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_atomics">Atomic intrinsics</a>
219 <li><a href="#int_memory_barrier"><tt>llvm.memory_barrier</tt></a></li>
220 <li><a href="#int_atomic_cmp_swap"><tt>llvm.atomic.cmp.swap</tt></a></li>
221 <li><a href="#int_atomic_swap"><tt>llvm.atomic.swap</tt></a></li>
222 <li><a href="#int_atomic_load_add"><tt>llvm.atomic.load.add</tt></a></li>
223 <li><a href="#int_atomic_load_sub"><tt>llvm.atomic.load.sub</tt></a></li>
224 <li><a href="#int_atomic_load_and"><tt>llvm.atomic.load.and</tt></a></li>
225 <li><a href="#int_atomic_load_nand"><tt>llvm.atomic.load.nand</tt></a></li>
226 <li><a href="#int_atomic_load_or"><tt>llvm.atomic.load.or</tt></a></li>
227 <li><a href="#int_atomic_load_xor"><tt>llvm.atomic.load.xor</tt></a></li>
228 <li><a href="#int_atomic_load_max"><tt>llvm.atomic.load.max</tt></a></li>
229 <li><a href="#int_atomic_load_min"><tt>llvm.atomic.load.min</tt></a></li>
230 <li><a href="#int_atomic_load_umax"><tt>llvm.atomic.load.umax</tt></a></li>
231 <li><a href="#int_atomic_load_umin"><tt>llvm.atomic.load.umin</tt></a></li>
234 <li><a href="#int_general">General intrinsics</a>
236 <li><a href="#int_var_annotation">
237 '<tt>llvm.var.annotation</tt>' Intrinsic</a></li>
238 <li><a href="#int_annotation">
239 '<tt>llvm.annotation.*</tt>' Intrinsic</a></li>
240 <li><a href="#int_trap">
241 '<tt>llvm.trap</tt>' Intrinsic</a></li>
242 <li><a href="#int_stackprotector">
243 '<tt>llvm.stackprotector</tt>' Intrinsic</a></li>
250 <div class="doc_author">
251 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
252 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
255 <!-- *********************************************************************** -->
256 <div class="doc_section"> <a name="abstract">Abstract </a></div>
257 <!-- *********************************************************************** -->
259 <div class="doc_text">
260 <p>This document is a reference manual for the LLVM assembly language.
261 LLVM is a Static Single Assignment (SSA) based representation that provides
262 type safety, low-level operations, flexibility, and the capability of
263 representing 'all' high-level languages cleanly. It is the common code
264 representation used throughout all phases of the LLVM compilation
268 <!-- *********************************************************************** -->
269 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
270 <!-- *********************************************************************** -->
272 <div class="doc_text">
274 <p>The LLVM code representation is designed to be used in three
275 different forms: as an in-memory compiler IR, as an on-disk bitcode
276 representation (suitable for fast loading by a Just-In-Time compiler),
277 and as a human readable assembly language representation. This allows
278 LLVM to provide a powerful intermediate representation for efficient
279 compiler transformations and analysis, while providing a natural means
280 to debug and visualize the transformations. The three different forms
281 of LLVM are all equivalent. This document describes the human readable
282 representation and notation.</p>
284 <p>The LLVM representation aims to be light-weight and low-level
285 while being expressive, typed, and extensible at the same time. It
286 aims to be a "universal IR" of sorts, by being at a low enough level
287 that high-level ideas may be cleanly mapped to it (similar to how
288 microprocessors are "universal IR's", allowing many source languages to
289 be mapped to them). By providing type information, LLVM can be used as
290 the target of optimizations: for example, through pointer analysis, it
291 can be proven that a C automatic variable is never accessed outside of
292 the current function... allowing it to be promoted to a simple SSA
293 value instead of a memory location.</p>
297 <!-- _______________________________________________________________________ -->
298 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
300 <div class="doc_text">
302 <p>It is important to note that this document describes 'well formed'
303 LLVM assembly language. There is a difference between what the parser
304 accepts and what is considered 'well formed'. For example, the
305 following instruction is syntactically okay, but not well formed:</p>
307 <div class="doc_code">
309 %x = <a href="#i_add">add</a> i32 1, %x
313 <p>...because the definition of <tt>%x</tt> does not dominate all of
314 its uses. The LLVM infrastructure provides a verification pass that may
315 be used to verify that an LLVM module is well formed. This pass is
316 automatically run by the parser after parsing input assembly and by
317 the optimizer before it outputs bitcode. The violations pointed out
318 by the verifier pass indicate bugs in transformation passes or input to
322 <!-- Describe the typesetting conventions here. -->
324 <!-- *********************************************************************** -->
325 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
326 <!-- *********************************************************************** -->
328 <div class="doc_text">
330 <p>LLVM identifiers come in two basic types: global and local. Global
331 identifiers (functions, global variables) begin with the @ character. Local
332 identifiers (register names, types) begin with the % character. Additionally,
333 there are three different formats for identifiers, for different purposes:</p>
336 <li>Named values are represented as a string of characters with their prefix.
337 For example, %foo, @DivisionByZero, %a.really.long.identifier. The actual
338 regular expression used is '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
339 Identifiers which require other characters in their names can be surrounded
340 with quotes. Special characters may be escaped using "\xx" where xx is the
341 ASCII code for the character in hexadecimal. In this way, any character can
342 be used in a name value, even quotes themselves.
344 <li>Unnamed values are represented as an unsigned numeric value with their
345 prefix. For example, %12, @2, %44.</li>
347 <li>Constants, which are described in a <a href="#constants">section about
348 constants</a>, below.</li>
351 <p>LLVM requires that values start with a prefix for two reasons: Compilers
352 don't need to worry about name clashes with reserved words, and the set of
353 reserved words may be expanded in the future without penalty. Additionally,
354 unnamed identifiers allow a compiler to quickly come up with a temporary
355 variable without having to avoid symbol table conflicts.</p>
357 <p>Reserved words in LLVM are very similar to reserved words in other
358 languages. There are keywords for different opcodes
359 ('<tt><a href="#i_add">add</a></tt>',
360 '<tt><a href="#i_bitcast">bitcast</a></tt>',
361 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
362 href="#t_void">void</a></tt>', '<tt><a href="#t_primitive">i32</a></tt>', etc...),
363 and others. These reserved words cannot conflict with variable names, because
364 none of them start with a prefix character ('%' or '@').</p>
366 <p>Here is an example of LLVM code to multiply the integer variable
367 '<tt>%X</tt>' by 8:</p>
371 <div class="doc_code">
373 %result = <a href="#i_mul">mul</a> i32 %X, 8
377 <p>After strength reduction:</p>
379 <div class="doc_code">
381 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
385 <p>And the hard way:</p>
387 <div class="doc_code">
389 <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
390 <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
391 %result = <a href="#i_add">add</a> i32 %1, %1
395 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
396 important lexical features of LLVM:</p>
400 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
403 <li>Unnamed temporaries are created when the result of a computation is not
404 assigned to a named value.</li>
406 <li>Unnamed temporaries are numbered sequentially</li>
410 <p>...and it also shows a convention that we follow in this document. When
411 demonstrating instructions, we will follow an instruction with a comment that
412 defines the type and name of value produced. Comments are shown in italic
417 <!-- *********************************************************************** -->
418 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
419 <!-- *********************************************************************** -->
421 <!-- ======================================================================= -->
422 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
425 <div class="doc_text">
427 <p>LLVM programs are composed of "Module"s, each of which is a
428 translation unit of the input programs. Each module consists of
429 functions, global variables, and symbol table entries. Modules may be
430 combined together with the LLVM linker, which merges function (and
431 global variable) definitions, resolves forward declarations, and merges
432 symbol table entries. Here is an example of the "hello world" module:</p>
434 <div class="doc_code">
435 <pre><i>; Declare the string constant as a global constant...</i>
436 <a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a
437 href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
439 <i>; External declaration of the puts function</i>
440 <a href="#functionstructure">declare</a> i32 @puts(i8 *) <i>; i32(i8 *)* </i>
442 <i>; Definition of main function</i>
443 define i32 @main() { <i>; i32()* </i>
444 <i>; Convert [13x i8 ]* to i8 *...</i>
446 href="#i_getelementptr">getelementptr</a> [13 x i8 ]* @.LC0, i64 0, i64 0 <i>; i8 *</i>
448 <i>; Call puts function to write out the string to stdout...</i>
450 href="#i_call">call</a> i32 @puts(i8 * %cast210) <i>; i32</i>
452 href="#i_ret">ret</a> i32 0<br>}<br>
456 <p>This example is made up of a <a href="#globalvars">global variable</a>
457 named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
458 function, and a <a href="#functionstructure">function definition</a>
459 for "<tt>main</tt>".</p>
461 <p>In general, a module is made up of a list of global values,
462 where both functions and global variables are global values. Global values are
463 represented by a pointer to a memory location (in this case, a pointer to an
464 array of char, and a pointer to a function), and have one of the following <a
465 href="#linkage">linkage types</a>.</p>
469 <!-- ======================================================================= -->
470 <div class="doc_subsection">
471 <a name="linkage">Linkage Types</a>
474 <div class="doc_text">
477 All Global Variables and Functions have one of the following types of linkage:
482 <dt><tt><b><a name="linkage_internal">internal</a></b></tt>: </dt>
484 <dd>Global values with internal linkage are only directly accessible by
485 objects in the current module. In particular, linking code into a module with
486 an internal global value may cause the internal to be renamed as necessary to
487 avoid collisions. Because the symbol is internal to the module, all
488 references can be updated. This corresponds to the notion of the
489 '<tt>static</tt>' keyword in C.
492 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
494 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
495 the same name when linkage occurs. This is typically used to implement
496 inline functions, templates, or other code which must be generated in each
497 translation unit that uses it. Unreferenced <tt>linkonce</tt> globals are
498 allowed to be discarded.
501 <dt><tt><b><a name="linkage_common">common</a></b></tt>: </dt>
503 <dd>"<tt>common</tt>" linkage is exactly the same as <tt>linkonce</tt>
504 linkage, except that unreferenced <tt>common</tt> globals may not be
505 discarded. This is used for globals that may be emitted in multiple
506 translation units, but that are not guaranteed to be emitted into every
507 translation unit that uses them. One example of this is tentative
508 definitions in C, such as "<tt>int X;</tt>" at global scope.
511 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
513 <dd>"<tt>weak</tt>" linkage is the same as <tt>common</tt> linkage, except
514 that some targets may choose to emit different assembly sequences for them
515 for target-dependent reasons. This is used for globals that are declared
516 "weak" in C source code.
519 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
521 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
522 pointer to array type. When two global variables with appending linkage are
523 linked together, the two global arrays are appended together. This is the
524 LLVM, typesafe, equivalent of having the system linker append together
525 "sections" with identical names when .o files are linked.
528 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt>: </dt>
529 <dd>The semantics of this linkage follow the ELF object file model: the
530 symbol is weak until linked, if not linked, the symbol becomes null instead
531 of being an undefined reference.
534 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
536 <dd>If none of the above identifiers are used, the global is externally
537 visible, meaning that it participates in linkage and can be used to resolve
538 external symbol references.
543 The next two types of linkage are targeted for Microsoft Windows platform
544 only. They are designed to support importing (exporting) symbols from (to)
545 DLLs (Dynamic Link Libraries).
549 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt>: </dt>
551 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
552 or variable via a global pointer to a pointer that is set up by the DLL
553 exporting the symbol. On Microsoft Windows targets, the pointer name is
554 formed by combining <code>_imp__</code> and the function or variable name.
557 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt>: </dt>
559 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
560 pointer to a pointer in a DLL, so that it can be referenced with the
561 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
562 name is formed by combining <code>_imp__</code> and the function or variable
568 <p>For example, since the "<tt>.LC0</tt>"
569 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
570 variable and was linked with this one, one of the two would be renamed,
571 preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
572 external (i.e., lacking any linkage declarations), they are accessible
573 outside of the current module.</p>
574 <p>It is illegal for a function <i>declaration</i>
575 to have any linkage type other than "externally visible", <tt>dllimport</tt>,
576 or <tt>extern_weak</tt>.</p>
577 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt> and <tt>weak</tt>
581 <!-- ======================================================================= -->
582 <div class="doc_subsection">
583 <a name="callingconv">Calling Conventions</a>
586 <div class="doc_text">
588 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
589 and <a href="#i_invoke">invokes</a> can all have an optional calling convention
590 specified for the call. The calling convention of any pair of dynamic
591 caller/callee must match, or the behavior of the program is undefined. The
592 following calling conventions are supported by LLVM, and more may be added in
596 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
598 <dd>This calling convention (the default if no other calling convention is
599 specified) matches the target C calling conventions. This calling convention
600 supports varargs function calls and tolerates some mismatch in the declared
601 prototype and implemented declaration of the function (as does normal C).
604 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
606 <dd>This calling convention attempts to make calls as fast as possible
607 (e.g. by passing things in registers). This calling convention allows the
608 target to use whatever tricks it wants to produce fast code for the target,
609 without having to conform to an externally specified ABI (Application Binary
610 Interface). Implementations of this convention should allow arbitrary
611 <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> to be
612 supported. This calling convention does not support varargs and requires the
613 prototype of all callees to exactly match the prototype of the function
617 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
619 <dd>This calling convention attempts to make code in the caller as efficient
620 as possible under the assumption that the call is not commonly executed. As
621 such, these calls often preserve all registers so that the call does not break
622 any live ranges in the caller side. This calling convention does not support
623 varargs and requires the prototype of all callees to exactly match the
624 prototype of the function definition.
627 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
629 <dd>Any calling convention may be specified by number, allowing
630 target-specific calling conventions to be used. Target specific calling
631 conventions start at 64.
635 <p>More calling conventions can be added/defined on an as-needed basis, to
636 support pascal conventions or any other well-known target-independent
641 <!-- ======================================================================= -->
642 <div class="doc_subsection">
643 <a name="visibility">Visibility Styles</a>
646 <div class="doc_text">
649 All Global Variables and Functions have one of the following visibility styles:
653 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
655 <dd>On targets that use the ELF object file format, default visibility means
656 that the declaration is visible to other
657 modules and, in shared libraries, means that the declared entity may be
658 overridden. On Darwin, default visibility means that the declaration is
659 visible to other modules. Default visibility corresponds to "external
660 linkage" in the language.
663 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
665 <dd>Two declarations of an object with hidden visibility refer to the same
666 object if they are in the same shared object. Usually, hidden visibility
667 indicates that the symbol will not be placed into the dynamic symbol table,
668 so no other module (executable or shared library) can reference it
672 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
674 <dd>On ELF, protected visibility indicates that the symbol will be placed in
675 the dynamic symbol table, but that references within the defining module will
676 bind to the local symbol. That is, the symbol cannot be overridden by another
683 <!-- ======================================================================= -->
684 <div class="doc_subsection">
685 <a name="globalvars">Global Variables</a>
688 <div class="doc_text">
690 <p>Global variables define regions of memory allocated at compilation time
691 instead of run-time. Global variables may optionally be initialized, may have
692 an explicit section to be placed in, and may have an optional explicit alignment
693 specified. A variable may be defined as "thread_local", which means that it
694 will not be shared by threads (each thread will have a separated copy of the
695 variable). A variable may be defined as a global "constant," which indicates
696 that the contents of the variable will <b>never</b> be modified (enabling better
697 optimization, allowing the global data to be placed in the read-only section of
698 an executable, etc). Note that variables that need runtime initialization
699 cannot be marked "constant" as there is a store to the variable.</p>
702 LLVM explicitly allows <em>declarations</em> of global variables to be marked
703 constant, even if the final definition of the global is not. This capability
704 can be used to enable slightly better optimization of the program, but requires
705 the language definition to guarantee that optimizations based on the
706 'constantness' are valid for the translation units that do not include the
710 <p>As SSA values, global variables define pointer values that are in
711 scope (i.e. they dominate) all basic blocks in the program. Global
712 variables always define a pointer to their "content" type because they
713 describe a region of memory, and all memory objects in LLVM are
714 accessed through pointers.</p>
716 <p>A global variable may be declared to reside in a target-specifc numbered
717 address space. For targets that support them, address spaces may affect how
718 optimizations are performed and/or what target instructions are used to access
719 the variable. The default address space is zero. The address space qualifier
720 must precede any other attributes.</p>
722 <p>LLVM allows an explicit section to be specified for globals. If the target
723 supports it, it will emit globals to the section specified.</p>
725 <p>An explicit alignment may be specified for a global. If not present, or if
726 the alignment is set to zero, the alignment of the global is set by the target
727 to whatever it feels convenient. If an explicit alignment is specified, the
728 global is forced to have at least that much alignment. All alignments must be
731 <p>For example, the following defines a global in a numbered address space with
732 an initializer, section, and alignment:</p>
734 <div class="doc_code">
736 @G = constant float 1.0 addrspace(5), section "foo", align 4
743 <!-- ======================================================================= -->
744 <div class="doc_subsection">
745 <a name="functionstructure">Functions</a>
748 <div class="doc_text">
750 <p>LLVM function definitions consist of the "<tt>define</tt>" keyord,
751 an optional <a href="#linkage">linkage type</a>, an optional
752 <a href="#visibility">visibility style</a>, an optional
753 <a href="#callingconv">calling convention</a>, a return type, an optional
754 <a href="#paramattrs">parameter attribute</a> for the return type, a function
755 name, a (possibly empty) argument list (each with optional
756 <a href="#paramattrs">parameter attributes</a>), optional
757 <a href="#fnattrs">function attributes</a>, an optional section,
758 an optional alignment, an optional <a href="#gc">garbage collector name</a>,
759 an opening curly brace, a list of basic blocks, and a closing curly brace.
761 LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
762 optional <a href="#linkage">linkage type</a>, an optional
763 <a href="#visibility">visibility style</a>, an optional
764 <a href="#callingconv">calling convention</a>, a return type, an optional
765 <a href="#paramattrs">parameter attribute</a> for the return type, a function
766 name, a possibly empty list of arguments, an optional alignment, and an optional
767 <a href="#gc">garbage collector name</a>.</p>
769 <p>A function definition contains a list of basic blocks, forming the CFG
770 (Control Flow Graph) for
771 the function. Each basic block may optionally start with a label (giving the
772 basic block a symbol table entry), contains a list of instructions, and ends
773 with a <a href="#terminators">terminator</a> instruction (such as a branch or
774 function return).</p>
776 <p>The first basic block in a function is special in two ways: it is immediately
777 executed on entrance to the function, and it is not allowed to have predecessor
778 basic blocks (i.e. there can not be any branches to the entry block of a
779 function). Because the block can have no predecessors, it also cannot have any
780 <a href="#i_phi">PHI nodes</a>.</p>
782 <p>LLVM allows an explicit section to be specified for functions. If the target
783 supports it, it will emit functions to the section specified.</p>
785 <p>An explicit alignment may be specified for a function. If not present, or if
786 the alignment is set to zero, the alignment of the function is set by the target
787 to whatever it feels convenient. If an explicit alignment is specified, the
788 function is forced to have at least that much alignment. All alignments must be
793 <div class="doc_code">
795 define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
796 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
797 <ResultType> @<FunctionName> ([argument list])
798 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
799 [<a href="#gc">gc</a>] { ... }
806 <!-- ======================================================================= -->
807 <div class="doc_subsection">
808 <a name="aliasstructure">Aliases</a>
810 <div class="doc_text">
811 <p>Aliases act as "second name" for the aliasee value (which can be either
812 function, global variable, another alias or bitcast of global value). Aliases
813 may have an optional <a href="#linkage">linkage type</a>, and an
814 optional <a href="#visibility">visibility style</a>.</p>
818 <div class="doc_code">
820 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
828 <!-- ======================================================================= -->
829 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
830 <div class="doc_text">
831 <p>The return type and each parameter of a function type may have a set of
832 <i>parameter attributes</i> associated with them. Parameter attributes are
833 used to communicate additional information about the result or parameters of
834 a function. Parameter attributes are considered to be part of the function,
835 not of the function type, so functions with different parameter attributes
836 can have the same function type.</p>
838 <p>Parameter attributes are simple keywords that follow the type specified. If
839 multiple parameter attributes are needed, they are space separated. For
842 <div class="doc_code">
844 declare i32 @printf(i8* noalias , ...)
845 declare i32 @atoi(i8 zeroext)
846 declare signext i8 @returns_signed_char()
850 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
851 <tt>readonly</tt>) come immediately after the argument list.</p>
853 <p>Currently, only the following parameter attributes are defined:</p>
855 <dt><tt>zeroext</tt></dt>
856 <dd>This indicates to the code generator that the parameter or return value
857 should be zero-extended to a 32-bit value by the caller (for a parameter)
858 or the callee (for a return value).</dd>
860 <dt><tt>signext</tt></dt>
861 <dd>This indicates to the code generator that the parameter or return value
862 should be sign-extended to a 32-bit value by the caller (for a parameter)
863 or the callee (for a return value).</dd>
865 <dt><tt>inreg</tt></dt>
866 <dd>This indicates that this parameter or return value should be treated
867 in a special target-dependent fashion during while emitting code for a
868 function call or return (usually, by putting it in a register as opposed
869 to memory, though some targets use it to distinguish between two different
870 kinds of registers). Use of this attribute is target-specific.</dd>
872 <dt><tt><a name="byval">byval</a></tt></dt>
873 <dd>This indicates that the pointer parameter should really be passed by
874 value to the function. The attribute implies that a hidden copy of the
875 pointee is made between the caller and the callee, so the callee is unable
876 to modify the value in the callee. This attribute is only valid on LLVM
877 pointer arguments. It is generally used to pass structs and arrays by
878 value, but is also valid on pointers to scalars. The copy is considered to
879 belong to the caller not the callee (for example,
880 <tt><a href="#readonly">readonly</a></tt> functions should not write to
881 <tt>byval</tt> parameters). This is not a valid attribute for return
884 <dt><tt>sret</tt></dt>
885 <dd>This indicates that the pointer parameter specifies the address of a
886 structure that is the return value of the function in the source program.
887 This pointer must be guaranteed by the caller to be valid: loads and stores
888 to the structure may be assumed by the callee to not to trap. This may only
889 be applied to the first parameter. This is not a valid attribute for
892 <dt><tt>noalias</tt></dt>
893 <dd>This indicates that the pointer does not alias any global or any other
894 parameter. The caller is responsible for ensuring that this is the
895 case. On a function return value, <tt>noalias</tt> additionally indicates
896 that the pointer does not alias any other pointers visible to the
897 caller. Note that this applies only to pointers that can be used to actually
898 load/store a value: NULL, unique pointers from malloc(0), and freed pointers
899 are considered to not alias anything.</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.
1001 <p>If a function that has an <tt>ssp</tt> attribute is inlined into a function
1002 that doesn't have an <tt>ssp</tt> attribute, then the resulting function will
1003 have an <tt>ssp</tt> attribute.</p></dd>
1005 <dt><tt>sspreq</tt></dt>
1006 <dd>This attribute indicates that the function should <em>always</em> emit a
1007 stack smashing protector. This overrides the <tt><a href="#ssp">ssp</a></tt>
1010 <p>If a function that has an <tt>sspreq</tt> attribute is inlined into a
1011 function that doesn't have an <tt>sspreq</tt> attribute or which has
1012 an <tt>ssp</tt> attribute, then the resulting function will have
1013 an <tt>sspreq</tt> attribute.</p></dd>
1018 <!-- ======================================================================= -->
1019 <div class="doc_subsection">
1020 <a name="moduleasm">Module-Level Inline Assembly</a>
1023 <div class="doc_text">
1025 Modules may contain "module-level inline asm" blocks, which corresponds to the
1026 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
1027 LLVM and treated as a single unit, but may be separated in the .ll file if
1028 desired. The syntax is very simple:
1031 <div class="doc_code">
1033 module asm "inline asm code goes here"
1034 module asm "more can go here"
1038 <p>The strings can contain any character by escaping non-printable characters.
1039 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1044 The inline asm code is simply printed to the machine code .s file when
1045 assembly code is generated.
1049 <!-- ======================================================================= -->
1050 <div class="doc_subsection">
1051 <a name="datalayout">Data Layout</a>
1054 <div class="doc_text">
1055 <p>A module may specify a target specific data layout string that specifies how
1056 data is to be laid out in memory. The syntax for the data layout is simply:</p>
1057 <pre> target datalayout = "<i>layout specification</i>"</pre>
1058 <p>The <i>layout specification</i> consists of a list of specifications
1059 separated by the minus sign character ('-'). Each specification starts with a
1060 letter and may include other information after the letter to define some
1061 aspect of the data layout. The specifications accepted are as follows: </p>
1064 <dd>Specifies that the target lays out data in big-endian form. That is, the
1065 bits with the most significance have the lowest address location.</dd>
1067 <dd>Specifies that the target lays out data in little-endian form. That is,
1068 the bits with the least significance have the lowest address location.</dd>
1069 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1070 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1071 <i>preferred</i> alignments. All sizes are in bits. Specifying the <i>pref</i>
1072 alignment is optional. If omitted, the preceding <tt>:</tt> should be omitted
1074 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1075 <dd>This specifies the alignment for an integer type of a given bit
1076 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1077 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1078 <dd>This specifies the alignment for a vector type of a given bit
1080 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1081 <dd>This specifies the alignment for a floating point type of a given bit
1082 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
1084 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1085 <dd>This specifies the alignment for an aggregate type of a given bit
1088 <p>When constructing the data layout for a given target, LLVM starts with a
1089 default set of specifications which are then (possibly) overriden by the
1090 specifications in the <tt>datalayout</tt> keyword. The default specifications
1091 are given in this list:</p>
1093 <li><tt>E</tt> - big endian</li>
1094 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
1095 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1096 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1097 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1098 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1099 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1100 alignment of 64-bits</li>
1101 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1102 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1103 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1104 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1105 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1107 <p>When LLVM is determining the alignment for a given type, it uses the
1108 following rules:</p>
1110 <li>If the type sought is an exact match for one of the specifications, that
1111 specification is used.</li>
1112 <li>If no match is found, and the type sought is an integer type, then the
1113 smallest integer type that is larger than the bitwidth of the sought type is
1114 used. If none of the specifications are larger than the bitwidth then the the
1115 largest integer type is used. For example, given the default specifications
1116 above, the i7 type will use the alignment of i8 (next largest) while both
1117 i65 and i256 will use the alignment of i64 (largest specified).</li>
1118 <li>If no match is found, and the type sought is a vector type, then the
1119 largest vector type that is smaller than the sought vector type will be used
1120 as a fall back. This happens because <128 x double> can be implemented
1121 in terms of 64 <2 x double>, for example.</li>
1125 <!-- *********************************************************************** -->
1126 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
1127 <!-- *********************************************************************** -->
1129 <div class="doc_text">
1131 <p>The LLVM type system is one of the most important features of the
1132 intermediate representation. Being typed enables a number of
1133 optimizations to be performed on the intermediate representation directly,
1134 without having to do
1135 extra analyses on the side before the transformation. A strong type
1136 system makes it easier to read the generated code and enables novel
1137 analyses and transformations that are not feasible to perform on normal
1138 three address code representations.</p>
1142 <!-- ======================================================================= -->
1143 <div class="doc_subsection"> <a name="t_classifications">Type
1144 Classifications</a> </div>
1145 <div class="doc_text">
1146 <p>The types fall into a few useful
1147 classifications:</p>
1149 <table border="1" cellspacing="0" cellpadding="4">
1151 <tr><th>Classification</th><th>Types</th></tr>
1153 <td><a href="#t_integer">integer</a></td>
1154 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1157 <td><a href="#t_floating">floating point</a></td>
1158 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1161 <td><a name="t_firstclass">first class</a></td>
1162 <td><a href="#t_integer">integer</a>,
1163 <a href="#t_floating">floating point</a>,
1164 <a href="#t_pointer">pointer</a>,
1165 <a href="#t_vector">vector</a>,
1166 <a href="#t_struct">structure</a>,
1167 <a href="#t_array">array</a>,
1168 <a href="#t_label">label</a>.
1172 <td><a href="#t_primitive">primitive</a></td>
1173 <td><a href="#t_label">label</a>,
1174 <a href="#t_void">void</a>,
1175 <a href="#t_floating">floating point</a>.</td>
1178 <td><a href="#t_derived">derived</a></td>
1179 <td><a href="#t_integer">integer</a>,
1180 <a href="#t_array">array</a>,
1181 <a href="#t_function">function</a>,
1182 <a href="#t_pointer">pointer</a>,
1183 <a href="#t_struct">structure</a>,
1184 <a href="#t_pstruct">packed structure</a>,
1185 <a href="#t_vector">vector</a>,
1186 <a href="#t_opaque">opaque</a>.
1192 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
1193 most important. Values of these types are the only ones which can be
1194 produced by instructions, passed as arguments, or used as operands to
1198 <!-- ======================================================================= -->
1199 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
1201 <div class="doc_text">
1202 <p>The primitive types are the fundamental building blocks of the LLVM
1207 <!-- _______________________________________________________________________ -->
1208 <div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1210 <div class="doc_text">
1213 <tr><th>Type</th><th>Description</th></tr>
1214 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1215 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1216 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1217 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1218 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1223 <!-- _______________________________________________________________________ -->
1224 <div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1226 <div class="doc_text">
1228 <p>The void type does not represent any value and has no size.</p>
1237 <!-- _______________________________________________________________________ -->
1238 <div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1240 <div class="doc_text">
1242 <p>The label type represents code labels.</p>
1252 <!-- ======================================================================= -->
1253 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1255 <div class="doc_text">
1257 <p>The real power in LLVM comes from the derived types in the system.
1258 This is what allows a programmer to represent arrays, functions,
1259 pointers, and other useful types. Note that these derived types may be
1260 recursive: For example, it is possible to have a two dimensional array.</p>
1264 <!-- _______________________________________________________________________ -->
1265 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1267 <div class="doc_text">
1270 <p>The integer type is a very simple derived type that simply specifies an
1271 arbitrary bit width for the integer type desired. Any bit width from 1 bit to
1272 2^23-1 (about 8 million) can be specified.</p>
1280 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1284 <table class="layout">
1287 <td><tt>i1</tt></td>
1288 <td>a single-bit integer.</td>
1290 <td><tt>i32</tt></td>
1291 <td>a 32-bit integer.</td>
1293 <td><tt>i1942652</tt></td>
1294 <td>a really big integer of over 1 million bits.</td>
1300 <!-- _______________________________________________________________________ -->
1301 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1303 <div class="doc_text">
1307 <p>The array type is a very simple derived type that arranges elements
1308 sequentially in memory. The array type requires a size (number of
1309 elements) and an underlying data type.</p>
1314 [<# elements> x <elementtype>]
1317 <p>The number of elements is a constant integer value; elementtype may
1318 be any type with a size.</p>
1321 <table class="layout">
1323 <td class="left"><tt>[40 x i32]</tt></td>
1324 <td class="left">Array of 40 32-bit integer values.</td>
1327 <td class="left"><tt>[41 x i32]</tt></td>
1328 <td class="left">Array of 41 32-bit integer values.</td>
1331 <td class="left"><tt>[4 x i8]</tt></td>
1332 <td class="left">Array of 4 8-bit integer values.</td>
1335 <p>Here are some examples of multidimensional arrays:</p>
1336 <table class="layout">
1338 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1339 <td class="left">3x4 array of 32-bit integer values.</td>
1342 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1343 <td class="left">12x10 array of single precision floating point values.</td>
1346 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1347 <td class="left">2x3x4 array of 16-bit integer values.</td>
1351 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1352 length array. Normally, accesses past the end of an array are undefined in
1353 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
1354 As a special case, however, zero length arrays are recognized to be variable
1355 length. This allows implementation of 'pascal style arrays' with the LLVM
1356 type "{ i32, [0 x float]}", for example.</p>
1360 <!-- _______________________________________________________________________ -->
1361 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1362 <div class="doc_text">
1366 <p>The function type can be thought of as a function signature. It
1367 consists of a return type and a list of formal parameter types. The
1368 return type of a function type is a scalar type, a void type, or a struct type.
1369 If the return type is a struct type then all struct elements must be of first
1370 class types, and the struct must have at least one element.</p>
1375 <returntype list> (<parameter list>)
1378 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1379 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1380 which indicates that the function takes a variable number of arguments.
1381 Variable argument functions can access their arguments with the <a
1382 href="#int_varargs">variable argument handling intrinsic</a> functions.
1383 '<tt><returntype list></tt>' is a comma-separated list of
1384 <a href="#t_firstclass">first class</a> type specifiers.</p>
1387 <table class="layout">
1389 <td class="left"><tt>i32 (i32)</tt></td>
1390 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1392 </tr><tr class="layout">
1393 <td class="left"><tt>float (i16 signext, i32 *) *
1395 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1396 an <tt>i16</tt> that should be sign extended and a
1397 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1400 </tr><tr class="layout">
1401 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1402 <td class="left">A vararg function that takes at least one
1403 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1404 which returns an integer. This is the signature for <tt>printf</tt> in
1407 </tr><tr class="layout">
1408 <td class="left"><tt>{i32, i32} (i32)</tt></td>
1409 <td class="left">A function taking an <tt>i32</tt>, returning two
1410 <tt>i32</tt> values as an aggregate of type <tt>{ i32, i32 }</tt>
1416 <!-- _______________________________________________________________________ -->
1417 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1418 <div class="doc_text">
1420 <p>The structure type is used to represent a collection of data members
1421 together in memory. The packing of the field types is defined to match
1422 the ABI of the underlying processor. The elements of a structure may
1423 be any type that has a size.</p>
1424 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1425 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1426 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1429 <pre> { <type list> }<br></pre>
1431 <table class="layout">
1433 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1434 <td class="left">A triple of three <tt>i32</tt> values</td>
1435 </tr><tr class="layout">
1436 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1437 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1438 second element is a <a href="#t_pointer">pointer</a> to a
1439 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1440 an <tt>i32</tt>.</td>
1445 <!-- _______________________________________________________________________ -->
1446 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1448 <div class="doc_text">
1450 <p>The packed structure type is used to represent a collection of data members
1451 together in memory. There is no padding between fields. Further, the alignment
1452 of a packed structure is 1 byte. The elements of a packed structure may
1453 be any type that has a size.</p>
1454 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1455 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1456 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1459 <pre> < { <type list> } > <br></pre>
1461 <table class="layout">
1463 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1464 <td class="left">A triple of three <tt>i32</tt> values</td>
1465 </tr><tr class="layout">
1467 <tt>< { float, i32 (i32)* } ></tt></td>
1468 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1469 second element is a <a href="#t_pointer">pointer</a> to a
1470 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1471 an <tt>i32</tt>.</td>
1476 <!-- _______________________________________________________________________ -->
1477 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1478 <div class="doc_text">
1480 <p>As in many languages, the pointer type represents a pointer or
1481 reference to another object, which must live in memory. Pointer types may have
1482 an optional address space attribute defining the target-specific numbered
1483 address space where the pointed-to object resides. The default address space is
1486 <pre> <type> *<br></pre>
1488 <table class="layout">
1490 <td class="left"><tt>[4x i32]*</tt></td>
1491 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1492 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1495 <td class="left"><tt>i32 (i32 *) *</tt></td>
1496 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
1497 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1501 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1502 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1503 that resides in address space #5.</td>
1508 <!-- _______________________________________________________________________ -->
1509 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1510 <div class="doc_text">
1514 <p>A vector type is a simple derived type that represents a vector
1515 of elements. Vector types are used when multiple primitive data
1516 are operated in parallel using a single instruction (SIMD).
1517 A vector type requires a size (number of
1518 elements) and an underlying primitive data type. Vectors must have a power
1519 of two length (1, 2, 4, 8, 16 ...). Vector types are
1520 considered <a href="#t_firstclass">first class</a>.</p>
1525 < <# elements> x <elementtype> >
1528 <p>The number of elements is a constant integer value; elementtype may
1529 be any integer or floating point type.</p>
1533 <table class="layout">
1535 <td class="left"><tt><4 x i32></tt></td>
1536 <td class="left">Vector of 4 32-bit integer values.</td>
1539 <td class="left"><tt><8 x float></tt></td>
1540 <td class="left">Vector of 8 32-bit floating-point values.</td>
1543 <td class="left"><tt><2 x i64></tt></td>
1544 <td class="left">Vector of 2 64-bit integer values.</td>
1549 <!-- _______________________________________________________________________ -->
1550 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1551 <div class="doc_text">
1555 <p>Opaque types are used to represent unknown types in the system. This
1556 corresponds (for example) to the C notion of a forward declared structure type.
1557 In LLVM, opaque types can eventually be resolved to any type (not just a
1558 structure type).</p>
1568 <table class="layout">
1570 <td class="left"><tt>opaque</tt></td>
1571 <td class="left">An opaque type.</td>
1577 <!-- *********************************************************************** -->
1578 <div class="doc_section"> <a name="constants">Constants</a> </div>
1579 <!-- *********************************************************************** -->
1581 <div class="doc_text">
1583 <p>LLVM has several different basic types of constants. This section describes
1584 them all and their syntax.</p>
1588 <!-- ======================================================================= -->
1589 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1591 <div class="doc_text">
1594 <dt><b>Boolean constants</b></dt>
1596 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1597 constants of the <tt><a href="#t_primitive">i1</a></tt> type.
1600 <dt><b>Integer constants</b></dt>
1602 <dd>Standard integers (such as '4') are constants of the <a
1603 href="#t_integer">integer</a> type. Negative numbers may be used with
1607 <dt><b>Floating point constants</b></dt>
1609 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1610 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1611 notation (see below). The assembler requires the exact decimal value of
1612 a floating-point constant. For example, the assembler accepts 1.25 but
1613 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
1614 constants must have a <a href="#t_floating">floating point</a> type. </dd>
1616 <dt><b>Null pointer constants</b></dt>
1618 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1619 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1623 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1624 of floating point constants. For example, the form '<tt>double
1625 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1626 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1627 (and the only time that they are generated by the disassembler) is when a
1628 floating point constant must be emitted but it cannot be represented as a
1629 decimal floating point number. For example, NaN's, infinities, and other
1630 special values are represented in their IEEE hexadecimal format so that
1631 assembly and disassembly do not cause any bits to change in the constants.</p>
1635 <!-- ======================================================================= -->
1636 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1639 <div class="doc_text">
1640 <p>Aggregate constants arise from aggregation of simple constants
1641 and smaller aggregate constants.</p>
1644 <dt><b>Structure constants</b></dt>
1646 <dd>Structure constants are represented with notation similar to structure
1647 type definitions (a comma separated list of elements, surrounded by braces
1648 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
1649 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>". Structure constants
1650 must have <a href="#t_struct">structure type</a>, and the number and
1651 types of elements must match those specified by the type.
1654 <dt><b>Array constants</b></dt>
1656 <dd>Array constants are represented with notation similar to array type
1657 definitions (a comma separated list of elements, surrounded by square brackets
1658 (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74 ]</tt>". Array
1659 constants must have <a href="#t_array">array type</a>, and the number and
1660 types of elements must match those specified by the type.
1663 <dt><b>Vector constants</b></dt>
1665 <dd>Vector constants are represented with notation similar to vector type
1666 definitions (a comma separated list of elements, surrounded by
1667 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32 42,
1668 i32 11, i32 74, i32 100 ></tt>". Vector constants must have <a
1669 href="#t_vector">vector type</a>, and the number and types of elements must
1670 match those specified by the type.
1673 <dt><b>Zero initialization</b></dt>
1675 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1676 value to zero of <em>any</em> type, including scalar and aggregate types.
1677 This is often used to avoid having to print large zero initializers (e.g. for
1678 large arrays) and is always exactly equivalent to using explicit zero
1685 <!-- ======================================================================= -->
1686 <div class="doc_subsection">
1687 <a name="globalconstants">Global Variable and Function Addresses</a>
1690 <div class="doc_text">
1692 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1693 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1694 constants. These constants are explicitly referenced when the <a
1695 href="#identifiers">identifier for the global</a> is used and always have <a
1696 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1699 <div class="doc_code">
1703 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1709 <!-- ======================================================================= -->
1710 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1711 <div class="doc_text">
1712 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1713 no specific value. Undefined values may be of any type and be used anywhere
1714 a constant is permitted.</p>
1716 <p>Undefined values indicate to the compiler that the program is well defined
1717 no matter what value is used, giving the compiler more freedom to optimize.
1721 <!-- ======================================================================= -->
1722 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1725 <div class="doc_text">
1727 <p>Constant expressions are used to allow expressions involving other constants
1728 to be used as constants. Constant expressions may be of any <a
1729 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1730 that does not have side effects (e.g. load and call are not supported). The
1731 following is the syntax for constant expressions:</p>
1734 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1735 <dd>Truncate a constant to another type. The bit size of CST must be larger
1736 than the bit size of TYPE. Both types must be integers.</dd>
1738 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1739 <dd>Zero extend a constant to another type. The bit size of CST must be
1740 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1742 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1743 <dd>Sign extend a constant to another type. The bit size of CST must be
1744 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1746 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1747 <dd>Truncate a floating point constant to another floating point type. The
1748 size of CST must be larger than the size of TYPE. Both types must be
1749 floating point.</dd>
1751 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1752 <dd>Floating point extend a constant to another type. The size of CST must be
1753 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1755 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
1756 <dd>Convert a floating point constant to the corresponding unsigned integer
1757 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1758 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1759 of the same number of elements. If the value won't fit in the integer type,
1760 the results are undefined.</dd>
1762 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
1763 <dd>Convert a floating point constant to the corresponding signed integer
1764 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1765 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1766 of the same number of elements. If the value won't fit in the integer type,
1767 the results are undefined.</dd>
1769 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
1770 <dd>Convert an unsigned integer constant to the corresponding floating point
1771 constant. TYPE must be a scalar or vector floating point type. CST must be of
1772 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1773 of the same number of elements. If the value won't fit in the floating point
1774 type, the results are undefined.</dd>
1776 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
1777 <dd>Convert a signed integer constant to the corresponding floating point
1778 constant. TYPE must be a scalar or vector floating point type. CST must be of
1779 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1780 of the same number of elements. If the value won't fit in the floating point
1781 type, the results are undefined.</dd>
1783 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
1784 <dd>Convert a pointer typed constant to the corresponding integer constant
1785 TYPE must be an integer type. CST must be of pointer type. The CST value is
1786 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1788 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
1789 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1790 pointer type. CST must be of integer type. The CST value is zero extended,
1791 truncated, or unchanged to make it fit in a pointer size. This one is
1792 <i>really</i> dangerous!</dd>
1794 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
1795 <dd>Convert a constant, CST, to another TYPE. The size of CST and TYPE must be
1796 identical (same number of bits). The conversion is done as if the CST value
1797 was stored to memory and read back as TYPE. In other words, no bits change
1798 with this operator, just the type. This can be used for conversion of
1799 vector types to any other type, as long as they have the same bit width. For
1800 pointers it is only valid to cast to another pointer type. It is not valid
1801 to bitcast to or from an aggregate type.
1804 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1806 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1807 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1808 instruction, the index list may have zero or more indexes, which are required
1809 to make sense for the type of "CSTPTR".</dd>
1811 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1813 <dd>Perform the <a href="#i_select">select operation</a> on
1816 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
1817 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
1819 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
1820 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
1822 <dt><b><tt>vicmp COND ( VAL1, VAL2 )</tt></b></dt>
1823 <dd>Performs the <a href="#i_vicmp">vicmp operation</a> on constants.</dd>
1825 <dt><b><tt>vfcmp COND ( VAL1, VAL2 )</tt></b></dt>
1826 <dd>Performs the <a href="#i_vfcmp">vfcmp operation</a> on constants.</dd>
1828 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1830 <dd>Perform the <a href="#i_extractelement">extractelement
1831 operation</a> on constants.</dd>
1833 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1835 <dd>Perform the <a href="#i_insertelement">insertelement
1836 operation</a> on constants.</dd>
1839 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
1841 <dd>Perform the <a href="#i_shufflevector">shufflevector
1842 operation</a> on constants.</dd>
1844 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1846 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1847 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1848 binary</a> operations. The constraints on operands are the same as those for
1849 the corresponding instruction (e.g. no bitwise operations on floating point
1850 values are allowed).</dd>
1854 <!-- *********************************************************************** -->
1855 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1856 <!-- *********************************************************************** -->
1858 <!-- ======================================================================= -->
1859 <div class="doc_subsection">
1860 <a name="inlineasm">Inline Assembler Expressions</a>
1863 <div class="doc_text">
1866 LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
1867 Module-Level Inline Assembly</a>) through the use of a special value. This
1868 value represents the inline assembler as a string (containing the instructions
1869 to emit), a list of operand constraints (stored as a string), and a flag that
1870 indicates whether or not the inline asm expression has side effects. An example
1871 inline assembler expression is:
1874 <div class="doc_code">
1876 i32 (i32) asm "bswap $0", "=r,r"
1881 Inline assembler expressions may <b>only</b> be used as the callee operand of
1882 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
1885 <div class="doc_code">
1887 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
1892 Inline asms with side effects not visible in the constraint list must be marked
1893 as having side effects. This is done through the use of the
1894 '<tt>sideeffect</tt>' keyword, like so:
1897 <div class="doc_code">
1899 call void asm sideeffect "eieio", ""()
1903 <p>TODO: The format of the asm and constraints string still need to be
1904 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1905 need to be documented). This is probably best done by reference to another
1906 document that covers inline asm from a holistic perspective.
1911 <!-- *********************************************************************** -->
1912 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1913 <!-- *********************************************************************** -->
1915 <div class="doc_text">
1917 <p>The LLVM instruction set consists of several different
1918 classifications of instructions: <a href="#terminators">terminator
1919 instructions</a>, <a href="#binaryops">binary instructions</a>,
1920 <a href="#bitwiseops">bitwise binary instructions</a>, <a
1921 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1922 instructions</a>.</p>
1926 <!-- ======================================================================= -->
1927 <div class="doc_subsection"> <a name="terminators">Terminator
1928 Instructions</a> </div>
1930 <div class="doc_text">
1932 <p>As mentioned <a href="#functionstructure">previously</a>, every
1933 basic block in a program ends with a "Terminator" instruction, which
1934 indicates which block should be executed after the current block is
1935 finished. These terminator instructions typically yield a '<tt>void</tt>'
1936 value: they produce control flow, not values (the one exception being
1937 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1938 <p>There are six different terminator instructions: the '<a
1939 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1940 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1941 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1942 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1943 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1947 <!-- _______________________________________________________________________ -->
1948 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1949 Instruction</a> </div>
1950 <div class="doc_text">
1953 ret <type> <value> <i>; Return a value from a non-void function</i>
1954 ret void <i>; Return from void function</i>
1959 <p>The '<tt>ret</tt>' instruction is used to return control flow (and
1960 optionally a value) from a function back to the caller.</p>
1961 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
1962 returns a value and then causes control flow, and one that just causes
1963 control flow to occur.</p>
1967 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument,
1968 the return value. The type of the return value must be a
1969 '<a href="#t_firstclass">first class</a>' type.</p>
1971 <p>A function is not <a href="#wellformed">well formed</a> if
1972 it it has a non-void return type and contains a '<tt>ret</tt>'
1973 instruction with no return value or a return value with a type that
1974 does not match its type, or if it has a void return type and contains
1975 a '<tt>ret</tt>' instruction with a return value.</p>
1979 <p>When the '<tt>ret</tt>' instruction is executed, control flow
1980 returns back to the calling function's context. If the caller is a "<a
1981 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1982 the instruction after the call. If the caller was an "<a
1983 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1984 at the beginning of the "normal" destination block. If the instruction
1985 returns a value, that value shall set the call or invoke instruction's
1991 ret i32 5 <i>; Return an integer value of 5</i>
1992 ret void <i>; Return from a void function</i>
1993 ret { i32, i8 } { i32 4, i8 2 } <i>; Return an aggregate of values 4 and 2</i>
1996 <!-- _______________________________________________________________________ -->
1997 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1998 <div class="doc_text">
2000 <pre> br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
2003 <p>The '<tt>br</tt>' instruction is used to cause control flow to
2004 transfer to a different basic block in the current function. There are
2005 two forms of this instruction, corresponding to a conditional branch
2006 and an unconditional branch.</p>
2008 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
2009 single '<tt>i1</tt>' value and two '<tt>label</tt>' values. The
2010 unconditional form of the '<tt>br</tt>' instruction takes a single
2011 '<tt>label</tt>' value as a target.</p>
2013 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
2014 argument is evaluated. If the value is <tt>true</tt>, control flows
2015 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
2016 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
2018 <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
2019 href="#i_ret">ret</a> i32 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> i32 0<br></pre>
2021 <!-- _______________________________________________________________________ -->
2022 <div class="doc_subsubsection">
2023 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
2026 <div class="doc_text">
2030 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
2035 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
2036 several different places. It is a generalization of the '<tt>br</tt>'
2037 instruction, allowing a branch to occur to one of many possible
2043 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
2044 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
2045 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
2046 table is not allowed to contain duplicate constant entries.</p>
2050 <p>The <tt>switch</tt> instruction specifies a table of values and
2051 destinations. When the '<tt>switch</tt>' instruction is executed, this
2052 table is searched for the given value. If the value is found, control flow is
2053 transfered to the corresponding destination; otherwise, control flow is
2054 transfered to the default destination.</p>
2056 <h5>Implementation:</h5>
2058 <p>Depending on properties of the target machine and the particular
2059 <tt>switch</tt> instruction, this instruction may be code generated in different
2060 ways. For example, it could be generated as a series of chained conditional
2061 branches or with a lookup table.</p>
2066 <i>; Emulate a conditional br instruction</i>
2067 %Val = <a href="#i_zext">zext</a> i1 %value to i32
2068 switch i32 %Val, label %truedest [i32 0, label %falsedest ]
2070 <i>; Emulate an unconditional br instruction</i>
2071 switch i32 0, label %dest [ ]
2073 <i>; Implement a jump table:</i>
2074 switch i32 %val, label %otherwise [ i32 0, label %onzero
2076 i32 2, label %ontwo ]
2080 <!-- _______________________________________________________________________ -->
2081 <div class="doc_subsubsection">
2082 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
2085 <div class="doc_text">
2090 <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>]
2091 to label <normal label> unwind label <exception label>
2096 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
2097 function, with the possibility of control flow transfer to either the
2098 '<tt>normal</tt>' label or the
2099 '<tt>exception</tt>' label. If the callee function returns with the
2100 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
2101 "normal" label. If the callee (or any indirect callees) returns with the "<a
2102 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
2103 continued at the dynamically nearest "exception" label.</p>
2107 <p>This instruction requires several arguments:</p>
2111 The optional "cconv" marker indicates which <a href="#callingconv">calling
2112 convention</a> the call should use. If none is specified, the call defaults
2113 to using C calling conventions.
2116 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
2117 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>',
2118 and '<tt>inreg</tt>' attributes are valid here.</li>
2120 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
2121 function value being invoked. In most cases, this is a direct function
2122 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
2123 an arbitrary pointer to function value.
2126 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
2127 function to be invoked. </li>
2129 <li>'<tt>function args</tt>': argument list whose types match the function
2130 signature argument types. If the function signature indicates the function
2131 accepts a variable number of arguments, the extra arguments can be
2134 <li>'<tt>normal label</tt>': the label reached when the called function
2135 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
2137 <li>'<tt>exception label</tt>': the label reached when a callee returns with
2138 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
2140 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
2141 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
2142 '<tt>readnone</tt>' attributes are valid here.</li>
2147 <p>This instruction is designed to operate as a standard '<tt><a
2148 href="#i_call">call</a></tt>' instruction in most regards. The primary
2149 difference is that it establishes an association with a label, which is used by
2150 the runtime library to unwind the stack.</p>
2152 <p>This instruction is used in languages with destructors to ensure that proper
2153 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
2154 exception. Additionally, this is important for implementation of
2155 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
2159 %retval = invoke i32 @Test(i32 15) to label %Continue
2160 unwind label %TestCleanup <i>; {i32}:retval set</i>
2161 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
2162 unwind label %TestCleanup <i>; {i32}:retval set</i>
2167 <!-- _______________________________________________________________________ -->
2169 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
2170 Instruction</a> </div>
2172 <div class="doc_text">
2181 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
2182 at the first callee in the dynamic call stack which used an <a
2183 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
2184 primarily used to implement exception handling.</p>
2188 <p>The '<tt>unwind</tt>' instruction causes execution of the current function to
2189 immediately halt. The dynamic call stack is then searched for the first <a
2190 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
2191 execution continues at the "exceptional" destination block specified by the
2192 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
2193 dynamic call chain, undefined behavior results.</p>
2196 <!-- _______________________________________________________________________ -->
2198 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
2199 Instruction</a> </div>
2201 <div class="doc_text">
2210 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
2211 instruction is used to inform the optimizer that a particular portion of the
2212 code is not reachable. This can be used to indicate that the code after a
2213 no-return function cannot be reached, and other facts.</p>
2217 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
2222 <!-- ======================================================================= -->
2223 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
2224 <div class="doc_text">
2225 <p>Binary operators are used to do most of the computation in a
2226 program. They require two operands of the same type, execute an operation on them, and
2227 produce a single value. The operands might represent
2228 multiple data, as is the case with the <a href="#t_vector">vector</a> data type.
2229 The result value has the same type as its operands.</p>
2230 <p>There are several different binary operators:</p>
2232 <!-- _______________________________________________________________________ -->
2233 <div class="doc_subsubsection">
2234 <a name="i_add">'<tt>add</tt>' Instruction</a>
2237 <div class="doc_text">
2242 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2247 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
2251 <p>The two arguments to the '<tt>add</tt>' instruction must be <a
2252 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>, or
2253 <a href="#t_vector">vector</a> values. Both arguments must have identical
2258 <p>The value produced is the integer or floating point sum of the two
2261 <p>If an integer sum has unsigned overflow, the result returned is the
2262 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2265 <p>Because LLVM integers use a two's complement representation, this
2266 instruction is appropriate for both signed and unsigned integers.</p>
2271 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
2274 <!-- _______________________________________________________________________ -->
2275 <div class="doc_subsubsection">
2276 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
2279 <div class="doc_text">
2284 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2289 <p>The '<tt>sub</tt>' instruction returns the difference of its two
2292 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
2293 '<tt>neg</tt>' instruction present in most other intermediate
2294 representations.</p>
2298 <p>The two arguments to the '<tt>sub</tt>' instruction must be <a
2299 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>,
2300 or <a href="#t_vector">vector</a> values. Both arguments must have identical
2305 <p>The value produced is the integer or floating point difference of
2306 the two operands.</p>
2308 <p>If an integer difference has unsigned overflow, the result returned is the
2309 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2312 <p>Because LLVM integers use a two's complement representation, this
2313 instruction is appropriate for both signed and unsigned integers.</p>
2317 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
2318 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
2322 <!-- _______________________________________________________________________ -->
2323 <div class="doc_subsubsection">
2324 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
2327 <div class="doc_text">
2330 <pre> <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2333 <p>The '<tt>mul</tt>' instruction returns the product of its two
2338 <p>The two arguments to the '<tt>mul</tt>' instruction must be <a
2339 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>,
2340 or <a href="#t_vector">vector</a> values. Both arguments must have identical
2345 <p>The value produced is the integer or floating point product of the
2348 <p>If the result of an integer multiplication has unsigned overflow,
2349 the result returned is the mathematical result modulo
2350 2<sup>n</sup>, where n is the bit width of the result.</p>
2351 <p>Because LLVM integers use a two's complement representation, and the
2352 result is the same width as the operands, this instruction returns the
2353 correct result for both signed and unsigned integers. If a full product
2354 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands
2355 should be sign-extended or zero-extended as appropriate to the
2356 width of the full product.</p>
2358 <pre> <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
2362 <!-- _______________________________________________________________________ -->
2363 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
2365 <div class="doc_text">
2367 <pre> <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2370 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two
2375 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
2376 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2377 values. Both arguments must have identical types.</p>
2381 <p>The value produced is the unsigned integer quotient of the two operands.</p>
2382 <p>Note that unsigned integer division and signed integer division are distinct
2383 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
2384 <p>Division by zero leads to undefined behavior.</p>
2386 <pre> <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2389 <!-- _______________________________________________________________________ -->
2390 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
2392 <div class="doc_text">
2395 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2400 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
2405 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
2406 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2407 values. Both arguments must have identical types.</p>
2410 <p>The value produced is the signed integer quotient of the two operands rounded towards zero.</p>
2411 <p>Note that signed integer division and unsigned integer division are distinct
2412 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
2413 <p>Division by zero leads to undefined behavior. Overflow also leads to
2414 undefined behavior; this is a rare case, but can occur, for example,
2415 by doing a 32-bit division of -2147483648 by -1.</p>
2417 <pre> <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2420 <!-- _______________________________________________________________________ -->
2421 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
2422 Instruction</a> </div>
2423 <div class="doc_text">
2426 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2430 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
2435 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
2436 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a>
2437 of floating point values. Both arguments must have identical types.</p>
2441 <p>The value produced is the floating point quotient of the two operands.</p>
2446 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
2450 <!-- _______________________________________________________________________ -->
2451 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
2453 <div class="doc_text">
2455 <pre> <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2458 <p>The '<tt>urem</tt>' instruction returns the remainder from the
2459 unsigned division of its two arguments.</p>
2461 <p>The two arguments to the '<tt>urem</tt>' instruction must be
2462 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2463 values. Both arguments must have identical types.</p>
2465 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
2466 This instruction always performs an unsigned division to get the remainder.</p>
2467 <p>Note that unsigned integer remainder and signed integer remainder are
2468 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
2469 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
2471 <pre> <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2475 <!-- _______________________________________________________________________ -->
2476 <div class="doc_subsubsection">
2477 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
2480 <div class="doc_text">
2485 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2490 <p>The '<tt>srem</tt>' instruction returns the remainder from the
2491 signed division of its two operands. This instruction can also take
2492 <a href="#t_vector">vector</a> versions of the values in which case
2493 the elements must be integers.</p>
2497 <p>The two arguments to the '<tt>srem</tt>' instruction must be
2498 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2499 values. Both arguments must have identical types.</p>
2503 <p>This instruction returns the <i>remainder</i> of a division (where the result
2504 has the same sign as the dividend, <tt>op1</tt>), not the <i>modulo</i>
2505 operator (where the result has the same sign as the divisor, <tt>op2</tt>) of
2506 a value. For more information about the difference, see <a
2507 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
2508 Math Forum</a>. For a table of how this is implemented in various languages,
2509 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
2510 Wikipedia: modulo operation</a>.</p>
2511 <p>Note that signed integer remainder and unsigned integer remainder are
2512 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
2513 <p>Taking the remainder of a division by zero leads to undefined behavior.
2514 Overflow also leads to undefined behavior; this is a rare case, but can occur,
2515 for example, by taking the remainder of a 32-bit division of -2147483648 by -1.
2516 (The remainder doesn't actually overflow, but this rule lets srem be
2517 implemented using instructions that return both the result of the division
2518 and the remainder.)</p>
2520 <pre> <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2524 <!-- _______________________________________________________________________ -->
2525 <div class="doc_subsubsection">
2526 <a name="i_frem">'<tt>frem</tt>' Instruction</a> </div>
2528 <div class="doc_text">
2531 <pre> <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2534 <p>The '<tt>frem</tt>' instruction returns the remainder from the
2535 division of its two operands.</p>
2537 <p>The two arguments to the '<tt>frem</tt>' instruction must be
2538 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a>
2539 of floating point values. Both arguments must have identical types.</p>
2543 <p>This instruction returns the <i>remainder</i> of a division.
2544 The remainder has the same sign as the dividend.</p>
2549 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
2553 <!-- ======================================================================= -->
2554 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
2555 Operations</a> </div>
2556 <div class="doc_text">
2557 <p>Bitwise binary operators are used to do various forms of
2558 bit-twiddling in a program. They are generally very efficient
2559 instructions and can commonly be strength reduced from other
2560 instructions. They require two operands of the same type, execute an operation on them,
2561 and produce a single value. The resulting value is the same type as its operands.</p>
2564 <!-- _______________________________________________________________________ -->
2565 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
2566 Instruction</a> </div>
2567 <div class="doc_text">
2569 <pre> <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2574 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2575 the left a specified number of bits.</p>
2579 <p>Both arguments to the '<tt>shl</tt>' instruction must be the same <a
2580 href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2581 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
2585 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod 2<sup>n</sup>,
2586 where n is the width of the result. If <tt>op2</tt> is (statically or dynamically) negative or
2587 equal to or larger than the number of bits in <tt>op1</tt>, the result is undefined.</p>
2589 <h5>Example:</h5><pre>
2590 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
2591 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
2592 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
2593 <result> = shl i32 1, 32 <i>; undefined</i>
2596 <!-- _______________________________________________________________________ -->
2597 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
2598 Instruction</a> </div>
2599 <div class="doc_text">
2601 <pre> <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2605 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
2606 operand shifted to the right a specified number of bits with zero fill.</p>
2609 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
2610 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2611 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
2615 <p>This instruction always performs a logical shift right operation. The most
2616 significant bits of the result will be filled with zero bits after the
2617 shift. If <tt>op2</tt> is (statically or dynamically) equal to or larger than
2618 the number of bits in <tt>op1</tt>, the result is undefined.</p>
2622 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
2623 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
2624 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
2625 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
2626 <result> = lshr i32 1, 32 <i>; undefined</i>
2630 <!-- _______________________________________________________________________ -->
2631 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
2632 Instruction</a> </div>
2633 <div class="doc_text">
2636 <pre> <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2640 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
2641 operand shifted to the right a specified number of bits with sign extension.</p>
2644 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
2645 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2646 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
2649 <p>This instruction always performs an arithmetic shift right operation,
2650 The most significant bits of the result will be filled with the sign bit
2651 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
2652 larger than the number of bits in <tt>op1</tt>, the result is undefined.
2657 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
2658 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
2659 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
2660 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
2661 <result> = ashr i32 1, 32 <i>; undefined</i>
2665 <!-- _______________________________________________________________________ -->
2666 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
2667 Instruction</a> </div>
2669 <div class="doc_text">
2674 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2679 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
2680 its two operands.</p>
2684 <p>The two arguments to the '<tt>and</tt>' instruction must be
2685 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2686 values. Both arguments must have identical types.</p>
2689 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
2692 <table border="1" cellspacing="0" cellpadding="4">
2724 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
2725 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
2726 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
2729 <!-- _______________________________________________________________________ -->
2730 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
2731 <div class="doc_text">
2733 <pre> <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2736 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
2737 or of its two operands.</p>
2740 <p>The two arguments to the '<tt>or</tt>' instruction must be
2741 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2742 values. Both arguments must have identical types.</p>
2744 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
2747 <table border="1" cellspacing="0" cellpadding="4">
2778 <pre> <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
2779 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
2780 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
2783 <!-- _______________________________________________________________________ -->
2784 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
2785 Instruction</a> </div>
2786 <div class="doc_text">
2788 <pre> <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2791 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
2792 or of its two operands. The <tt>xor</tt> is used to implement the
2793 "one's complement" operation, which is the "~" operator in C.</p>
2795 <p>The two arguments to the '<tt>xor</tt>' instruction must be
2796 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2797 values. Both arguments must have identical types.</p>
2801 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
2804 <table border="1" cellspacing="0" cellpadding="4">
2836 <pre> <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
2837 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
2838 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
2839 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
2843 <!-- ======================================================================= -->
2844 <div class="doc_subsection">
2845 <a name="vectorops">Vector Operations</a>
2848 <div class="doc_text">
2850 <p>LLVM supports several instructions to represent vector operations in a
2851 target-independent manner. These instructions cover the element-access and
2852 vector-specific operations needed to process vectors effectively. While LLVM
2853 does directly support these vector operations, many sophisticated algorithms
2854 will want to use target-specific intrinsics to take full advantage of a specific
2859 <!-- _______________________________________________________________________ -->
2860 <div class="doc_subsubsection">
2861 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2864 <div class="doc_text">
2869 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
2875 The '<tt>extractelement</tt>' instruction extracts a single scalar
2876 element from a vector at a specified index.
2883 The first operand of an '<tt>extractelement</tt>' instruction is a
2884 value of <a href="#t_vector">vector</a> type. The second operand is
2885 an index indicating the position from which to extract the element.
2886 The index may be a variable.</p>
2891 The result is a scalar of the same type as the element type of
2892 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2893 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2894 results are undefined.
2900 %result = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
2905 <!-- _______________________________________________________________________ -->
2906 <div class="doc_subsubsection">
2907 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
2910 <div class="doc_text">
2915 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
2921 The '<tt>insertelement</tt>' instruction inserts a scalar
2922 element into a vector at a specified index.
2929 The first operand of an '<tt>insertelement</tt>' instruction is a
2930 value of <a href="#t_vector">vector</a> type. The second operand is a
2931 scalar value whose type must equal the element type of the first
2932 operand. The third operand is an index indicating the position at
2933 which to insert the value. The index may be a variable.</p>
2938 The result is a vector of the same type as <tt>val</tt>. Its
2939 element values are those of <tt>val</tt> except at position
2940 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
2941 exceeds the length of <tt>val</tt>, the results are undefined.
2947 %result = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
2951 <!-- _______________________________________________________________________ -->
2952 <div class="doc_subsubsection">
2953 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
2956 <div class="doc_text">
2961 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
2967 The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
2968 from two input vectors, returning a vector with the same element type as
2969 the input and length that is the same as the shuffle mask.
2975 The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
2976 with types that match each other. The third argument is a shuffle mask whose
2977 element type is always 'i32'. The result of the instruction is a vector whose
2978 length is the same as the shuffle mask and whose element type is the same as
2979 the element type of the first two operands.
2983 The shuffle mask operand is required to be a constant vector with either
2984 constant integer or undef values.
2990 The elements of the two input vectors are numbered from left to right across
2991 both of the vectors. The shuffle mask operand specifies, for each element of
2992 the result vector, which element of the two input vectors the result element
2993 gets. The element selector may be undef (meaning "don't care") and the second
2994 operand may be undef if performing a shuffle from only one vector.
3000 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
3001 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
3002 %result = shufflevector <4 x i32> %v1, <4 x i32> undef,
3003 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
3004 %result = shufflevector <8 x i32> %v1, <8 x i32> undef,
3005 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i>
3006 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
3007 <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>
3012 <!-- ======================================================================= -->
3013 <div class="doc_subsection">
3014 <a name="aggregateops">Aggregate Operations</a>
3017 <div class="doc_text">
3019 <p>LLVM supports several instructions for working with aggregate values.
3024 <!-- _______________________________________________________________________ -->
3025 <div class="doc_subsubsection">
3026 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
3029 <div class="doc_text">
3034 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
3040 The '<tt>extractvalue</tt>' instruction extracts the value of a struct field
3041 or array element from an aggregate value.
3048 The first operand of an '<tt>extractvalue</tt>' instruction is a
3049 value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a>
3050 type. The operands are constant indices to specify which value to extract
3051 in a similar manner as indices in a
3052 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
3058 The result is the value at the position in the aggregate specified by
3065 %result = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
3070 <!-- _______________________________________________________________________ -->
3071 <div class="doc_subsubsection">
3072 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
3075 <div class="doc_text">
3080 <result> = insertvalue <aggregate type> <val>, <ty> <val>, <idx> <i>; yields <n x <ty>></i>
3086 The '<tt>insertvalue</tt>' instruction inserts a value
3087 into a struct field or array element in an aggregate.
3094 The first operand of an '<tt>insertvalue</tt>' instruction is a
3095 value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a> type.
3096 The second operand is a first-class value to insert.
3097 The following operands are constant indices
3098 indicating the position at which to insert the value in a similar manner as
3100 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
3101 The value to insert must have the same type as the value identified
3108 The result is an aggregate of the same type as <tt>val</tt>. Its
3109 value is that of <tt>val</tt> except that the value at the position
3110 specified by the indices is that of <tt>elt</tt>.
3116 %result = insertvalue {i32, float} %agg, i32 1, 0 <i>; yields {i32, float}</i>
3121 <!-- ======================================================================= -->
3122 <div class="doc_subsection">
3123 <a name="memoryops">Memory Access and Addressing Operations</a>
3126 <div class="doc_text">
3128 <p>A key design point of an SSA-based representation is how it
3129 represents memory. In LLVM, no memory locations are in SSA form, which
3130 makes things very simple. This section describes how to read, write,
3131 allocate, and free memory in LLVM.</p>
3135 <!-- _______________________________________________________________________ -->
3136 <div class="doc_subsubsection">
3137 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
3140 <div class="doc_text">
3145 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3150 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
3151 heap and returns a pointer to it. The object is always allocated in the generic
3152 address space (address space zero).</p>
3156 <p>The '<tt>malloc</tt>' instruction allocates
3157 <tt>sizeof(<type>)*NumElements</tt>
3158 bytes of memory from the operating system and returns a pointer of the
3159 appropriate type to the program. If "NumElements" is specified, it is the
3160 number of elements allocated, otherwise "NumElements" is defaulted to be one.
3161 If a constant alignment is specified, the value result of the allocation is guaranteed to
3162 be aligned to at least that boundary. If not specified, or if zero, the target can
3163 choose to align the allocation on any convenient boundary.</p>
3165 <p>'<tt>type</tt>' must be a sized type.</p>
3169 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
3170 a pointer is returned. The result of a zero byte allocation is undefined. The
3171 result is null if there is insufficient memory available.</p>
3176 %array = malloc [4 x i8 ] <i>; yields {[%4 x i8]*}:array</i>
3178 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
3179 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
3180 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
3181 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
3182 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
3186 <!-- _______________________________________________________________________ -->
3187 <div class="doc_subsubsection">
3188 <a name="i_free">'<tt>free</tt>' Instruction</a>
3191 <div class="doc_text">
3196 free <type> <value> <i>; yields {void}</i>
3201 <p>The '<tt>free</tt>' instruction returns memory back to the unused
3202 memory heap to be reallocated in the future.</p>
3206 <p>'<tt>value</tt>' shall be a pointer value that points to a value
3207 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
3212 <p>Access to the memory pointed to by the pointer is no longer defined
3213 after this instruction executes. If the pointer is null, the operation
3219 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
3220 free [4 x i8]* %array
3224 <!-- _______________________________________________________________________ -->
3225 <div class="doc_subsubsection">
3226 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
3229 <div class="doc_text">
3234 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3239 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
3240 currently executing function, to be automatically released when this function
3241 returns to its caller. The object is always allocated in the generic address
3242 space (address space zero).</p>
3246 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
3247 bytes of memory on the runtime stack, returning a pointer of the
3248 appropriate type to the program. If "NumElements" is specified, it is the
3249 number of elements allocated, otherwise "NumElements" is defaulted to be one.
3250 If a constant alignment is specified, the value result of the allocation is guaranteed
3251 to be aligned to at least that boundary. If not specified, or if zero, the target
3252 can choose to align the allocation on any convenient boundary.</p>
3254 <p>'<tt>type</tt>' may be any sized type.</p>
3258 <p>Memory is allocated; a pointer is returned. The operation is undefiend if
3259 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
3260 memory is automatically released when the function returns. The '<tt>alloca</tt>'
3261 instruction is commonly used to represent automatic variables that must
3262 have an address available. When the function returns (either with the <tt><a
3263 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
3264 instructions), the memory is reclaimed. Allocating zero bytes
3265 is legal, but the result is undefined.</p>
3270 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
3271 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
3272 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
3273 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
3277 <!-- _______________________________________________________________________ -->
3278 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
3279 Instruction</a> </div>
3280 <div class="doc_text">
3282 <pre> <result> = load <ty>* <pointer>[, align <alignment>]<br> <result> = volatile load <ty>* <pointer>[, align <alignment>]<br></pre>
3284 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
3286 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
3287 address from which to load. The pointer must point to a <a
3288 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
3289 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
3290 the number or order of execution of this <tt>load</tt> with other
3291 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
3294 The optional constant "align" argument specifies the alignment of the operation
3295 (that is, the alignment of the memory address). A value of 0 or an
3296 omitted "align" argument means that the operation has the preferential
3297 alignment for the target. It is the responsibility of the code emitter
3298 to ensure that the alignment information is correct. Overestimating
3299 the alignment results in an undefined behavior. Underestimating the
3300 alignment may produce less efficient code. An alignment of 1 is always
3304 <p>The location of memory pointed to is loaded.</p>
3306 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3308 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
3309 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
3312 <!-- _______________________________________________________________________ -->
3313 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
3314 Instruction</a> </div>
3315 <div class="doc_text">
3317 <pre> store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3318 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3321 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
3323 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
3324 to store and an address at which to store it. The type of the '<tt><pointer></tt>'
3325 operand must be a pointer to the <a href="#t_firstclass">first class</a> type
3326 of the '<tt><value></tt>'
3327 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
3328 optimizer is not allowed to modify the number or order of execution of
3329 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
3330 href="#i_store">store</a></tt> instructions.</p>
3332 The optional constant "align" argument specifies the alignment of the operation
3333 (that is, the alignment of the memory address). A value of 0 or an
3334 omitted "align" argument means that the operation has the preferential
3335 alignment for the target. It is the responsibility of the code emitter
3336 to ensure that the alignment information is correct. Overestimating
3337 the alignment results in an undefined behavior. Underestimating the
3338 alignment may produce less efficient code. An alignment of 1 is always
3342 <p>The contents of memory are updated to contain '<tt><value></tt>'
3343 at the location specified by the '<tt><pointer></tt>' operand.</p>
3345 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3346 store i32 3, i32* %ptr <i>; yields {void}</i>
3347 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
3351 <!-- _______________________________________________________________________ -->
3352 <div class="doc_subsubsection">
3353 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
3356 <div class="doc_text">
3359 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
3365 The '<tt>getelementptr</tt>' instruction is used to get the address of a
3366 subelement of an aggregate data structure. It performs address calculation only
3367 and does not access memory.</p>
3371 <p>The first argument is always a pointer, and forms the basis of the
3372 calculation. The remaining arguments are indices, that indicate which of the
3373 elements of the aggregate object are indexed. The interpretation of each index
3374 is dependent on the type being indexed into. The first index always indexes the
3375 pointer value given as the first argument, the second index indexes a value of
3376 the type pointed to (not necessarily the value directly pointed to, since the
3377 first index can be non-zero), etc. The first type indexed into must be a pointer
3378 value, subsequent types can be arrays, vectors and structs. Note that subsequent
3379 types being indexed into can never be pointers, since that would require loading
3380 the pointer before continuing calculation.</p>
3382 <p>The type of each index argument depends on the type it is indexing into.
3383 When indexing into a (packed) structure, only <tt>i32</tt> integer
3384 <b>constants</b> are allowed. When indexing into an array, pointer or vector,
3385 only integers of 32 or 64 bits are allowed (also non-constants). 32-bit values
3386 will be sign extended to 64-bits if required.</p>
3388 <p>For example, let's consider a C code fragment and how it gets
3389 compiled to LLVM:</p>
3391 <div class="doc_code">
3404 int *foo(struct ST *s) {
3405 return &s[1].Z.B[5][13];
3410 <p>The LLVM code generated by the GCC frontend is:</p>
3412 <div class="doc_code">
3414 %RT = type { i8 , [10 x [20 x i32]], i8 }
3415 %ST = type { i32, double, %RT }
3417 define i32* %foo(%ST* %s) {
3419 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
3427 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
3428 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
3429 }</tt>' type, a structure. The second index indexes into the third element of
3430 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
3431 i8 }</tt>' type, another structure. The third index indexes into the second
3432 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
3433 array. The two dimensions of the array are subscripted into, yielding an
3434 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
3435 to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
3437 <p>Note that it is perfectly legal to index partially through a
3438 structure, returning a pointer to an inner element. Because of this,
3439 the LLVM code for the given testcase is equivalent to:</p>
3442 define i32* %foo(%ST* %s) {
3443 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
3444 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
3445 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
3446 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
3447 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
3452 <p>Note that it is undefined to access an array out of bounds: array and
3453 pointer indexes must always be within the defined bounds of the array type.
3454 The one exception for this rule is zero length arrays. These arrays are
3455 defined to be accessible as variable length arrays, which requires access
3456 beyond the zero'th element.</p>
3458 <p>The getelementptr instruction is often confusing. For some more insight
3459 into how it works, see <a href="GetElementPtr.html">the getelementptr
3465 <i>; yields [12 x i8]*:aptr</i>
3466 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
3467 <i>; yields i8*:vptr</i>
3468 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
3469 <i>; yields i8*:eptr</i>
3470 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
3474 <!-- ======================================================================= -->
3475 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
3477 <div class="doc_text">
3478 <p>The instructions in this category are the conversion instructions (casting)
3479 which all take a single operand and a type. They perform various bit conversions
3483 <!-- _______________________________________________________________________ -->
3484 <div class="doc_subsubsection">
3485 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
3487 <div class="doc_text">
3491 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
3496 The '<tt>trunc</tt>' instruction truncates its operand to the type <tt>ty2</tt>.
3501 The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
3502 be an <a href="#t_integer">integer</a> type, and a type that specifies the size
3503 and type of the result, which must be an <a href="#t_integer">integer</a>
3504 type. The bit size of <tt>value</tt> must be larger than the bit size of
3505 <tt>ty2</tt>. Equal sized types are not allowed.</p>
3509 The '<tt>trunc</tt>' instruction truncates the high order bits in <tt>value</tt>
3510 and converts the remaining bits to <tt>ty2</tt>. Since the source size must be
3511 larger than the destination size, <tt>trunc</tt> cannot be a <i>no-op cast</i>.
3512 It will always truncate bits.</p>
3516 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
3517 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
3518 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
3522 <!-- _______________________________________________________________________ -->
3523 <div class="doc_subsubsection">
3524 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
3526 <div class="doc_text">
3530 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
3534 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
3539 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
3540 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3541 also be of <a href="#t_integer">integer</a> type. The bit size of the
3542 <tt>value</tt> must be smaller than the bit size of the destination type,
3546 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
3547 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
3549 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
3553 %X = zext i32 257 to i64 <i>; yields i64:257</i>
3554 %Y = zext i1 true to i32 <i>; yields i32:1</i>
3558 <!-- _______________________________________________________________________ -->
3559 <div class="doc_subsubsection">
3560 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
3562 <div class="doc_text">
3566 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
3570 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
3574 The '<tt>sext</tt>' instruction takes a value to cast, which must be of
3575 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3576 also be of <a href="#t_integer">integer</a> type. The bit size of the
3577 <tt>value</tt> must be smaller than the bit size of the destination type,
3582 The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
3583 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size of
3584 the type <tt>ty2</tt>.</p>
3586 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
3590 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
3591 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
3595 <!-- _______________________________________________________________________ -->
3596 <div class="doc_subsubsection">
3597 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
3600 <div class="doc_text">
3605 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
3609 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
3614 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
3615 point</a> value to cast and a <a href="#t_floating">floating point</a> type to
3616 cast it to. The size of <tt>value</tt> must be larger than the size of
3617 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
3618 <i>no-op cast</i>.</p>
3621 <p> The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
3622 <a href="#t_floating">floating point</a> type to a smaller
3623 <a href="#t_floating">floating point</a> type. If the value cannot fit within
3624 the destination type, <tt>ty2</tt>, then the results are undefined.</p>
3628 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
3629 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
3633 <!-- _______________________________________________________________________ -->
3634 <div class="doc_subsubsection">
3635 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
3637 <div class="doc_text">
3641 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
3645 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
3646 floating point value.</p>
3649 <p>The '<tt>fpext</tt>' instruction takes a
3650 <a href="#t_floating">floating point</a> <tt>value</tt> to cast,
3651 and a <a href="#t_floating">floating point</a> type to cast it to. The source
3652 type must be smaller than the destination type.</p>
3655 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
3656 <a href="#t_floating">floating point</a> type to a larger
3657 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
3658 used to make a <i>no-op cast</i> because it always changes bits. Use
3659 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
3663 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
3664 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
3668 <!-- _______________________________________________________________________ -->
3669 <div class="doc_subsubsection">
3670 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
3672 <div class="doc_text">
3676 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
3680 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
3681 unsigned integer equivalent of type <tt>ty2</tt>.
3685 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
3686 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3687 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3688 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3689 vector integer type with the same number of elements as <tt>ty</tt></p>
3692 <p> The '<tt>fptoui</tt>' instruction converts its
3693 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3694 towards zero) unsigned integer value. If the value cannot fit in <tt>ty2</tt>,
3695 the results are undefined.</p>
3699 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
3700 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
3701 %X = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
3705 <!-- _______________________________________________________________________ -->
3706 <div class="doc_subsubsection">
3707 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
3709 <div class="doc_text">
3713 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
3717 <p>The '<tt>fptosi</tt>' instruction converts
3718 <a href="#t_floating">floating point</a> <tt>value</tt> to type <tt>ty2</tt>.
3722 <p> The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
3723 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3724 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3725 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3726 vector integer type with the same number of elements as <tt>ty</tt></p>
3729 <p>The '<tt>fptosi</tt>' instruction converts its
3730 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3731 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
3732 the results are undefined.</p>
3736 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
3737 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
3738 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
3742 <!-- _______________________________________________________________________ -->
3743 <div class="doc_subsubsection">
3744 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
3746 <div class="doc_text">
3750 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3754 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
3755 integer and converts that value to the <tt>ty2</tt> type.</p>
3758 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
3759 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3760 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3761 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3762 floating point type with the same number of elements as <tt>ty</tt></p>
3765 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
3766 integer quantity and converts it to the corresponding floating point value. If
3767 the value cannot fit in the floating point value, the results are undefined.</p>
3771 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
3772 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
3776 <!-- _______________________________________________________________________ -->
3777 <div class="doc_subsubsection">
3778 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
3780 <div class="doc_text">
3784 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3788 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed
3789 integer and converts that value to the <tt>ty2</tt> type.</p>
3792 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
3793 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3794 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3795 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3796 floating point type with the same number of elements as <tt>ty</tt></p>
3799 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed
3800 integer quantity and converts it to the corresponding floating point value. If
3801 the value cannot fit in the floating point value, the results are undefined.</p>
3805 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
3806 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
3810 <!-- _______________________________________________________________________ -->
3811 <div class="doc_subsubsection">
3812 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
3814 <div class="doc_text">
3818 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
3822 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
3823 the integer type <tt>ty2</tt>.</p>
3826 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
3827 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
3828 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.</p>
3831 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
3832 <tt>ty2</tt> by interpreting the pointer value as an integer and either
3833 truncating or zero extending that value to the size of the integer type. If
3834 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
3835 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
3836 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
3841 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
3842 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
3846 <!-- _______________________________________________________________________ -->
3847 <div class="doc_subsubsection">
3848 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
3850 <div class="doc_text">
3854 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
3858 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to
3859 a pointer type, <tt>ty2</tt>.</p>
3862 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
3863 value to cast, and a type to cast it to, which must be a
3864 <a href="#t_pointer">pointer</a> type.</p>
3867 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
3868 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
3869 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
3870 size of a pointer then a truncation is done. If <tt>value</tt> is smaller than
3871 the size of a pointer then a zero extension is done. If they are the same size,
3872 nothing is done (<i>no-op cast</i>).</p>
3876 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
3877 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
3878 %Y = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
3882 <!-- _______________________________________________________________________ -->
3883 <div class="doc_subsubsection">
3884 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
3886 <div class="doc_text">
3890 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
3895 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3896 <tt>ty2</tt> without changing any bits.</p>
3900 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be
3901 a non-aggregate first class value, and a type to cast it to, which must also be
3902 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes of
3904 and the destination type, <tt>ty2</tt>, must be identical. If the source
3905 type is a pointer, the destination type must also be a pointer. This
3906 instruction supports bitwise conversion of vectors to integers and to vectors
3907 of other types (as long as they have the same size).</p>
3910 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3911 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
3912 this conversion. The conversion is done as if the <tt>value</tt> had been
3913 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only be
3914 converted to other pointer types with this instruction. To convert pointers to
3915 other types, use the <a href="#i_inttoptr">inttoptr</a> or
3916 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
3920 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
3921 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
3922 %Z = bitcast <2 x int> %V to i64; <i>; yields i64: %V</i>
3926 <!-- ======================================================================= -->
3927 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
3928 <div class="doc_text">
3929 <p>The instructions in this category are the "miscellaneous"
3930 instructions, which defy better classification.</p>
3933 <!-- _______________________________________________________________________ -->
3934 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
3936 <div class="doc_text">
3938 <pre> <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
3941 <p>The '<tt>icmp</tt>' instruction returns a boolean value or
3942 a vector of boolean values based on comparison
3943 of its two integer, integer vector, or pointer operands.</p>
3945 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
3946 the condition code indicating the kind of comparison to perform. It is not
3947 a value, just a keyword. The possible condition code are:
3950 <li><tt>eq</tt>: equal</li>
3951 <li><tt>ne</tt>: not equal </li>
3952 <li><tt>ugt</tt>: unsigned greater than</li>
3953 <li><tt>uge</tt>: unsigned greater or equal</li>
3954 <li><tt>ult</tt>: unsigned less than</li>
3955 <li><tt>ule</tt>: unsigned less or equal</li>
3956 <li><tt>sgt</tt>: signed greater than</li>
3957 <li><tt>sge</tt>: signed greater or equal</li>
3958 <li><tt>slt</tt>: signed less than</li>
3959 <li><tt>sle</tt>: signed less or equal</li>
3961 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
3962 <a href="#t_pointer">pointer</a>
3963 or integer <a href="#t_vector">vector</a> typed.
3964 They must also be identical types.</p>
3966 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to
3967 the condition code given as <tt>cond</tt>. The comparison performed always
3968 yields either an <a href="#t_primitive"><tt>i1</tt></a> or vector of <tt>i1</tt> result, as follows:
3971 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
3972 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3974 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
3975 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.</li>
3976 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
3977 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
3978 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
3979 <tt>true</tt> if <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
3980 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
3981 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
3982 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
3983 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
3984 <li><tt>sgt</tt>: interprets the operands as signed values and yields
3985 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
3986 <li><tt>sge</tt>: interprets the operands as signed values and yields
3987 <tt>true</tt> if <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
3988 <li><tt>slt</tt>: interprets the operands as signed values and yields
3989 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
3990 <li><tt>sle</tt>: interprets the operands as signed values and yields
3991 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
3993 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
3994 values are compared as if they were integers.</p>
3995 <p>If the operands are integer vectors, then they are compared
3996 element by element. The result is an <tt>i1</tt> vector with
3997 the same number of elements as the values being compared.
3998 Otherwise, the result is an <tt>i1</tt>.
4002 <pre> <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
4003 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
4004 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
4005 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
4006 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
4007 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
4011 <!-- _______________________________________________________________________ -->
4012 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
4014 <div class="doc_text">
4016 <pre> <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
4019 <p>The '<tt>fcmp</tt>' instruction returns a boolean value
4020 or vector of boolean values based on comparison
4021 of its operands.</p>
4023 If the operands are floating point scalars, then the result
4024 type is a boolean (<a href="#t_primitive"><tt>i1</tt></a>).
4026 <p>If the operands are floating point vectors, then the result type
4027 is a vector of boolean with the same number of elements as the
4028 operands being compared.</p>
4030 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
4031 the condition code indicating the kind of comparison to perform. It is not
4032 a value, just a keyword. The possible condition code are:</p>
4034 <li><tt>false</tt>: no comparison, always returns false</li>
4035 <li><tt>oeq</tt>: ordered and equal</li>
4036 <li><tt>ogt</tt>: ordered and greater than </li>
4037 <li><tt>oge</tt>: ordered and greater than or equal</li>
4038 <li><tt>olt</tt>: ordered and less than </li>
4039 <li><tt>ole</tt>: ordered and less than or equal</li>
4040 <li><tt>one</tt>: ordered and not equal</li>
4041 <li><tt>ord</tt>: ordered (no nans)</li>
4042 <li><tt>ueq</tt>: unordered or equal</li>
4043 <li><tt>ugt</tt>: unordered or greater than </li>
4044 <li><tt>uge</tt>: unordered or greater than or equal</li>
4045 <li><tt>ult</tt>: unordered or less than </li>
4046 <li><tt>ule</tt>: unordered or less than or equal</li>
4047 <li><tt>une</tt>: unordered or not equal</li>
4048 <li><tt>uno</tt>: unordered (either nans)</li>
4049 <li><tt>true</tt>: no comparison, always returns true</li>
4051 <p><i>Ordered</i> means that neither operand is a QNAN while
4052 <i>unordered</i> means that either operand may be a QNAN.</p>
4053 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be
4054 either a <a href="#t_floating">floating point</a> type
4055 or a <a href="#t_vector">vector</a> of floating point type.
4056 They must have identical types.</p>
4058 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4059 according to the condition code given as <tt>cond</tt>.
4060 If the operands are vectors, then the vectors are compared
4062 Each comparison performed
4063 always yields an <a href="#t_primitive">i1</a> result, as follows:</p>
4065 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
4066 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4067 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
4068 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4069 <tt>op1</tt> is greather than <tt>op2</tt>.</li>
4070 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4071 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4072 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4073 <tt>op1</tt> is less than <tt>op2</tt>.</li>
4074 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4075 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4076 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4077 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
4078 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
4079 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
4080 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
4081 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
4082 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4083 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
4084 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4085 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
4086 <tt>op1</tt> is less than <tt>op2</tt>.</li>
4087 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
4088 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4089 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
4090 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
4091 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
4092 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
4096 <pre> <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
4097 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
4098 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
4099 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
4103 <!-- _______________________________________________________________________ -->
4104 <div class="doc_subsubsection">
4105 <a name="i_vicmp">'<tt>vicmp</tt>' Instruction</a>
4107 <div class="doc_text">
4109 <pre> <result> = vicmp <cond> <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4112 <p>The '<tt>vicmp</tt>' instruction returns an integer vector value based on
4113 element-wise comparison of its two integer vector operands.</p>
4115 <p>The '<tt>vicmp</tt>' instruction takes three operands. The first operand is
4116 the condition code indicating the kind of comparison to perform. It is not
4117 a value, just a keyword. The possible condition code are:</p>
4119 <li><tt>eq</tt>: equal</li>
4120 <li><tt>ne</tt>: not equal </li>
4121 <li><tt>ugt</tt>: unsigned greater than</li>
4122 <li><tt>uge</tt>: unsigned greater or equal</li>
4123 <li><tt>ult</tt>: unsigned less than</li>
4124 <li><tt>ule</tt>: unsigned less or equal</li>
4125 <li><tt>sgt</tt>: signed greater than</li>
4126 <li><tt>sge</tt>: signed greater or equal</li>
4127 <li><tt>slt</tt>: signed less than</li>
4128 <li><tt>sle</tt>: signed less or equal</li>
4130 <p>The remaining two arguments must be <a href="#t_vector">vector</a> or
4131 <a href="#t_integer">integer</a> typed. They must also be identical types.</p>
4133 <p>The '<tt>vicmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4134 according to the condition code given as <tt>cond</tt>. The comparison yields a
4135 <a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, of
4136 identical type as the values being compared. The most significant bit in each
4137 element is 1 if the element-wise comparison evaluates to true, and is 0
4138 otherwise. All other bits of the result are undefined. The condition codes
4139 are evaluated identically to the <a href="#i_icmp">'<tt>icmp</tt>'
4140 instruction</a>.</p>
4144 <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>
4145 <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>
4149 <!-- _______________________________________________________________________ -->
4150 <div class="doc_subsubsection">
4151 <a name="i_vfcmp">'<tt>vfcmp</tt>' Instruction</a>
4153 <div class="doc_text">
4155 <pre> <result> = vfcmp <cond> <ty> <op1>, <op2></pre>
4157 <p>The '<tt>vfcmp</tt>' instruction returns an integer vector value based on
4158 element-wise comparison of its two floating point vector operands. The output
4159 elements have the same width as the input elements.</p>
4161 <p>The '<tt>vfcmp</tt>' instruction takes three operands. The first operand is
4162 the condition code indicating the kind of comparison to perform. It is not
4163 a value, just a keyword. The possible condition code are:</p>
4165 <li><tt>false</tt>: no comparison, always returns false</li>
4166 <li><tt>oeq</tt>: ordered and equal</li>
4167 <li><tt>ogt</tt>: ordered and greater than </li>
4168 <li><tt>oge</tt>: ordered and greater than or equal</li>
4169 <li><tt>olt</tt>: ordered and less than </li>
4170 <li><tt>ole</tt>: ordered and less than or equal</li>
4171 <li><tt>one</tt>: ordered and not equal</li>
4172 <li><tt>ord</tt>: ordered (no nans)</li>
4173 <li><tt>ueq</tt>: unordered or equal</li>
4174 <li><tt>ugt</tt>: unordered or greater than </li>
4175 <li><tt>uge</tt>: unordered or greater than or equal</li>
4176 <li><tt>ult</tt>: unordered or less than </li>
4177 <li><tt>ule</tt>: unordered or less than or equal</li>
4178 <li><tt>une</tt>: unordered or not equal</li>
4179 <li><tt>uno</tt>: unordered (either nans)</li>
4180 <li><tt>true</tt>: no comparison, always returns true</li>
4182 <p>The remaining two arguments must be <a href="#t_vector">vector</a> of
4183 <a href="#t_floating">floating point</a> typed. They must also be identical
4186 <p>The '<tt>vfcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4187 according to the condition code given as <tt>cond</tt>. The comparison yields a
4188 <a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, with
4189 an identical number of elements as the values being compared, and each element
4190 having identical with to the width of the floating point elements. The most
4191 significant bit in each element is 1 if the element-wise comparison evaluates to
4192 true, and is 0 otherwise. All other bits of the result are undefined. The
4193 condition codes are evaluated identically to the
4194 <a href="#i_fcmp">'<tt>fcmp</tt>' instruction</a>.</p>
4198 <i>; yields: result=<2 x i32> < i32 0, i32 -1 ></i>
4199 <result> = vfcmp oeq <2 x float> < float 4, float 0 >, < float 5, float 0 >
4201 <i>; yields: result=<2 x i64> < i64 -1, i64 0 ></i>
4202 <result> = vfcmp ult <2 x double> < double 1, double 2 >, < double 2, double 2>
4206 <!-- _______________________________________________________________________ -->
4207 <div class="doc_subsubsection">
4208 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
4211 <div class="doc_text">
4215 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
4217 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
4218 the SSA graph representing the function.</p>
4221 <p>The type of the incoming values is specified with the first type
4222 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
4223 as arguments, with one pair for each predecessor basic block of the
4224 current block. Only values of <a href="#t_firstclass">first class</a>
4225 type may be used as the value arguments to the PHI node. Only labels
4226 may be used as the label arguments.</p>
4228 <p>There must be no non-phi instructions between the start of a basic
4229 block and the PHI instructions: i.e. PHI instructions must be first in
4234 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
4235 specified by the pair corresponding to the predecessor basic block that executed
4236 just prior to the current block.</p>
4240 Loop: ; Infinite loop that counts from 0 on up...
4241 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
4242 %nextindvar = add i32 %indvar, 1
4247 <!-- _______________________________________________________________________ -->
4248 <div class="doc_subsubsection">
4249 <a name="i_select">'<tt>select</tt>' Instruction</a>
4252 <div class="doc_text">
4257 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
4259 <i>selty</i> is either i1 or {<N x i1>}
4265 The '<tt>select</tt>' instruction is used to choose one value based on a
4266 condition, without branching.
4273 The '<tt>select</tt>' instruction requires an 'i1' value or
4274 a vector of 'i1' values indicating the
4275 condition, and two values of the same <a href="#t_firstclass">first class</a>
4276 type. If the val1/val2 are vectors and
4277 the condition is a scalar, then entire vectors are selected, not
4278 individual elements.
4284 If the condition is an i1 and it evaluates to 1, the instruction returns the first
4285 value argument; otherwise, it returns the second value argument.
4288 If the condition is a vector of i1, then the value arguments must
4289 be vectors of the same size, and the selection is done element
4296 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
4301 <!-- _______________________________________________________________________ -->
4302 <div class="doc_subsubsection">
4303 <a name="i_call">'<tt>call</tt>' Instruction</a>
4306 <div class="doc_text">
4310 <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>]
4315 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
4319 <p>This instruction requires several arguments:</p>
4323 <p>The optional "tail" marker indicates whether the callee function accesses
4324 any allocas or varargs in the caller. If the "tail" marker is present, the
4325 function call is eligible for tail call optimization. Note that calls may
4326 be marked "tail" even if they do not occur before a <a
4327 href="#i_ret"><tt>ret</tt></a> instruction.</p>
4330 <p>The optional "cconv" marker indicates which <a href="#callingconv">calling
4331 convention</a> the call should use. If none is specified, the call defaults
4332 to using C calling conventions.</p>
4336 <p>The optional <a href="#paramattrs">Parameter Attributes</a> list for
4337 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>',
4338 and '<tt>inreg</tt>' attributes are valid here.</p>
4342 <p>'<tt>ty</tt>': the type of the call instruction itself which is also
4343 the type of the return value. Functions that return no value are marked
4344 <tt><a href="#t_void">void</a></tt>.</p>
4347 <p>'<tt>fnty</tt>': shall be the signature of the pointer to function
4348 value being invoked. The argument types must match the types implied by
4349 this signature. This type can be omitted if the function is not varargs
4350 and if the function type does not return a pointer to a function.</p>
4353 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
4354 be invoked. In most cases, this is a direct function invocation, but
4355 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
4356 to function value.</p>
4359 <p>'<tt>function args</tt>': argument list whose types match the
4360 function signature argument types. All arguments must be of
4361 <a href="#t_firstclass">first class</a> type. If the function signature
4362 indicates the function accepts a variable number of arguments, the extra
4363 arguments can be specified.</p>
4366 <p>The optional <a href="#fnattrs">function attributes</a> list. Only
4367 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
4368 '<tt>readnone</tt>' attributes are valid here.</p>
4374 <p>The '<tt>call</tt>' instruction is used to cause control flow to
4375 transfer to a specified function, with its incoming arguments bound to
4376 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
4377 instruction in the called function, control flow continues with the
4378 instruction after the function call, and the return value of the
4379 function is bound to the result argument.</p>
4384 %retval = call i32 @test(i32 %argc)
4385 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42) <i>; yields i32</i>
4386 %X = tail call i32 @foo() <i>; yields i32</i>
4387 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
4388 call void %foo(i8 97 signext)
4390 %struct.A = type { i32, i8 }
4391 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
4392 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
4393 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
4394 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
4395 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
4400 <!-- _______________________________________________________________________ -->
4401 <div class="doc_subsubsection">
4402 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
4405 <div class="doc_text">
4410 <resultval> = va_arg <va_list*> <arglist>, <argty>
4415 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
4416 the "variable argument" area of a function call. It is used to implement the
4417 <tt>va_arg</tt> macro in C.</p>
4421 <p>This instruction takes a <tt>va_list*</tt> value and the type of
4422 the argument. It returns a value of the specified argument type and
4423 increments the <tt>va_list</tt> to point to the next argument. The
4424 actual type of <tt>va_list</tt> is target specific.</p>
4428 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
4429 type from the specified <tt>va_list</tt> and causes the
4430 <tt>va_list</tt> to point to the next argument. For more information,
4431 see the variable argument handling <a href="#int_varargs">Intrinsic
4434 <p>It is legal for this instruction to be called in a function which does not
4435 take a variable number of arguments, for example, the <tt>vfprintf</tt>
4438 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
4439 href="#intrinsics">intrinsic function</a> because it takes a type as an
4444 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
4448 <!-- *********************************************************************** -->
4449 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
4450 <!-- *********************************************************************** -->
4452 <div class="doc_text">
4454 <p>LLVM supports the notion of an "intrinsic function". These functions have
4455 well known names and semantics and are required to follow certain restrictions.
4456 Overall, these intrinsics represent an extension mechanism for the LLVM
4457 language that does not require changing all of the transformations in LLVM when
4458 adding to the language (or the bitcode reader/writer, the parser, etc...).</p>
4460 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
4461 prefix is reserved in LLVM for intrinsic names; thus, function names may not
4462 begin with this prefix. Intrinsic functions must always be external functions:
4463 you cannot define the body of intrinsic functions. Intrinsic functions may
4464 only be used in call or invoke instructions: it is illegal to take the address
4465 of an intrinsic function. Additionally, because intrinsic functions are part
4466 of the LLVM language, it is required if any are added that they be documented
4469 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
4470 a family of functions that perform the same operation but on different data
4471 types. Because LLVM can represent over 8 million different integer types,
4472 overloading is used commonly to allow an intrinsic function to operate on any
4473 integer type. One or more of the argument types or the result type can be
4474 overloaded to accept any integer type. Argument types may also be defined as
4475 exactly matching a previous argument's type or the result type. This allows an
4476 intrinsic function which accepts multiple arguments, but needs all of them to
4477 be of the same type, to only be overloaded with respect to a single argument or
4480 <p>Overloaded intrinsics will have the names of its overloaded argument types
4481 encoded into its function name, each preceded by a period. Only those types
4482 which are overloaded result in a name suffix. Arguments whose type is matched
4483 against another type do not. For example, the <tt>llvm.ctpop</tt> function can
4484 take an integer of any width and returns an integer of exactly the same integer
4485 width. This leads to a family of functions such as
4486 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29 %val)</tt>.
4487 Only one type, the return type, is overloaded, and only one type suffix is
4488 required. Because the argument's type is matched against the return type, it
4489 does not require its own name suffix.</p>
4491 <p>To learn how to add an intrinsic function, please see the
4492 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.
4497 <!-- ======================================================================= -->
4498 <div class="doc_subsection">
4499 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
4502 <div class="doc_text">
4504 <p>Variable argument support is defined in LLVM with the <a
4505 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
4506 intrinsic functions. These functions are related to the similarly
4507 named macros defined in the <tt><stdarg.h></tt> header file.</p>
4509 <p>All of these functions operate on arguments that use a
4510 target-specific value type "<tt>va_list</tt>". The LLVM assembly
4511 language reference manual does not define what this type is, so all
4512 transformations should be prepared to handle these functions regardless of
4515 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
4516 instruction and the variable argument handling intrinsic functions are
4519 <div class="doc_code">
4521 define i32 @test(i32 %X, ...) {
4522 ; Initialize variable argument processing
4524 %ap2 = bitcast i8** %ap to i8*
4525 call void @llvm.va_start(i8* %ap2)
4527 ; Read a single integer argument
4528 %tmp = va_arg i8** %ap, i32
4530 ; Demonstrate usage of llvm.va_copy and llvm.va_end
4532 %aq2 = bitcast i8** %aq to i8*
4533 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
4534 call void @llvm.va_end(i8* %aq2)
4536 ; Stop processing of arguments.
4537 call void @llvm.va_end(i8* %ap2)
4541 declare void @llvm.va_start(i8*)
4542 declare void @llvm.va_copy(i8*, i8*)
4543 declare void @llvm.va_end(i8*)
4549 <!-- _______________________________________________________________________ -->
4550 <div class="doc_subsubsection">
4551 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
4555 <div class="doc_text">
4557 <pre> declare void %llvm.va_start(i8* <arglist>)<br></pre>
4559 <p>The '<tt>llvm.va_start</tt>' intrinsic initializes
4560 <tt>*<arglist></tt> for subsequent use by <tt><a
4561 href="#i_va_arg">va_arg</a></tt>.</p>
4565 <p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
4569 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
4570 macro available in C. In a target-dependent way, it initializes the
4571 <tt>va_list</tt> element to which the argument points, so that the next call to
4572 <tt>va_arg</tt> will produce the first variable argument passed to the function.
4573 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
4574 last argument of the function as the compiler can figure that out.</p>
4578 <!-- _______________________________________________________________________ -->
4579 <div class="doc_subsubsection">
4580 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
4583 <div class="doc_text">
4585 <pre> declare void @llvm.va_end(i8* <arglist>)<br></pre>
4588 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
4589 which has been initialized previously with <tt><a href="#int_va_start">llvm.va_start</a></tt>
4590 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
4594 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
4598 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
4599 macro available in C. In a target-dependent way, it destroys the
4600 <tt>va_list</tt> element to which the argument points. Calls to <a
4601 href="#int_va_start"><tt>llvm.va_start</tt></a> and <a href="#int_va_copy">
4602 <tt>llvm.va_copy</tt></a> must be matched exactly with calls to
4603 <tt>llvm.va_end</tt>.</p>
4607 <!-- _______________________________________________________________________ -->
4608 <div class="doc_subsubsection">
4609 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
4612 <div class="doc_text">
4617 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
4622 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
4623 from the source argument list to the destination argument list.</p>
4627 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
4628 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
4633 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
4634 macro available in C. In a target-dependent way, it copies the source
4635 <tt>va_list</tt> element into the destination <tt>va_list</tt> element. This
4636 intrinsic is necessary because the <tt><a href="#int_va_start">
4637 llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
4638 example, memory allocation.</p>
4642 <!-- ======================================================================= -->
4643 <div class="doc_subsection">
4644 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
4647 <div class="doc_text">
4650 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
4651 Collection</a> (GC) requires the implementation and generation of these
4653 These intrinsics allow identification of <a href="#int_gcroot">GC roots on the
4654 stack</a>, as well as garbage collector implementations that require <a
4655 href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a> barriers.
4656 Front-ends for type-safe garbage collected languages should generate these
4657 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
4658 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
4661 <p>The garbage collection intrinsics only operate on objects in the generic
4662 address space (address space zero).</p>
4666 <!-- _______________________________________________________________________ -->
4667 <div class="doc_subsubsection">
4668 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
4671 <div class="doc_text">
4676 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
4681 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
4682 the code generator, and allows some metadata to be associated with it.</p>
4686 <p>The first argument specifies the address of a stack object that contains the
4687 root pointer. The second pointer (which must be either a constant or a global
4688 value address) contains the meta-data to be associated with the root.</p>
4692 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
4693 location. At compile-time, the code generator generates information to allow
4694 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
4695 intrinsic may only be used in a function which <a href="#gc">specifies a GC
4701 <!-- _______________________________________________________________________ -->
4702 <div class="doc_subsubsection">
4703 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
4706 <div class="doc_text">
4711 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
4716 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
4717 locations, allowing garbage collector implementations that require read
4722 <p>The second argument is the address to read from, which should be an address
4723 allocated from the garbage collector. The first object is a pointer to the
4724 start of the referenced object, if needed by the language runtime (otherwise
4729 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
4730 instruction, but may be replaced with substantially more complex code by the
4731 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
4732 may only be used in a function which <a href="#gc">specifies a GC
4738 <!-- _______________________________________________________________________ -->
4739 <div class="doc_subsubsection">
4740 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
4743 <div class="doc_text">
4748 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
4753 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
4754 locations, allowing garbage collector implementations that require write
4755 barriers (such as generational or reference counting collectors).</p>
4759 <p>The first argument is the reference to store, the second is the start of the
4760 object to store it to, and the third is the address of the field of Obj to
4761 store to. If the runtime does not require a pointer to the object, Obj may be
4766 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
4767 instruction, but may be replaced with substantially more complex code by the
4768 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
4769 may only be used in a function which <a href="#gc">specifies a GC
4776 <!-- ======================================================================= -->
4777 <div class="doc_subsection">
4778 <a name="int_codegen">Code Generator Intrinsics</a>
4781 <div class="doc_text">
4783 These intrinsics are provided by LLVM to expose special features that may only
4784 be implemented with code generator support.
4789 <!-- _______________________________________________________________________ -->
4790 <div class="doc_subsubsection">
4791 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
4794 <div class="doc_text">
4798 declare i8 *@llvm.returnaddress(i32 <level>)
4804 The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
4805 target-specific value indicating the return address of the current function
4806 or one of its callers.
4812 The argument to this intrinsic indicates which function to return the address
4813 for. Zero indicates the calling function, one indicates its caller, etc. The
4814 argument is <b>required</b> to be a constant integer value.
4820 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
4821 the return address of the specified call frame, or zero if it cannot be
4822 identified. The value returned by this intrinsic is likely to be incorrect or 0
4823 for arguments other than zero, so it should only be used for debugging purposes.
4827 Note that calling this intrinsic does not prevent function inlining or other
4828 aggressive transformations, so the value returned may not be that of the obvious
4829 source-language caller.
4834 <!-- _______________________________________________________________________ -->
4835 <div class="doc_subsubsection">
4836 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
4839 <div class="doc_text">
4843 declare i8 *@llvm.frameaddress(i32 <level>)
4849 The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
4850 target-specific frame pointer value for the specified stack frame.
4856 The argument to this intrinsic indicates which function to return the frame
4857 pointer for. Zero indicates the calling function, one indicates its caller,
4858 etc. The argument is <b>required</b> to be a constant integer value.
4864 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
4865 the frame address of the specified call frame, or zero if it cannot be
4866 identified. The value returned by this intrinsic is likely to be incorrect or 0
4867 for arguments other than zero, so it should only be used for debugging purposes.
4871 Note that calling this intrinsic does not prevent function inlining or other
4872 aggressive transformations, so the value returned may not be that of the obvious
4873 source-language caller.
4877 <!-- _______________________________________________________________________ -->
4878 <div class="doc_subsubsection">
4879 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
4882 <div class="doc_text">
4886 declare i8 *@llvm.stacksave()
4892 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
4893 the function stack, for use with <a href="#int_stackrestore">
4894 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
4895 features like scoped automatic variable sized arrays in C99.
4901 This intrinsic returns a opaque pointer value that can be passed to <a
4902 href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
4903 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
4904 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
4905 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
4906 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
4907 that were allocated after the <tt>llvm.stacksave</tt> was executed.
4912 <!-- _______________________________________________________________________ -->
4913 <div class="doc_subsubsection">
4914 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
4917 <div class="doc_text">
4921 declare void @llvm.stackrestore(i8 * %ptr)
4927 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
4928 the function stack to the state it was in when the corresponding <a
4929 href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
4930 useful for implementing language features like scoped automatic variable sized
4937 See the description for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.
4943 <!-- _______________________________________________________________________ -->
4944 <div class="doc_subsubsection">
4945 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
4948 <div class="doc_text">
4952 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>)
4959 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
4960 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
4962 effect on the behavior of the program but can change its performance
4969 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
4970 determining if the fetch should be for a read (0) or write (1), and
4971 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
4972 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
4973 <tt>locality</tt> arguments must be constant integers.
4979 This intrinsic does not modify the behavior of the program. In particular,
4980 prefetches cannot trap and do not produce a value. On targets that support this
4981 intrinsic, the prefetch can provide hints to the processor cache for better
4987 <!-- _______________________________________________________________________ -->
4988 <div class="doc_subsubsection">
4989 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
4992 <div class="doc_text">
4996 declare void @llvm.pcmarker(i32 <id>)
5003 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
5005 code to simulators and other tools. The method is target specific, but it is
5006 expected that the marker will use exported symbols to transmit the PC of the
5008 The marker makes no guarantees that it will remain with any specific instruction
5009 after optimizations. It is possible that the presence of a marker will inhibit
5010 optimizations. The intended use is to be inserted after optimizations to allow
5011 correlations of simulation runs.
5017 <tt>id</tt> is a numerical id identifying the marker.
5023 This intrinsic does not modify the behavior of the program. Backends that do not
5024 support this intrinisic may ignore it.
5029 <!-- _______________________________________________________________________ -->
5030 <div class="doc_subsubsection">
5031 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
5034 <div class="doc_text">
5038 declare i64 @llvm.readcyclecounter( )
5045 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
5046 counter register (or similar low latency, high accuracy clocks) on those targets
5047 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
5048 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
5049 should only be used for small timings.
5055 When directly supported, reading the cycle counter should not modify any memory.
5056 Implementations are allowed to either return a application specific value or a
5057 system wide value. On backends without support, this is lowered to a constant 0.
5062 <!-- ======================================================================= -->
5063 <div class="doc_subsection">
5064 <a name="int_libc">Standard C Library Intrinsics</a>
5067 <div class="doc_text">
5069 LLVM provides intrinsics for a few important standard C library functions.
5070 These intrinsics allow source-language front-ends to pass information about the
5071 alignment of the pointer arguments to the code generator, providing opportunity
5072 for more efficient code generation.
5077 <!-- _______________________________________________________________________ -->
5078 <div class="doc_subsubsection">
5079 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
5082 <div class="doc_text">
5085 <p>This is an overloaded intrinsic. You can use llvm.memcpy on any integer bit
5086 width. Not all targets support all bit widths however.</p>
5088 declare void @llvm.memcpy.i8(i8 * <dest>, i8 * <src>,
5089 i8 <len>, i32 <align>)
5090 declare void @llvm.memcpy.i16(i8 * <dest>, i8 * <src>,
5091 i16 <len>, i32 <align>)
5092 declare void @llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
5093 i32 <len>, i32 <align>)
5094 declare void @llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
5095 i64 <len>, i32 <align>)
5101 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
5102 location to the destination location.
5106 Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
5107 intrinsics do not return a value, and takes an extra alignment argument.
5113 The first argument is a pointer to the destination, the second is a pointer to
5114 the source. The third argument is an integer argument
5115 specifying the number of bytes to copy, and the fourth argument is the alignment
5116 of the source and destination locations.
5120 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5121 the caller guarantees that both the source and destination pointers are aligned
5128 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
5129 location to the destination location, which are not allowed to overlap. It
5130 copies "len" bytes of memory over. If the argument is known to be aligned to
5131 some boundary, this can be specified as the fourth argument, otherwise it should
5137 <!-- _______________________________________________________________________ -->
5138 <div class="doc_subsubsection">
5139 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
5142 <div class="doc_text">
5145 <p>This is an overloaded intrinsic. You can use llvm.memmove on any integer bit
5146 width. Not all targets support all bit widths however.</p>
5148 declare void @llvm.memmove.i8(i8 * <dest>, i8 * <src>,
5149 i8 <len>, i32 <align>)
5150 declare void @llvm.memmove.i16(i8 * <dest>, i8 * <src>,
5151 i16 <len>, i32 <align>)
5152 declare void @llvm.memmove.i32(i8 * <dest>, i8 * <src>,
5153 i32 <len>, i32 <align>)
5154 declare void @llvm.memmove.i64(i8 * <dest>, i8 * <src>,
5155 i64 <len>, i32 <align>)
5161 The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
5162 location to the destination location. It is similar to the
5163 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to overlap.
5167 Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
5168 intrinsics do not return a value, and takes an extra alignment argument.
5174 The first argument is a pointer to the destination, the second is a pointer to
5175 the source. The third argument is an integer argument
5176 specifying the number of bytes to copy, and the fourth argument is the alignment
5177 of the source and destination locations.
5181 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5182 the caller guarantees that the source and destination pointers are aligned to
5189 The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
5190 location to the destination location, which may overlap. It
5191 copies "len" bytes of memory over. If the argument is known to be aligned to
5192 some boundary, this can be specified as the fourth argument, otherwise it should
5198 <!-- _______________________________________________________________________ -->
5199 <div class="doc_subsubsection">
5200 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
5203 <div class="doc_text">
5206 <p>This is an overloaded intrinsic. You can use llvm.memset on any integer bit
5207 width. Not all targets support all bit widths however.</p>
5209 declare void @llvm.memset.i8(i8 * <dest>, i8 <val>,
5210 i8 <len>, i32 <align>)
5211 declare void @llvm.memset.i16(i8 * <dest>, i8 <val>,
5212 i16 <len>, i32 <align>)
5213 declare void @llvm.memset.i32(i8 * <dest>, i8 <val>,
5214 i32 <len>, i32 <align>)
5215 declare void @llvm.memset.i64(i8 * <dest>, i8 <val>,
5216 i64 <len>, i32 <align>)
5222 The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
5227 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
5228 does not return a value, and takes an extra alignment argument.
5234 The first argument is a pointer to the destination to fill, the second is the
5235 byte value to fill it with, the third argument is an integer
5236 argument specifying the number of bytes to fill, and the fourth argument is the
5237 known alignment of destination location.
5241 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5242 the caller guarantees that the destination pointer is aligned to that boundary.
5248 The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
5250 destination location. If the argument is known to be aligned to some boundary,
5251 this can be specified as the fourth argument, otherwise it should be set to 0 or
5257 <!-- _______________________________________________________________________ -->
5258 <div class="doc_subsubsection">
5259 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
5262 <div class="doc_text">
5265 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
5266 floating point or vector of floating point type. Not all targets support all
5269 declare float @llvm.sqrt.f32(float %Val)
5270 declare double @llvm.sqrt.f64(double %Val)
5271 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
5272 declare fp128 @llvm.sqrt.f128(fp128 %Val)
5273 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
5279 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
5280 returning the same value as the libm '<tt>sqrt</tt>' functions would. Unlike
5281 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
5282 negative numbers other than -0.0 (which allows for better optimization, because
5283 there is no need to worry about errno being set). <tt>llvm.sqrt(-0.0)</tt> is
5284 defined to return -0.0 like IEEE sqrt.
5290 The argument and return value are floating point numbers of the same type.
5296 This function returns the sqrt of the specified operand if it is a nonnegative
5297 floating point number.
5301 <!-- _______________________________________________________________________ -->
5302 <div class="doc_subsubsection">
5303 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
5306 <div class="doc_text">
5309 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
5310 floating point or vector of floating point type. Not all targets support all
5313 declare float @llvm.powi.f32(float %Val, i32 %power)
5314 declare double @llvm.powi.f64(double %Val, i32 %power)
5315 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
5316 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
5317 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
5323 The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
5324 specified (positive or negative) power. The order of evaluation of
5325 multiplications is not defined. When a vector of floating point type is
5326 used, the second argument remains a scalar integer value.
5332 The second argument is an integer power, and the first is a value to raise to
5339 This function returns the first value raised to the second power with an
5340 unspecified sequence of rounding operations.</p>
5343 <!-- _______________________________________________________________________ -->
5344 <div class="doc_subsubsection">
5345 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
5348 <div class="doc_text">
5351 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
5352 floating point or vector of floating point type. Not all targets support all
5355 declare float @llvm.sin.f32(float %Val)
5356 declare double @llvm.sin.f64(double %Val)
5357 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
5358 declare fp128 @llvm.sin.f128(fp128 %Val)
5359 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
5365 The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.
5371 The argument and return value are floating point numbers of the same type.
5377 This function returns the sine of the specified operand, returning the
5378 same values as the libm <tt>sin</tt> functions would, and handles error
5379 conditions in the same way.</p>
5382 <!-- _______________________________________________________________________ -->
5383 <div class="doc_subsubsection">
5384 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
5387 <div class="doc_text">
5390 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
5391 floating point or vector of floating point type. Not all targets support all
5394 declare float @llvm.cos.f32(float %Val)
5395 declare double @llvm.cos.f64(double %Val)
5396 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
5397 declare fp128 @llvm.cos.f128(fp128 %Val)
5398 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
5404 The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.
5410 The argument and return value are floating point numbers of the same type.
5416 This function returns the cosine of the specified operand, returning the
5417 same values as the libm <tt>cos</tt> functions would, and handles error
5418 conditions in the same way.</p>
5421 <!-- _______________________________________________________________________ -->
5422 <div class="doc_subsubsection">
5423 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
5426 <div class="doc_text">
5429 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
5430 floating point or vector of floating point type. Not all targets support all
5433 declare float @llvm.pow.f32(float %Val, float %Power)
5434 declare double @llvm.pow.f64(double %Val, double %Power)
5435 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
5436 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
5437 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
5443 The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
5444 specified (positive or negative) power.
5450 The second argument is a floating point power, and the first is a value to
5451 raise to that power.
5457 This function returns the first value raised to the second power,
5459 same values as the libm <tt>pow</tt> functions would, and handles error
5460 conditions in the same way.</p>
5464 <!-- ======================================================================= -->
5465 <div class="doc_subsection">
5466 <a name="int_manip">Bit Manipulation Intrinsics</a>
5469 <div class="doc_text">
5471 LLVM provides intrinsics for a few important bit manipulation operations.
5472 These allow efficient code generation for some algorithms.
5477 <!-- _______________________________________________________________________ -->
5478 <div class="doc_subsubsection">
5479 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
5482 <div class="doc_text">
5485 <p>This is an overloaded intrinsic function. You can use bswap on any integer
5486 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
5488 declare i16 @llvm.bswap.i16(i16 <id>)
5489 declare i32 @llvm.bswap.i32(i32 <id>)
5490 declare i64 @llvm.bswap.i64(i64 <id>)
5496 The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
5497 values with an even number of bytes (positive multiple of 16 bits). These are
5498 useful for performing operations on data that is not in the target's native
5505 The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
5506 and low byte of the input i16 swapped. Similarly, the <tt>llvm.bswap.i32</tt>
5507 intrinsic returns an i32 value that has the four bytes of the input i32
5508 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the returned
5509 i32 will have its bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i48</tt>,
5510 <tt>llvm.bswap.i64</tt> and other intrinsics extend this concept to
5511 additional even-byte lengths (6 bytes, 8 bytes and more, respectively).
5516 <!-- _______________________________________________________________________ -->
5517 <div class="doc_subsubsection">
5518 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
5521 <div class="doc_text">
5524 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
5525 width. Not all targets support all bit widths however.</p>
5527 declare i8 @llvm.ctpop.i8 (i8 <src>)
5528 declare i16 @llvm.ctpop.i16(i16 <src>)
5529 declare i32 @llvm.ctpop.i32(i32 <src>)
5530 declare i64 @llvm.ctpop.i64(i64 <src>)
5531 declare i256 @llvm.ctpop.i256(i256 <src>)
5537 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
5544 The only argument is the value to be counted. The argument may be of any
5545 integer type. The return type must match the argument type.
5551 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
5555 <!-- _______________________________________________________________________ -->
5556 <div class="doc_subsubsection">
5557 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
5560 <div class="doc_text">
5563 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
5564 integer bit width. Not all targets support all bit widths however.</p>
5566 declare i8 @llvm.ctlz.i8 (i8 <src>)
5567 declare i16 @llvm.ctlz.i16(i16 <src>)
5568 declare i32 @llvm.ctlz.i32(i32 <src>)
5569 declare i64 @llvm.ctlz.i64(i64 <src>)
5570 declare i256 @llvm.ctlz.i256(i256 <src>)
5576 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
5577 leading zeros in a variable.
5583 The only argument is the value to be counted. The argument may be of any
5584 integer type. The return type must match the argument type.
5590 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
5591 in a variable. If the src == 0 then the result is the size in bits of the type
5592 of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.
5598 <!-- _______________________________________________________________________ -->
5599 <div class="doc_subsubsection">
5600 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
5603 <div class="doc_text">
5606 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
5607 integer bit width. Not all targets support all bit widths however.</p>
5609 declare i8 @llvm.cttz.i8 (i8 <src>)
5610 declare i16 @llvm.cttz.i16(i16 <src>)
5611 declare i32 @llvm.cttz.i32(i32 <src>)
5612 declare i64 @llvm.cttz.i64(i64 <src>)
5613 declare i256 @llvm.cttz.i256(i256 <src>)
5619 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
5626 The only argument is the value to be counted. The argument may be of any
5627 integer type. The return type must match the argument type.
5633 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
5634 in a variable. If the src == 0 then the result is the size in bits of the type
5635 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
5639 <!-- _______________________________________________________________________ -->
5640 <div class="doc_subsubsection">
5641 <a name="int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic</a>
5644 <div class="doc_text">
5647 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.select</tt>
5648 on any integer bit width.</p>
5650 declare i17 @llvm.part.select.i17 (i17 %val, i32 %loBit, i32 %hiBit)
5651 declare i29 @llvm.part.select.i29 (i29 %val, i32 %loBit, i32 %hiBit)
5655 <p>The '<tt>llvm.part.select</tt>' family of intrinsic functions selects a
5656 range of bits from an integer value and returns them in the same bit width as
5657 the original value.</p>
5660 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5661 any bit width but they must have the same bit width. The second and third
5662 arguments must be <tt>i32</tt> type since they specify only a bit index.</p>
5665 <p>The operation of the '<tt>llvm.part.select</tt>' intrinsic has two modes
5666 of operation: forwards and reverse. If <tt>%loBit</tt> is greater than
5667 <tt>%hiBits</tt> then the intrinsic operates in reverse mode. Otherwise it
5668 operates in forward mode.</p>
5669 <p>In forward mode, this intrinsic is the equivalent of shifting <tt>%val</tt>
5670 right by <tt>%loBit</tt> bits and then ANDing it with a mask with
5671 only the <tt>%hiBit - %loBit</tt> bits set, as follows:</p>
5673 <li>The <tt>%val</tt> is shifted right (LSHR) by the number of bits specified
5674 by <tt>%loBits</tt>. This normalizes the value to the low order bits.</li>
5675 <li>The <tt>%loBits</tt> value is subtracted from the <tt>%hiBits</tt> value
5676 to determine the number of bits to retain.</li>
5677 <li>A mask of the retained bits is created by shifting a -1 value.</li>
5678 <li>The mask is ANDed with <tt>%val</tt> to produce the result.</li>
5680 <p>In reverse mode, a similar computation is made except that the bits are
5681 returned in the reverse order. So, for example, if <tt>X</tt> has the value
5682 <tt>i16 0x0ACF (101011001111)</tt> and we apply
5683 <tt>part.select(i16 X, 8, 3)</tt> to it, we get back the value
5684 <tt>i16 0x0026 (000000100110)</tt>.</p>
5687 <div class="doc_subsubsection">
5688 <a name="int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic</a>
5691 <div class="doc_text">
5694 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.set</tt>
5695 on any integer bit width.</p>
5697 declare i17 @llvm.part.set.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
5698 declare i29 @llvm.part.set.i29.i9 (i29 %val, i9 %repl, i32 %lo, i32 %hi)
5702 <p>The '<tt>llvm.part.set</tt>' family of intrinsic functions replaces a range
5703 of bits in an integer value with another integer value. It returns the integer
5704 with the replaced bits.</p>
5707 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5708 any bit width but they must have the same bit width. <tt>%val</tt> is the value
5709 whose bits will be replaced. The second argument, <tt>%repl</tt> may be an
5710 integer of any bit width. The third and fourth arguments must be <tt>i32</tt>
5711 type since they specify only a bit index.</p>
5714 <p>The operation of the '<tt>llvm.part.set</tt>' intrinsic has two modes
5715 of operation: forwards and reverse. If <tt>%lo</tt> is greater than
5716 <tt>%hi</tt> then the intrinsic operates in reverse mode. Otherwise it
5717 operates in forward mode.</p>
5718 <p>For both modes, the <tt>%repl</tt> value is prepared for use by either
5719 truncating it down to the size of the replacement area or zero extending it
5720 up to that size.</p>
5721 <p>In forward mode, the bits between <tt>%lo</tt> and <tt>%hi</tt> (inclusive)
5722 are replaced with corresponding bits from <tt>%repl</tt>. That is the 0th bit
5723 in <tt>%repl</tt> replaces the <tt>%lo</tt>th bit in <tt>%val</tt> and etc. up
5724 to the <tt>%hi</tt>th bit.</p>
5725 <p>In reverse mode, a similar computation is made except that the bits are
5726 reversed. That is, the <tt>0</tt>th bit in <tt>%repl</tt> replaces the
5727 <tt>%hi</tt> bit in <tt>%val</tt> and etc. down to the <tt>%lo</tt>th bit.</p>
5730 llvm.part.set(0xFFFF, 0, 4, 7) -> 0xFF0F
5731 llvm.part.set(0xFFFF, 0, 7, 4) -> 0xFF0F
5732 llvm.part.set(0xFFFF, 1, 7, 4) -> 0xFF8F
5733 llvm.part.set(0xFFFF, F, 8, 3) -> 0xFFE7
5734 llvm.part.set(0xFFFF, 0, 3, 8) -> 0xFE07
5738 <!-- ======================================================================= -->
5739 <div class="doc_subsection">
5740 <a name="int_debugger">Debugger Intrinsics</a>
5743 <div class="doc_text">
5745 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
5746 are described in the <a
5747 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
5748 Debugging</a> document.
5753 <!-- ======================================================================= -->
5754 <div class="doc_subsection">
5755 <a name="int_eh">Exception Handling Intrinsics</a>
5758 <div class="doc_text">
5759 <p> The LLVM exception handling intrinsics (which all start with
5760 <tt>llvm.eh.</tt> prefix), are described in the <a
5761 href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
5762 Handling</a> document. </p>
5765 <!-- ======================================================================= -->
5766 <div class="doc_subsection">
5767 <a name="int_trampoline">Trampoline Intrinsic</a>
5770 <div class="doc_text">
5772 This intrinsic makes it possible to excise one parameter, marked with
5773 the <tt>nest</tt> attribute, from a function. The result is a callable
5774 function pointer lacking the nest parameter - the caller does not need
5775 to provide a value for it. Instead, the value to use is stored in
5776 advance in a "trampoline", a block of memory usually allocated
5777 on the stack, which also contains code to splice the nest value into the
5778 argument list. This is used to implement the GCC nested function address
5782 For example, if the function is
5783 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
5784 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as follows:</p>
5786 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
5787 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
5788 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
5789 %fp = bitcast i8* %p to i32 (i32, i32)*
5791 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
5792 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
5795 <!-- _______________________________________________________________________ -->
5796 <div class="doc_subsubsection">
5797 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
5799 <div class="doc_text">
5802 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
5806 This fills the memory pointed to by <tt>tramp</tt> with code
5807 and returns a function pointer suitable for executing it.
5811 The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
5812 pointers. The <tt>tramp</tt> argument must point to a sufficiently large
5813 and sufficiently aligned block of memory; this memory is written to by the
5814 intrinsic. Note that the size and the alignment are target-specific - LLVM
5815 currently provides no portable way of determining them, so a front-end that
5816 generates this intrinsic needs to have some target-specific knowledge.
5817 The <tt>func</tt> argument must hold a function bitcast to an <tt>i8*</tt>.
5821 The block of memory pointed to by <tt>tramp</tt> is filled with target
5822 dependent code, turning it into a function. A pointer to this function is
5823 returned, but needs to be bitcast to an
5824 <a href="#int_trampoline">appropriate function pointer type</a>
5825 before being called. The new function's signature is the same as that of
5826 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
5827 removed. At most one such <tt>nest</tt> argument is allowed, and it must be
5828 of pointer type. Calling the new function is equivalent to calling
5829 <tt>func</tt> with the same argument list, but with <tt>nval</tt> used for the
5830 missing <tt>nest</tt> argument. If, after calling
5831 <tt>llvm.init.trampoline</tt>, the memory pointed to by <tt>tramp</tt> is
5832 modified, then the effect of any later call to the returned function pointer is
5837 <!-- ======================================================================= -->
5838 <div class="doc_subsection">
5839 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
5842 <div class="doc_text">
5844 These intrinsic functions expand the "universal IR" of LLVM to represent
5845 hardware constructs for atomic operations and memory synchronization. This
5846 provides an interface to the hardware, not an interface to the programmer. It
5847 is aimed at a low enough level to allow any programming models or APIs
5848 (Application Programming Interfaces) which
5849 need atomic behaviors to map cleanly onto it. It is also modeled primarily on
5850 hardware behavior. Just as hardware provides a "universal IR" for source
5851 languages, it also provides a starting point for developing a "universal"
5852 atomic operation and synchronization IR.
5855 These do <em>not</em> form an API such as high-level threading libraries,
5856 software transaction memory systems, atomic primitives, and intrinsic
5857 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
5858 application libraries. The hardware interface provided by LLVM should allow
5859 a clean implementation of all of these APIs and parallel programming models.
5860 No one model or paradigm should be selected above others unless the hardware
5861 itself ubiquitously does so.
5866 <!-- _______________________________________________________________________ -->
5867 <div class="doc_subsubsection">
5868 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
5870 <div class="doc_text">
5873 declare void @llvm.memory.barrier( i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>,
5879 The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
5880 specific pairs of memory access types.
5884 The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
5885 The first four arguments enables a specific barrier as listed below. The fith
5886 argument specifies that the barrier applies to io or device or uncached memory.
5890 <li><tt>ll</tt>: load-load barrier</li>
5891 <li><tt>ls</tt>: load-store barrier</li>
5892 <li><tt>sl</tt>: store-load barrier</li>
5893 <li><tt>ss</tt>: store-store barrier</li>
5894 <li><tt>device</tt>: barrier applies to device and uncached memory also.</li>
5898 This intrinsic causes the system to enforce some ordering constraints upon
5899 the loads and stores of the program. This barrier does not indicate
5900 <em>when</em> any events will occur, it only enforces an <em>order</em> in
5901 which they occur. For any of the specified pairs of load and store operations
5902 (f.ex. load-load, or store-load), all of the first operations preceding the
5903 barrier will complete before any of the second operations succeeding the
5904 barrier begin. Specifically the semantics for each pairing is as follows:
5907 <li><tt>ll</tt>: All loads before the barrier must complete before any load
5908 after the barrier begins.</li>
5910 <li><tt>ls</tt>: All loads before the barrier must complete before any
5911 store after the barrier begins.</li>
5912 <li><tt>ss</tt>: All stores before the barrier must complete before any
5913 store after the barrier begins.</li>
5914 <li><tt>sl</tt>: All stores before the barrier must complete before any
5915 load after the barrier begins.</li>
5918 These semantics are applied with a logical "and" behavior when more than one
5919 is enabled in a single memory barrier intrinsic.
5922 Backends may implement stronger barriers than those requested when they do not
5923 support as fine grained a barrier as requested. Some architectures do not
5924 need all types of barriers and on such architectures, these become noops.
5931 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
5932 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
5933 <i>; guarantee the above finishes</i>
5934 store i32 8, %ptr <i>; before this begins</i>
5938 <!-- _______________________________________________________________________ -->
5939 <div class="doc_subsubsection">
5940 <a name="int_atomic_cmp_swap">'<tt>llvm.atomic.cmp.swap.*</tt>' Intrinsic</a>
5942 <div class="doc_text">
5945 This is an overloaded intrinsic. You can use <tt>llvm.atomic.cmp.swap</tt> on
5946 any integer bit width and for different address spaces. Not all targets
5947 support all bit widths however.</p>
5950 declare i8 @llvm.atomic.cmp.swap.i8.p0i8( i8* <ptr>, i8 <cmp>, i8 <val> )
5951 declare i16 @llvm.atomic.cmp.swap.i16.p0i16( i16* <ptr>, i16 <cmp>, i16 <val> )
5952 declare i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* <ptr>, i32 <cmp>, i32 <val> )
5953 declare i64 @llvm.atomic.cmp.swap.i64.p0i64( i64* <ptr>, i64 <cmp>, i64 <val> )
5958 This loads a value in memory and compares it to a given value. If they are
5959 equal, it stores a new value into the memory.
5963 The <tt>llvm.atomic.cmp.swap</tt> intrinsic takes three arguments. The result as
5964 well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
5965 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
5966 this integer type. While any bit width integer may be used, targets may only
5967 lower representations they support in hardware.
5972 This entire intrinsic must be executed atomically. It first loads the value
5973 in memory pointed to by <tt>ptr</tt> and compares it with the value
5974 <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the memory. The
5975 loaded value is yielded in all cases. This provides the equivalent of an
5976 atomic compare-and-swap operation within the SSA framework.
5984 %val1 = add i32 4, 4
5985 %result1 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 4, %val1 )
5986 <i>; yields {i32}:result1 = 4</i>
5987 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
5988 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
5990 %val2 = add i32 1, 1
5991 %result2 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 5, %val2 )
5992 <i>; yields {i32}:result2 = 8</i>
5993 %stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
5995 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
5999 <!-- _______________________________________________________________________ -->
6000 <div class="doc_subsubsection">
6001 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
6003 <div class="doc_text">
6007 This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
6008 integer bit width. Not all targets support all bit widths however.</p>
6010 declare i8 @llvm.atomic.swap.i8.p0i8( i8* <ptr>, i8 <val> )
6011 declare i16 @llvm.atomic.swap.i16.p0i16( i16* <ptr>, i16 <val> )
6012 declare i32 @llvm.atomic.swap.i32.p0i32( i32* <ptr>, i32 <val> )
6013 declare i64 @llvm.atomic.swap.i64.p0i64( i64* <ptr>, i64 <val> )
6018 This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
6019 the value from memory. It then stores the value in <tt>val</tt> in the memory
6025 The <tt>llvm.atomic.swap</tt> intrinsic takes two arguments. Both the
6026 <tt>val</tt> argument and the result must be integers of the same bit width.
6027 The first argument, <tt>ptr</tt>, must be a pointer to a value of this
6028 integer type. The targets may only lower integer representations they
6033 This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
6034 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
6035 equivalent of an atomic swap operation within the SSA framework.
6043 %val1 = add i32 4, 4
6044 %result1 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val1 )
6045 <i>; yields {i32}:result1 = 4</i>
6046 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
6047 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
6049 %val2 = add i32 1, 1
6050 %result2 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val2 )
6051 <i>; yields {i32}:result2 = 8</i>
6053 %stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
6054 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
6058 <!-- _______________________________________________________________________ -->
6059 <div class="doc_subsubsection">
6060 <a name="int_atomic_load_add">'<tt>llvm.atomic.load.add.*</tt>' Intrinsic</a>
6063 <div class="doc_text">
6066 This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.add</tt> on any
6067 integer bit width. Not all targets support all bit widths however.</p>
6069 declare i8 @llvm.atomic.load.add.i8..p0i8( i8* <ptr>, i8 <delta> )
6070 declare i16 @llvm.atomic.load.add.i16..p0i16( i16* <ptr>, i16 <delta> )
6071 declare i32 @llvm.atomic.load.add.i32..p0i32( i32* <ptr>, i32 <delta> )
6072 declare i64 @llvm.atomic.load.add.i64..p0i64( i64* <ptr>, i64 <delta> )
6077 This intrinsic adds <tt>delta</tt> to the value stored in memory at
6078 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
6083 The intrinsic takes two arguments, the first a pointer to an integer value
6084 and the second an integer value. The result is also an integer value. These
6085 integer types can have any bit width, but they must all have the same bit
6086 width. The targets may only lower integer representations they support.
6090 This intrinsic does a series of operations atomically. It first loads the
6091 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
6092 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
6099 %result1 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 4 )
6100 <i>; yields {i32}:result1 = 4</i>
6101 %result2 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 2 )
6102 <i>; yields {i32}:result2 = 8</i>
6103 %result3 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 5 )
6104 <i>; yields {i32}:result3 = 10</i>
6105 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
6109 <!-- _______________________________________________________________________ -->
6110 <div class="doc_subsubsection">
6111 <a name="int_atomic_load_sub">'<tt>llvm.atomic.load.sub.*</tt>' Intrinsic</a>
6114 <div class="doc_text">
6117 This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.sub</tt> on
6118 any integer bit width and for different address spaces. Not all targets
6119 support all bit widths however.</p>
6121 declare i8 @llvm.atomic.load.sub.i8.p0i32( i8* <ptr>, i8 <delta> )
6122 declare i16 @llvm.atomic.load.sub.i16.p0i32( i16* <ptr>, i16 <delta> )
6123 declare i32 @llvm.atomic.load.sub.i32.p0i32( i32* <ptr>, i32 <delta> )
6124 declare i64 @llvm.atomic.load.sub.i64.p0i32( i64* <ptr>, i64 <delta> )
6129 This intrinsic subtracts <tt>delta</tt> to the value stored in memory at
6130 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
6135 The intrinsic takes two arguments, the first a pointer to an integer value
6136 and the second an integer value. The result is also an integer value. These
6137 integer types can have any bit width, but they must all have the same bit
6138 width. The targets may only lower integer representations they support.
6142 This intrinsic does a series of operations atomically. It first loads the
6143 value stored at <tt>ptr</tt>. It then subtracts <tt>delta</tt>, stores the
6144 result to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
6151 %result1 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 4 )
6152 <i>; yields {i32}:result1 = 8</i>
6153 %result2 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 2 )
6154 <i>; yields {i32}:result2 = 4</i>
6155 %result3 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 5 )
6156 <i>; yields {i32}:result3 = 2</i>
6157 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = -3</i>
6161 <!-- _______________________________________________________________________ -->
6162 <div class="doc_subsubsection">
6163 <a name="int_atomic_load_and">'<tt>llvm.atomic.load.and.*</tt>' Intrinsic</a><br>
6164 <a name="int_atomic_load_nand">'<tt>llvm.atomic.load.nand.*</tt>' Intrinsic</a><br>
6165 <a name="int_atomic_load_or">'<tt>llvm.atomic.load.or.*</tt>' Intrinsic</a><br>
6166 <a name="int_atomic_load_xor">'<tt>llvm.atomic.load.xor.*</tt>' Intrinsic</a><br>
6169 <div class="doc_text">
6172 These are overloaded intrinsics. You can use <tt>llvm.atomic.load_and</tt>,
6173 <tt>llvm.atomic.load_nand</tt>, <tt>llvm.atomic.load_or</tt>, and
6174 <tt>llvm.atomic.load_xor</tt> on any integer bit width and for different
6175 address spaces. Not all targets support all bit widths however.</p>
6177 declare i8 @llvm.atomic.load.and.i8.p0i8( i8* <ptr>, i8 <delta> )
6178 declare i16 @llvm.atomic.load.and.i16.p0i16( i16* <ptr>, i16 <delta> )
6179 declare i32 @llvm.atomic.load.and.i32.p0i32( i32* <ptr>, i32 <delta> )
6180 declare i64 @llvm.atomic.load.and.i64.p0i64( i64* <ptr>, i64 <delta> )
6185 declare i8 @llvm.atomic.load.or.i8.p0i8( i8* <ptr>, i8 <delta> )
6186 declare i16 @llvm.atomic.load.or.i16.p0i16( i16* <ptr>, i16 <delta> )
6187 declare i32 @llvm.atomic.load.or.i32.p0i32( i32* <ptr>, i32 <delta> )
6188 declare i64 @llvm.atomic.load.or.i64.p0i64( i64* <ptr>, i64 <delta> )
6193 declare i8 @llvm.atomic.load.nand.i8.p0i32( i8* <ptr>, i8 <delta> )
6194 declare i16 @llvm.atomic.load.nand.i16.p0i32( i16* <ptr>, i16 <delta> )
6195 declare i32 @llvm.atomic.load.nand.i32.p0i32( i32* <ptr>, i32 <delta> )
6196 declare i64 @llvm.atomic.load.nand.i64.p0i32( i64* <ptr>, i64 <delta> )
6201 declare i8 @llvm.atomic.load.xor.i8.p0i32( i8* <ptr>, i8 <delta> )
6202 declare i16 @llvm.atomic.load.xor.i16.p0i32( i16* <ptr>, i16 <delta> )
6203 declare i32 @llvm.atomic.load.xor.i32.p0i32( i32* <ptr>, i32 <delta> )
6204 declare i64 @llvm.atomic.load.xor.i64.p0i32( i64* <ptr>, i64 <delta> )
6209 These intrinsics bitwise the operation (and, nand, or, xor) <tt>delta</tt> to
6210 the value stored in memory at <tt>ptr</tt>. It yields the original value
6216 These intrinsics take two arguments, the first a pointer to an integer value
6217 and the second an integer value. The result is also an integer value. These
6218 integer types can have any bit width, but they must all have the same bit
6219 width. The targets may only lower integer representations they support.
6223 These intrinsics does a series of operations atomically. They first load the
6224 value stored at <tt>ptr</tt>. They then do the bitwise operation
6225 <tt>delta</tt>, store the result to <tt>ptr</tt>. They yield the original
6226 value stored at <tt>ptr</tt>.
6232 store i32 0x0F0F, %ptr
6233 %result0 = call i32 @llvm.atomic.load.nand.i32.p0i32( i32* %ptr, i32 0xFF )
6234 <i>; yields {i32}:result0 = 0x0F0F</i>
6235 %result1 = call i32 @llvm.atomic.load.and.i32.p0i32( i32* %ptr, i32 0xFF )
6236 <i>; yields {i32}:result1 = 0xFFFFFFF0</i>
6237 %result2 = call i32 @llvm.atomic.load.or.i32.p0i32( i32* %ptr, i32 0F )
6238 <i>; yields {i32}:result2 = 0xF0</i>
6239 %result3 = call i32 @llvm.atomic.load.xor.i32.p0i32( i32* %ptr, i32 0F )
6240 <i>; yields {i32}:result3 = FF</i>
6241 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = F0</i>
6246 <!-- _______________________________________________________________________ -->
6247 <div class="doc_subsubsection">
6248 <a name="int_atomic_load_max">'<tt>llvm.atomic.load.max.*</tt>' Intrinsic</a><br>
6249 <a name="int_atomic_load_min">'<tt>llvm.atomic.load.min.*</tt>' Intrinsic</a><br>
6250 <a name="int_atomic_load_umax">'<tt>llvm.atomic.load.umax.*</tt>' Intrinsic</a><br>
6251 <a name="int_atomic_load_umin">'<tt>llvm.atomic.load.umin.*</tt>' Intrinsic</a><br>
6254 <div class="doc_text">
6257 These are overloaded intrinsics. You can use <tt>llvm.atomic.load_max</tt>,
6258 <tt>llvm.atomic.load_min</tt>, <tt>llvm.atomic.load_umax</tt>, and
6259 <tt>llvm.atomic.load_umin</tt> on any integer bit width and for different
6260 address spaces. Not all targets
6261 support all bit widths however.</p>
6263 declare i8 @llvm.atomic.load.max.i8.p0i8( i8* <ptr>, i8 <delta> )
6264 declare i16 @llvm.atomic.load.max.i16.p0i16( i16* <ptr>, i16 <delta> )
6265 declare i32 @llvm.atomic.load.max.i32.p0i32( i32* <ptr>, i32 <delta> )
6266 declare i64 @llvm.atomic.load.max.i64.p0i64( i64* <ptr>, i64 <delta> )
6271 declare i8 @llvm.atomic.load.min.i8.p0i8( i8* <ptr>, i8 <delta> )
6272 declare i16 @llvm.atomic.load.min.i16.p0i16( i16* <ptr>, i16 <delta> )
6273 declare i32 @llvm.atomic.load.min.i32..p0i32( i32* <ptr>, i32 <delta> )
6274 declare i64 @llvm.atomic.load.min.i64..p0i64( i64* <ptr>, i64 <delta> )
6279 declare i8 @llvm.atomic.load.umax.i8.p0i8( i8* <ptr>, i8 <delta> )
6280 declare i16 @llvm.atomic.load.umax.i16.p0i16( i16* <ptr>, i16 <delta> )
6281 declare i32 @llvm.atomic.load.umax.i32.p0i32( i32* <ptr>, i32 <delta> )
6282 declare i64 @llvm.atomic.load.umax.i64.p0i64( i64* <ptr>, i64 <delta> )
6287 declare i8 @llvm.atomic.load.umin.i8..p0i8( i8* <ptr>, i8 <delta> )
6288 declare i16 @llvm.atomic.load.umin.i16.p0i16( i16* <ptr>, i16 <delta> )
6289 declare i32 @llvm.atomic.load.umin.i32..p0i32( i32* <ptr>, i32 <delta> )
6290 declare i64 @llvm.atomic.load.umin.i64..p0i64( i64* <ptr>, i64 <delta> )
6295 These intrinsics takes the signed or unsigned minimum or maximum of
6296 <tt>delta</tt> and the value stored in memory at <tt>ptr</tt>. It yields the
6297 original value at <tt>ptr</tt>.
6302 These intrinsics take two arguments, the first a pointer to an integer value
6303 and the second an integer value. The result is also an integer value. These
6304 integer types can have any bit width, but they must all have the same bit
6305 width. The targets may only lower integer representations they support.
6309 These intrinsics does a series of operations atomically. They first load the
6310 value stored at <tt>ptr</tt>. They then do the signed or unsigned min or max
6311 <tt>delta</tt> and the value, store the result to <tt>ptr</tt>. They yield
6312 the original value stored at <tt>ptr</tt>.
6319 %result0 = call i32 @llvm.atomic.load.min.i32.p0i32( i32* %ptr, i32 -2 )
6320 <i>; yields {i32}:result0 = 7</i>
6321 %result1 = call i32 @llvm.atomic.load.max.i32.p0i32( i32* %ptr, i32 8 )
6322 <i>; yields {i32}:result1 = -2</i>
6323 %result2 = call i32 @llvm.atomic.load.umin.i32.p0i32( i32* %ptr, i32 10 )
6324 <i>; yields {i32}:result2 = 8</i>
6325 %result3 = call i32 @llvm.atomic.load.umax.i32.p0i32( i32* %ptr, i32 30 )
6326 <i>; yields {i32}:result3 = 8</i>
6327 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 30</i>
6331 <!-- ======================================================================= -->
6332 <div class="doc_subsection">
6333 <a name="int_general">General Intrinsics</a>
6336 <div class="doc_text">
6337 <p> This class of intrinsics is designed to be generic and has
6338 no specific purpose. </p>
6341 <!-- _______________________________________________________________________ -->
6342 <div class="doc_subsubsection">
6343 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
6346 <div class="doc_text">
6350 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
6356 The '<tt>llvm.var.annotation</tt>' intrinsic
6362 The first argument is a pointer to a value, the second is a pointer to a
6363 global string, the third is a pointer to a global string which is the source
6364 file name, and the last argument is the line number.
6370 This intrinsic allows annotation of local variables with arbitrary strings.
6371 This can be useful for special purpose optimizations that want to look for these
6372 annotations. These have no other defined use, they are ignored by code
6373 generation and optimization.
6377 <!-- _______________________________________________________________________ -->
6378 <div class="doc_subsubsection">
6379 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
6382 <div class="doc_text">
6385 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
6386 any integer bit width.
6389 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
6390 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
6391 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
6392 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
6393 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
6399 The '<tt>llvm.annotation</tt>' intrinsic.
6405 The first argument is an integer value (result of some expression),
6406 the second is a pointer to a global string, the third is a pointer to a global
6407 string which is the source file name, and the last argument is the line number.
6408 It returns the value of the first argument.
6414 This intrinsic allows annotations to be put on arbitrary expressions
6415 with arbitrary strings. This can be useful for special purpose optimizations
6416 that want to look for these annotations. These have no other defined use, they
6417 are ignored by code generation and optimization.
6421 <!-- _______________________________________________________________________ -->
6422 <div class="doc_subsubsection">
6423 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
6426 <div class="doc_text">
6430 declare void @llvm.trap()
6436 The '<tt>llvm.trap</tt>' intrinsic
6448 This intrinsics is lowered to the target dependent trap instruction. If the
6449 target does not have a trap instruction, this intrinsic will be lowered to the
6450 call of the abort() function.
6454 <!-- _______________________________________________________________________ -->
6455 <div class="doc_subsubsection">
6456 <a name="int_stackprotector">'<tt>llvm.stackprotector</tt>' Intrinsic</a>
6458 <div class="doc_text">
6461 declare void @llvm.stackprotector( i8* <guard>, i8** <slot> )
6466 The <tt>llvm.stackprotector</tt> intrinsic takes the <tt>guard</tt> and stores
6467 it onto the stack at <tt>slot</tt>. The stack slot is adjusted to ensure that
6468 it is placed on the stack before local variables.
6472 The <tt>llvm.stackprotector</tt> intrinsic requires two pointer arguments. The
6473 first argument is the value loaded from the stack guard
6474 <tt>@__stack_chk_guard</tt>. The second variable is an <tt>alloca</tt> that
6475 has enough space to hold the value of the guard.
6479 This intrinsic causes the prologue/epilogue inserter to force the position of
6480 the <tt>AllocaInst</tt> stack slot to be before local variables on the
6481 stack. This is to ensure that if a local variable on the stack is overwritten,
6482 it will destroy the value of the guard. When the function exits, the guard on
6483 the stack is checked against the original guard. If they're different, then
6484 the program aborts by calling the <tt>__stack_chk_fail()</tt> function.
6488 <!-- *********************************************************************** -->
6491 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
6492 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
6493 <a href="http://validator.w3.org/check/referer"><img
6494 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
6496 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
6497 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
6498 Last modified: $Date$