_vector">vector type</a>, and the number and types of elements must
1661 match those specified by the type.
1664 <dt><b>Zero initialization</b></dt>
1666 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1667 value to zero of <em>any</em> type, including scalar and aggregate types.
1668 This is often used to avoid having to print large zero initializers (e.g. for
1669 large arrays) and is always exactly equivalent to using explicit zero
1676 <!-- ======================================================================= -->
1677 <div class="doc_subsection">
1678 <a name="globalconstants">Global Variable and Function Addresses</a>
1681 <div class="doc_text">
1683 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1684 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1685 constants. These constants are explicitly referenced when the <a
1686 href="#identifiers">identifier for the global</a> is used and always have <a
1687 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1690 <div class="doc_code">
1694 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1700 <!-- ======================================================================= -->
1701 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1702 <div class="doc_text">
1703 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1704 no specific value. Undefined values may be of any type and be used anywhere
1705 a constant is permitted.</p>
1707 <p>Undefined values indicate to the compiler that the program is well defined
1708 no matter what value is used, giving the compiler more freedom to optimize.
1712 <!-- ======================================================================= -->
1713 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1716 <div class="doc_text">
1718 <p>Constant expressions are used to allow expressions involving other constants
1719 to be used as constants. Constant expressions may be of any <a
1720 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1721 that does not have side effects (e.g. load and call are not supported). The
1722 following is the syntax for constant expressions:</p>
1725 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1726 <dd>Truncate a constant to another type. The bit size of CST must be larger
1727 than the bit size of TYPE. Both types must be integers.</dd>
1729 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1730 <dd>Zero extend a constant to another type. The bit size of CST must be
1731 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1733 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1734 <dd>Sign extend a constant to another type. The bit size of CST must be
1735 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1737 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1738 <dd>Truncate a floating point constant to another floating point type. The
1739 size of CST must be larger than the size of TYPE. Both types must be
1740 floating point.</dd>
1742 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1743 <dd>Floating point extend a constant to another type. The size of CST must be
1744 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1746 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
1747 <dd>Convert a floating point constant to the corresponding unsigned integer
1748 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1749 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1750 of the same number of elements. If the value won't fit in the integer type,
1751 the results are undefined.</dd>
1753 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
1754 <dd>Convert a floating point constant to the corresponding signed integer
1755 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1756 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1757 of the same number of elements. If the value won't fit in the integer type,
1758 the results are undefined.</dd>
1760 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
1761 <dd>Convert an unsigned integer constant to the corresponding floating point
1762 constant. TYPE must be a scalar or vector floating point type. CST must be of
1763 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1764 of the same number of elements. If the value won't fit in the floating point
1765 type, the results are undefined.</dd>
1767 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
1768 <dd>Convert a signed integer constant to the corresponding floating point
1769 constant. TYPE must be a scalar or vector floating point type. CST must be of
1770 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1771 of the same number of elements. If the value won't fit in the floating point
1772 type, the results are undefined.</dd>
1774 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
1775 <dd>Convert a pointer typed constant to the corresponding integer constant
1776 TYPE must be an integer type. CST must be of pointer type. The CST value is
1777 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1779 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
1780 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1781 pointer type. CST must be of integer type. The CST value is zero extended,
1782 truncated, or unchanged to make it fit in a pointer size. This one is
1783 <i>really</i> dangerous!</dd>
1785 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
1786 <dd>Convert a constant, CST, to another TYPE. The size of CST and TYPE must be
1787 identical (same number of bits). The conversion is done as if the CST value
1788 was stored to memory and read back as TYPE. In other words, no bits change
1789 with this operator, just the type. This can be used for conversion of
1790 vector types to any other type, as long as they have the same bit width. For
1791 pointers it is only valid to cast to another pointer type. It is not valid
1792 to bitcast to or from an aggregate type.
1795 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1797 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1798 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1799 instruction, the index list may have zero or more indexes, which are required
1800 to make sense for the type of "CSTPTR".</dd>
1802 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1804 <dd>Perform the <a href="#i_select">select operation</a> on
1807 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
1808 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
1810 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
1811 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
1813 <dt><b><tt>vicmp COND ( VAL1, VAL2 )</tt></b></dt>
1814 <dd>Performs the <a href="#i_vicmp">vicmp operation</a> on constants.</dd>
1816 <dt><b><tt>vfcmp COND ( VAL1, VAL2 )</tt></b></dt>
1817 <dd>Performs the <a href="#i_vfcmp">vfcmp operation</a> on constants.</dd>
1819 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1821 <dd>Perform the <a href="#i_extractelement">extractelement
1822 operation</a> on constants.</dd>
1824 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1826 <dd>Perform the <a href="#i_insertelement">insertelement
1827 operation</a> on constants.</dd>
1830 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
1832 <dd>Perform the <a href="#i_shufflevector">shufflevector
1833 operation</a> on constants.</dd>
1835 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1837 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1838 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1839 binary</a> operations. The constraints on operands are the same as those for
1840 the corresponding instruction (e.g. no bitwise operations on floating point
1841 values are allowed).</dd>
1845 <!-- *********************************************************************** -->
1846 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1847 <!-- *********************************************************************** -->
1849 <!-- ======================================================================= -->
1850 <div class="doc_subsection">
1851 <a name="inlineasm">Inline Assembler Expressions</a>
1854 <div class="doc_text">
1857 LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
1858 Module-Level Inline Assembly</a>) through the use of a special value. This
1859 value represents the inline assembler as a string (containing the instructions
1860 to emit), a list of operand constraints (stored as a string), and a flag that
1861 indicates whether or not the inline asm expression has side effects. An example
1862 inline assembler expression is:
1865 <div class="doc_code">
1867 i32 (i32) asm "bswap $0", "=r,r"
1872 Inline assembler expressions may <b>only</b> be used as the callee operand of
1873 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
1876 <div class="doc_code">
1878 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
1883 Inline asms with side effects not visible in the constraint list must be marked
1884 as having side effects. This is done through the use of the
1885 '<tt>sideeffect</tt>' keyword, like so:
1888 <div class="doc_code">
1890 call void asm sideeffect "eieio", ""()
1894 <p>TODO: The format of the asm and constraints string still need to be
1895 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1896 need to be documented). This is probably best done by reference to another
1897 document that covers inline asm from a holistic perspective.
1902 <!-- *********************************************************************** -->
1903 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1904 <!-- *********************************************************************** -->
1906 <div class="doc_text">
1908 <p>The LLVM instruction set consists of several different
1909 classifications of instructions: <a href="#terminators">terminator
1910 instructions</a>, <a href="#binaryops">binary instructions</a>,
1911 <a href="#bitwiseops">bitwise binary instructions</a>, <a
1912 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1913 instructions</a>.</p>
1917 <!-- ======================================================================= -->
1918 <div class="doc_subsection"> <a name="terminators">Terminator
1919 Instructions</a> </div>
1921 <div class="doc_text">
1923 <p>As mentioned <a href="#functionstructure">previously</a>, every
1924 basic block in a program ends with a "Terminator" instruction, which
1925 indicates which block should be executed after the current block is
1926 finished. These terminator instructions typically yield a '<tt>void</tt>'
1927 value: they produce control flow, not values (the one exception being
1928 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1929 <p>There are six different terminator instructions: the '<a
1930 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1931 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1932 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1933 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1934 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1938 <!-- _______________________________________________________________________ -->
1939 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1940 Instruction</a> </div>
1941 <div class="doc_text">
1944 ret <type> <value> <i>; Return a value from a non-void function</i>
1945 ret void <i>; Return from void function</i>
1950 <p>The '<tt>ret</tt>' instruction is used to return control flow (and
1951 optionally a value) from a function back to the caller.</p>
1952 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
1953 returns a value and then causes control flow, and one that just causes
1954 control flow to occur.</p>
1958 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument,
1959 the return value. The type of the return value must be a
1960 '<a href="#t_firstclass">first class</a>' type.</p>
1962 <p>A function is not <a href="#wellformed">well formed</a> if
1963 it it has a non-void return type and contains a '<tt>ret</tt>'
1964 instruction with no return value or a return value with a type that
1965 does not match its type, or if it has a void return type and contains
1966 a '<tt>ret</tt>' instruction with a return value.</p>
1970 <p>When the '<tt>ret</tt>' instruction is executed, control flow
1971 returns back to the calling function's context. If the caller is a "<a
1972 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1973 the instruction after the call. If the caller was an "<a
1974 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1975 at the beginning of the "normal" destination block. If the instruction
1976 returns a value, that value shall set the call or invoke instruction's
1982 ret i32 5 <i>; Return an integer value of 5</i>
1983 ret void <i>; Return from a void function</i>
1984 ret { i32, i8 } { i32 4, i8 2 } <i>; Return an aggregate of values 4 and 2</i>
1987 <!-- _______________________________________________________________________ -->
1988 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1989 <div class="doc_text">
1991 <pre> br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1994 <p>The '<tt>br</tt>' instruction is used to cause control flow to
1995 transfer to a different basic block in the current function. There are
1996 two forms of this instruction, corresponding to a conditional branch
1997 and an unconditional branch.</p>
1999 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
2000 single '<tt>i1</tt>' value and two '<tt>label</tt>' values. The
2001 unconditional form of the '<tt>br</tt>' instruction takes a single
2002 '<tt>label</tt>' value as a target.</p>
2004 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
2005 argument is evaluated. If the value is <tt>true</tt>, control flows
2006 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
2007 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
2009 <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
2010 href="#i_ret">ret</a> i32 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> i32 0<br></pre>
2012 <!-- _______________________________________________________________________ -->
2013 <div class="doc_subsubsection">
2014 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
2017 <div class="doc_text">
2021 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
2026 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
2027 several different places. It is a generalization of the '<tt>br</tt>'
2028 instruction, allowing a branch to occur to one of many possible
2034 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
2035 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
2036 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
2037 table is not allowed to contain duplicate constant entries.</p>
2041 <p>The <tt>switch</tt> instruction specifies a table of values and
2042 destinations. When the '<tt>switch</tt>' instruction is executed, this
2043 table is searched for the given value. If the value is found, control flow is
2044 transfered to the corresponding destination; otherwise, control flow is
2045 transfered to the default destination.</p>
2047 <h5>Implementation:</h5>
2049 <p>Depending on properties of the target machine and the particular
2050 <tt>switch</tt> instruction, this instruction may be code generated in different
2051 ways. For example, it could be generated as a series of chained conditional
2052 branches or with a lookup table.</p>
2057 <i>; Emulate a conditional br instruction</i>
2058 %Val = <a href="#i_zext">zext</a> i1 %value to i32
2059 switch i32 %Val, label %truedest [i32 0, label %falsedest ]
2061 <i>; Emulate an unconditional br instruction</i>
2062 switch i32 0, label %dest [ ]
2064 <i>; Implement a jump table:</i>
2065 switch i32 %val, label %otherwise [ i32 0, label %onzero
2067 i32 2, label %ontwo ]
2071 <!-- _______________________________________________________________________ -->
2072 <div class="doc_subsubsection">
2073 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
2076 <div class="doc_text">
2081 <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>]
2082 to label <normal label> unwind label <exception label>
2087 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
2088 function, with the possibility of control flow transfer to either the
2089 '<tt>normal</tt>' label or the
2090 '<tt>exception</tt>' label. If the callee function returns with the
2091 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
2092 "normal" label. If the callee (or any indirect callees) returns with the "<a
2093 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
2094 continued at the dynamically nearest "exception" label.</p>
2098 <p>This instruction requires several arguments:</p>
2102 The optional "cconv" marker indicates which <a href="#callingconv">calling
2103 convention</a> the call should use. If none is specified, the call defaults
2104 to using C calling conventions.
2107 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
2108 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>',
2109 and '<tt>inreg</tt>' attributes are valid here.</li>
2111 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
2112 function value being invoked. In most cases, this is a direct function
2113 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
2114 an arbitrary pointer to function value.
2117 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
2118 function to be invoked. </li>
2120 <li>'<tt>function args</tt>': argument list whose types match the function
2121 signature argument types. If the function signature indicates the function
2122 accepts a variable number of arguments, the extra arguments can be
2125 <li>'<tt>normal label</tt>': the label reached when the called function
2126 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
2128 <li>'<tt>exception label</tt>': the label reached when a callee returns with
2129 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
2131 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
2132 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
2133 '<tt>readnone</tt>' attributes are valid here.</li>
2138 <p>This instruction is designed to operate as a standard '<tt><a
2139 href="#i_call">call</a></tt>' instruction in most regards. The primary
2140 difference is that it establishes an association with a label, which is used by
2141 the runtime library to unwind the stack.</p>
2143 <p>This instruction is used in languages with destructors to ensure that proper
2144 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
2145 exception. Additionally, this is important for implementation of
2146 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
2150 %retval = invoke i32 @Test(i32 15) to label %Continue
2151 unwind label %TestCleanup <i>; {i32}:retval set</i>
2152 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
2153 unwind label %TestCleanup <i>; {i32}:retval set</i>
2158 <!-- _______________________________________________________________________ -->
2160 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
2161 Instruction</a> </div>
2163 <div class="doc_text">
2172 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
2173 at the first callee in the dynamic call stack which used an <a
2174 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
2175 primarily used to implement exception handling.</p>
2179 <p>The '<tt>unwind</tt>' instruction causes execution of the current function to
2180 immediately halt. The dynamic call stack is then searched for the first <a
2181 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
2182 execution continues at the "exceptional" destination block specified by the
2183 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
2184 dynamic call chain, undefined behavior results.</p>
2187 <!-- _______________________________________________________________________ -->
2189 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
2190 Instruction</a> </div>
2192 <div class="doc_text">
2201 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
2202 instruction is used to inform the optimizer that a particular portion of the
2203 code is not reachable. This can be used to indicate that the code after a
2204 no-return function cannot be reached, and other facts.</p>
2208 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
2213 <!-- ======================================================================= -->
2214 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
2215 <div class="doc_text">
2216 <p>Binary operators are used to do most of the computation in a
2217 program. They require two operands of the same type, execute an operation on them, and
2218 produce a single value. The operands might represent
2219 multiple data, as is the case with the <a href="#t_vector">vector</a> data type.
2220 The result value has the same type as its operands.</p>
2221 <p>There are several different binary operators:</p>
2223 <!-- _______________________________________________________________________ -->
2224 <div class="doc_subsubsection">
2225 <a name="i_add">'<tt>add</tt>' Instruction</a>
2228 <div class="doc_text">
2233 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2238 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
2242 <p>The two arguments to the '<tt>add</tt>' instruction must be <a
2243 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>, or
2244 <a href="#t_vector">vector</a> values. Both arguments must have identical
2249 <p>The value produced is the integer or floating point sum of the two
2252 <p>If an integer sum has unsigned overflow, the result returned is the
2253 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2256 <p>Because LLVM integers use a two's complement representation, this
2257 instruction is appropriate for both signed and unsigned integers.</p>
2262 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
2265 <!-- _______________________________________________________________________ -->
2266 <div class="doc_subsubsection">
2267 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
2270 <div class="doc_text">
2275 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2280 <p>The '<tt>sub</tt>' instruction returns the difference of its two
2283 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
2284 '<tt>neg</tt>' instruction present in most other intermediate
2285 representations.</p>
2289 <p>The two arguments to the '<tt>sub</tt>' instruction must be <a
2290 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>,
2291 or <a href="#t_vector">vector</a> values. Both arguments must have identical
2296 <p>The value produced is the integer or floating point difference of
2297 the two operands.</p>
2299 <p>If an integer difference has unsigned overflow, the result returned is the
2300 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2303 <p>Because LLVM integers use a two's complement representation, this
2304 instruction is appropriate for both signed and unsigned integers.</p>
2308 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
2309 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
2313 <!-- _______________________________________________________________________ -->
2314 <div class="doc_subsubsection">
2315 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
2318 <div class="doc_text">
2321 <pre> <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2324 <p>The '<tt>mul</tt>' instruction returns the product of its two
2329 <p>The two arguments to the '<tt>mul</tt>' instruction must be <a
2330 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>,
2331 or <a href="#t_vector">vector</a> values. Both arguments must have identical
2336 <p>The value produced is the integer or floating point product of the
2339 <p>If the result of an integer multiplication has unsigned overflow,
2340 the result returned is the mathematical result modulo
2341 2<sup>n</sup>, where n is the bit width of the result.</p>
2342 <p>Because LLVM integers use a two's complement representation, and the
2343 result is the same width as the operands, this instruction returns the
2344 correct result for both signed and unsigned integers. If a full product
2345 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands
2346 should be sign-extended or zero-extended as appropriate to the
2347 width of the full product.</p>
2349 <pre> <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
2353 <!-- _______________________________________________________________________ -->
2354 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
2356 <div class="doc_text">
2358 <pre> <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2361 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two
2366 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
2367 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2368 values. Both arguments must have identical types.</p>
2372 <p>The value produced is the unsigned integer quotient of the two operands.</p>
2373 <p>Note that unsigned integer division and signed integer division are distinct
2374 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
2375 <p>Division by zero leads to undefined behavior.</p>
2377 <pre> <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2380 <!-- _______________________________________________________________________ -->
2381 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
2383 <div class="doc_text">
2386 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2391 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
2396 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
2397 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2398 values. Both arguments must have identical types.</p>
2401 <p>The value produced is the signed integer quotient of the two operands rounded towards zero.</p>
2402 <p>Note that signed integer division and unsigned integer division are distinct
2403 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
2404 <p>Division by zero leads to undefined behavior. Overflow also leads to
2405 undefined behavior; this is a rare case, but can occur, for example,
2406 by doing a 32-bit division of -2147483648 by -1.</p>
2408 <pre> <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2411 <!-- _______________________________________________________________________ -->
2412 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
2413 Instruction</a> </div>
2414 <div class="doc_text">
2417 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2421 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
2426 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
2427 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a>
2428 of floating point values. Both arguments must have identical types.</p>
2432 <p>The value produced is the floating point quotient of the two operands.</p>
2437 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
2441 <!-- _______________________________________________________________________ -->
2442 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
2444 <div class="doc_text">
2446 <pre> <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2449 <p>The '<tt>urem</tt>' instruction returns the remainder from the
2450 unsigned division of its two arguments.</p>
2452 <p>The two arguments to the '<tt>urem</tt>' instruction must be
2453 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2454 values. Both arguments must have identical types.</p>
2456 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
2457 This instruction always performs an unsigned division to get the remainder.</p>
2458 <p>Note that unsigned integer remainder and signed integer remainder are
2459 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
2460 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
2462 <pre> <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2466 <!-- _______________________________________________________________________ -->
2467 <div class="doc_subsubsection">
2468 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
2471 <div class="doc_text">
2476 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2481 <p>The '<tt>srem</tt>' instruction returns the remainder from the
2482 signed division of its two operands. This instruction can also take
2483 <a href="#t_vector">vector</a> versions of the values in which case
2484 the elements must be integers.</p>
2488 <p>The two arguments to the '<tt>srem</tt>' instruction must be
2489 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2490 values. Both arguments must have identical types.</p>
2494 <p>This instruction returns the <i>remainder</i> of a division (where the result
2495 has the same sign as the dividend, <tt>op1</tt>), not the <i>modulo</i>
2496 operator (where the result has the same sign as the divisor, <tt>op2</tt>) of
2497 a value. For more information about the difference, see <a
2498 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
2499 Math Forum</a>. For a table of how this is implemented in various languages,
2500 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
2501 Wikipedia: modulo operation</a>.</p>
2502 <p>Note that signed integer remainder and unsigned integer remainder are
2503 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
2504 <p>Taking the remainder of a division by zero leads to undefined behavior.
2505 Overflow also leads to undefined behavior; this is a rare case, but can occur,
2506 for example, by taking the remainder of a 32-bit division of -2147483648 by -1.
2507 (The remainder doesn't actually overflow, but this rule lets srem be
2508 implemented using instructions that return both the result of the division
2509 and the remainder.)</p>
2511 <pre> <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2515 <!-- _______________________________________________________________________ -->
2516 <div class="doc_subsubsection">
2517 <a name="i_frem">'<tt>frem</tt>' Instruction</a> </div>
2519 <div class="doc_text">
2522 <pre> <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2525 <p>The '<tt>frem</tt>' instruction returns the remainder from the
2526 division of its two operands.</p>
2528 <p>The two arguments to the '<tt>frem</tt>' instruction must be
2529 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a>
2530 of floating point values. Both arguments must have identical types.</p>
2534 <p>This instruction returns the <i>remainder</i> of a division.
2535 The remainder has the same sign as the dividend.</p>
2540 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
2544 <!-- ======================================================================= -->
2545 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
2546 Operations</a> </div>
2547 <div class="doc_text">
2548 <p>Bitwise binary operators are used to do various forms of
2549 bit-twiddling in a program. They are generally very efficient
2550 instructions and can commonly be strength reduced from other
2551 instructions. They require two operands of the same type, execute an operation on them,
2552 and produce a single value. The resulting value is the same type as its operands.</p>
2555 <!-- _______________________________________________________________________ -->
2556 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
2557 Instruction</a> </div>
2558 <div class="doc_text">
2560 <pre> <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2565 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2566 the left a specified number of bits.</p>
2570 <p>Both arguments to the '<tt>shl</tt>' instruction must be the same <a
2571 href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2572 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
2576 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod 2<sup>n</sup>,
2577 where n is the width of the result. If <tt>op2</tt> is (statically or dynamically) negative or
2578 equal to or larger than the number of bits in <tt>op1</tt>, the result is undefined.</p>
2580 <h5>Example:</h5><pre>
2581 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
2582 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
2583 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
2584 <result> = shl i32 1, 32 <i>; undefined</i>
2587 <!-- _______________________________________________________________________ -->
2588 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
2589 Instruction</a> </div>
2590 <div class="doc_text">
2592 <pre> <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2596 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
2597 operand shifted to the right a specified number of bits with zero fill.</p>
2600 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
2601 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2602 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
2606 <p>This instruction always performs a logical shift right operation. The most
2607 significant bits of the result will be filled with zero bits after the
2608 shift. If <tt>op2</tt> is (statically or dynamically) equal to or larger than
2609 the number of bits in <tt>op1</tt>, the result is undefined.</p>
2613 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
2614 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
2615 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
2616 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
2617 <result> = lshr i32 1, 32 <i>; undefined</i>
2621 <!-- _______________________________________________________________________ -->
2622 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
2623 Instruction</a> </div>
2624 <div class="doc_text">
2627 <pre> <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2631 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
2632 operand shifted to the right a specified number of bits with sign extension.</p>
2635 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
2636 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2637 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
2640 <p>This instruction always performs an arithmetic shift right operation,
2641 The most significant bits of the result will be filled with the sign bit
2642 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
2643 larger than the number of bits in <tt>op1</tt>, the result is undefined.
2648 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
2649 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
2650 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
2651 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
2652 <result> = ashr i32 1, 32 <i>; undefined</i>
2656 <!-- _______________________________________________________________________ -->
2657 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
2658 Instruction</a> </div>
2660 <div class="doc_text">
2665 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2670 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
2671 its two operands.</p>
2675 <p>The two arguments to the '<tt>and</tt>' instruction must be
2676 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2677 values. Both arguments must have identical types.</p>
2680 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
2683 <table border="1" cellspacing="0" cellpadding="4">
2715 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
2716 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
2717 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
2720 <!-- _______________________________________________________________________ -->
2721 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
2722 <div class="doc_text">
2724 <pre> <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2727 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
2728 or of its two operands.</p>
2731 <p>The two arguments to the '<tt>or</tt>' instruction must be
2732 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2733 values. Both arguments must have identical types.</p>
2735 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
2738 <table border="1" cellspacing="0" cellpadding="4">
2769 <pre> <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
2770 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
2771 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
2774 <!-- _______________________________________________________________________ -->
2775 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
2776 Instruction</a> </div>
2777 <div class="doc_text">
2779 <pre> <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2782 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
2783 or of its two operands. The <tt>xor</tt> is used to implement the
2784 "one's complement" operation, which is the "~" operator in C.</p>
2786 <p>The two arguments to the '<tt>xor</tt>' instruction must be
2787 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2788 values. Both arguments must have identical types.</p>
2792 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
2795 <table border="1" cellspacing="0" cellpadding="4">
2827 <pre> <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
2828 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
2829 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
2830 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
2834 <!-- ======================================================================= -->
2835 <div class="doc_subsection">
2836 <a name="vectorops">Vector Operations</a>
2839 <div class="doc_text">
2841 <p>LLVM supports several instructions to represent vector operations in a
2842 target-independent manner. These instructions cover the element-access and
2843 vector-specific operations needed to process vectors effectively. While LLVM
2844 does directly support these vector operations, many sophisticated algorithms
2845 will want to use target-specific intrinsics to take full advantage of a specific
2850 <!-- _______________________________________________________________________ -->
2851 <div class="doc_subsubsection">
2852 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2855 <div class="doc_text">
2860 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
2866 The '<tt>extractelement</tt>' instruction extracts a single scalar
2867 element from a vector at a specified index.
2874 The first operand of an '<tt>extractelement</tt>' instruction is a
2875 value of <a href="#t_vector">vector</a> type. The second operand is
2876 an index indicating the position from which to extract the element.
2877 The index may be a variable.</p>
2882 The result is a scalar of the same type as the element type of
2883 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2884 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2885 results are undefined.
2891 %result = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
2896 <!-- _______________________________________________________________________ -->
2897 <div class="doc_subsubsection">
2898 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
2901 <div class="doc_text">
2906 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
2912 The '<tt>insertelement</tt>' instruction inserts a scalar
2913 element into a vector at a specified index.
2920 The first operand of an '<tt>insertelement</tt>' instruction is a
2921 value of <a href="#t_vector">vector</a> type. The second operand is a
2922 scalar value whose type must equal the element type of the first
2923 operand. The third operand is an index indicating the position at
2924 which to insert the value. The index may be a variable.</p>
2929 The result is a vector of the same type as <tt>val</tt>. Its
2930 element values are those of <tt>val</tt> except at position
2931 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
2932 exceeds the length of <tt>val</tt>, the results are undefined.
2938 %result = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
2942 <!-- _______________________________________________________________________ -->
2943 <div class="doc_subsubsection">
2944 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
2947 <div class="doc_text">
2952 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
2958 The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
2959 from two input vectors, returning a vector with the same element type as
2960 the input and length that is the same as the shuffle mask.
2966 The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
2967 with types that match each other. The third argument is a shuffle mask whose
2968 element type is always 'i32'. The result of the instruction is a vector whose
2969 length is the same as the shuffle mask and whose element type is the same as
2970 the element type of the first two operands.
2974 The shuffle mask operand is required to be a constant vector with either
2975 constant integer or undef values.
2981 The elements of the two input vectors are numbered from left to right across
2982 both of the vectors. The shuffle mask operand specifies, for each element of
2983 the result vector, which element of the two input vectors the result element
2984 gets. The element selector may be undef (meaning "don't care") and the second
2985 operand may be undef if performing a shuffle from only one vector.
2991 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
2992 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
2993 %result = shufflevector <4 x i32> %v1, <4 x i32> undef,
2994 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
2995 %result = shufflevector <8 x i32> %v1, <8 x i32> undef,
2996 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i>
2997 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
2998 <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>
3003 <!-- ======================================================================= -->
3004 <div class="doc_subsection">
3005 <a name="aggregateops">Aggregate Operations</a>
3008 <div class="doc_text">
3010 <p>LLVM supports several instructions for working with aggregate values.
3015 <!-- _______________________________________________________________________ -->
3016 <div class="doc_subsubsection">
3017 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
3020 <div class="doc_text">
3025 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
3031 The '<tt>extractvalue</tt>' instruction extracts the value of a struct field
3032 or array element from an aggregate value.
3039 The first operand of an '<tt>extractvalue</tt>' instruction is a
3040 value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a>
3041 type. The operands are constant indices to specify which value to extract
3042 in a similar manner as indices in a
3043 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
3049 The result is the value at the position in the aggregate specified by
3056 %result = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
3061 <!-- _______________________________________________________________________ -->
3062 <div class="doc_subsubsection">
3063 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
3066 <div class="doc_text">
3071 <result> = insertvalue <aggregate type> <val>, <ty> <val>, <idx> <i>; yields <n x <ty>></i>
3077 The '<tt>insertvalue</tt>' instruction inserts a value
3078 into a struct field or array element in an aggregate.
3085 The first operand of an '<tt>insertvalue</tt>' instruction is a
3086 value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a> type.
3087 The second operand is a first-class value to insert.
3088 The following operands are constant indices
3089 indicating the position at which to insert the value in a similar manner as
3091 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
3092 The value to insert must have the same type as the value identified
3099 The result is an aggregate of the same type as <tt>val</tt>. Its
3100 value is that of <tt>val</tt> except that the value at the position
3101 specified by the indices is that of <tt>elt</tt>.
3107 %result = insertvalue {i32, float} %agg, i32 1, 0 <i>; yields {i32, float}</i>
3112 <!-- ======================================================================= -->
3113 <div class="doc_subsection">
3114 <a name="memoryops">Memory Access and Addressing Operations</a>
3117 <div class="doc_text">
3119 <p>A key design point of an SSA-based representation is how it
3120 represents memory. In LLVM, no memory locations are in SSA form, which
3121 makes things very simple. This section describes how to read, write,
3122 allocate, and free memory in LLVM.</p>
3126 <!-- _______________________________________________________________________ -->
3127 <div class="doc_subsubsection">
3128 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
3131 <div class="doc_text">
3136 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3141 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
3142 heap and returns a pointer to it. The object is always allocated in the generic
3143 address space (address space zero).</p>
3147 <p>The '<tt>malloc</tt>' instruction allocates
3148 <tt>sizeof(<type>)*NumElements</tt>
3149 bytes of memory from the operating system and returns a pointer of the
3150 appropriate type to the program. If "NumElements" is specified, it is the
3151 number of elements allocated, otherwise "NumElements" is defaulted to be one.
3152 If a constant alignment is specified, the value result of the allocation is guaranteed to
3153 be aligned to at least that boundary. If not specified, or if zero, the target can
3154 choose to align the allocation on any convenient boundary.</p>
3156 <p>'<tt>type</tt>' must be a sized type.</p>
3160 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
3161 a pointer is returned. The result of a zero byte allocattion is undefined. The
3162 result is null if there is insufficient memory available.</p>
3167 %array = malloc [4 x i8 ] <i>; yields {[%4 x i8]*}:array</i>
3169 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
3170 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
3171 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
3172 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
3173 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
3177 <!-- _______________________________________________________________________ -->
3178 <div class="doc_subsubsection">
3179 <a name="i_free">'<tt>free</tt>' Instruction</a>
3182 <div class="doc_text">
3187 free <type> <value> <i>; yields {void}</i>
3192 <p>The '<tt>free</tt>' instruction returns memory back to the unused
3193 memory heap to be reallocated in the future.</p>
3197 <p>'<tt>value</tt>' shall be a pointer value that points to a value
3198 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
3203 <p>Access to the memory pointed to by the pointer is no longer defined
3204 after this instruction executes. If the pointer is null, the operation
3210 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
3211 free [4 x i8]* %array
3215 <!-- _______________________________________________________________________ -->
3216 <div class="doc_subsubsection">
3217 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
3220 <div class="doc_text">
3225 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3230 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
3231 currently executing function, to be automatically released when this function
3232 returns to its caller. The object is always allocated in the generic address
3233 space (address space zero).</p>
3237 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
3238 bytes of memory on the runtime stack, returning a pointer of the
3239 appropriate type to the program. If "NumElements" is specified, it is the
3240 number of elements allocated, otherwise "NumElements" is defaulted to be one.
3241 If a constant alignment is specified, the value result of the allocation is guaranteed
3242 to be aligned to at least that boundary. If not specified, or if zero, the target
3243 can choose to align the allocation on any convenient boundary.</p>
3245 <p>'<tt>type</tt>' may be any sized type.</p>
3249 <p>Memory is allocated; a pointer is returned. The operation is undefiend if
3250 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
3251 memory is automatically released when the function returns. The '<tt>alloca</tt>'
3252 instruction is commonly used to represent automatic variables that must
3253 have an address available. When the function returns (either with the <tt><a
3254 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
3255 instructions), the memory is reclaimed. Allocating zero bytes
3256 is legal, but the result is undefined.</p>
3261 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
3262 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
3263 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
3264 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
3268 <!-- _______________________________________________________________________ -->
3269 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
3270 Instruction</a> </div>
3271 <div class="doc_text">
3273 <pre> <result> = load <ty>* <pointer>[, align <alignment>]<br> <result> = volatile load <ty>* <pointer>[, align <alignment>]<br></pre>
3275 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
3277 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
3278 address from which to load. The pointer must point to a <a
3279 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
3280 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
3281 the number or order of execution of this <tt>load</tt> with other
3282 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
3285 The optional constant "align" argument specifies the alignment of the operation
3286 (that is, the alignment of the memory address). A value of 0 or an
3287 omitted "align" argument means that the operation has the preferential
3288 alignment for the target. It is the responsibility of the code emitter
3289 to ensure that the alignment information is correct. Overestimating
3290 the alignment results in an undefined behavior. Underestimating the
3291 alignment may produce less efficient code. An alignment of 1 is always
3295 <p>The location of memory pointed to is loaded.</p>
3297 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3299 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
3300 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
3303 <!-- _______________________________________________________________________ -->
3304 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
3305 Instruction</a> </div>
3306 <div class="doc_text">
3308 <pre> store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3309 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3312 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
3314 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
3315 to store and an address at which to store it. The type of the '<tt><pointer></tt>'
3316 operand must be a pointer to the <a href="#t_firstclass">first class</a> type
3317 of the '<tt><value></tt>'
3318 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
3319 optimizer is not allowed to modify the number or order of execution of
3320 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
3321 href="#i_store">store</a></tt> instructions.</p>
3323 The optional constant "align" argument specifies the alignment of the operation
3324 (that is, the alignment of the memory address). A value of 0 or an
3325 omitted "align" argument means that the operation has the preferential
3326 alignment for the target. It is the responsibility of the code emitter
3327 to ensure that the alignment information is correct. Overestimating
3328 the alignment results in an undefined behavior. Underestimating the
3329 alignment may produce less efficient code. An alignment of 1 is always
3333 <p>The contents of memory are updated to contain '<tt><value></tt>'
3334 at the location specified by the '<tt><pointer></tt>' operand.</p>
3336 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3337 store i32 3, i32* %ptr <i>; yields {void}</i>
3338 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
3342 <!-- _______________________________________________________________________ -->
3343 <div class="doc_subsubsection">
3344 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
3347 <div class="doc_text">
3350 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
3356 The '<tt>getelementptr</tt>' instruction is used to get the address of a
3357 subelement of an aggregate data structure. It performs address calculation only
3358 and does not access memory.</p>
3362 <p>The first argument is always a pointer, and forms the basis of the
3363 calculation. The remaining arguments are indices, that indicate which of the
3364 elements of the aggregate object are indexed. The interpretation of each index
3365 is dependent on the type being indexed into. The first index always indexes the
3366 pointer value given as the first argument, the second index indexes a value of
3367 the type pointed to (not necessarily the value directly pointed to, since the
3368 first index can be non-zero), etc. The first type indexed into must be a pointer
3369 value, subsequent types can be arrays, vectors and structs. Note that subsequent
3370 types being indexed into can never be pointers, since that would require loading
3371 the pointer before continuing calculation.</p>
3373 <p>The type of each index argument depends on the type it is indexing into.
3374 When indexing into a (packed) structure, only <tt>i32</tt> integer
3375 <b>constants</b> are allowed. When indexing into an array, pointer or vector,
3376 only integers of 32 or 64 bits are allowed (also non-constants). 32-bit values
3377 will be sign extended to 64-bits if required.</p>
3379 <p>For example, let's consider a C code fragment and how it gets
3380 compiled to LLVM:</p>
3382 <div class="doc_code">
3395 int *foo(struct ST *s) {
3396 return &s[1].Z.B[5][13];
3401 <p>The LLVM code generated by the GCC frontend is:</p>
3403 <div class="doc_code">
3405 %RT = type { i8 , [10 x [20 x i32]], i8 }
3406 %ST = type { i32, double, %RT }
3408 define i32* %foo(%ST* %s) {
3410 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
3418 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
3419 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
3420 }</tt>' type, a structure. The second index indexes into the third element of
3421 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
3422 i8 }</tt>' type, another structure. The third index indexes into the second
3423 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
3424 array. The two dimensions of the array are subscripted into, yielding an
3425 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
3426 to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
3428 <p>Note that it is perfectly legal to index partially through a
3429 structure, returning a pointer to an inner element. Because of this,
3430 the LLVM code for the given testcase is equivalent to:</p>
3433 define i32* %foo(%ST* %s) {
3434 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
3435 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
3436 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
3437 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
3438 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
3443 <p>Note that it is undefined to access an array out of bounds: array and
3444 pointer indexes must always be within the defined bounds of the array type.
3445 The one exception for this rule is zero length arrays. These arrays are
3446 defined to be accessible as variable length arrays, which requires access
3447 beyond the zero'th element.</p>
3449 <p>The getelementptr instruction is often confusing. For some more insight
3450 into how it works, see <a href="GetElementPtr.html">the getelementptr
3456 <i>; yields [12 x i8]*:aptr</i>
3457 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
3458 <i>; yields i8*:vptr</i>
3459 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
3460 <i>; yields i8*:eptr</i>
3461 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
3465 <!-- ======================================================================= -->
3466 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
3468 <div class="doc_text">
3469 <p>The instructions in this category are the conversion instructions (casting)
3470 which all take a single operand and a type. They perform various bit conversions
3474 <!-- _______________________________________________________________________ -->
3475 <div class="doc_subsubsection">
3476 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
3478 <div class="doc_text">
3482 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
3487 The '<tt>trunc</tt>' instruction truncates its operand to the type <tt>ty2</tt>.
3492 The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
3493 be an <a href="#t_integer">integer</a> type, and a type that specifies the size
3494 and type of the result, which must be an <a href="#t_integer">integer</a>
3495 type. The bit size of <tt>value</tt> must be larger than the bit size of
3496 <tt>ty2</tt>. Equal sized types are not allowed.</p>
3500 The '<tt>trunc</tt>' instruction truncates the high order bits in <tt>value</tt>
3501 and converts the remaining bits to <tt>ty2</tt>. Since the source size must be
3502 larger than the destination size, <tt>trunc</tt> cannot be a <i>no-op cast</i>.
3503 It will always truncate bits.</p>
3507 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
3508 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
3509 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
3513 <!-- _______________________________________________________________________ -->
3514 <div class="doc_subsubsection">
3515 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
3517 <div class="doc_text">
3521 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
3525 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
3530 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
3531 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3532 also be of <a href="#t_integer">integer</a> type. The bit size of the
3533 <tt>value</tt> must be smaller than the bit size of the destination type,
3537 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
3538 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
3540 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
3544 %X = zext i32 257 to i64 <i>; yields i64:257</i>
3545 %Y = zext i1 true to i32 <i>; yields i32:1</i>
3549 <!-- _______________________________________________________________________ -->
3550 <div class="doc_subsubsection">
3551 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
3553 <div class="doc_text">
3557 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
3561 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
3565 The '<tt>sext</tt>' instruction takes a value to cast, which must be of
3566 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3567 also be of <a href="#t_integer">integer</a> type. The bit size of the
3568 <tt>value</tt> must be smaller than the bit size of the destination type,
3573 The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
3574 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size of
3575 the type <tt>ty2</tt>.</p>
3577 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
3581 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
3582 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
3586 <!-- _______________________________________________________________________ -->
3587 <div class="doc_subsubsection">
3588 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
3591 <div class="doc_text">
3596 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
3600 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
3605 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
3606 point</a> value to cast and a <a href="#t_floating">floating point</a> type to
3607 cast it to. The size of <tt>value</tt> must be larger than the size of
3608 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
3609 <i>no-op cast</i>.</p>
3612 <p> The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
3613 <a href="#t_floating">floating point</a> type to a smaller
3614 <a href="#t_floating">floating point</a> type. If the value cannot fit within
3615 the destination type, <tt>ty2</tt>, then the results are undefined.</p>
3619 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
3620 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
3624 <!-- _______________________________________________________________________ -->
3625 <div class="doc_subsubsection">
3626 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
3628 <div class="doc_text">
3632 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
3636 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
3637 floating point value.</p>
3640 <p>The '<tt>fpext</tt>' instruction takes a
3641 <a href="#t_floating">floating point</a> <tt>value</tt> to cast,
3642 and a <a href="#t_floating">floating point</a> type to cast it to. The source
3643 type must be smaller than the destination type.</p>
3646 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
3647 <a href="#t_floating">floating point</a> type to a larger
3648 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
3649 used to make a <i>no-op cast</i> because it always changes bits. Use
3650 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
3654 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
3655 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
3659 <!-- _______________________________________________________________________ -->
3660 <div class="doc_subsubsection">
3661 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
3663 <div class="doc_text">
3667 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
3671 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
3672 unsigned integer equivalent of type <tt>ty2</tt>.
3676 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
3677 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3678 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3679 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3680 vector integer type with the same number of elements as <tt>ty</tt></p>
3683 <p> The '<tt>fptoui</tt>' instruction converts its
3684 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3685 towards zero) unsigned integer value. If the value cannot fit in <tt>ty2</tt>,
3686 the results are undefined.</p>
3690 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
3691 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
3692 %X = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
3696 <!-- _______________________________________________________________________ -->
3697 <div class="doc_subsubsection">
3698 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
3700 <div class="doc_text">
3704 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
3708 <p>The '<tt>fptosi</tt>' instruction converts
3709 <a href="#t_floating">floating point</a> <tt>value</tt> to type <tt>ty2</tt>.
3713 <p> The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
3714 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3715 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3716 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3717 vector integer type with the same number of elements as <tt>ty</tt></p>
3720 <p>The '<tt>fptosi</tt>' instruction converts its
3721 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3722 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
3723 the results are undefined.</p>
3727 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
3728 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
3729 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
3733 <!-- _______________________________________________________________________ -->
3734 <div class="doc_subsubsection">
3735 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
3737 <div class="doc_text">
3741 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3745 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
3746 integer and converts that value to the <tt>ty2</tt> type.</p>
3749 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
3750 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3751 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3752 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3753 floating point type with the same number of elements as <tt>ty</tt></p>
3756 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
3757 integer quantity and converts it to the corresponding floating point value. If
3758 the value cannot fit in the floating point value, the results are undefined.</p>
3762 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
3763 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
3767 <!-- _______________________________________________________________________ -->
3768 <div class="doc_subsubsection">
3769 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
3771 <div class="doc_text">
3775 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3779 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed
3780 integer and converts that value to the <tt>ty2</tt> type.</p>
3783 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
3784 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3785 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3786 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3787 floating point type with the same number of elements as <tt>ty</tt></p>
3790 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed
3791 integer quantity and converts it to the corresponding floating point value. If
3792 the value cannot fit in the floating point value, the results are undefined.</p>
3796 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
3797 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
3801 <!-- _______________________________________________________________________ -->
3802 <div class="doc_subsubsection">
3803 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
3805 <div class="doc_text">
3809 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
3813 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
3814 the integer type <tt>ty2</tt>.</p>
3817 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
3818 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
3819 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.</p>
3822 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
3823 <tt>ty2</tt> by interpreting the pointer value as an integer and either
3824 truncating or zero extending that value to the size of the integer type. If
3825 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
3826 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
3827 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
3832 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
3833 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
3837 <!-- _______________________________________________________________________ -->
3838 <div class="doc_subsubsection">
3839 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
3841 <div class="doc_text">
3845 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
3849 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to
3850 a pointer type, <tt>ty2</tt>.</p>
3853 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
3854 value to cast, and a type to cast it to, which must be a
3855 <a href="#t_pointer">pointer</a> type.</p>
3858 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
3859 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
3860 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
3861 size of a pointer then a truncation is done. If <tt>value</tt> is smaller than
3862 the size of a pointer then a zero extension is done. If they are the same size,
3863 nothing is done (<i>no-op cast</i>).</p>
3867 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
3868 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
3869 %Y = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
3873 <!-- _______________________________________________________________________ -->
3874 <div class="doc_subsubsection">
3875 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
3877 <div class="doc_text">
3881 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
3886 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3887 <tt>ty2</tt> without changing any bits.</p>
3891 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be
3892 a non-aggregate first class value, and a type to cast it to, which must also be
3893 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes of
3895 and the destination type, <tt>ty2</tt>, must be identical. If the source
3896 type is a pointer, the destination type must also be a pointer. This
3897 instruction supports bitwise conversion of vectors to integers and to vectors
3898 of other types (as long as they have the same size).</p>
3901 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3902 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
3903 this conversion. The conversion is done as if the <tt>value</tt> had been
3904 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only be
3905 converted to other pointer types with this instruction. To convert pointers to
3906 other types, use the <a href="#i_inttoptr">inttoptr</a> or
3907 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
3911 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
3912 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
3913 %Z = bitcast <2 x int> %V to i64; <i>; yields i64: %V</i>
3917 <!-- ======================================================================= -->
3918 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
3919 <div class="doc_text">
3920 <p>The instructions in this category are the "miscellaneous"
3921 instructions, which defy better classification.</p>
3924 <!-- _______________________________________________________________________ -->
3925 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
3927 <div class="doc_text">
3929 <pre> <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
3932 <p>The '<tt>icmp</tt>' instruction returns a boolean value or
3933 a vector of boolean values based on comparison
3934 of its two integer, integer vector, or pointer operands.</p>
3936 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
3937 the condition code indicating the kind of comparison to perform. It is not
3938 a value, just a keyword. The possible condition code are:
3941 <li><tt>eq</tt>: equal</li>
3942 <li><tt>ne</tt>: not equal </li>
3943 <li><tt>ugt</tt>: unsigned greater than</li>
3944 <li><tt>uge</tt>: unsigned greater or equal</li>
3945 <li><tt>ult</tt>: unsigned less than</li>
3946 <li><tt>ule</tt>: unsigned less or equal</li>
3947 <li><tt>sgt</tt>: signed greater than</li>
3948 <li><tt>sge</tt>: signed greater or equal</li>
3949 <li><tt>slt</tt>: signed less than</li>
3950 <li><tt>sle</tt>: signed less or equal</li>
3952 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
3953 <a href="#t_pointer">pointer</a>
3954 or integer <a href="#t_vector">vector</a> typed.
3955 They must also be identical types.</p>
3957 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to
3958 the condition code given as <tt>cond</tt>. The comparison performed always
3959 yields either an <a href="#t_primitive"><tt>i1</tt></a> or vector of <tt>i1</tt> result, as follows:
3962 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
3963 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3965 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
3966 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.</li>
3967 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
3968 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
3969 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
3970 <tt>true</tt> if <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
3971 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
3972 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
3973 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
3974 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
3975 <li><tt>sgt</tt>: interprets the operands as signed values and yields
3976 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
3977 <li><tt>sge</tt>: interprets the operands as signed values and yields
3978 <tt>true</tt> if <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
3979 <li><tt>slt</tt>: interprets the operands as signed values and yields
3980 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
3981 <li><tt>sle</tt>: interprets the operands as signed values and yields
3982 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
3984 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
3985 values are compared as if they were integers.</p>
3986 <p>If the operands are integer vectors, then they are compared
3987 element by element. The result is an <tt>i1</tt> vector with
3988 the same number of elements as the values being compared.
3989 Otherwise, the result is an <tt>i1</tt>.
3993 <pre> <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
3994 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
3995 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
3996 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
3997 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
3998 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
4002 <!-- _______________________________________________________________________ -->
4003 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
4005 <div class="doc_text">
4007 <pre> <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4010 <p>The '<tt>fcmp</tt>' instruction returns a boolean value
4011 or vector of boolean values based on comparison
4012 of its operands.</p>
4014 If the operands are floating point scalars, then the result
4015 type is a boolean (<a href="#t_primitive"><tt>i1</tt></a>).
4017 <p>If the operands are floating point vectors, then the result type
4018 is a vector of boolean with the same number of elements as the
4019 operands being compared.</p>
4021 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
4022 the condition code indicating the kind of comparison to perform. It is not
4023 a value, just a keyword. The possible condition code are:</p>
4025 <li><tt>false</tt>: no comparison, always returns false</li>
4026 <li><tt>oeq</tt>: ordered and equal</li>
4027 <li><tt>ogt</tt>: ordered and greater than </li>
4028 <li><tt>oge</tt>: ordered and greater than or equal</li>
4029 <li><tt>olt</tt>: ordered and less than </li>
4030 <li><tt>ole</tt>: ordered and less than or equal</li>
4031 <li><tt>one</tt>: ordered and not equal</li>
4032 <li><tt>ord</tt>: ordered (no nans)</li>
4033 <li><tt>ueq</tt>: unordered or equal</li>
4034 <li><tt>ugt</tt>: unordered or greater than </li>
4035 <li><tt>uge</tt>: unordered or greater than or equal</li>
4036 <li><tt>ult</tt>: unordered or less than </li>
4037 <li><tt>ule</tt>: unordered or less than or equal</li>
4038 <li><tt>une</tt>: unordered or not equal</li>
4039 <li><tt>uno</tt>: unordered (either nans)</li>
4040 <li><tt>true</tt>: no comparison, always returns true</li>
4042 <p><i>Ordered</i> means that neither operand is a QNAN while
4043 <i>unordered</i> means that either operand may be a QNAN.</p>
4044 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be
4045 either a <a href="#t_floating">floating point</a> type
4046 or a <a href="#t_vector">vector</a> of floating point type.
4047 They must have identical types.</p>
4049 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4050 according to the condition code given as <tt>cond</tt>.
4051 If the operands are vectors, then the vectors are compared
4053 Each comparison performed
4054 always yields an <a href="#t_primitive">i1</a> result, as follows:</p>
4056 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
4057 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4058 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
4059 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4060 <tt>op1</tt> is greather than <tt>op2</tt>.</li>
4061 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4062 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4063 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4064 <tt>op1</tt> is less than <tt>op2</tt>.</li>
4065 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4066 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4067 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4068 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
4069 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
4070 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
4071 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
4072 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
4073 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4074 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
4075 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4076 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
4077 <tt>op1</tt> is less than <tt>op2</tt>.</li>
4078 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
4079 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4080 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
4081 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
4082 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
4083 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
4087 <pre> <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
4088 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
4089 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
4090 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
4094 <!-- _______________________________________________________________________ -->
4095 <div class="doc_subsubsection">
4096 <a name="i_vicmp">'<tt>vicmp</tt>' Instruction</a>
4098 <div class="doc_text">
4100 <pre> <result> = vicmp <cond> <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4103 <p>The '<tt>vicmp</tt>' instruction returns an integer vector value based on
4104 element-wise comparison of its two integer vector operands.</p>
4106 <p>The '<tt>vicmp</tt>' instruction takes three operands. The first operand is
4107 the condition code indicating the kind of comparison to perform. It is not
4108 a value, just a keyword. The possible condition code are:</p>
4110 <li><tt>eq</tt>: equal</li>
4111 <li><tt>ne</tt>: not equal </li>
4112 <li><tt>ugt</tt>: unsigned greater than</li>
4113 <li><tt>uge</tt>: unsigned greater or equal</li>
4114 <li><tt>ult</tt>: unsigned less than</li>
4115 <li><tt>ule</tt>: unsigned less or equal</li>
4116 <li><tt>sgt</tt>: signed greater than</li>
4117 <li><tt>sge</tt>: signed greater or equal</li>
4118 <li><tt>slt</tt>: signed less than</li>
4119 <li><tt>sle</tt>: signed less or equal</li>
4121 <p>The remaining two arguments must be <a href="#t_vector">vector</a> or
4122 <a href="#t_integer">integer</a> typed. They must also be identical types.</p>
4124 <p>The '<tt>vicmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4125 according to the condition code given as <tt>cond</tt>. The comparison yields a
4126 <a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, of
4127 identical type as the values being compared. The most significant bit in each
4128 element is 1 if the element-wise comparison evaluates to true, and is 0
4129 otherwise. All other bits of the result are undefined. The condition codes
4130 are evaluated identically to the <a href="#i_icmp">'<tt>icmp</tt>'
4131 instruction</a>.</p>
4135 <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>
4136 <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>
4140 <!-- _______________________________________________________________________ -->
4141 <div class="doc_subsubsection">
4142 <a name="i_vfcmp">'<tt>vfcmp</tt>' Instruction</a>
4144 <div class="doc_text">
4146 <pre> <result> = vfcmp <cond> <ty> <op1>, <op2></pre>
4148 <p>The '<tt>vfcmp</tt>' instruction returns an integer vector value based on
4149 element-wise comparison of its two floating point vector operands. The output
4150 elements have the same width as the input elements.</p>
4152 <p>The '<tt>vfcmp</tt>' instruction takes three operands. The first operand is
4153 the condition code indicating the kind of comparison to perform. It is not
4154 a value, just a keyword. The possible condition code are:</p>
4156 <li><tt>false</tt>: no comparison, always returns false</li>
4157 <li><tt>oeq</tt>: ordered and equal</li>
4158 <li><tt>ogt</tt>: ordered and greater than </li>
4159 <li><tt>oge</tt>: ordered and greater than or equal</li>
4160 <li><tt>olt</tt>: ordered and less than </li>
4161 <li><tt>ole</tt>: ordered and less than or equal</li>
4162 <li><tt>one</tt>: ordered and not equal</li>
4163 <li><tt>ord</tt>: ordered (no nans)</li>
4164 <li><tt>ueq</tt>: unordered or equal</li>
4165 <li><tt>ugt</tt>: unordered or greater than </li>
4166 <li><tt>uge</tt>: unordered or greater than or equal</li>
4167 <li><tt>ult</tt>: unordered or less than </li>
4168 <li><tt>ule</tt>: unordered or less than or equal</li>
4169 <li><tt>une</tt>: unordered or not equal</li>
4170 <li><tt>uno</tt>: unordered (either nans)</li>
4171 <li><tt>true</tt>: no comparison, always returns true</li>
4173 <p>The remaining two arguments must be <a href="#t_vector">vector</a> of
4174 <a href="#t_floating">floating point</a> typed. They must also be identical
4177 <p>The '<tt>vfcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4178 according to the condition code given as <tt>cond</tt>. The comparison yields a
4179 <a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, with
4180 an identical number of elements as the values being compared, and each element
4181 having identical with to the width of the floating point elements. The most
4182 significant bit in each element is 1 if the element-wise comparison evaluates to
4183 true, and is 0 otherwise. All other bits of the result are undefined. The
4184 condition codes are evaluated identically to the
4185 <a href="#i_fcmp">'<tt>fcmp</tt>' instruction</a>.</p>
4189 <i>; yields: result=<2 x i32> < i32 0, i32 -1 ></i>
4190 <result> = vfcmp oeq <2 x float> < float 4, float 0 >, < float 5, float 0 >
4192 <i>; yields: result=<2 x i64> < i64 -1, i64 0 ></i>
4193 <result> = vfcmp ult <2 x double> < double 1, double 2 >, < double 2, double 2>
4197 <!-- _______________________________________________________________________ -->
4198 <div class="doc_subsubsection">
4199 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
4202 <div class="doc_text">
4206 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
4208 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
4209 the SSA graph representing the function.</p>
4212 <p>The type of the incoming values is specified with the first type
4213 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
4214 as arguments, with one pair for each predecessor basic block of the
4215 current block. Only values of <a href="#t_firstclass">first class</a>
4216 type may be used as the value arguments to the PHI node. Only labels
4217 may be used as the label arguments.</p>
4219 <p>There must be no non-phi instructions between the start of a basic
4220 block and the PHI instructions: i.e. PHI instructions must be first in
4225 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
4226 specified by the pair corresponding to the predecessor basic block that executed
4227 just prior to the current block.</p>
4231 Loop: ; Infinite loop that counts from 0 on up...
4232 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
4233 %nextindvar = add i32 %indvar, 1
4238 <!-- _______________________________________________________________________ -->
4239 <div class="doc_subsubsection">
4240 <a name="i_select">'<tt>select</tt>' Instruction</a>
4243 <div class="doc_text">
4248 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
4250 <i>selty</i> is either i1 or {<N x i1>}
4256 The '<tt>select</tt>' instruction is used to choose one value based on a
4257 condition, without branching.
4264 The '<tt>select</tt>' instruction requires an 'i1' value or
4265 a vector of 'i1' values indicating the
4266 condition, and two values of the same <a href="#t_firstclass">first class</a>
4267 type. If the val1/val2 are vectors and
4268 the condition is a scalar, then entire vectors are selected, not
4269 individual elements.
4275 If the condition is an i1 and it evaluates to 1, the instruction returns the first
4276 value argument; otherwise, it returns the second value argument.
4279 If the condition is a vector of i1, then the value arguments must
4280 be vectors of the same size, and the selection is done element
4287 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
4292 <!-- _______________________________________________________________________ -->
4293 <div class="doc_subsubsection">
4294 <a name="i_call">'<tt>call</tt>' Instruction</a>
4297 <div class="doc_text">
4301 <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>]
4306 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
4310 <p>This instruction requires several arguments:</p>
4314 <p>The optional "tail" marker indicates whether the callee function accesses
4315 any allocas or varargs in the caller. If the "tail" marker is present, the
4316 function call is eligible for tail call optimization. Note that calls may
4317 be marked "tail" even if they do not occur before a <a
4318 href="#i_ret"><tt>ret</tt></a> instruction.</p>
4321 <p>The optional "cconv" marker indicates which <a href="#callingconv">calling
4322 convention</a> the call should use. If none is specified, the call defaults
4323 to using C calling conventions.</p>
4327 <p>The optional <a href="#paramattrs">Parameter Attributes</a> list for
4328 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>',
4329 and '<tt>inreg</tt>' attributes are valid here.</p>
4333 <p>'<tt>ty</tt>': the type of the call instruction itself which is also
4334 the type of the return value. Functions that return no value are marked
4335 <tt><a href="#t_void">void</a></tt>.</p>
4338 <p>'<tt>fnty</tt>': shall be the signature of the pointer to function
4339 value being invoked. The argument types must match the types implied by
4340 this signature. This type can be omitted if the function is not varargs
4341 and if the function type does not return a pointer to a function.</p>
4344 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
4345 be invoked. In most cases, this is a direct function invocation, but
4346 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
4347 to function value.</p>
4350 <p>'<tt>function args</tt>': argument list whose types match the
4351 function signature argument types. All arguments must be of
4352 <a href="#t_firstclass">first class</a> type. If the function signature
4353 indicates the function accepts a variable number of arguments, the extra
4354 arguments can be specified.</p>
4357 <p>The optional <a href="#fnattrs">function attributes</a> list. Only
4358 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
4359 '<tt>readnone</tt>' attributes are valid here.</p>
4365 <p>The '<tt>call</tt>' instruction is used to cause control flow to
4366 transfer to a specified function, with its incoming arguments bound to
4367 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
4368 instruction in the called function, control flow continues with the
4369 instruction after the function call, and the return value of the
4370 function is bound to the result argument.</p>
4375 %retval = call i32 @test(i32 %argc)
4376 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42) <i>; yields i32</i>
4377 %X = tail call i32 @foo() <i>; yields i32</i>
4378 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
4379 call void %foo(i8 97 signext)
4381 %struct.A = type { i32, i8 }
4382 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
4383 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
4384 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
4385 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
4386 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
4391 <!-- _______________________________________________________________________ -->
4392 <div class="doc_subsubsection">
4393 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
4396 <div class="doc_text">
4401 <resultval> = va_arg <va_list*> <arglist>, <argty>
4406 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
4407 the "variable argument" area of a function call. It is used to implement the
4408 <tt>va_arg</tt> macro in C.</p>
4412 <p>This instruction takes a <tt>va_list*</tt> value and the type of
4413 the argument. It returns a value of the specified argument type and
4414 increments the <tt>va_list</tt> to point to the next argument. The
4415 actual type of <tt>va_list</tt> is target specific.</p>
4419 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
4420 type from the specified <tt>va_list</tt> and causes the
4421 <tt>va_list</tt> to point to the next argument. For more information,
4422 see the variable argument handling <a href="#int_varargs">Intrinsic
4425 <p>It is legal for this instruction to be called in a function which does not
4426 take a variable number of arguments, for example, the <tt>vfprintf</tt>
4429 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
4430 href="#intrinsics">intrinsic function</a> because it takes a type as an
4435 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
4439 <!-- *********************************************************************** -->
4440 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
4441 <!-- *********************************************************************** -->
4443 <div class="doc_text">
4445 <p>LLVM supports the notion of an "intrinsic function". These functions have
4446 well known names and semantics and are required to follow certain restrictions.
4447 Overall, these intrinsics represent an extension mechanism for the LLVM
4448 language that does not require changing all of the transformations in LLVM when
4449 adding to the language (or the bitcode reader/writer, the parser, etc...).</p>
4451 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
4452 prefix is reserved in LLVM for intrinsic names; thus, function names may not
4453 begin with this prefix. Intrinsic functions must always be external functions:
4454 you cannot define the body of intrinsic functions. Intrinsic functions may
4455 only be used in call or invoke instructions: it is illegal to take the address
4456 of an intrinsic function. Additionally, because intrinsic functions are part
4457 of the LLVM language, it is required if any are added that they be documented
4460 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
4461 a family of functions that perform the same operation but on different data
4462 types. Because LLVM can represent over 8 million different integer types,
4463 overloading is used commonly to allow an intrinsic function to operate on any
4464 integer type. One or more of the argument types or the result type can be
4465 overloaded to accept any integer type. Argument types may also be defined as
4466 exactly matching a previous argument's type or the result type. This allows an
4467 intrinsic function which accepts multiple arguments, but needs all of them to
4468 be of the same type, to only be overloaded with respect to a single argument or
4471 <p>Overloaded intrinsics will have the names of its overloaded argument types
4472 encoded into its function name, each preceded by a period. Only those types
4473 which are overloaded result in a name suffix. Arguments whose type is matched
4474 against another type do not. For example, the <tt>llvm.ctpop</tt> function can
4475 take an integer of any width and returns an integer of exactly the same integer
4476 width. This leads to a family of functions such as
4477 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29 %val)</tt>.
4478 Only one type, the return type, is overloaded, and only one type suffix is
4479 required. Because the argument's type is matched against the return type, it
4480 does not require its own name suffix.</p>
4482 <p>To learn how to add an intrinsic function, please see the
4483 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.
4488 <!-- ======================================================================= -->
4489 <div class="doc_subsection">
4490 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
4493 <div class="doc_text">
4495 <p>Variable argument support is defined in LLVM with the <a
4496 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
4497 intrinsic functions. These functions are related to the similarly
4498 named macros defined in the <tt><stdarg.h></tt> header file.</p>
4500 <p>All of these functions operate on arguments that use a
4501 target-specific value type "<tt>va_list</tt>". The LLVM assembly
4502 language reference manual does not define what this type is, so all
4503 transformations should be prepared to handle these functions regardless of
4506 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
4507 instruction and the variable argument handling intrinsic functions are
4510 <div class="doc_code">
4512 define i32 @test(i32 %X, ...) {
4513 ; Initialize variable argument processing
4515 %ap2 = bitcast i8** %ap to i8*
4516 call void @llvm.va_start(i8* %ap2)
4518 ; Read a single integer argument
4519 %tmp = va_arg i8** %ap, i32
4521 ; Demonstrate usage of llvm.va_copy and llvm.va_end
4523 %aq2 = bitcast i8** %aq to i8*
4524 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
4525 call void @llvm.va_end(i8* %aq2)
4527 ; Stop processing of arguments.
4528 call void @llvm.va_end(i8* %ap2)
4532 declare void @llvm.va_start(i8*)
4533 declare void @llvm.va_copy(i8*, i8*)
4534 declare void @llvm.va_end(i8*)
4540 <!-- _______________________________________________________________________ -->
4541 <div class="doc_subsubsection">
4542 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
4546 <div class="doc_text">
4548 <pre> declare void %llvm.va_start(i8* <arglist>)<br></pre>
4550 <p>The '<tt>llvm.va_start</tt>' intrinsic initializes
4551 <tt>*<arglist></tt> for subsequent use by <tt><a
4552 href="#i_va_arg">va_arg</a></tt>.</p>
4556 <p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
4560 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
4561 macro available in C. In a target-dependent way, it initializes the
4562 <tt>va_list</tt> element to which the argument points, so that the next call to
4563 <tt>va_arg</tt> will produce the first variable argument passed to the function.
4564 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
4565 last argument of the function as the compiler can figure that out.</p>
4569 <!-- _______________________________________________________________________ -->
4570 <div class="doc_subsubsection">
4571 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
4574 <div class="doc_text">
4576 <pre> declare void @llvm.va_end(i8* <arglist>)<br></pre>
4579 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
4580 which has been initialized previously with <tt><a href="#int_va_start">llvm.va_start</a></tt>
4581 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
4585 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
4589 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
4590 macro available in C. In a target-dependent way, it destroys the
4591 <tt>va_list</tt> element to which the argument points. Calls to <a
4592 href="#int_va_start"><tt>llvm.va_start</tt></a> and <a href="#int_va_copy">
4593 <tt>llvm.va_copy</tt></a> must be matched exactly with calls to
4594 <tt>llvm.va_end</tt>.</p>
4598 <!-- _______________________________________________________________________ -->
4599 <div class="doc_subsubsection">
4600 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
4603 <div class="doc_text">
4608 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
4613 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
4614 from the source argument list to the destination argument list.</p>
4618 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
4619 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
4624 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
4625 macro available in C. In a target-dependent way, it copies the source
4626 <tt>va_list</tt> element into the destination <tt>va_list</tt> element. This
4627 intrinsic is necessary because the <tt><a href="#int_va_start">
4628 llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
4629 example, memory allocation.</p>
4633 <!-- ======================================================================= -->
4634 <div class="doc_subsection">
4635 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
4638 <div class="doc_text">
4641 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
4642 Collection</a> (GC) requires the implementation and generation of these
4644 These intrinsics allow identification of <a href="#int_gcroot">GC roots on the
4645 stack</a>, as well as garbage collector implementations that require <a
4646 href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a> barriers.
4647 Front-ends for type-safe garbage collected languages should generate these
4648 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
4649 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
4652 <p>The garbage collection intrinsics only operate on objects in the generic
4653 address space (address space zero).</p>
4657 <!-- _______________________________________________________________________ -->
4658 <div class="doc_subsubsection">
4659 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
4662 <div class="doc_text">
4667 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
4672 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
4673 the code generator, and allows some metadata to be associated with it.</p>
4677 <p>The first argument specifies the address of a stack object that contains the
4678 root pointer. The second pointer (which must be either a constant or a global
4679 value address) contains the meta-data to be associated with the root.</p>
4683 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
4684 location. At compile-time, the code generator generates information to allow
4685 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
4686 intrinsic may only be used in a function which <a href="#gc">specifies a GC
4692 <!-- _______________________________________________________________________ -->
4693 <div class="doc_subsubsection">
4694 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
4697 <div class="doc_text">
4702 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
4707 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
4708 locations, allowing garbage collector implementations that require read
4713 <p>The second argument is the address to read from, which should be an address
4714 allocated from the garbage collector. The first object is a pointer to the
4715 start of the referenced object, if needed by the language runtime (otherwise
4720 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
4721 instruction, but may be replaced with substantially more complex code by the
4722 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
4723 may only be used in a function which <a href="#gc">specifies a GC
4729 <!-- _______________________________________________________________________ -->
4730 <div class="doc_subsubsection">
4731 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
4734 <div class="doc_text">
4739 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
4744 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
4745 locations, allowing garbage collector implementations that require write
4746 barriers (such as generational or reference counting collectors).</p>
4750 <p>The first argument is the reference to store, the second is the start of the
4751 object to store it to, and the third is the address of the field of Obj to
4752 store to. If the runtime does not require a pointer to the object, Obj may be
4757 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
4758 instruction, but may be replaced with substantially more complex code by the
4759 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
4760 may only be used in a function which <a href="#gc">specifies a GC
4767 <!-- ======================================================================= -->
4768 <div class="doc_subsection">
4769 <a name="int_codegen">Code Generator Intrinsics</a>
4772 <div class="doc_text">
4774 These intrinsics are provided by LLVM to expose special features that may only
4775 be implemented with code generator support.
4780 <!-- _______________________________________________________________________ -->
4781 <div class="doc_subsubsection">
4782 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
4785 <div class="doc_text">
4789 declare i8 *@llvm.returnaddress(i32 <level>)
4795 The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
4796 target-specific value indicating the return address of the current function
4797 or one of its callers.
4803 The argument to this intrinsic indicates which function to return the address
4804 for. Zero indicates the calling function, one indicates its caller, etc. The
4805 argument is <b>required</b> to be a constant integer value.
4811 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
4812 the return address of the specified call frame, or zero if it cannot be
4813 identified. The value returned by this intrinsic is likely to be incorrect or 0
4814 for arguments other than zero, so it should only be used for debugging purposes.
4818 Note that calling this intrinsic does not prevent function inlining or other
4819 aggressive transformations, so the value returned may not be that of the obvious
4820 source-language caller.
4825 <!-- _______________________________________________________________________ -->
4826 <div class="doc_subsubsection">
4827 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
4830 <div class="doc_text">
4834 declare i8 *@llvm.frameaddress(i32 <level>)
4840 The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
4841 target-specific frame pointer value for the specified stack frame.
4847 The argument to this intrinsic indicates which function to return the frame
4848 pointer for. Zero indicates the calling function, one indicates its caller,
4849 etc. The argument is <b>required</b> to be a constant integer value.
4855 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
4856 the frame address of the specified call frame, or zero if it cannot be
4857 identified. The value returned by this intrinsic is likely to be incorrect or 0
4858 for arguments other than zero, so it should only be used for debugging purposes.
4862 Note that calling this intrinsic does not prevent function inlining or other
4863 aggressive transformations, so the value returned may not be that of the obvious
4864 source-language caller.
4868 <!-- _______________________________________________________________________ -->
4869 <div class="doc_subsubsection">
4870 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
4873 <div class="doc_text">
4877 declare i8 *@llvm.stacksave()
4883 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
4884 the function stack, for use with <a href="#int_stackrestore">
4885 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
4886 features like scoped automatic variable sized arrays in C99.
4892 This intrinsic returns a opaque pointer value that can be passed to <a
4893 href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
4894 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
4895 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
4896 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
4897 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
4898 that were allocated after the <tt>llvm.stacksave</tt> was executed.
4903 <!-- _______________________________________________________________________ -->
4904 <div class="doc_subsubsection">
4905 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
4908 <div class="doc_text">
4912 declare void @llvm.stackrestore(i8 * %ptr)
4918 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
4919 the function stack to the state it was in when the corresponding <a
4920 href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
4921 useful for implementing language features like scoped automatic variable sized
4928 See the description for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.
4934 <!-- _______________________________________________________________________ -->
4935 <div class="doc_subsubsection">
4936 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
4939 <div class="doc_text">
4943 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>)
4950 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
4951 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
4953 effect on the behavior of the program but can change its performance
4960 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
4961 determining if the fetch should be for a read (0) or write (1), and
4962 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
4963 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
4964 <tt>locality</tt> arguments must be constant integers.
4970 This intrinsic does not modify the behavior of the program. In particular,
4971 prefetches cannot trap and do not produce a value. On targets that support this
4972 intrinsic, the prefetch can provide hints to the processor cache for better
4978 <!-- _______________________________________________________________________ -->
4979 <div class="doc_subsubsection">
4980 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
4983 <div class="doc_text">
4987 declare void @llvm.pcmarker(i32 <id>)
4994 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
4996 code to simulators and other tools. The method is target specific, but it is
4997 expected that the marker will use exported symbols to transmit the PC of the
4999 The marker makes no guarantees that it will remain with any specific instruction
5000 after optimizations. It is possible that the presence of a marker will inhibit
5001 optimizations. The intended use is to be inserted after optimizations to allow
5002 correlations of simulation runs.
5008 <tt>id</tt> is a numerical id identifying the marker.
5014 This intrinsic does not modify the behavior of the program. Backends that do not
5015 support this intrinisic may ignore it.
5020 <!-- _______________________________________________________________________ -->
5021 <div class="doc_subsubsection">
5022 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
5025 <div class="doc_text">
5029 declare i64 @llvm.readcyclecounter( )
5036 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
5037 counter register (or similar low latency, high accuracy clocks) on those targets
5038 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
5039 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
5040 should only be used for small timings.
5046 When directly supported, reading the cycle counter should not modify any memory.
5047 Implementations are allowed to either return a application specific value or a
5048 system wide value. On backends without support, this is lowered to a constant 0.
5053 <!-- ======================================================================= -->
5054 <div class="doc_subsection">
5055 <a name="int_libc">Standard C Library Intrinsics</a>
5058 <div class="doc_text">
5060 LLVM provides intrinsics for a few important standard C library functions.
5061 These intrinsics allow source-language front-ends to pass information about the
5062 alignment of the pointer arguments to the code generator, providing opportunity
5063 for more efficient code generation.
5068 <!-- _______________________________________________________________________ -->
5069 <div class="doc_subsubsection">
5070 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
5073 <div class="doc_text">
5077 declare void @llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
5078 i32 <len>, i32 <align>)
5079 declare void @llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
5080 i64 <len>, i32 <align>)
5086 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
5087 location to the destination location.
5091 Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
5092 intrinsics do not return a value, and takes an extra alignment argument.
5098 The first argument is a pointer to the destination, the second is a pointer to
5099 the source. The third argument is an integer argument
5100 specifying the number of bytes to copy, and the fourth argument is the alignment
5101 of the source and destination locations.
5105 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5106 the caller guarantees that both the source and destination pointers are aligned
5113 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
5114 location to the destination location, which are not allowed to overlap. It
5115 copies "len" bytes of memory over. If the argument is known to be aligned to
5116 some boundary, this can be specified as the fourth argument, otherwise it should
5122 <!-- _______________________________________________________________________ -->
5123 <div class="doc_subsubsection">
5124 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
5127 <div class="doc_text">
5131 declare void @llvm.memmove.i32(i8 * <dest>, i8 * <src>,
5132 i32 <len>, i32 <align>)
5133 declare void @llvm.memmove.i64(i8 * <dest>, i8 * <src>,
5134 i64 <len>, i32 <align>)
5140 The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
5141 location to the destination location. It is similar to the
5142 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to overlap.
5146 Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
5147 intrinsics do not return a value, and takes an extra alignment argument.
5153 The first argument is a pointer to the destination, the second is a pointer to
5154 the source. The third argument is an integer argument
5155 specifying the number of bytes to copy, and the fourth argument is the alignment
5156 of the source and destination locations.
5160 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5161 the caller guarantees that the source and destination pointers are aligned to
5168 The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
5169 location to the destination location, which may overlap. It
5170 copies "len" bytes of memory over. If the argument is known to be aligned to
5171 some boundary, this can be specified as the fourth argument, otherwise it should
5177 <!-- _______________________________________________________________________ -->
5178 <div class="doc_subsubsection">
5179 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
5182 <div class="doc_text">
5186 declare void @llvm.memset.i32(i8 * <dest>, i8 <val>,
5187 i32 <len>, i32 <align>)
5188 declare void @llvm.memset.i64(i8 * <dest>, i8 <val>,
5189 i64 <len>, i32 <align>)
5195 The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
5200 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
5201 does not return a value, and takes an extra alignment argument.
5207 The first argument is a pointer to the destination to fill, the second is the
5208 byte value to fill it with, the third argument is an integer
5209 argument specifying the number of bytes to fill, and the fourth argument is the
5210 known alignment of destination location.
5214 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5215 the caller guarantees that the destination pointer is aligned to that boundary.
5221 The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
5223 destination location. If the argument is known to be aligned to some boundary,
5224 this can be specified as the fourth argument, otherwise it should be set to 0 or
5230 <!-- _______________________________________________________________________ -->
5231 <div class="doc_subsubsection">
5232 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
5235 <div class="doc_text">
5238 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
5239 floating point or vector of floating point type. Not all targets support all
5242 declare float @llvm.sqrt.f32(float %Val)
5243 declare double @llvm.sqrt.f64(double %Val)
5244 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
5245 declare fp128 @llvm.sqrt.f128(fp128 %Val)
5246 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
5252 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
5253 returning the same value as the libm '<tt>sqrt</tt>' functions would. Unlike
5254 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
5255 negative numbers other than -0.0 (which allows for better optimization, because
5256 there is no need to worry about errno being set). <tt>llvm.sqrt(-0.0)</tt> is
5257 defined to return -0.0 like IEEE sqrt.
5263 The argument and return value are floating point numbers of the same type.
5269 This function returns the sqrt of the specified operand if it is a nonnegative
5270 floating point number.
5274 <!-- _______________________________________________________________________ -->
5275 <div class="doc_subsubsection">
5276 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
5279 <div class="doc_text">
5282 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
5283 floating point or vector of floating point type. Not all targets support all
5286 declare float @llvm.powi.f32(float %Val, i32 %power)
5287 declare double @llvm.powi.f64(double %Val, i32 %power)
5288 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
5289 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
5290 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
5296 The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
5297 specified (positive or negative) power. The order of evaluation of
5298 multiplications is not defined. When a vector of floating point type is
5299 used, the second argument remains a scalar integer value.
5305 The second argument is an integer power, and the first is a value to raise to
5312 This function returns the first value raised to the second power with an
5313 unspecified sequence of rounding operations.</p>
5316 <!-- _______________________________________________________________________ -->
5317 <div class="doc_subsubsection">
5318 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
5321 <div class="doc_text">
5324 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
5325 floating point or vector of floating point type. Not all targets support all
5328 declare float @llvm.sin.f32(float %Val)
5329 declare double @llvm.sin.f64(double %Val)
5330 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
5331 declare fp128 @llvm.sin.f128(fp128 %Val)
5332 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
5338 The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.
5344 The argument and return value are floating point numbers of the same type.
5350 This function returns the sine of the specified operand, returning the
5351 same values as the libm <tt>sin</tt> functions would, and handles error
5352 conditions in the same way.</p>
5355 <!-- _______________________________________________________________________ -->
5356 <div class="doc_subsubsection">
5357 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
5360 <div class="doc_text">
5363 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
5364 floating point or vector of floating point type. Not all targets support all
5367 declare float @llvm.cos.f32(float %Val)
5368 declare double @llvm.cos.f64(double %Val)
5369 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
5370 declare fp128 @llvm.cos.f128(fp128 %Val)
5371 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
5377 The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.
5383 The argument and return value are floating point numbers of the same type.
5389 This function returns the cosine of the specified operand, returning the
5390 same values as the libm <tt>cos</tt> functions would, and handles error
5391 conditions in the same way.</p>
5394 <!-- _______________________________________________________________________ -->
5395 <div class="doc_subsubsection">
5396 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
5399 <div class="doc_text">
5402 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
5403 floating point or vector of floating point type. Not all targets support all
5406 declare float @llvm.pow.f32(float %Val, float %Power)
5407 declare double @llvm.pow.f64(double %Val, double %Power)
5408 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
5409 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
5410 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
5416 The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
5417 specified (positive or negative) power.
5423 The second argument is a floating point power, and the first is a value to
5424 raise to that power.
5430 This function returns the first value raised to the second power,
5432 same values as the libm <tt>pow</tt> functions would, and handles error
5433 conditions in the same way.</p>
5437 <!-- ======================================================================= -->
5438 <div class="doc_subsection">
5439 <a name="int_manip">Bit Manipulation Intrinsics</a>
5442 <div class="doc_text">
5444 LLVM provides intrinsics for a few important bit manipulation operations.
5445 These allow efficient code generation for some algorithms.
5450 <!-- _______________________________________________________________________ -->
5451 <div class="doc_subsubsection">
5452 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
5455 <div class="doc_text">
5458 <p>This is an overloaded intrinsic function. You can use bswap on any integer
5459 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
5461 declare i16 @llvm.bswap.i16(i16 <id>)
5462 declare i32 @llvm.bswap.i32(i32 <id>)
5463 declare i64 @llvm.bswap.i64(i64 <id>)
5469 The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
5470 values with an even number of bytes (positive multiple of 16 bits). These are
5471 useful for performing operations on data that is not in the target's native
5478 The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
5479 and low byte of the input i16 swapped. Similarly, the <tt>llvm.bswap.i32</tt>
5480 intrinsic returns an i32 value that has the four bytes of the input i32
5481 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the returned
5482 i32 will have its bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i48</tt>,
5483 <tt>llvm.bswap.i64</tt> and other intrinsics extend this concept to
5484 additional even-byte lengths (6 bytes, 8 bytes and more, respectively).
5489 <!-- _______________________________________________________________________ -->
5490 <div class="doc_subsubsection">
5491 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
5494 <div class="doc_text">
5497 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
5498 width. Not all targets support all bit widths however.</p>
5500 declare i8 @llvm.ctpop.i8 (i8 <src>)
5501 declare i16 @llvm.ctpop.i16(i16 <src>)
5502 declare i32 @llvm.ctpop.i32(i32 <src>)
5503 declare i64 @llvm.ctpop.i64(i64 <src>)
5504 declare i256 @llvm.ctpop.i256(i256 <src>)
5510 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
5517 The only argument is the value to be counted. The argument may be of any
5518 integer type. The return type must match the argument type.
5524 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
5528 <!-- _______________________________________________________________________ -->
5529 <div class="doc_subsubsection">
5530 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
5533 <div class="doc_text">
5536 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
5537 integer bit width. Not all targets support all bit widths however.</p>
5539 declare i8 @llvm.ctlz.i8 (i8 <src>)
5540 declare i16 @llvm.ctlz.i16(i16 <src>)
5541 declare i32 @llvm.ctlz.i32(i32 <src>)
5542 declare i64 @llvm.ctlz.i64(i64 <src>)
5543 declare i256 @llvm.ctlz.i256(i256 <src>)
5549 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
5550 leading zeros in a variable.
5556 The only argument is the value to be counted. The argument may be of any
5557 integer type. The return type must match the argument type.
5563 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
5564 in a variable. If the src == 0 then the result is the size in bits of the type
5565 of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.
5571 <!-- _______________________________________________________________________ -->
5572 <div class="doc_subsubsection">
5573 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
5576 <div class="doc_text">
5579 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
5580 integer bit width. Not all targets support all bit widths however.</p>
5582 declare i8 @llvm.cttz.i8 (i8 <src>)
5583 declare i16 @llvm.cttz.i16(i16 <src>)
5584 declare i32 @llvm.cttz.i32(i32 <src>)
5585 declare i64 @llvm.cttz.i64(i64 <src>)
5586 declare i256 @llvm.cttz.i256(i256 <src>)
5592 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
5599 The only argument is the value to be counted. The argument may be of any
5600 integer type. The return type must match the argument type.
5606 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
5607 in a variable. If the src == 0 then the result is the size in bits of the type
5608 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
5612 <!-- _______________________________________________________________________ -->
5613 <div class="doc_subsubsection">
5614 <a name="int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic</a>
5617 <div class="doc_text">
5620 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.select</tt>
5621 on any integer bit width.</p>
5623 declare i17 @llvm.part.select.i17 (i17 %val, i32 %loBit, i32 %hiBit)
5624 declare i29 @llvm.part.select.i29 (i29 %val, i32 %loBit, i32 %hiBit)
5628 <p>The '<tt>llvm.part.select</tt>' family of intrinsic functions selects a
5629 range of bits from an integer value and returns them in the same bit width as
5630 the original value.</p>
5633 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5634 any bit width but they must have the same bit width. The second and third
5635 arguments must be <tt>i32</tt> type since they specify only a bit index.</p>
5638 <p>The operation of the '<tt>llvm.part.select</tt>' intrinsic has two modes
5639 of operation: forwards and reverse. If <tt>%loBit</tt> is greater than
5640 <tt>%hiBits</tt> then the intrinsic operates in reverse mode. Otherwise it
5641 operates in forward mode.</p>
5642 <p>In forward mode, this intrinsic is the equivalent of shifting <tt>%val</tt>
5643 right by <tt>%loBit</tt> bits and then ANDing it with a mask with
5644 only the <tt>%hiBit - %loBit</tt> bits set, as follows:</p>
5646 <li>The <tt>%val</tt> is shifted right (LSHR) by the number of bits specified
5647 by <tt>%loBits</tt>. This normalizes the value to the low order bits.</li>
5648 <li>The <tt>%loBits</tt> value is subtracted from the <tt>%hiBits</tt> value
5649 to determine the number of bits to retain.</li>
5650 <li>A mask of the retained bits is created by shifting a -1 value.</li>
5651 <li>The mask is ANDed with <tt>%val</tt> to produce the result.</li>
5653 <p>In reverse mode, a similar computation is made except that the bits are
5654 returned in the reverse order. So, for example, if <tt>X</tt> has the value
5655 <tt>i16 0x0ACF (101011001111)</tt> and we apply
5656 <tt>part.select(i16 X, 8, 3)</tt> to it, we get back the value
5657 <tt>i16 0x0026 (000000100110)</tt>.</p>
5660 <div class="doc_subsubsection">
5661 <a name="int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic</a>
5664 <div class="doc_text">
5667 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.set</tt>
5668 on any integer bit width.</p>
5670 declare i17 @llvm.part.set.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
5671 declare i29 @llvm.part.set.i29.i9 (i29 %val, i9 %repl, i32 %lo, i32 %hi)
5675 <p>The '<tt>llvm.part.set</tt>' family of intrinsic functions replaces a range
5676 of bits in an integer value with another integer value. It returns the integer
5677 with the replaced bits.</p>
5680 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5681 any bit width but they must have the same bit width. <tt>%val</tt> is the value
5682 whose bits will be replaced. The second argument, <tt>%repl</tt> may be an
5683 integer of any bit width. The third and fourth arguments must be <tt>i32</tt>
5684 type since they specify only a bit index.</p>
5687 <p>The operation of the '<tt>llvm.part.set</tt>' intrinsic has two modes
5688 of operation: forwards and reverse. If <tt>%lo</tt> is greater than
5689 <tt>%hi</tt> then the intrinsic operates in reverse mode. Otherwise it
5690 operates in forward mode.</p>
5691 <p>For both modes, the <tt>%repl</tt> value is prepared for use by either
5692 truncating it down to the size of the replacement area or zero extending it
5693 up to that size.</p>
5694 <p>In forward mode, the bits between <tt>%lo</tt> and <tt>%hi</tt> (inclusive)
5695 are replaced with corresponding bits from <tt>%repl</tt>. That is the 0th bit
5696 in <tt>%repl</tt> replaces the <tt>%lo</tt>th bit in <tt>%val</tt> and etc. up
5697 to the <tt>%hi</tt>th bit.</p>
5698 <p>In reverse mode, a similar computation is made except that the bits are
5699 reversed. That is, the <tt>0</tt>th bit in <tt>%repl</tt> replaces the
5700 <tt>%hi</tt> bit in <tt>%val</tt> and etc. down to the <tt>%lo</tt>th bit.</p>
5703 llvm.part.set(0xFFFF, 0, 4, 7) -> 0xFF0F
5704 llvm.part.set(0xFFFF, 0, 7, 4) -> 0xFF0F
5705 llvm.part.set(0xFFFF, 1, 7, 4) -> 0xFF8F
5706 llvm.part.set(0xFFFF, F, 8, 3) -> 0xFFE7
5707 llvm.part.set(0xFFFF, 0, 3, 8) -> 0xFE07
5711 <!-- ======================================================================= -->
5712 <div class="doc_subsection">
5713 <a name="int_debugger">Debugger Intrinsics</a>
5716 <div class="doc_text">
5718 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
5719 are described in the <a
5720 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
5721 Debugging</a> document.
5726 <!-- ======================================================================= -->
5727 <div class="doc_subsection">
5728 <a name="int_eh">Exception Handling Intrinsics</a>
5731 <div class="doc_text">
5732 <p> The LLVM exception handling intrinsics (which all start with
5733 <tt>llvm.eh.</tt> prefix), are described in the <a
5734 href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
5735 Handling</a> document. </p>
5738 <!-- ======================================================================= -->
5739 <div class="doc_subsection">
5740 <a name="int_trampoline">Trampoline Intrinsic</a>
5743 <div class="doc_text">
5745 This intrinsic makes it possible to excise one parameter, marked with
5746 the <tt>nest</tt> attribute, from a function. The result is a callable
5747 function pointer lacking the nest parameter - the caller does not need
5748 to provide a value for it. Instead, the value to use is stored in
5749 advance in a "trampoline", a block of memory usually allocated
5750 on the stack, which also contains code to splice the nest value into the
5751 argument list. This is used to implement the GCC nested function address
5755 For example, if the function is
5756 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
5757 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as follows:</p>
5759 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
5760 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
5761 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
5762 %fp = bitcast i8* %p to i32 (i32, i32)*
5764 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
5765 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
5768 <!-- _______________________________________________________________________ -->
5769 <div class="doc_subsubsection">
5770 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
5772 <div class="doc_text">
5775 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
5779 This fills the memory pointed to by <tt>tramp</tt> with code
5780 and returns a function pointer suitable for executing it.
5784 The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
5785 pointers. The <tt>tramp</tt> argument must point to a sufficiently large
5786 and sufficiently aligned block of memory; this memory is written to by the
5787 intrinsic. Note that the size and the alignment are target-specific - LLVM
5788 currently provides no portable way of determining them, so a front-end that
5789 generates this intrinsic needs to have some target-specific knowledge.
5790 The <tt>func</tt> argument must hold a function bitcast to an <tt>i8*</tt>.
5794 The block of memory pointed to by <tt>tramp</tt> is filled with target
5795 dependent code, turning it into a function. A pointer to this function is
5796 returned, but needs to be bitcast to an
5797 <a href="#int_trampoline">appropriate function pointer type</a>
5798 before being called. The new function's signature is the same as that of
5799 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
5800 removed. At most one such <tt>nest</tt> argument is allowed, and it must be
5801 of pointer type. Calling the new function is equivalent to calling
5802 <tt>func</tt> with the same argument list, but with <tt>nval</tt> used for the
5803 missing <tt>nest</tt> argument. If, after calling
5804 <tt>llvm.init.trampoline</tt>, the memory pointed to by <tt>tramp</tt> is
5805 modified, then the effect of any later call to the returned function pointer is
5810 <!-- ======================================================================= -->
5811 <div class="doc_subsection">
5812 <a name="int_stackprotect">Stack Protector Intrinsic</a>
5815 <div class="doc_text">
5817 This intrinsic is used when stack protectors are required. LLVM generates a
5818 call to load the randomized stack protector guard's value. The intrinsic is
5819 used so that LLVM can ensure that the stack guard is placed onto the stack in
5820 the appropriate place—before local variables are allocated on the stack.
5824 <!-- _______________________________________________________________________ -->
5825 <div class="doc_subsubsection">
5826 <a name="int_ssp">'<tt>llvm.stackprotector</tt>' Intrinsic</a>
5828 <div class="doc_text">
5831 declare void @llvm.stackprotector( i8* <guard>, i8** <slot> )
5836 The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and stores
5837 it onto the stack at <tt>slot</tt>. The stack slot is adjusted to ensure that
5838 it's before local variables are allocated on the stack.
5842 The <tt>llvm.stackprotector</tt> intrinsic requires two pointer arguments. The
5843 first argument is the value loaded from the stack guard
5844 <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt> that
5845 has enough space to hold the value of the guard.
5849 This intrinsic causes the prologue/epilogue inserter to force the position of
5850 the <tt>AllocaInst</tt> stack slot to be before local variables on the
5851 stack. This is to ensure that if a local variable on the stack is overwritten,
5852 it will destroy the value of the guard. When the function exits, the guard on
5853 the stack is checked against the original guard. If they're different, then
5854 the program aborts by calling the <tt>__stack_chk_fail()</tt> function.
5858 <!-- ======================================================================= -->
5859 <div class="doc_subsection">
5860 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
5863 <div class="doc_text">
5865 These intrinsic functions expand the "universal IR" of LLVM to represent
5866 hardware constructs for atomic operations and memory synchronization. This
5867 provides an interface to the hardware, not an interface to the programmer. It
5868 is aimed at a low enough level to allow any programming models or APIs
5869 (Application Programming Interfaces) which
5870 need atomic behaviors to map cleanly onto it. It is also modeled primarily on
5871 hardware behavior. Just as hardware provides a "universal IR" for source
5872 languages, it also provides a starting point for developing a "universal"
5873 atomic operation and synchronization IR.
5876 These do <em>not</em> form an API such as high-level threading libraries,
5877 software transaction memory systems, atomic primitives, and intrinsic
5878 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
5879 application libraries. The hardware interface provided by LLVM should allow
5880 a clean implementation of all of these APIs and parallel programming models.
5881 No one model or paradigm should be selected above others unless the hardware
5882 itself ubiquitously does so.
5887 <!-- _______________________________________________________________________ -->
5888 <div class="doc_subsubsection">
5889 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
5891 <div class="doc_text">
5894 declare void @llvm.memory.barrier( i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>,
5900 The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
5901 specific pairs of memory access types.
5905 The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
5906 The first four arguments enables a specific barrier as listed below. The fith
5907 argument specifies that the barrier applies to io or device or uncached memory.
5911 <li><tt>ll</tt>: load-load barrier</li>
5912 <li><tt>ls</tt>: load-store barrier</li>
5913 <li><tt>sl</tt>: store-load barrier</li>
5914 <li><tt>ss</tt>: store-store barrier</li>
5915 <li><tt>device</tt>: barrier applies to device and uncached memory also.</li>
5919 This intrinsic causes the system to enforce some ordering constraints upon
5920 the loads and stores of the program. This barrier does not indicate
5921 <em>when</em> any events will occur, it only enforces an <em>order</em> in
5922 which they occur. For any of the specified pairs of load and store operations
5923 (f.ex. load-load, or store-load), all of the first operations preceding the
5924 barrier will complete before any of the second operations succeeding the
5925 barrier begin. Specifically the semantics for each pairing is as follows:
5928 <li><tt>ll</tt>: All loads before the barrier must complete before any load
5929 after the barrier begins.</li>
5931 <li><tt>ls</tt>: All loads before the barrier must complete before any
5932 store after the barrier begins.</li>
5933 <li><tt>ss</tt>: All stores before the barrier must complete before any
5934 store after the barrier begins.</li>
5935 <li><tt>sl</tt>: All stores before the barrier must complete before any
5936 load after the barrier begins.</li>
5939 These semantics are applied with a logical "and" behavior when more than one
5940 is enabled in a single memory barrier intrinsic.
5943 Backends may implement stronger barriers than those requested when they do not
5944 support as fine grained a barrier as requested. Some architectures do not
5945 need all types of barriers and on such architectures, these become noops.
5952 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
5953 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
5954 <i>; guarantee the above finishes</i>
5955 store i32 8, %ptr <i>; before this begins</i>
5959 <!-- _______________________________________________________________________ -->
5960 <div class="doc_subsubsection">
5961 <a name="int_atomic_cmp_swap">'<tt>llvm.atomic.cmp.swap.*</tt>' Intrinsic</a>
5963 <div class="doc_text">
5966 This is an overloaded intrinsic. You can use <tt>llvm.atomic.cmp.swap</tt> on
5967 any integer bit width and for different address spaces. Not all targets
5968 support all bit widths however.</p>
5971 declare i8 @llvm.atomic.cmp.swap.i8.p0i8( i8* <ptr>, i8 <cmp>, i8 <val> )
5972 declare i16 @llvm.atomic.cmp.swap.i16.p0i16( i16* <ptr>, i16 <cmp>, i16 <val> )
5973 declare i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* <ptr>, i32 <cmp>, i32 <val> )
5974 declare i64 @llvm.atomic.cmp.swap.i64.p0i64( i64* <ptr>, i64 <cmp>, i64 <val> )
5979 This loads a value in memory and compares it to a given value. If they are
5980 equal, it stores a new value into the memory.
5984 The <tt>llvm.atomic.cmp.swap</tt> intrinsic takes three arguments. The result as
5985 well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
5986 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
5987 this integer type. While any bit width integer may be used, targets may only
5988 lower representations they support in hardware.
5993 This entire intrinsic must be executed atomically. It first loads the value
5994 in memory pointed to by <tt>ptr</tt> and compares it with the value
5995 <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the memory. The
5996 loaded value is yielded in all cases. This provides the equivalent of an
5997 atomic compare-and-swap operation within the SSA framework.
6005 %val1 = add i32 4, 4
6006 %result1 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 4, %val1 )
6007 <i>; yields {i32}:result1 = 4</i>
6008 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
6009 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
6011 %val2 = add i32 1, 1
6012 %result2 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 5, %val2 )
6013 <i>; yields {i32}:result2 = 8</i>
6014 %stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
6016 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
6020 <!-- _______________________________________________________________________ -->
6021 <div class="doc_subsubsection">
6022 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
6024 <div class="doc_text">
6028 This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
6029 integer bit width. Not all targets support all bit widths however.</p>
6031 declare i8 @llvm.atomic.swap.i8.p0i8( i8* <ptr>, i8 <val> )
6032 declare i16 @llvm.atomic.swap.i16.p0i16( i16* <ptr>, i16 <val> )
6033 declare i32 @llvm.atomic.swap.i32.p0i32( i32* <ptr>, i32 <val> )
6034 declare i64 @llvm.atomic.swap.i64.p0i64( i64* <ptr>, i64 <val> )
6039 This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
6040 the value from memory. It then stores the value in <tt>val</tt> in the memory
6046 The <tt>llvm.atomic.swap</tt> intrinsic takes two arguments. Both the
6047 <tt>val</tt> argument and the result must be integers of the same bit width.
6048 The first argument, <tt>ptr</tt>, must be a pointer to a value of this
6049 integer type. The targets may only lower integer representations they
6054 This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
6055 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
6056 equivalent of an atomic swap operation within the SSA framework.
6064 %val1 = add i32 4, 4
6065 %result1 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val1 )
6066 <i>; yields {i32}:result1 = 4</i>
6067 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
6068 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
6070 %val2 = add i32 1, 1
6071 %result2 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val2 )
6072 <i>; yields {i32}:result2 = 8</i>
6074 %stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
6075 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
6079 <!-- _______________________________________________________________________ -->
6080 <div class="doc_subsubsection">
6081 <a name="int_atomic_load_add">'<tt>llvm.atomic.load.add.*</tt>' Intrinsic</a>
6084 <div class="doc_text">
6087 This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.add</tt> on any
6088 integer bit width. Not all targets support all bit widths however.</p>
6090 declare i8 @llvm.atomic.load.add.i8..p0i8( i8* <ptr>, i8 <delta> )
6091 declare i16 @llvm.atomic.load.add.i16..p0i16( i16* <ptr>, i16 <delta> )
6092 declare i32 @llvm.atomic.load.add.i32..p0i32( i32* <ptr>, i32 <delta> )
6093 declare i64 @llvm.atomic.load.add.i64..p0i64( i64* <ptr>, i64 <delta> )
6098 This intrinsic adds <tt>delta</tt> to the value stored in memory at
6099 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
6104 The intrinsic takes two arguments, the first a pointer to an integer value
6105 and the second an integer value. The result is also an integer value. These
6106 integer types can have any bit width, but they must all have the same bit
6107 width. The targets may only lower integer representations they support.
6111 This intrinsic does a series of operations atomically. It first loads the
6112 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
6113 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
6120 %result1 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 4 )
6121 <i>; yields {i32}:result1 = 4</i>
6122 %result2 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 2 )
6123 <i>; yields {i32}:result2 = 8</i>
6124 %result3 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 5 )
6125 <i>; yields {i32}:result3 = 10</i>
6126 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
6130 <!-- _______________________________________________________________________ -->
6131 <div class="doc_subsubsection">
6132 <a name="int_atomic_load_sub">'<tt>llvm.atomic.load.sub.*</tt>' Intrinsic</a>
6135 <div class="doc_text">
6138 This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.sub</tt> on
6139 any integer bit width and for different address spaces. Not all targets
6140 support all bit widths however.</p>
6142 declare i8 @llvm.atomic.load.sub.i8.p0i32( i8* <ptr>, i8 <delta> )
6143 declare i16 @llvm.atomic.load.sub.i16.p0i32( i16* <ptr>, i16 <delta> )
6144 declare i32 @llvm.atomic.load.sub.i32.p0i32( i32* <ptr>, i32 <delta> )
6145 declare i64 @llvm.atomic.load.sub.i64.p0i32( i64* <ptr>, i64 <delta> )
6150 This intrinsic subtracts <tt>delta</tt> to the value stored in memory at
6151 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
6156 The intrinsic takes two arguments, the first a pointer to an integer value
6157 and the second an integer value. The result is also an integer value. These
6158 integer types can have any bit width, but they must all have the same bit
6159 width. The targets may only lower integer representations they support.
6163 This intrinsic does a series of operations atomically. It first loads the
6164 value stored at <tt>ptr</tt>. It then subtracts <tt>delta</tt>, stores the
6165 result to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
6172 %result1 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 4 )
6173 <i>; yields {i32}:result1 = 8</i>
6174 %result2 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 2 )
6175 <i>; yields {i32}:result2 = 4</i>
6176 %result3 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 5 )
6177 <i>; yields {i32}:result3 = 2</i>
6178 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = -3</i>
6182 <!-- _______________________________________________________________________ -->
6183 <div class="doc_subsubsection">
6184 <a name="int_atomic_load_and">'<tt>llvm.atomic.load.and.*</tt>' Intrinsic</a><br>
6185 <a name="int_atomic_load_nand">'<tt>llvm.atomic.load.nand.*</tt>' Intrinsic</a><br>
6186 <a name="int_atomic_load_or">'<tt>llvm.atomic.load.or.*</tt>' Intrinsic</a><br>
6187 <a name="int_atomic_load_xor">'<tt>llvm.atomic.load.xor.*</tt>' Intrinsic</a><br>
6190 <div class="doc_text">
6193 These are overloaded intrinsics. You can use <tt>llvm.atomic.load_and</tt>,
6194 <tt>llvm.atomic.load_nand</tt>, <tt>llvm.atomic.load_or</tt>, and
6195 <tt>llvm.atomic.load_xor</tt> on any integer bit width and for different
6196 address spaces. Not all targets support all bit widths however.</p>
6198 declare i8 @llvm.atomic.load.and.i8.p0i8( i8* <ptr>, i8 <delta> )
6199 declare i16 @llvm.atomic.load.and.i16.p0i16( i16* <ptr>, i16 <delta> )
6200 declare i32 @llvm.atomic.load.and.i32.p0i32( i32* <ptr>, i32 <delta> )
6201 declare i64 @llvm.atomic.load.and.i64.p0i64( i64* <ptr>, i64 <delta> )
6206 declare i8 @llvm.atomic.load.or.i8.p0i8( i8* <ptr>, i8 <delta> )
6207 declare i16 @llvm.atomic.load.or.i16.p0i16( i16* <ptr>, i16 <delta> )
6208 declare i32 @llvm.atomic.load.or.i32.p0i32( i32* <ptr>, i32 <delta> )
6209 declare i64 @llvm.atomic.load.or.i64.p0i64( i64* <ptr>, i64 <delta> )
6214 declare i8 @llvm.atomic.load.nand.i8.p0i32( i8* <ptr>, i8 <delta> )
6215 declare i16 @llvm.atomic.load.nand.i16.p0i32( i16* <ptr>, i16 <delta> )
6216 declare i32 @llvm.atomic.load.nand.i32.p0i32( i32* <ptr>, i32 <delta> )
6217 declare i64 @llvm.atomic.load.nand.i64.p0i32( i64* <ptr>, i64 <delta> )
6222 declare i8 @llvm.atomic.load.xor.i8.p0i32( i8* <ptr>, i8 <delta> )
6223 declare i16 @llvm.atomic.load.xor.i16.p0i32( i16* <ptr>, i16 <delta> )
6224 declare i32 @llvm.atomic.load.xor.i32.p0i32( i32* <ptr>, i32 <delta> )
6225 declare i64 @llvm.atomic.load.xor.i64.p0i32( i64* <ptr>, i64 <delta> )
6230 These intrinsics bitwise the operation (and, nand, or, xor) <tt>delta</tt> to
6231 the value stored in memory at <tt>ptr</tt>. It yields the original value
6237 These intrinsics take two arguments, the first a pointer to an integer value
6238 and the second an integer value. The result is also an integer value. These
6239 integer types can have any bit width, but they must all have the same bit
6240 width. The targets may only lower integer representations they support.
6244 These intrinsics does a series of operations atomically. They first load the
6245 value stored at <tt>ptr</tt>. They then do the bitwise operation
6246 <tt>delta</tt>, store the result to <tt>ptr</tt>. They yield the original
6247 value stored at <tt>ptr</tt>.
6253 store i32 0x0F0F, %ptr
6254 %result0 = call i32 @llvm.atomic.load.nand.i32.p0i32( i32* %ptr, i32 0xFF )
6255 <i>; yields {i32}:result0 = 0x0F0F</i>
6256 %result1 = call i32 @llvm.atomic.load.and.i32.p0i32( i32* %ptr, i32 0xFF )
6257 <i>; yields {i32}:result1 = 0xFFFFFFF0</i>
6258 %result2 = call i32 @llvm.atomic.load.or.i32.p0i32( i32* %ptr, i32 0F )
6259 <i>; yields {i32}:result2 = 0xF0</i>
6260 %result3 = call i32 @llvm.atomic.load.xor.i32.p0i32( i32* %ptr, i32 0F )
6261 <i>; yields {i32}:result3 = FF</i>
6262 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = F0</i>
6267 <!-- _______________________________________________________________________ -->
6268 <div class="doc_subsubsection">
6269 <a name="int_atomic_load_max">'<tt>llvm.atomic.load.max.*</tt>' Intrinsic</a><br>
6270 <a name="int_atomic_load_min">'<tt>llvm.atomic.load.min.*</tt>' Intrinsic</a><br>
6271 <a name="int_atomic_load_umax">'<tt>llvm.atomic.load.umax.*</tt>' Intrinsic</a><br>
6272 <a name="int_atomic_load_umin">'<tt>llvm.atomic.load.umin.*</tt>' Intrinsic</a><br>
6275 <div class="doc_text">
6278 These are overloaded intrinsics. You can use <tt>llvm.atomic.load_max</tt>,
6279 <tt>llvm.atomic.load_min</tt>, <tt>llvm.atomic.load_umax</tt>, and
6280 <tt>llvm.atomic.load_umin</tt> on any integer bit width and for different
6281 address spaces. Not all targets
6282 support all bit widths however.</p>
6284 declare i8 @llvm.atomic.load.max.i8.p0i8( i8* <ptr>, i8 <delta> )
6285 declare i16 @llvm.atomic.load.max.i16.p0i16( i16* <ptr>, i16 <delta> )
6286 declare i32 @llvm.atomic.load.max.i32.p0i32( i32* <ptr>, i32 <delta> )
6287 declare i64 @llvm.atomic.load.max.i64.p0i64( i64* <ptr>, i64 <delta> )
6292 declare i8 @llvm.atomic.load.min.i8.p0i8( i8* <ptr>, i8 <delta> )
6293 declare i16 @llvm.atomic.load.min.i16.p0i16( i16* <ptr>, i16 <delta> )
6294 declare i32 @llvm.atomic.load.min.i32..p0i32( i32* <ptr>, i32 <delta> )
6295 declare i64 @llvm.atomic.load.min.i64..p0i64( i64* <ptr>, i64 <delta> )
6300 declare i8 @llvm.atomic.load.umax.i8.p0i8( i8* <ptr>, i8 <delta> )
6301 declare i16 @llvm.atomic.load.umax.i16.p0i16( i16* <ptr>, i16 <delta> )
6302 declare i32 @llvm.atomic.load.umax.i32.p0i32( i32* <ptr>, i32 <delta> )
6303 declare i64 @llvm.atomic.load.umax.i64.p0i64( i64* <ptr>, i64 <delta> )
6308 declare i8 @llvm.atomic.load.umin.i8..p0i8( i8* <ptr>, i8 <delta> )
6309 declare i16 @llvm.atomic.load.umin.i16.p0i16( i16* <ptr>, i16 <delta> )
6310 declare i32 @llvm.atomic.load.umin.i32..p0i32( i32* <ptr>, i32 <delta> )
6311 declare i64 @llvm.atomic.load.umin.i64..p0i64( i64* <ptr>, i64 <delta> )
6316 These intrinsics takes the signed or unsigned minimum or maximum of
6317 <tt>delta</tt> and the value stored in memory at <tt>ptr</tt>. It yields the
6318 original value at <tt>ptr</tt>.
6323 These intrinsics take two arguments, the first a pointer to an integer value
6324 and the second an integer value. The result is also an integer value. These
6325 integer types can have any bit width, but they must all have the same bit
6326 width. The targets may only lower integer representations they support.
6330 These intrinsics does a series of operations atomically. They first load the
6331 value stored at <tt>ptr</tt>. They then do the signed or unsigned min or max
6332 <tt>delta</tt> and the value, store the result to <tt>ptr</tt>. They yield
6333 the original value stored at <tt>ptr</tt>.
6340 %result0 = call i32 @llvm.atomic.load.min.i32.p0i32( i32* %ptr, i32 -2 )
6341 <i>; yields {i32}:result0 = 7</i>
6342 %result1 = call i32 @llvm.atomic.load.max.i32.p0i32( i32* %ptr, i32 8 )
6343 <i>; yields {i32}:result1 = -2</i>
6344 %result2 = call i32 @llvm.atomic.load.umin.i32.p0i32( i32* %ptr, i32 10 )
6345 <i>; yields {i32}:result2 = 8</i>
6346 %result3 = call i32 @llvm.atomic.load.umax.i32.p0i32( i32* %ptr, i32 30 )
6347 <i>; yields {i32}:result3 = 8</i>
6348 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 30</i>
6352 <!-- ======================================================================= -->
6353 <div class="doc_subsection">
6354 <a name="int_general">General Intrinsics</a>
6357 <div class="doc_text">
6358 <p> This class of intrinsics is designed to be generic and has
6359 no specific purpose. </p>
6362 <!-- _______________________________________________________________________ -->
6363 <div class="doc_subsubsection">
6364 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
6367 <div class="doc_text">
6371 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
6377 The '<tt>llvm.var.annotation</tt>' intrinsic
6383 The first argument is a pointer to a value, the second is a pointer to a
6384 global string, the third is a pointer to a global string which is the source
6385 file name, and the last argument is the line number.
6391 This intrinsic allows annotation of local variables with arbitrary strings.
6392 This can be useful for special purpose optimizations that want to look for these
6393 annotations. These have no other defined use, they are ignored by code
6394 generation and optimization.
6398 <!-- _______________________________________________________________________ -->
6399 <div class="doc_subsubsection">
6400 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
6403 <div class="doc_text">
6406 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
6407 any integer bit width.
6410 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
6411 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
6412 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
6413 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
6414 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
6420 The '<tt>llvm.annotation</tt>' intrinsic.
6426 The first argument is an integer value (result of some expression),
6427 the second is a pointer to a global string, the third is a pointer to a global
6428 string which is the source file name, and the last argument is the line number.
6429 It returns the value of the first argument.
6435 This intrinsic allows annotations to be put on arbitrary expressions
6436 with arbitrary strings. This can be useful for special purpose optimizations
6437 that want to look for these annotations. These have no other defined use, they
6438 are ignored by code generation and optimization.
6442 <!-- _______________________________________________________________________ -->
6443 <div class="doc_subsubsection">
6444 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
6447 <div class="doc_text">
6451 declare void @llvm.trap()
6457 The '<tt>llvm.trap</tt>' intrinsic
6469 This intrinsics is lowered to the target dependent trap instruction. If the
6470 target does not have a trap instruction, this intrinsic will be lowered to the
6471 call of the abort() function.
6475 <!-- *********************************************************************** -->
6478 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
6479 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
6480 <a href="http://validator.w3.org/check/referer"><img
6481 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
6483 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
6484 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
6485 Last modified: $Date$