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>
248 <div class="doc_author">
249 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
250 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
253 <!-- *********************************************************************** -->
254 <div class="doc_section"> <a name="abstract">Abstract </a></div>
255 <!-- *********************************************************************** -->
257 <div class="doc_text">
258 <p>This document is a reference manual for the LLVM assembly language.
259 LLVM is a Static Single Assignment (SSA) based representation that provides
260 type safety, low-level operations, flexibility, and the capability of
261 representing 'all' high-level languages cleanly. It is the common code
262 representation used throughout all phases of the LLVM compilation
266 <!-- *********************************************************************** -->
267 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
268 <!-- *********************************************************************** -->
270 <div class="doc_text">
272 <p>The LLVM code representation is designed to be used in three
273 different forms: as an in-memory compiler IR, as an on-disk bitcode
274 representation (suitable for fast loading by a Just-In-Time compiler),
275 and as a human readable assembly language representation. This allows
276 LLVM to provide a powerful intermediate representation for efficient
277 compiler transformations and analysis, while providing a natural means
278 to debug and visualize the transformations. The three different forms
279 of LLVM are all equivalent. This document describes the human readable
280 representation and notation.</p>
282 <p>The LLVM representation aims to be light-weight and low-level
283 while being expressive, typed, and extensible at the same time. It
284 aims to be a "universal IR" of sorts, by being at a low enough level
285 that high-level ideas may be cleanly mapped to it (similar to how
286 microprocessors are "universal IR's", allowing many source languages to
287 be mapped to them). By providing type information, LLVM can be used as
288 the target of optimizations: for example, through pointer analysis, it
289 can be proven that a C automatic variable is never accessed outside of
290 the current function... allowing it to be promoted to a simple SSA
291 value instead of a memory location.</p>
295 <!-- _______________________________________________________________________ -->
296 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
298 <div class="doc_text">
300 <p>It is important to note that this document describes 'well formed'
301 LLVM assembly language. There is a difference between what the parser
302 accepts and what is considered 'well formed'. For example, the
303 following instruction is syntactically okay, but not well formed:</p>
305 <div class="doc_code">
307 %x = <a href="#i_add">add</a> i32 1, %x
311 <p>...because the definition of <tt>%x</tt> does not dominate all of
312 its uses. The LLVM infrastructure provides a verification pass that may
313 be used to verify that an LLVM module is well formed. This pass is
314 automatically run by the parser after parsing input assembly and by
315 the optimizer before it outputs bitcode. The violations pointed out
316 by the verifier pass indicate bugs in transformation passes or input to
320 <!-- Describe the typesetting conventions here. -->
322 <!-- *********************************************************************** -->
323 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
324 <!-- *********************************************************************** -->
326 <div class="doc_text">
328 <p>LLVM identifiers come in two basic types: global and local. Global
329 identifiers (functions, global variables) begin with the @ character. Local
330 identifiers (register names, types) begin with the % character. Additionally,
331 there are three different formats for identifiers, for different purposes:</p>
334 <li>Named values are represented as a string of characters with their prefix.
335 For example, %foo, @DivisionByZero, %a.really.long.identifier. The actual
336 regular expression used is '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
337 Identifiers which require other characters in their names can be surrounded
338 with quotes. Special characters may be escaped using "\xx" where xx is the
339 ASCII code for the character in hexadecimal. In this way, any character can
340 be used in a name value, even quotes themselves.
342 <li>Unnamed values are represented as an unsigned numeric value with their
343 prefix. For example, %12, @2, %44.</li>
345 <li>Constants, which are described in a <a href="#constants">section about
346 constants</a>, below.</li>
349 <p>LLVM requires that values start with a prefix for two reasons: Compilers
350 don't need to worry about name clashes with reserved words, and the set of
351 reserved words may be expanded in the future without penalty. Additionally,
352 unnamed identifiers allow a compiler to quickly come up with a temporary
353 variable without having to avoid symbol table conflicts.</p>
355 <p>Reserved words in LLVM are very similar to reserved words in other
356 languages. There are keywords for different opcodes
357 ('<tt><a href="#i_add">add</a></tt>',
358 '<tt><a href="#i_bitcast">bitcast</a></tt>',
359 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
360 href="#t_void">void</a></tt>', '<tt><a href="#t_primitive">i32</a></tt>', etc...),
361 and others. These reserved words cannot conflict with variable names, because
362 none of them start with a prefix character ('%' or '@').</p>
364 <p>Here is an example of LLVM code to multiply the integer variable
365 '<tt>%X</tt>' by 8:</p>
369 <div class="doc_code">
371 %result = <a href="#i_mul">mul</a> i32 %X, 8
375 <p>After strength reduction:</p>
377 <div class="doc_code">
379 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
383 <p>And the hard way:</p>
385 <div class="doc_code">
387 <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
388 <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
389 %result = <a href="#i_add">add</a> i32 %1, %1
393 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
394 important lexical features of LLVM:</p>
398 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
401 <li>Unnamed temporaries are created when the result of a computation is not
402 assigned to a named value.</li>
404 <li>Unnamed temporaries are numbered sequentially</li>
408 <p>...and it also shows a convention that we follow in this document. When
409 demonstrating instructions, we will follow an instruction with a comment that
410 defines the type and name of value produced. Comments are shown in italic
415 <!-- *********************************************************************** -->
416 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
417 <!-- *********************************************************************** -->
419 <!-- ======================================================================= -->
420 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
423 <div class="doc_text">
425 <p>LLVM programs are composed of "Module"s, each of which is a
426 translation unit of the input programs. Each module consists of
427 functions, global variables, and symbol table entries. Modules may be
428 combined together with the LLVM linker, which merges function (and
429 global variable) definitions, resolves forward declarations, and merges
430 symbol table entries. Here is an example of the "hello world" module:</p>
432 <div class="doc_code">
433 <pre><i>; Declare the string constant as a global constant...</i>
434 <a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a
435 href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
437 <i>; External declaration of the puts function</i>
438 <a href="#functionstructure">declare</a> i32 @puts(i8 *) <i>; i32(i8 *)* </i>
440 <i>; Definition of main function</i>
441 define i32 @main() { <i>; i32()* </i>
442 <i>; Convert [13x i8 ]* to i8 *...</i>
444 href="#i_getelementptr">getelementptr</a> [13 x i8 ]* @.LC0, i64 0, i64 0 <i>; i8 *</i>
446 <i>; Call puts function to write out the string to stdout...</i>
448 href="#i_call">call</a> i32 @puts(i8 * %cast210) <i>; i32</i>
450 href="#i_ret">ret</a> i32 0<br>}<br>
454 <p>This example is made up of a <a href="#globalvars">global variable</a>
455 named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
456 function, and a <a href="#functionstructure">function definition</a>
457 for "<tt>main</tt>".</p>
459 <p>In general, a module is made up of a list of global values,
460 where both functions and global variables are global values. Global values are
461 represented by a pointer to a memory location (in this case, a pointer to an
462 array of char, and a pointer to a function), and have one of the following <a
463 href="#linkage">linkage types</a>.</p>
467 <!-- ======================================================================= -->
468 <div class="doc_subsection">
469 <a name="linkage">Linkage Types</a>
472 <div class="doc_text">
475 All Global Variables and Functions have one of the following types of linkage:
480 <dt><tt><b><a name="linkage_internal">internal</a></b></tt>: </dt>
482 <dd>Global values with internal linkage are only directly accessible by
483 objects in the current module. In particular, linking code into a module with
484 an internal global value may cause the internal to be renamed as necessary to
485 avoid collisions. Because the symbol is internal to the module, all
486 references can be updated. This corresponds to the notion of the
487 '<tt>static</tt>' keyword in C.
490 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
492 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
493 the same name when linkage occurs. This is typically used to implement
494 inline functions, templates, or other code which must be generated in each
495 translation unit that uses it. Unreferenced <tt>linkonce</tt> globals are
496 allowed to be discarded.
499 <dt><tt><b><a name="linkage_common">common</a></b></tt>: </dt>
501 <dd>"<tt>common</tt>" linkage is exactly the same as <tt>linkonce</tt>
502 linkage, except that unreferenced <tt>common</tt> globals may not be
503 discarded. This is used for globals that may be emitted in multiple
504 translation units, but that are not guaranteed to be emitted into every
505 translation unit that uses them. One example of this is tentative
506 definitions in C, such as "<tt>int X;</tt>" at global scope.
509 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
511 <dd>"<tt>weak</tt>" linkage is the same as <tt>common</tt> linkage, except
512 that some targets may choose to emit different assembly sequences for them
513 for target-dependent reasons. This is used for globals that are declared
514 "weak" in C source code.
517 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
519 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
520 pointer to array type. When two global variables with appending linkage are
521 linked together, the two global arrays are appended together. This is the
522 LLVM, typesafe, equivalent of having the system linker append together
523 "sections" with identical names when .o files are linked.
526 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt>: </dt>
527 <dd>The semantics of this linkage follow the ELF object file model: the
528 symbol is weak until linked, if not linked, the symbol becomes null instead
529 of being an undefined reference.
532 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
534 <dd>If none of the above identifiers are used, the global is externally
535 visible, meaning that it participates in linkage and can be used to resolve
536 external symbol references.
541 The next two types of linkage are targeted for Microsoft Windows platform
542 only. They are designed to support importing (exporting) symbols from (to)
543 DLLs (Dynamic Link Libraries).
547 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt>: </dt>
549 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
550 or variable via a global pointer to a pointer that is set up by the DLL
551 exporting the symbol. On Microsoft Windows targets, the pointer name is
552 formed by combining <code>_imp__</code> and the function or variable name.
555 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt>: </dt>
557 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
558 pointer to a pointer in a DLL, so that it can be referenced with the
559 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
560 name is formed by combining <code>_imp__</code> and the function or variable
566 <p><a name="linkage_external"></a>For example, since the "<tt>.LC0</tt>"
567 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
568 variable and was linked with this one, one of the two would be renamed,
569 preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
570 external (i.e., lacking any linkage declarations), they are accessible
571 outside of the current module.</p>
572 <p>It is illegal for a function <i>declaration</i>
573 to have any linkage type other than "externally visible", <tt>dllimport</tt>,
574 or <tt>extern_weak</tt>.</p>
575 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt> and <tt>weak</tt>
579 <!-- ======================================================================= -->
580 <div class="doc_subsection">
581 <a name="callingconv">Calling Conventions</a>
584 <div class="doc_text">
586 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
587 and <a href="#i_invoke">invokes</a> can all have an optional calling convention
588 specified for the call. The calling convention of any pair of dynamic
589 caller/callee must match, or the behavior of the program is undefined. The
590 following calling conventions are supported by LLVM, and more may be added in
594 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
596 <dd>This calling convention (the default if no other calling convention is
597 specified) matches the target C calling conventions. This calling convention
598 supports varargs function calls and tolerates some mismatch in the declared
599 prototype and implemented declaration of the function (as does normal C).
602 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
604 <dd>This calling convention attempts to make calls as fast as possible
605 (e.g. by passing things in registers). This calling convention allows the
606 target to use whatever tricks it wants to produce fast code for the target,
607 without having to conform to an externally specified ABI (Application Binary
608 Interface). Implementations of this convention should allow arbitrary
609 <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> to be
610 supported. This calling convention does not support varargs and requires the
611 prototype of all callees to exactly match the prototype of the function
615 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
617 <dd>This calling convention attempts to make code in the caller as efficient
618 as possible under the assumption that the call is not commonly executed. As
619 such, these calls often preserve all registers so that the call does not break
620 any live ranges in the caller side. This calling convention does not support
621 varargs and requires the prototype of all callees to exactly match the
622 prototype of the function definition.
625 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
627 <dd>Any calling convention may be specified by number, allowing
628 target-specific calling conventions to be used. Target specific calling
629 conventions start at 64.
633 <p>More calling conventions can be added/defined on an as-needed basis, to
634 support pascal conventions or any other well-known target-independent
639 <!-- ======================================================================= -->
640 <div class="doc_subsection">
641 <a name="visibility">Visibility Styles</a>
644 <div class="doc_text">
647 All Global Variables and Functions have one of the following visibility styles:
651 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
653 <dd>On targets that use the ELF object file format, default visibility means
654 that the declaration is visible to other
655 modules and, in shared libraries, means that the declared entity may be
656 overridden. On Darwin, default visibility means that the declaration is
657 visible to other modules. Default visibility corresponds to "external
658 linkage" in the language.
661 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
663 <dd>Two declarations of an object with hidden visibility refer to the same
664 object if they are in the same shared object. Usually, hidden visibility
665 indicates that the symbol will not be placed into the dynamic symbol table,
666 so no other module (executable or shared library) can reference it
670 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
672 <dd>On ELF, protected visibility indicates that the symbol will be placed in
673 the dynamic symbol table, but that references within the defining module will
674 bind to the local symbol. That is, the symbol cannot be overridden by another
681 <!-- ======================================================================= -->
682 <div class="doc_subsection">
683 <a name="globalvars">Global Variables</a>
686 <div class="doc_text">
688 <p>Global variables define regions of memory allocated at compilation time
689 instead of run-time. Global variables may optionally be initialized, may have
690 an explicit section to be placed in, and may have an optional explicit alignment
691 specified. A variable may be defined as "thread_local", which means that it
692 will not be shared by threads (each thread will have a separated copy of the
693 variable). A variable may be defined as a global "constant," which indicates
694 that the contents of the variable will <b>never</b> be modified (enabling better
695 optimization, allowing the global data to be placed in the read-only section of
696 an executable, etc). Note that variables that need runtime initialization
697 cannot be marked "constant" as there is a store to the variable.</p>
700 LLVM explicitly allows <em>declarations</em> of global variables to be marked
701 constant, even if the final definition of the global is not. This capability
702 can be used to enable slightly better optimization of the program, but requires
703 the language definition to guarantee that optimizations based on the
704 'constantness' are valid for the translation units that do not include the
708 <p>As SSA values, global variables define pointer values that are in
709 scope (i.e. they dominate) all basic blocks in the program. Global
710 variables always define a pointer to their "content" type because they
711 describe a region of memory, and all memory objects in LLVM are
712 accessed through pointers.</p>
714 <p>A global variable may be declared to reside in a target-specifc numbered
715 address space. For targets that support them, address spaces may affect how
716 optimizations are performed and/or what target instructions are used to access
717 the variable. The default address space is zero. The address space qualifier
718 must precede any other attributes.</p>
720 <p>LLVM allows an explicit section to be specified for globals. If the target
721 supports it, it will emit globals to the section specified.</p>
723 <p>An explicit alignment may be specified for a global. If not present, or if
724 the alignment is set to zero, the alignment of the global is set by the target
725 to whatever it feels convenient. If an explicit alignment is specified, the
726 global is forced to have at least that much alignment. All alignments must be
729 <p>For example, the following defines a global in a numbered address space with
730 an initializer, section, and alignment:</p>
732 <div class="doc_code">
734 @G = constant float 1.0 addrspace(5), section "foo", align 4
741 <!-- ======================================================================= -->
742 <div class="doc_subsection">
743 <a name="functionstructure">Functions</a>
746 <div class="doc_text">
748 <p>LLVM function definitions consist of the "<tt>define</tt>" keyord,
749 an optional <a href="#linkage">linkage type</a>, an optional
750 <a href="#visibility">visibility style</a>, an optional
751 <a href="#callingconv">calling convention</a>, a return type, an optional
752 <a href="#paramattrs">parameter attribute</a> for the return type, a function
753 name, a (possibly empty) argument list (each with optional
754 <a href="#paramattrs">parameter attributes</a>), optional
755 <a href="#fnattrs">function attributes</a>, an optional section,
756 an optional alignment, an optional <a href="#gc">garbage collector name</a>,
757 an opening curly brace, a list of basic blocks, and a closing curly brace.
759 LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
760 optional <a href="#linkage">linkage type</a>, an optional
761 <a href="#visibility">visibility style</a>, an optional
762 <a href="#callingconv">calling convention</a>, a return type, an optional
763 <a href="#paramattrs">parameter attribute</a> for the return type, a function
764 name, a possibly empty list of arguments, an optional alignment, and an optional
765 <a href="#gc">garbage collector name</a>.</p>
767 <p>A function definition contains a list of basic blocks, forming the CFG
768 (Control Flow Graph) for
769 the function. Each basic block may optionally start with a label (giving the
770 basic block a symbol table entry), contains a list of instructions, and ends
771 with a <a href="#terminators">terminator</a> instruction (such as a branch or
772 function return).</p>
774 <p>The first basic block in a function is special in two ways: it is immediately
775 executed on entrance to the function, and it is not allowed to have predecessor
776 basic blocks (i.e. there can not be any branches to the entry block of a
777 function). Because the block can have no predecessors, it also cannot have any
778 <a href="#i_phi">PHI nodes</a>.</p>
780 <p>LLVM allows an explicit section to be specified for functions. If the target
781 supports it, it will emit functions to the section specified.</p>
783 <p>An explicit alignment may be specified for a function. If not present, or if
784 the alignment is set to zero, the alignment of the function is set by the target
785 to whatever it feels convenient. If an explicit alignment is specified, the
786 function is forced to have at least that much alignment. All alignments must be
791 <div class="doc_code">
793 define [<a href="#linkage">linkage</a>] [<a href="#visibility">visibility</a>]
794 [<a href="#callingconv">cconv</a>] [<a href="#paramattrs">ret attrs</a>]
795 <ResultType> @<FunctionName> ([argument list])
796 [<a href="#fnattrs">fn Attrs</a>] [section "name"] [align N]
797 [<a href="#gc">gc</a>] { ... }
804 <!-- ======================================================================= -->
805 <div class="doc_subsection">
806 <a name="aliasstructure">Aliases</a>
808 <div class="doc_text">
809 <p>Aliases act as "second name" for the aliasee value (which can be either
810 function, global variable, another alias or bitcast of global value). Aliases
811 may have an optional <a href="#linkage">linkage type</a>, and an
812 optional <a href="#visibility">visibility style</a>.</p>
816 <div class="doc_code">
818 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
826 <!-- ======================================================================= -->
827 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
828 <div class="doc_text">
829 <p>The return type and each parameter of a function type may have a set of
830 <i>parameter attributes</i> associated with them. Parameter attributes are
831 used to communicate additional information about the result or parameters of
832 a function. Parameter attributes are considered to be part of the function,
833 not of the function type, so functions with different parameter attributes
834 can have the same function type.</p>
836 <p>Parameter attributes are simple keywords that follow the type specified. If
837 multiple parameter attributes are needed, they are space separated. For
840 <div class="doc_code">
842 declare i32 @printf(i8* noalias , ...)
843 declare i32 @atoi(i8 zeroext)
844 declare signext i8 @returns_signed_char()
848 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
849 <tt>readonly</tt>) come immediately after the argument list.</p>
851 <p>Currently, only the following parameter attributes are defined:</p>
853 <dt><tt>zeroext</tt></dt>
854 <dd>This indicates to the code generator that the parameter or return value
855 should be zero-extended to a 32-bit value by the caller (for a parameter)
856 or the callee (for a return value).</dd>
858 <dt><tt>signext</tt></dt>
859 <dd>This indicates to the code generator that the parameter or return value
860 should be sign-extended to a 32-bit value by the caller (for a parameter)
861 or the callee (for a return value).</dd>
863 <dt><tt>inreg</tt></dt>
864 <dd>This indicates that this parameter or return value should be treated
865 in a special target-dependent fashion during while emitting code for a
866 function call or return (usually, by putting it in a register as opposed
867 to memory, though some targets use it to distinguish between two different
868 kinds of registers). Use of this attribute is target-specific.</dd>
870 <dt><tt><a name="byval">byval</a></tt></dt>
871 <dd>This indicates that the pointer parameter should really be passed by
872 value to the function. The attribute implies that a hidden copy of the
873 pointee is made between the caller and the callee, so the callee is unable
874 to modify the value in the callee. This attribute is only valid on LLVM
875 pointer arguments. It is generally used to pass structs and arrays by
876 value, but is also valid on pointers to scalars. The copy is considered to
877 belong to the caller not the callee (for example,
878 <tt><a href="#readonly">readonly</a></tt> functions should not write to
879 <tt>byval</tt> parameters). This is not a valid attribute for return
882 <dt><tt>sret</tt></dt>
883 <dd>This indicates that the pointer parameter specifies the address of a
884 structure that is the return value of the function in the source program.
885 This pointer must be guaranteed by the caller to be valid: loads and stores
886 to the structure may be assumed by the callee to not to trap. This may only
887 be applied to the first parameter. This is not a valid attribute for
890 <dt><tt>noalias</tt></dt>
891 <dd>This indicates that the parameter does not alias any global or any other
892 parameter. The caller is responsible for ensuring that this is the case,
893 usually by placing the value in a stack allocation. This is not a valid
894 attribute for return values.</dd>
896 <dt><tt>nest</tt></dt>
897 <dd>This indicates that the pointer parameter can be excised using the
898 <a href="#int_trampoline">trampoline intrinsics</a>. This is not a valid
899 attribute for return values.</dd>
904 <!-- ======================================================================= -->
905 <div class="doc_subsection">
906 <a name="gc">Garbage Collector Names</a>
909 <div class="doc_text">
910 <p>Each function may specify a garbage collector name, which is simply a
913 <div class="doc_code"><pre
914 >define void @f() gc "name" { ...</pre></div>
916 <p>The compiler declares the supported values of <i>name</i>. Specifying a
917 collector which will cause the compiler to alter its output in order to support
918 the named garbage collection algorithm.</p>
921 <!-- ======================================================================= -->
922 <div class="doc_subsection">
923 <a name="fnattrs">Function Attributes</a>
926 <div class="doc_text">
928 <p>Function attributes are set to communicate additional information about
929 a function. Function attributes are considered to be part of the function,
930 not of the function type, so functions with different parameter attributes
931 can have the same function type.</p>
933 <p>Function attributes are simple keywords that follow the type specified. If
934 multiple attributes are needed, they are space separated. For
937 <div class="doc_code">
939 define void @f() noinline { ... }
940 define void @f() alwaysinline { ... }
941 define void @f() alwaysinline optsize { ... }
942 define void @f() optsize
947 <dt><tt>alwaysinline</tt></dt>
948 <dd>This attribute indicates that the inliner should attempt to inline this
949 function into callers whenever possible, ignoring any active inlining size
950 threshold for this caller.</dd>
952 <dt><tt>noinline</tt></dt>
953 <dd>This attribute indicates that the inliner should never inline this function
954 in any situation. This attribute may not be used together with the
955 <tt>alwaysinline</tt> attribute.</dd>
957 <dt><tt>optsize</tt></dt>
958 <dd>This attribute suggests that optimization passes and code generator passes
959 make choices that keep the code size of this function low, and otherwise do
960 optimizations specifically to reduce code size.</dd>
962 <dt><tt>noreturn</tt></dt>
963 <dd>This function attribute indicates that the function never returns normally.
964 This produces undefined behavior at runtime if the function ever does
965 dynamically return.</dd>
967 <dt><tt>nounwind</tt></dt>
968 <dd>This function attribute indicates that the function never returns with an
969 unwind or exceptional control flow. If the function does unwind, its runtime
970 behavior is undefined.</dd>
972 <dt><tt>readnone</tt></dt>
973 <dd>This attribute indicates that the function computes its result (or the
974 exception it throws) based strictly on its arguments, without dereferencing any
975 pointer arguments or otherwise accessing any mutable state (e.g. memory, control
976 registers, etc) visible to caller functions. It does not write through any
977 pointer arguments (including <tt><a href="#byval">byval</a></tt> arguments) and
978 never changes any state visible to callers.</dd>
980 <dt><tt><a name="readonly">readonly</a></tt></dt>
981 <dd>This attribute indicates that the function does not write through any
982 pointer arguments (including <tt><a href="#byval">byval</a></tt> arguments)
983 or otherwise modify any state (e.g. memory, control registers, etc) visible to
984 caller functions. It may dereference pointer arguments and read state that may
985 be set in the caller. A readonly function always returns the same value (or
986 throws the same exception) when called with the same set of arguments and global
992 <!-- ======================================================================= -->
993 <div class="doc_subsection">
994 <a name="moduleasm">Module-Level Inline Assembly</a>
997 <div class="doc_text">
999 Modules may contain "module-level inline asm" blocks, which corresponds to the
1000 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
1001 LLVM and treated as a single unit, but may be separated in the .ll file if
1002 desired. The syntax is very simple:
1005 <div class="doc_code">
1007 module asm "inline asm code goes here"
1008 module asm "more can go here"
1012 <p>The strings can contain any character by escaping non-printable characters.
1013 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
1018 The inline asm code is simply printed to the machine code .s file when
1019 assembly code is generated.
1023 <!-- ======================================================================= -->
1024 <div class="doc_subsection">
1025 <a name="datalayout">Data Layout</a>
1028 <div class="doc_text">
1029 <p>A module may specify a target specific data layout string that specifies how
1030 data is to be laid out in memory. The syntax for the data layout is simply:</p>
1031 <pre> target datalayout = "<i>layout specification</i>"</pre>
1032 <p>The <i>layout specification</i> consists of a list of specifications
1033 separated by the minus sign character ('-'). Each specification starts with a
1034 letter and may include other information after the letter to define some
1035 aspect of the data layout. The specifications accepted are as follows: </p>
1038 <dd>Specifies that the target lays out data in big-endian form. That is, the
1039 bits with the most significance have the lowest address location.</dd>
1041 <dd>Specifies that the target lays out data in little-endian form. That is,
1042 the bits with the least significance have the lowest address location.</dd>
1043 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1044 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1045 <i>preferred</i> alignments. All sizes are in bits. Specifying the <i>pref</i>
1046 alignment is optional. If omitted, the preceding <tt>:</tt> should be omitted
1048 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1049 <dd>This specifies the alignment for an integer type of a given bit
1050 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1051 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1052 <dd>This specifies the alignment for a vector type of a given bit
1054 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1055 <dd>This specifies the alignment for a floating point type of a given bit
1056 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
1058 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1059 <dd>This specifies the alignment for an aggregate type of a given bit
1062 <p>When constructing the data layout for a given target, LLVM starts with a
1063 default set of specifications which are then (possibly) overriden by the
1064 specifications in the <tt>datalayout</tt> keyword. The default specifications
1065 are given in this list:</p>
1067 <li><tt>E</tt> - big endian</li>
1068 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
1069 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1070 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1071 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1072 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1073 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1074 alignment of 64-bits</li>
1075 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1076 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1077 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1078 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1079 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1081 <p>When LLVM is determining the alignment for a given type, it uses the
1082 following rules:</p>
1084 <li>If the type sought is an exact match for one of the specifications, that
1085 specification is used.</li>
1086 <li>If no match is found, and the type sought is an integer type, then the
1087 smallest integer type that is larger than the bitwidth of the sought type is
1088 used. If none of the specifications are larger than the bitwidth then the the
1089 largest integer type is used. For example, given the default specifications
1090 above, the i7 type will use the alignment of i8 (next largest) while both
1091 i65 and i256 will use the alignment of i64 (largest specified).</li>
1092 <li>If no match is found, and the type sought is a vector type, then the
1093 largest vector type that is smaller than the sought vector type will be used
1094 as a fall back. This happens because <128 x double> can be implemented
1095 in terms of 64 <2 x double>, for example.</li>
1099 <!-- *********************************************************************** -->
1100 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
1101 <!-- *********************************************************************** -->
1103 <div class="doc_text">
1105 <p>The LLVM type system is one of the most important features of the
1106 intermediate representation. Being typed enables a number of
1107 optimizations to be performed on the intermediate representation directly,
1108 without having to do
1109 extra analyses on the side before the transformation. A strong type
1110 system makes it easier to read the generated code and enables novel
1111 analyses and transformations that are not feasible to perform on normal
1112 three address code representations.</p>
1116 <!-- ======================================================================= -->
1117 <div class="doc_subsection"> <a name="t_classifications">Type
1118 Classifications</a> </div>
1119 <div class="doc_text">
1120 <p>The types fall into a few useful
1121 classifications:</p>
1123 <table border="1" cellspacing="0" cellpadding="4">
1125 <tr><th>Classification</th><th>Types</th></tr>
1127 <td><a href="#t_integer">integer</a></td>
1128 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1131 <td><a href="#t_floating">floating point</a></td>
1132 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1135 <td><a name="t_firstclass">first class</a></td>
1136 <td><a href="#t_integer">integer</a>,
1137 <a href="#t_floating">floating point</a>,
1138 <a href="#t_pointer">pointer</a>,
1139 <a href="#t_vector">vector</a>,
1140 <a href="#t_struct">structure</a>,
1141 <a href="#t_array">array</a>,
1142 <a href="#t_label">label</a>.
1146 <td><a href="#t_primitive">primitive</a></td>
1147 <td><a href="#t_label">label</a>,
1148 <a href="#t_void">void</a>,
1149 <a href="#t_floating">floating point</a>.</td>
1152 <td><a href="#t_derived">derived</a></td>
1153 <td><a href="#t_integer">integer</a>,
1154 <a href="#t_array">array</a>,
1155 <a href="#t_function">function</a>,
1156 <a href="#t_pointer">pointer</a>,
1157 <a href="#t_struct">structure</a>,
1158 <a href="#t_pstruct">packed structure</a>,
1159 <a href="#t_vector">vector</a>,
1160 <a href="#t_opaque">opaque</a>.
1166 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
1167 most important. Values of these types are the only ones which can be
1168 produced by instructions, passed as arguments, or used as operands to
1172 <!-- ======================================================================= -->
1173 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
1175 <div class="doc_text">
1176 <p>The primitive types are the fundamental building blocks of the LLVM
1181 <!-- _______________________________________________________________________ -->
1182 <div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1184 <div class="doc_text">
1187 <tr><th>Type</th><th>Description</th></tr>
1188 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1189 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1190 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1191 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1192 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1197 <!-- _______________________________________________________________________ -->
1198 <div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1200 <div class="doc_text">
1202 <p>The void type does not represent any value and has no size.</p>
1211 <!-- _______________________________________________________________________ -->
1212 <div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1214 <div class="doc_text">
1216 <p>The label type represents code labels.</p>
1226 <!-- ======================================================================= -->
1227 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1229 <div class="doc_text">
1231 <p>The real power in LLVM comes from the derived types in the system.
1232 This is what allows a programmer to represent arrays, functions,
1233 pointers, and other useful types. Note that these derived types may be
1234 recursive: For example, it is possible to have a two dimensional array.</p>
1238 <!-- _______________________________________________________________________ -->
1239 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1241 <div class="doc_text">
1244 <p>The integer type is a very simple derived type that simply specifies an
1245 arbitrary bit width for the integer type desired. Any bit width from 1 bit to
1246 2^23-1 (about 8 million) can be specified.</p>
1254 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1258 <table class="layout">
1261 <td><tt>i1</tt></td>
1262 <td>a single-bit integer.</td>
1264 <td><tt>i32</tt></td>
1265 <td>a 32-bit integer.</td>
1267 <td><tt>i1942652</tt></td>
1268 <td>a really big integer of over 1 million bits.</td>
1274 <!-- _______________________________________________________________________ -->
1275 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1277 <div class="doc_text">
1281 <p>The array type is a very simple derived type that arranges elements
1282 sequentially in memory. The array type requires a size (number of
1283 elements) and an underlying data type.</p>
1288 [<# elements> x <elementtype>]
1291 <p>The number of elements is a constant integer value; elementtype may
1292 be any type with a size.</p>
1295 <table class="layout">
1297 <td class="left"><tt>[40 x i32]</tt></td>
1298 <td class="left">Array of 40 32-bit integer values.</td>
1301 <td class="left"><tt>[41 x i32]</tt></td>
1302 <td class="left">Array of 41 32-bit integer values.</td>
1305 <td class="left"><tt>[4 x i8]</tt></td>
1306 <td class="left">Array of 4 8-bit integer values.</td>
1309 <p>Here are some examples of multidimensional arrays:</p>
1310 <table class="layout">
1312 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1313 <td class="left">3x4 array of 32-bit integer values.</td>
1316 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1317 <td class="left">12x10 array of single precision floating point values.</td>
1320 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1321 <td class="left">2x3x4 array of 16-bit integer values.</td>
1325 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1326 length array. Normally, accesses past the end of an array are undefined in
1327 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
1328 As a special case, however, zero length arrays are recognized to be variable
1329 length. This allows implementation of 'pascal style arrays' with the LLVM
1330 type "{ i32, [0 x float]}", for example.</p>
1334 <!-- _______________________________________________________________________ -->
1335 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1336 <div class="doc_text">
1340 <p>The function type can be thought of as a function signature. It
1341 consists of a return type and a list of formal parameter types. The
1342 return type of a function type is a scalar type, a void type, or a struct type.
1343 If the return type is a struct type then all struct elements must be of first
1344 class types, and the struct must have at least one element.</p>
1349 <returntype list> (<parameter list>)
1352 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1353 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1354 which indicates that the function takes a variable number of arguments.
1355 Variable argument functions can access their arguments with the <a
1356 href="#int_varargs">variable argument handling intrinsic</a> functions.
1357 '<tt><returntype list></tt>' is a comma-separated list of
1358 <a href="#t_firstclass">first class</a> type specifiers.</p>
1361 <table class="layout">
1363 <td class="left"><tt>i32 (i32)</tt></td>
1364 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1366 </tr><tr class="layout">
1367 <td class="left"><tt>float (i16 signext, i32 *) *
1369 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1370 an <tt>i16</tt> that should be sign extended and a
1371 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1374 </tr><tr class="layout">
1375 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1376 <td class="left">A vararg function that takes at least one
1377 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1378 which returns an integer. This is the signature for <tt>printf</tt> in
1381 </tr><tr class="layout">
1382 <td class="left"><tt>{i32, i32} (i32)</tt></td>
1383 <td class="left">A function taking an <tt>i32></tt>, returning two
1384 <tt> i32 </tt> values as an aggregate of type <tt>{ i32, i32 }</tt>
1390 <!-- _______________________________________________________________________ -->
1391 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1392 <div class="doc_text">
1394 <p>The structure type is used to represent a collection of data members
1395 together in memory. The packing of the field types is defined to match
1396 the ABI of the underlying processor. The elements of a structure may
1397 be any type that has a size.</p>
1398 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1399 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1400 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1403 <pre> { <type list> }<br></pre>
1405 <table class="layout">
1407 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1408 <td class="left">A triple of three <tt>i32</tt> values</td>
1409 </tr><tr class="layout">
1410 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1411 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1412 second element is a <a href="#t_pointer">pointer</a> to a
1413 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1414 an <tt>i32</tt>.</td>
1419 <!-- _______________________________________________________________________ -->
1420 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1422 <div class="doc_text">
1424 <p>The packed structure type is used to represent a collection of data members
1425 together in memory. There is no padding between fields. Further, the alignment
1426 of a packed structure is 1 byte. The elements of a packed structure may
1427 be any type that has a size.</p>
1428 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1429 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1430 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1433 <pre> < { <type list> } > <br></pre>
1435 <table class="layout">
1437 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1438 <td class="left">A triple of three <tt>i32</tt> values</td>
1439 </tr><tr class="layout">
1441 <tt>< { float, i32 (i32)* } ></tt></td>
1442 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1443 second element is a <a href="#t_pointer">pointer</a> to a
1444 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1445 an <tt>i32</tt>.</td>
1450 <!-- _______________________________________________________________________ -->
1451 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1452 <div class="doc_text">
1454 <p>As in many languages, the pointer type represents a pointer or
1455 reference to another object, which must live in memory. Pointer types may have
1456 an optional address space attribute defining the target-specific numbered
1457 address space where the pointed-to object resides. The default address space is
1460 <pre> <type> *<br></pre>
1462 <table class="layout">
1464 <td class="left"><tt>[4x i32]*</tt></td>
1465 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1466 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1469 <td class="left"><tt>i32 (i32 *) *</tt></td>
1470 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
1471 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1475 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1476 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1477 that resides in address space #5.</td>
1482 <!-- _______________________________________________________________________ -->
1483 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1484 <div class="doc_text">
1488 <p>A vector type is a simple derived type that represents a vector
1489 of elements. Vector types are used when multiple primitive data
1490 are operated in parallel using a single instruction (SIMD).
1491 A vector type requires a size (number of
1492 elements) and an underlying primitive data type. Vectors must have a power
1493 of two length (1, 2, 4, 8, 16 ...). Vector types are
1494 considered <a href="#t_firstclass">first class</a>.</p>
1499 < <# elements> x <elementtype> >
1502 <p>The number of elements is a constant integer value; elementtype may
1503 be any integer or floating point type.</p>
1507 <table class="layout">
1509 <td class="left"><tt><4 x i32></tt></td>
1510 <td class="left">Vector of 4 32-bit integer values.</td>
1513 <td class="left"><tt><8 x float></tt></td>
1514 <td class="left">Vector of 8 32-bit floating-point values.</td>
1517 <td class="left"><tt><2 x i64></tt></td>
1518 <td class="left">Vector of 2 64-bit integer values.</td>
1523 <!-- _______________________________________________________________________ -->
1524 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1525 <div class="doc_text">
1529 <p>Opaque types are used to represent unknown types in the system. This
1530 corresponds (for example) to the C notion of a forward declared structure type.
1531 In LLVM, opaque types can eventually be resolved to any type (not just a
1532 structure type).</p>
1542 <table class="layout">
1544 <td class="left"><tt>opaque</tt></td>
1545 <td class="left">An opaque type.</td>
1551 <!-- *********************************************************************** -->
1552 <div class="doc_section"> <a name="constants">Constants</a> </div>
1553 <!-- *********************************************************************** -->
1555 <div class="doc_text">
1557 <p>LLVM has several different basic types of constants. This section describes
1558 them all and their syntax.</p>
1562 <!-- ======================================================================= -->
1563 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1565 <div class="doc_text">
1568 <dt><b>Boolean constants</b></dt>
1570 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1571 constants of the <tt><a href="#t_primitive">i1</a></tt> type.
1574 <dt><b>Integer constants</b></dt>
1576 <dd>Standard integers (such as '4') are constants of the <a
1577 href="#t_integer">integer</a> type. Negative numbers may be used with
1581 <dt><b>Floating point constants</b></dt>
1583 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1584 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1585 notation (see below). The assembler requires the exact decimal value of
1586 a floating-point constant. For example, the assembler accepts 1.25 but
1587 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
1588 constants must have a <a href="#t_floating">floating point</a> type. </dd>
1590 <dt><b>Null pointer constants</b></dt>
1592 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1593 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1597 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1598 of floating point constants. For example, the form '<tt>double
1599 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1600 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1601 (and the only time that they are generated by the disassembler) is when a
1602 floating point constant must be emitted but it cannot be represented as a
1603 decimal floating point number. For example, NaN's, infinities, and other
1604 special values are represented in their IEEE hexadecimal format so that
1605 assembly and disassembly do not cause any bits to change in the constants.</p>
1609 <!-- ======================================================================= -->
1610 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1613 <div class="doc_text">
1614 <p>Aggregate constants arise from aggregation of simple constants
1615 and smaller aggregate constants.</p>
1618 <dt><b>Structure constants</b></dt>
1620 <dd>Structure constants are represented with notation similar to structure
1621 type definitions (a comma separated list of elements, surrounded by braces
1622 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
1623 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>". Structure constants
1624 must have <a href="#t_struct">structure type</a>, and the number and
1625 types of elements must match those specified by the type.
1628 <dt><b>Array constants</b></dt>
1630 <dd>Array constants are represented with notation similar to array type
1631 definitions (a comma separated list of elements, surrounded by square brackets
1632 (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74 ]</tt>". Array
1633 constants must have <a href="#t_array">array type</a>, and the number and
1634 types of elements must match those specified by the type.
1637 <dt><b>Vector constants</b></dt>
1639 <dd>Vector constants are represented with notation similar to vector type
1640 definitions (a comma separated list of elements, surrounded by
1641 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32 42,
1642 i32 11, i32 74, i32 100 ></tt>". Vector constants must have <a
1643 href="#t_vector">vector type</a>, and the number and types of elements must
1644 match those specified by the type.
1647 <dt><b>Zero initialization</b></dt>
1649 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1650 value to zero of <em>any</em> type, including scalar and aggregate types.
1651 This is often used to avoid having to print large zero initializers (e.g. for
1652 large arrays) and is always exactly equivalent to using explicit zero
1659 <!-- ======================================================================= -->
1660 <div class="doc_subsection">
1661 <a name="globalconstants">Global Variable and Function Addresses</a>
1664 <div class="doc_text">
1666 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1667 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1668 constants. These constants are explicitly referenced when the <a
1669 href="#identifiers">identifier for the global</a> is used and always have <a
1670 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1673 <div class="doc_code">
1677 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1683 <!-- ======================================================================= -->
1684 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1685 <div class="doc_text">
1686 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1687 no specific value. Undefined values may be of any type and be used anywhere
1688 a constant is permitted.</p>
1690 <p>Undefined values indicate to the compiler that the program is well defined
1691 no matter what value is used, giving the compiler more freedom to optimize.
1695 <!-- ======================================================================= -->
1696 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1699 <div class="doc_text">
1701 <p>Constant expressions are used to allow expressions involving other constants
1702 to be used as constants. Constant expressions may be of any <a
1703 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1704 that does not have side effects (e.g. load and call are not supported). The
1705 following is the syntax for constant expressions:</p>
1708 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1709 <dd>Truncate a constant to another type. The bit size of CST must be larger
1710 than the bit size of TYPE. Both types must be integers.</dd>
1712 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1713 <dd>Zero extend a constant to another type. The bit size of CST must be
1714 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1716 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1717 <dd>Sign extend a constant to another type. The bit size of CST must be
1718 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1720 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1721 <dd>Truncate a floating point constant to another floating point type. The
1722 size of CST must be larger than the size of TYPE. Both types must be
1723 floating point.</dd>
1725 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1726 <dd>Floating point extend a constant to another type. The size of CST must be
1727 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1729 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
1730 <dd>Convert a floating point constant to the corresponding unsigned integer
1731 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1732 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1733 of the same number of elements. If the value won't fit in the integer type,
1734 the results are undefined.</dd>
1736 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
1737 <dd>Convert a floating point constant to the corresponding signed integer
1738 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1739 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1740 of the same number of elements. If the value won't fit in the integer type,
1741 the results are undefined.</dd>
1743 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
1744 <dd>Convert an unsigned integer constant to the corresponding floating point
1745 constant. TYPE must be a scalar or vector floating point type. CST must be of
1746 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1747 of the same number of elements. If the value won't fit in the floating point
1748 type, the results are undefined.</dd>
1750 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
1751 <dd>Convert a signed integer constant to the corresponding floating point
1752 constant. TYPE must be a scalar or vector floating point type. CST must be of
1753 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1754 of the same number of elements. If the value won't fit in the floating point
1755 type, the results are undefined.</dd>
1757 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
1758 <dd>Convert a pointer typed constant to the corresponding integer constant
1759 TYPE must be an integer type. CST must be of pointer type. The CST value is
1760 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1762 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
1763 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1764 pointer type. CST must be of integer type. The CST value is zero extended,
1765 truncated, or unchanged to make it fit in a pointer size. This one is
1766 <i>really</i> dangerous!</dd>
1768 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
1769 <dd>Convert a constant, CST, to another TYPE. The size of CST and TYPE must be
1770 identical (same number of bits). The conversion is done as if the CST value
1771 was stored to memory and read back as TYPE. In other words, no bits change
1772 with this operator, just the type. This can be used for conversion of
1773 vector types to any other type, as long as they have the same bit width. For
1774 pointers it is only valid to cast to another pointer type. It is not valid
1775 to bitcast to or from an aggregate type.
1778 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1780 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1781 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1782 instruction, the index list may have zero or more indexes, which are required
1783 to make sense for the type of "CSTPTR".</dd>
1785 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1787 <dd>Perform the <a href="#i_select">select operation</a> on
1790 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
1791 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
1793 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
1794 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
1796 <dt><b><tt>vicmp COND ( VAL1, VAL2 )</tt></b></dt>
1797 <dd>Performs the <a href="#i_vicmp">vicmp operation</a> on constants.</dd>
1799 <dt><b><tt>vfcmp COND ( VAL1, VAL2 )</tt></b></dt>
1800 <dd>Performs the <a href="#i_vfcmp">vfcmp operation</a> on constants.</dd>
1802 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1804 <dd>Perform the <a href="#i_extractelement">extractelement
1805 operation</a> on constants.</dd>
1807 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1809 <dd>Perform the <a href="#i_insertelement">insertelement
1810 operation</a> on constants.</dd>
1813 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
1815 <dd>Perform the <a href="#i_shufflevector">shufflevector
1816 operation</a> on constants.</dd>
1818 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1820 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1821 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1822 binary</a> operations. The constraints on operands are the same as those for
1823 the corresponding instruction (e.g. no bitwise operations on floating point
1824 values are allowed).</dd>
1828 <!-- *********************************************************************** -->
1829 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1830 <!-- *********************************************************************** -->
1832 <!-- ======================================================================= -->
1833 <div class="doc_subsection">
1834 <a name="inlineasm">Inline Assembler Expressions</a>
1837 <div class="doc_text">
1840 LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
1841 Module-Level Inline Assembly</a>) through the use of a special value. This
1842 value represents the inline assembler as a string (containing the instructions
1843 to emit), a list of operand constraints (stored as a string), and a flag that
1844 indicates whether or not the inline asm expression has side effects. An example
1845 inline assembler expression is:
1848 <div class="doc_code">
1850 i32 (i32) asm "bswap $0", "=r,r"
1855 Inline assembler expressions may <b>only</b> be used as the callee operand of
1856 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
1859 <div class="doc_code">
1861 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
1866 Inline asms with side effects not visible in the constraint list must be marked
1867 as having side effects. This is done through the use of the
1868 '<tt>sideeffect</tt>' keyword, like so:
1871 <div class="doc_code">
1873 call void asm sideeffect "eieio", ""()
1877 <p>TODO: The format of the asm and constraints string still need to be
1878 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1879 need to be documented). This is probably best done by reference to another
1880 document that covers inline asm from a holistic perspective.
1885 <!-- *********************************************************************** -->
1886 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1887 <!-- *********************************************************************** -->
1889 <div class="doc_text">
1891 <p>The LLVM instruction set consists of several different
1892 classifications of instructions: <a href="#terminators">terminator
1893 instructions</a>, <a href="#binaryops">binary instructions</a>,
1894 <a href="#bitwiseops">bitwise binary instructions</a>, <a
1895 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1896 instructions</a>.</p>
1900 <!-- ======================================================================= -->
1901 <div class="doc_subsection"> <a name="terminators">Terminator
1902 Instructions</a> </div>
1904 <div class="doc_text">
1906 <p>As mentioned <a href="#functionstructure">previously</a>, every
1907 basic block in a program ends with a "Terminator" instruction, which
1908 indicates which block should be executed after the current block is
1909 finished. These terminator instructions typically yield a '<tt>void</tt>'
1910 value: they produce control flow, not values (the one exception being
1911 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1912 <p>There are six different terminator instructions: the '<a
1913 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1914 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1915 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1916 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1917 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1921 <!-- _______________________________________________________________________ -->
1922 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1923 Instruction</a> </div>
1924 <div class="doc_text">
1927 ret <type> <value> <i>; Return a value from a non-void function</i>
1928 ret void <i>; Return from void function</i>
1933 <p>The '<tt>ret</tt>' instruction is used to return control flow (and
1934 optionally a value) from a function back to the caller.</p>
1935 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
1936 returns a value and then causes control flow, and one that just causes
1937 control flow to occur.</p>
1941 <p>The '<tt>ret</tt>' instruction optionally accepts a single argument,
1942 the return value. The type of the return value must be a
1943 '<a href="#t_firstclass">first class</a>' type.</p>
1945 <p>A function is not <a href="#wellformed">well formed</a> if
1946 it it has a non-void return type and contains a '<tt>ret</tt>'
1947 instruction with no return value or a return value with a type that
1948 does not match its type, or if it has a void return type and contains
1949 a '<tt>ret</tt>' instruction with a return value.</p>
1953 <p>When the '<tt>ret</tt>' instruction is executed, control flow
1954 returns back to the calling function's context. If the caller is a "<a
1955 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1956 the instruction after the call. If the caller was an "<a
1957 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1958 at the beginning of the "normal" destination block. If the instruction
1959 returns a value, that value shall set the call or invoke instruction's
1965 ret i32 5 <i>; Return an integer value of 5</i>
1966 ret void <i>; Return from a void function</i>
1967 ret { i32, i8 } { i32 4, i8 2 } <i>; Return an aggregate of values 4 and 2</i>
1970 <!-- _______________________________________________________________________ -->
1971 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1972 <div class="doc_text">
1974 <pre> br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1977 <p>The '<tt>br</tt>' instruction is used to cause control flow to
1978 transfer to a different basic block in the current function. There are
1979 two forms of this instruction, corresponding to a conditional branch
1980 and an unconditional branch.</p>
1982 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1983 single '<tt>i1</tt>' value and two '<tt>label</tt>' values. The
1984 unconditional form of the '<tt>br</tt>' instruction takes a single
1985 '<tt>label</tt>' value as a target.</p>
1987 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
1988 argument is evaluated. If the value is <tt>true</tt>, control flows
1989 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1990 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1992 <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
1993 href="#i_ret">ret</a> i32 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> i32 0<br></pre>
1995 <!-- _______________________________________________________________________ -->
1996 <div class="doc_subsubsection">
1997 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
2000 <div class="doc_text">
2004 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
2009 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
2010 several different places. It is a generalization of the '<tt>br</tt>'
2011 instruction, allowing a branch to occur to one of many possible
2017 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
2018 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
2019 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
2020 table is not allowed to contain duplicate constant entries.</p>
2024 <p>The <tt>switch</tt> instruction specifies a table of values and
2025 destinations. When the '<tt>switch</tt>' instruction is executed, this
2026 table is searched for the given value. If the value is found, control flow is
2027 transfered to the corresponding destination; otherwise, control flow is
2028 transfered to the default destination.</p>
2030 <h5>Implementation:</h5>
2032 <p>Depending on properties of the target machine and the particular
2033 <tt>switch</tt> instruction, this instruction may be code generated in different
2034 ways. For example, it could be generated as a series of chained conditional
2035 branches or with a lookup table.</p>
2040 <i>; Emulate a conditional br instruction</i>
2041 %Val = <a href="#i_zext">zext</a> i1 %value to i32
2042 switch i32 %Val, label %truedest [i32 0, label %falsedest ]
2044 <i>; Emulate an unconditional br instruction</i>
2045 switch i32 0, label %dest [ ]
2047 <i>; Implement a jump table:</i>
2048 switch i32 %val, label %otherwise [ i32 0, label %onzero
2050 i32 2, label %ontwo ]
2054 <!-- _______________________________________________________________________ -->
2055 <div class="doc_subsubsection">
2056 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
2059 <div class="doc_text">
2064 <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>]
2065 to label <normal label> unwind label <exception label>
2070 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
2071 function, with the possibility of control flow transfer to either the
2072 '<tt>normal</tt>' label or the
2073 '<tt>exception</tt>' label. If the callee function returns with the
2074 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
2075 "normal" label. If the callee (or any indirect callees) returns with the "<a
2076 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
2077 continued at the dynamically nearest "exception" label.</p>
2081 <p>This instruction requires several arguments:</p>
2085 The optional "cconv" marker indicates which <a href="#callingconv">calling
2086 convention</a> the call should use. If none is specified, the call defaults
2087 to using C calling conventions.
2090 <li>The optional <a href="#paramattrs">Parameter Attributes</a> list for
2091 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>',
2092 and '<tt>inreg</tt>' attributes are valid here.</li>
2094 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
2095 function value being invoked. In most cases, this is a direct function
2096 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
2097 an arbitrary pointer to function value.
2100 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
2101 function to be invoked. </li>
2103 <li>'<tt>function args</tt>': argument list whose types match the function
2104 signature argument types. If the function signature indicates the function
2105 accepts a variable number of arguments, the extra arguments can be
2108 <li>'<tt>normal label</tt>': the label reached when the called function
2109 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
2111 <li>'<tt>exception label</tt>': the label reached when a callee returns with
2112 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
2114 <li>The optional <a href="#fnattrs">function attributes</a> list. Only
2115 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
2116 '<tt>readnone</tt>' attributes are valid here.</li>
2121 <p>This instruction is designed to operate as a standard '<tt><a
2122 href="#i_call">call</a></tt>' instruction in most regards. The primary
2123 difference is that it establishes an association with a label, which is used by
2124 the runtime library to unwind the stack.</p>
2126 <p>This instruction is used in languages with destructors to ensure that proper
2127 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
2128 exception. Additionally, this is important for implementation of
2129 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
2133 %retval = invoke i32 @Test(i32 15) to label %Continue
2134 unwind label %TestCleanup <i>; {i32}:retval set</i>
2135 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
2136 unwind label %TestCleanup <i>; {i32}:retval set</i>
2141 <!-- _______________________________________________________________________ -->
2143 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
2144 Instruction</a> </div>
2146 <div class="doc_text">
2155 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
2156 at the first callee in the dynamic call stack which used an <a
2157 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
2158 primarily used to implement exception handling.</p>
2162 <p>The '<tt>unwind</tt>' instruction causes execution of the current function to
2163 immediately halt. The dynamic call stack is then searched for the first <a
2164 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
2165 execution continues at the "exceptional" destination block specified by the
2166 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
2167 dynamic call chain, undefined behavior results.</p>
2170 <!-- _______________________________________________________________________ -->
2172 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
2173 Instruction</a> </div>
2175 <div class="doc_text">
2184 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
2185 instruction is used to inform the optimizer that a particular portion of the
2186 code is not reachable. This can be used to indicate that the code after a
2187 no-return function cannot be reached, and other facts.</p>
2191 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
2196 <!-- ======================================================================= -->
2197 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
2198 <div class="doc_text">
2199 <p>Binary operators are used to do most of the computation in a
2200 program. They require two operands of the same type, execute an operation on them, and
2201 produce a single value. The operands might represent
2202 multiple data, as is the case with the <a href="#t_vector">vector</a> data type.
2203 The result value has the same type as its operands.</p>
2204 <p>There are several different binary operators:</p>
2206 <!-- _______________________________________________________________________ -->
2207 <div class="doc_subsubsection">
2208 <a name="i_add">'<tt>add</tt>' Instruction</a>
2211 <div class="doc_text">
2216 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2221 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
2225 <p>The two arguments to the '<tt>add</tt>' instruction must be <a
2226 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>, or
2227 <a href="#t_vector">vector</a> values. Both arguments must have identical
2232 <p>The value produced is the integer or floating point sum of the two
2235 <p>If an integer sum has unsigned overflow, the result returned is the
2236 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2239 <p>Because LLVM integers use a two's complement representation, this
2240 instruction is appropriate for both signed and unsigned integers.</p>
2245 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
2248 <!-- _______________________________________________________________________ -->
2249 <div class="doc_subsubsection">
2250 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
2253 <div class="doc_text">
2258 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2263 <p>The '<tt>sub</tt>' instruction returns the difference of its two
2266 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
2267 '<tt>neg</tt>' instruction present in most other intermediate
2268 representations.</p>
2272 <p>The two arguments to the '<tt>sub</tt>' instruction must be <a
2273 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>,
2274 or <a href="#t_vector">vector</a> values. Both arguments must have identical
2279 <p>The value produced is the integer or floating point difference of
2280 the two operands.</p>
2282 <p>If an integer difference has unsigned overflow, the result returned is the
2283 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2286 <p>Because LLVM integers use a two's complement representation, this
2287 instruction is appropriate for both signed and unsigned integers.</p>
2291 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
2292 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
2296 <!-- _______________________________________________________________________ -->
2297 <div class="doc_subsubsection">
2298 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
2301 <div class="doc_text">
2304 <pre> <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2307 <p>The '<tt>mul</tt>' instruction returns the product of its two
2312 <p>The two arguments to the '<tt>mul</tt>' instruction must be <a
2313 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>,
2314 or <a href="#t_vector">vector</a> values. Both arguments must have identical
2319 <p>The value produced is the integer or floating point product of the
2322 <p>If the result of an integer multiplication has unsigned overflow,
2323 the result returned is the mathematical result modulo
2324 2<sup>n</sup>, where n is the bit width of the result.</p>
2325 <p>Because LLVM integers use a two's complement representation, and the
2326 result is the same width as the operands, this instruction returns the
2327 correct result for both signed and unsigned integers. If a full product
2328 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands
2329 should be sign-extended or zero-extended as appropriate to the
2330 width of the full product.</p>
2332 <pre> <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
2336 <!-- _______________________________________________________________________ -->
2337 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
2339 <div class="doc_text">
2341 <pre> <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2344 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two
2349 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
2350 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2351 values. Both arguments must have identical types.</p>
2355 <p>The value produced is the unsigned integer quotient of the two operands.</p>
2356 <p>Note that unsigned integer division and signed integer division are distinct
2357 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
2358 <p>Division by zero leads to undefined behavior.</p>
2360 <pre> <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2363 <!-- _______________________________________________________________________ -->
2364 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
2366 <div class="doc_text">
2369 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2374 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
2379 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
2380 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2381 values. Both arguments must have identical types.</p>
2384 <p>The value produced is the signed integer quotient of the two operands rounded towards zero.</p>
2385 <p>Note that signed integer division and unsigned integer division are distinct
2386 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
2387 <p>Division by zero leads to undefined behavior. Overflow also leads to
2388 undefined behavior; this is a rare case, but can occur, for example,
2389 by doing a 32-bit division of -2147483648 by -1.</p>
2391 <pre> <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2394 <!-- _______________________________________________________________________ -->
2395 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
2396 Instruction</a> </div>
2397 <div class="doc_text">
2400 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2404 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
2409 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
2410 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a>
2411 of floating point values. Both arguments must have identical types.</p>
2415 <p>The value produced is the floating point quotient of the two operands.</p>
2420 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
2424 <!-- _______________________________________________________________________ -->
2425 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
2427 <div class="doc_text">
2429 <pre> <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2432 <p>The '<tt>urem</tt>' instruction returns the remainder from the
2433 unsigned division of its two arguments.</p>
2435 <p>The two arguments to the '<tt>urem</tt>' instruction must be
2436 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2437 values. Both arguments must have identical types.</p>
2439 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
2440 This instruction always performs an unsigned division to get the remainder.</p>
2441 <p>Note that unsigned integer remainder and signed integer remainder are
2442 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
2443 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
2445 <pre> <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2449 <!-- _______________________________________________________________________ -->
2450 <div class="doc_subsubsection">
2451 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
2454 <div class="doc_text">
2459 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2464 <p>The '<tt>srem</tt>' instruction returns the remainder from the
2465 signed division of its two operands. This instruction can also take
2466 <a href="#t_vector">vector</a> versions of the values in which case
2467 the elements must be integers.</p>
2471 <p>The two arguments to the '<tt>srem</tt>' instruction must be
2472 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2473 values. Both arguments must have identical types.</p>
2477 <p>This instruction returns the <i>remainder</i> of a division (where the result
2478 has the same sign as the dividend, <tt>op1</tt>), not the <i>modulo</i>
2479 operator (where the result has the same sign as the divisor, <tt>op2</tt>) of
2480 a value. For more information about the difference, see <a
2481 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
2482 Math Forum</a>. For a table of how this is implemented in various languages,
2483 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
2484 Wikipedia: modulo operation</a>.</p>
2485 <p>Note that signed integer remainder and unsigned integer remainder are
2486 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
2487 <p>Taking the remainder of a division by zero leads to undefined behavior.
2488 Overflow also leads to undefined behavior; this is a rare case, but can occur,
2489 for example, by taking the remainder of a 32-bit division of -2147483648 by -1.
2490 (The remainder doesn't actually overflow, but this rule lets srem be
2491 implemented using instructions that return both the result of the division
2492 and the remainder.)</p>
2494 <pre> <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2498 <!-- _______________________________________________________________________ -->
2499 <div class="doc_subsubsection">
2500 <a name="i_frem">'<tt>frem</tt>' Instruction</a> </div>
2502 <div class="doc_text">
2505 <pre> <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2508 <p>The '<tt>frem</tt>' instruction returns the remainder from the
2509 division of its two operands.</p>
2511 <p>The two arguments to the '<tt>frem</tt>' instruction must be
2512 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a>
2513 of floating point values. Both arguments must have identical types.</p>
2517 <p>This instruction returns the <i>remainder</i> of a division.
2518 The remainder has the same sign as the dividend.</p>
2523 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
2527 <!-- ======================================================================= -->
2528 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
2529 Operations</a> </div>
2530 <div class="doc_text">
2531 <p>Bitwise binary operators are used to do various forms of
2532 bit-twiddling in a program. They are generally very efficient
2533 instructions and can commonly be strength reduced from other
2534 instructions. They require two operands of the same type, execute an operation on them,
2535 and produce a single value. The resulting value is the same type as its operands.</p>
2538 <!-- _______________________________________________________________________ -->
2539 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
2540 Instruction</a> </div>
2541 <div class="doc_text">
2543 <pre> <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2548 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2549 the left a specified number of bits.</p>
2553 <p>Both arguments to the '<tt>shl</tt>' instruction must be the same <a
2554 href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2555 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
2559 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod 2<sup>n</sup>,
2560 where n is the width of the result. If <tt>op2</tt> is (statically or dynamically) negative or
2561 equal to or larger than the number of bits in <tt>op1</tt>, the result is undefined.</p>
2563 <h5>Example:</h5><pre>
2564 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
2565 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
2566 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
2567 <result> = shl i32 1, 32 <i>; undefined</i>
2570 <!-- _______________________________________________________________________ -->
2571 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
2572 Instruction</a> </div>
2573 <div class="doc_text">
2575 <pre> <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2579 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
2580 operand shifted to the right a specified number of bits with zero fill.</p>
2583 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
2584 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2585 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
2589 <p>This instruction always performs a logical shift right operation. The most
2590 significant bits of the result will be filled with zero bits after the
2591 shift. If <tt>op2</tt> is (statically or dynamically) equal to or larger than
2592 the number of bits in <tt>op1</tt>, the result is undefined.</p>
2596 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
2597 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
2598 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
2599 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
2600 <result> = lshr i32 1, 32 <i>; undefined</i>
2604 <!-- _______________________________________________________________________ -->
2605 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
2606 Instruction</a> </div>
2607 <div class="doc_text">
2610 <pre> <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2614 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
2615 operand shifted to the right a specified number of bits with sign extension.</p>
2618 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
2619 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2620 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
2623 <p>This instruction always performs an arithmetic shift right operation,
2624 The most significant bits of the result will be filled with the sign bit
2625 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
2626 larger than the number of bits in <tt>op1</tt>, the result is undefined.
2631 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
2632 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
2633 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
2634 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
2635 <result> = ashr i32 1, 32 <i>; undefined</i>
2639 <!-- _______________________________________________________________________ -->
2640 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
2641 Instruction</a> </div>
2643 <div class="doc_text">
2648 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2653 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
2654 its two operands.</p>
2658 <p>The two arguments to the '<tt>and</tt>' instruction must be
2659 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2660 values. Both arguments must have identical types.</p>
2663 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
2666 <table border="1" cellspacing="0" cellpadding="4">
2698 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
2699 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
2700 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
2703 <!-- _______________________________________________________________________ -->
2704 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
2705 <div class="doc_text">
2707 <pre> <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2710 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
2711 or of its two operands.</p>
2714 <p>The two arguments to the '<tt>or</tt>' instruction must be
2715 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2716 values. Both arguments must have identical types.</p>
2718 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
2721 <table border="1" cellspacing="0" cellpadding="4">
2752 <pre> <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
2753 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
2754 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
2757 <!-- _______________________________________________________________________ -->
2758 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
2759 Instruction</a> </div>
2760 <div class="doc_text">
2762 <pre> <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2765 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
2766 or of its two operands. The <tt>xor</tt> is used to implement the
2767 "one's complement" operation, which is the "~" operator in C.</p>
2769 <p>The two arguments to the '<tt>xor</tt>' instruction must be
2770 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2771 values. Both arguments must have identical types.</p>
2775 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
2778 <table border="1" cellspacing="0" cellpadding="4">
2810 <pre> <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
2811 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
2812 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
2813 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
2817 <!-- ======================================================================= -->
2818 <div class="doc_subsection">
2819 <a name="vectorops">Vector Operations</a>
2822 <div class="doc_text">
2824 <p>LLVM supports several instructions to represent vector operations in a
2825 target-independent manner. These instructions cover the element-access and
2826 vector-specific operations needed to process vectors effectively. While LLVM
2827 does directly support these vector operations, many sophisticated algorithms
2828 will want to use target-specific intrinsics to take full advantage of a specific
2833 <!-- _______________________________________________________________________ -->
2834 <div class="doc_subsubsection">
2835 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2838 <div class="doc_text">
2843 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
2849 The '<tt>extractelement</tt>' instruction extracts a single scalar
2850 element from a vector at a specified index.
2857 The first operand of an '<tt>extractelement</tt>' instruction is a
2858 value of <a href="#t_vector">vector</a> type. The second operand is
2859 an index indicating the position from which to extract the element.
2860 The index may be a variable.</p>
2865 The result is a scalar of the same type as the element type of
2866 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2867 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2868 results are undefined.
2874 %result = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
2879 <!-- _______________________________________________________________________ -->
2880 <div class="doc_subsubsection">
2881 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
2884 <div class="doc_text">
2889 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
2895 The '<tt>insertelement</tt>' instruction inserts a scalar
2896 element into a vector at a specified index.
2903 The first operand of an '<tt>insertelement</tt>' instruction is a
2904 value of <a href="#t_vector">vector</a> type. The second operand is a
2905 scalar value whose type must equal the element type of the first
2906 operand. The third operand is an index indicating the position at
2907 which to insert the value. The index may be a variable.</p>
2912 The result is a vector of the same type as <tt>val</tt>. Its
2913 element values are those of <tt>val</tt> except at position
2914 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
2915 exceeds the length of <tt>val</tt>, the results are undefined.
2921 %result = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
2925 <!-- _______________________________________________________________________ -->
2926 <div class="doc_subsubsection">
2927 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
2930 <div class="doc_text">
2935 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> <i>; yields <m x <ty>></i>
2941 The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
2942 from two input vectors, returning a vector with the same element type as
2943 the input and length that is the same as the shuffle mask.
2949 The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
2950 with types that match each other. The third argument is a shuffle mask whose
2951 element type is always 'i32'. The result of the instruction is a vector whose
2952 length is the same as the shuffle mask and whose element type is the same as
2953 the element type of the first two operands.
2957 The shuffle mask operand is required to be a constant vector with either
2958 constant integer or undef values.
2964 The elements of the two input vectors are numbered from left to right across
2965 both of the vectors. The shuffle mask operand specifies, for each element of
2966 the result vector, which element of the two input vectors the result element
2967 gets. The element selector may be undef (meaning "don't care") and the second
2968 operand may be undef if performing a shuffle from only one vector.
2974 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
2975 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
2976 %result = shufflevector <4 x i32> %v1, <4 x i32> undef,
2977 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
2978 %result = shufflevector <8 x i32> %v1, <8 x i32> undef,
2979 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i>
2980 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
2981 <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>
2986 <!-- ======================================================================= -->
2987 <div class="doc_subsection">
2988 <a name="aggregateops">Aggregate Operations</a>
2991 <div class="doc_text">
2993 <p>LLVM supports several instructions for working with aggregate values.
2998 <!-- _______________________________________________________________________ -->
2999 <div class="doc_subsubsection">
3000 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
3003 <div class="doc_text">
3008 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
3014 The '<tt>extractvalue</tt>' instruction extracts the value of a struct field
3015 or array element from an aggregate value.
3022 The first operand of an '<tt>extractvalue</tt>' instruction is a
3023 value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a>
3024 type. The operands are constant indices to specify which value to extract
3025 in a similar manner as indices in a
3026 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
3032 The result is the value at the position in the aggregate specified by
3039 %result = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
3044 <!-- _______________________________________________________________________ -->
3045 <div class="doc_subsubsection">
3046 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
3049 <div class="doc_text">
3054 <result> = insertvalue <aggregate type> <val>, <ty> <val>, <idx> <i>; yields <n x <ty>></i>
3060 The '<tt>insertvalue</tt>' instruction inserts a value
3061 into a struct field or array element in an aggregate.
3068 The first operand of an '<tt>insertvalue</tt>' instruction is a
3069 value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a> type.
3070 The second operand is a first-class value to insert.
3071 The following operands are constant indices
3072 indicating the position at which to insert the value in a similar manner as
3074 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
3075 The value to insert must have the same type as the value identified
3082 The result is an aggregate of the same type as <tt>val</tt>. Its
3083 value is that of <tt>val</tt> except that the value at the position
3084 specified by the indices is that of <tt>elt</tt>.
3090 %result = insertvalue {i32, float} %agg, i32 1, 0 <i>; yields {i32, float}</i>
3095 <!-- ======================================================================= -->
3096 <div class="doc_subsection">
3097 <a name="memoryops">Memory Access and Addressing Operations</a>
3100 <div class="doc_text">
3102 <p>A key design point of an SSA-based representation is how it
3103 represents memory. In LLVM, no memory locations are in SSA form, which
3104 makes things very simple. This section describes how to read, write,
3105 allocate, and free memory in LLVM.</p>
3109 <!-- _______________________________________________________________________ -->
3110 <div class="doc_subsubsection">
3111 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
3114 <div class="doc_text">
3119 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3124 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
3125 heap and returns a pointer to it. The object is always allocated in the generic
3126 address space (address space zero).</p>
3130 <p>The '<tt>malloc</tt>' instruction allocates
3131 <tt>sizeof(<type>)*NumElements</tt>
3132 bytes of memory from the operating system and returns a pointer of the
3133 appropriate type to the program. If "NumElements" is specified, it is the
3134 number of elements allocated, otherwise "NumElements" is defaulted to be one.
3135 If a constant alignment is specified, the value result of the allocation is guaranteed to
3136 be aligned to at least that boundary. If not specified, or if zero, the target can
3137 choose to align the allocation on any convenient boundary.</p>
3139 <p>'<tt>type</tt>' must be a sized type.</p>
3143 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
3144 a pointer is returned. The result of a zero byte allocattion is undefined. The
3145 result is null if there is insufficient memory available.</p>
3150 %array = malloc [4 x i8 ] <i>; yields {[%4 x i8]*}:array</i>
3152 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
3153 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
3154 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
3155 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
3156 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
3160 <!-- _______________________________________________________________________ -->
3161 <div class="doc_subsubsection">
3162 <a name="i_free">'<tt>free</tt>' Instruction</a>
3165 <div class="doc_text">
3170 free <type> <value> <i>; yields {void}</i>
3175 <p>The '<tt>free</tt>' instruction returns memory back to the unused
3176 memory heap to be reallocated in the future.</p>
3180 <p>'<tt>value</tt>' shall be a pointer value that points to a value
3181 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
3186 <p>Access to the memory pointed to by the pointer is no longer defined
3187 after this instruction executes. If the pointer is null, the operation
3193 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
3194 free [4 x i8]* %array
3198 <!-- _______________________________________________________________________ -->
3199 <div class="doc_subsubsection">
3200 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
3203 <div class="doc_text">
3208 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3213 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
3214 currently executing function, to be automatically released when this function
3215 returns to its caller. The object is always allocated in the generic address
3216 space (address space zero).</p>
3220 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
3221 bytes of memory on the runtime stack, returning a pointer of the
3222 appropriate type to the program. If "NumElements" is specified, it is the
3223 number of elements allocated, otherwise "NumElements" is defaulted to be one.
3224 If a constant alignment is specified, the value result of the allocation is guaranteed
3225 to be aligned to at least that boundary. If not specified, or if zero, the target
3226 can choose to align the allocation on any convenient boundary.</p>
3228 <p>'<tt>type</tt>' may be any sized type.</p>
3232 <p>Memory is allocated; a pointer is returned. The operation is undefiend if
3233 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
3234 memory is automatically released when the function returns. The '<tt>alloca</tt>'
3235 instruction is commonly used to represent automatic variables that must
3236 have an address available. When the function returns (either with the <tt><a
3237 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
3238 instructions), the memory is reclaimed. Allocating zero bytes
3239 is legal, but the result is undefined.</p>
3244 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
3245 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
3246 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
3247 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
3251 <!-- _______________________________________________________________________ -->
3252 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
3253 Instruction</a> </div>
3254 <div class="doc_text">
3256 <pre> <result> = load <ty>* <pointer>[, align <alignment>]<br> <result> = volatile load <ty>* <pointer>[, align <alignment>]<br></pre>
3258 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
3260 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
3261 address from which to load. The pointer must point to a <a
3262 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
3263 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
3264 the number or order of execution of this <tt>load</tt> with other
3265 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
3268 The optional constant "align" argument specifies the alignment of the operation
3269 (that is, the alignment of the memory address). A value of 0 or an
3270 omitted "align" argument means that the operation has the preferential
3271 alignment for the target. It is the responsibility of the code emitter
3272 to ensure that the alignment information is correct. Overestimating
3273 the alignment results in an undefined behavior. Underestimating the
3274 alignment may produce less efficient code. An alignment of 1 is always
3278 <p>The location of memory pointed to is loaded.</p>
3280 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3282 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
3283 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
3286 <!-- _______________________________________________________________________ -->
3287 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
3288 Instruction</a> </div>
3289 <div class="doc_text">
3291 <pre> store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3292 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3295 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
3297 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
3298 to store and an address at which to store it. The type of the '<tt><pointer></tt>'
3299 operand must be a pointer to the <a href="#t_firstclass">first class</a> type
3300 of the '<tt><value></tt>'
3301 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
3302 optimizer is not allowed to modify the number or order of execution of
3303 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
3304 href="#i_store">store</a></tt> instructions.</p>
3306 The optional constant "align" argument specifies the alignment of the operation
3307 (that is, the alignment of the memory address). A value of 0 or an
3308 omitted "align" argument means that the operation has the preferential
3309 alignment for the target. It is the responsibility of the code emitter
3310 to ensure that the alignment information is correct. Overestimating
3311 the alignment results in an undefined behavior. Underestimating the
3312 alignment may produce less efficient code. An alignment of 1 is always
3316 <p>The contents of memory are updated to contain '<tt><value></tt>'
3317 at the location specified by the '<tt><pointer></tt>' operand.</p>
3319 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3320 store i32 3, i32* %ptr <i>; yields {void}</i>
3321 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
3325 <!-- _______________________________________________________________________ -->
3326 <div class="doc_subsubsection">
3327 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
3330 <div class="doc_text">
3333 <result> = getelementptr <pty>* <ptrval>{, <ty> <idx>}*
3339 The '<tt>getelementptr</tt>' instruction is used to get the address of a
3340 subelement of an aggregate data structure. It performs address calculation only
3341 and does not access memory.</p>
3345 <p>The first argument is always a pointer, and forms the basis of the
3346 calculation. The remaining arguments are indices, that indicate which of the
3347 elements of the aggregate object are indexed. The interpretation of each index
3348 is dependent on the type being indexed into. The first index always indexes the
3349 pointer value given as the first argument, the second index indexes a value of
3350 the type pointed to (not necessarily the value directly pointed to, since the
3351 first index can be non-zero), etc. The first type indexed into must be a pointer
3352 value, subsequent types can be arrays, vectors and structs. Note that subsequent
3353 types being indexed into can never be pointers, since that would require loading
3354 the pointer before continuing calculation.</p>
3356 <p>The type of each index argument depends on the type it is indexing into.
3357 When indexing into a (packed) structure, only <tt>i32</tt> integer
3358 <b>constants</b> are allowed. When indexing into an array, pointer or vector,
3359 only integers of 32 or 64 bits are allowed (also non-constants). 32-bit values
3360 will be sign extended to 64-bits if required.</p>
3362 <p>For example, let's consider a C code fragment and how it gets
3363 compiled to LLVM:</p>
3365 <div class="doc_code">
3378 int *foo(struct ST *s) {
3379 return &s[1].Z.B[5][13];
3384 <p>The LLVM code generated by the GCC frontend is:</p>
3386 <div class="doc_code">
3388 %RT = type { i8 , [10 x [20 x i32]], i8 }
3389 %ST = type { i32, double, %RT }
3391 define i32* %foo(%ST* %s) {
3393 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
3401 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
3402 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
3403 }</tt>' type, a structure. The second index indexes into the third element of
3404 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
3405 i8 }</tt>' type, another structure. The third index indexes into the second
3406 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
3407 array. The two dimensions of the array are subscripted into, yielding an
3408 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
3409 to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
3411 <p>Note that it is perfectly legal to index partially through a
3412 structure, returning a pointer to an inner element. Because of this,
3413 the LLVM code for the given testcase is equivalent to:</p>
3416 define i32* %foo(%ST* %s) {
3417 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
3418 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
3419 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
3420 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
3421 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
3426 <p>Note that it is undefined to access an array out of bounds: array and
3427 pointer indexes must always be within the defined bounds of the array type.
3428 The one exception for this rule is zero length arrays. These arrays are
3429 defined to be accessible as variable length arrays, which requires access
3430 beyond the zero'th element.</p>
3432 <p>The getelementptr instruction is often confusing. For some more insight
3433 into how it works, see <a href="GetElementPtr.html">the getelementptr
3439 <i>; yields [12 x i8]*:aptr</i>
3440 %aptr = getelementptr {i32, [12 x i8]}* %saptr, i64 0, i32 1
3441 <i>; yields i8*:vptr</i>
3442 %vptr = getelementptr {i32, <2 x i8>}* %svptr, i64 0, i32 1, i32 1
3443 <i>; yields i8*:eptr</i>
3444 %eptr = getelementptr [12 x i8]* %aptr, i64 0, i32 1
3448 <!-- ======================================================================= -->
3449 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
3451 <div class="doc_text">
3452 <p>The instructions in this category are the conversion instructions (casting)
3453 which all take a single operand and a type. They perform various bit conversions
3457 <!-- _______________________________________________________________________ -->
3458 <div class="doc_subsubsection">
3459 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
3461 <div class="doc_text">
3465 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
3470 The '<tt>trunc</tt>' instruction truncates its operand to the type <tt>ty2</tt>.
3475 The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
3476 be an <a href="#t_integer">integer</a> type, and a type that specifies the size
3477 and type of the result, which must be an <a href="#t_integer">integer</a>
3478 type. The bit size of <tt>value</tt> must be larger than the bit size of
3479 <tt>ty2</tt>. Equal sized types are not allowed.</p>
3483 The '<tt>trunc</tt>' instruction truncates the high order bits in <tt>value</tt>
3484 and converts the remaining bits to <tt>ty2</tt>. Since the source size must be
3485 larger than the destination size, <tt>trunc</tt> cannot be a <i>no-op cast</i>.
3486 It will always truncate bits.</p>
3490 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
3491 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
3492 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
3496 <!-- _______________________________________________________________________ -->
3497 <div class="doc_subsubsection">
3498 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
3500 <div class="doc_text">
3504 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
3508 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
3513 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
3514 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3515 also be of <a href="#t_integer">integer</a> type. The bit size of the
3516 <tt>value</tt> must be smaller than the bit size of the destination type,
3520 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
3521 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
3523 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
3527 %X = zext i32 257 to i64 <i>; yields i64:257</i>
3528 %Y = zext i1 true to i32 <i>; yields i32:1</i>
3532 <!-- _______________________________________________________________________ -->
3533 <div class="doc_subsubsection">
3534 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
3536 <div class="doc_text">
3540 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
3544 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
3548 The '<tt>sext</tt>' instruction takes a value to cast, which must be of
3549 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3550 also be of <a href="#t_integer">integer</a> type. The bit size of the
3551 <tt>value</tt> must be smaller than the bit size of the destination type,
3556 The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
3557 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size of
3558 the type <tt>ty2</tt>.</p>
3560 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
3564 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
3565 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
3569 <!-- _______________________________________________________________________ -->
3570 <div class="doc_subsubsection">
3571 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
3574 <div class="doc_text">
3579 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
3583 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
3588 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
3589 point</a> value to cast and a <a href="#t_floating">floating point</a> type to
3590 cast it to. The size of <tt>value</tt> must be larger than the size of
3591 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
3592 <i>no-op cast</i>.</p>
3595 <p> The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
3596 <a href="#t_floating">floating point</a> type to a smaller
3597 <a href="#t_floating">floating point</a> type. If the value cannot fit within
3598 the destination type, <tt>ty2</tt>, then the results are undefined.</p>
3602 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
3603 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
3607 <!-- _______________________________________________________________________ -->
3608 <div class="doc_subsubsection">
3609 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
3611 <div class="doc_text">
3615 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
3619 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
3620 floating point value.</p>
3623 <p>The '<tt>fpext</tt>' instruction takes a
3624 <a href="#t_floating">floating point</a> <tt>value</tt> to cast,
3625 and a <a href="#t_floating">floating point</a> type to cast it to. The source
3626 type must be smaller than the destination type.</p>
3629 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
3630 <a href="#t_floating">floating point</a> type to a larger
3631 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
3632 used to make a <i>no-op cast</i> because it always changes bits. Use
3633 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
3637 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
3638 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
3642 <!-- _______________________________________________________________________ -->
3643 <div class="doc_subsubsection">
3644 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
3646 <div class="doc_text">
3650 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
3654 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
3655 unsigned integer equivalent of type <tt>ty2</tt>.
3659 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
3660 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3661 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3662 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3663 vector integer type with the same number of elements as <tt>ty</tt></p>
3666 <p> The '<tt>fptoui</tt>' instruction converts its
3667 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3668 towards zero) unsigned integer value. If the value cannot fit in <tt>ty2</tt>,
3669 the results are undefined.</p>
3673 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
3674 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
3675 %X = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
3679 <!-- _______________________________________________________________________ -->
3680 <div class="doc_subsubsection">
3681 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
3683 <div class="doc_text">
3687 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
3691 <p>The '<tt>fptosi</tt>' instruction converts
3692 <a href="#t_floating">floating point</a> <tt>value</tt> to type <tt>ty2</tt>.
3696 <p> The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
3697 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3698 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3699 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3700 vector integer type with the same number of elements as <tt>ty</tt></p>
3703 <p>The '<tt>fptosi</tt>' instruction converts its
3704 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3705 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
3706 the results are undefined.</p>
3710 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
3711 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
3712 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
3716 <!-- _______________________________________________________________________ -->
3717 <div class="doc_subsubsection">
3718 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
3720 <div class="doc_text">
3724 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3728 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
3729 integer and converts that value to the <tt>ty2</tt> type.</p>
3732 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
3733 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3734 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3735 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3736 floating point type with the same number of elements as <tt>ty</tt></p>
3739 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
3740 integer quantity and converts it to the corresponding floating point value. If
3741 the value cannot fit in the floating point value, the results are undefined.</p>
3745 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
3746 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
3750 <!-- _______________________________________________________________________ -->
3751 <div class="doc_subsubsection">
3752 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
3754 <div class="doc_text">
3758 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3762 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed
3763 integer and converts that value to the <tt>ty2</tt> type.</p>
3766 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
3767 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3768 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3769 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3770 floating point type with the same number of elements as <tt>ty</tt></p>
3773 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed
3774 integer quantity and converts it to the corresponding floating point value. If
3775 the value cannot fit in the floating point value, the results are undefined.</p>
3779 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
3780 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
3784 <!-- _______________________________________________________________________ -->
3785 <div class="doc_subsubsection">
3786 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
3788 <div class="doc_text">
3792 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
3796 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
3797 the integer type <tt>ty2</tt>.</p>
3800 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
3801 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
3802 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.</p>
3805 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
3806 <tt>ty2</tt> by interpreting the pointer value as an integer and either
3807 truncating or zero extending that value to the size of the integer type. If
3808 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
3809 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
3810 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
3815 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
3816 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
3820 <!-- _______________________________________________________________________ -->
3821 <div class="doc_subsubsection">
3822 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
3824 <div class="doc_text">
3828 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
3832 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to
3833 a pointer type, <tt>ty2</tt>.</p>
3836 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
3837 value to cast, and a type to cast it to, which must be a
3838 <a href="#t_pointer">pointer</a> type.</p>
3841 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
3842 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
3843 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
3844 size of a pointer then a truncation is done. If <tt>value</tt> is smaller than
3845 the size of a pointer then a zero extension is done. If they are the same size,
3846 nothing is done (<i>no-op cast</i>).</p>
3850 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
3851 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
3852 %Y = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
3856 <!-- _______________________________________________________________________ -->
3857 <div class="doc_subsubsection">
3858 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
3860 <div class="doc_text">
3864 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
3869 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3870 <tt>ty2</tt> without changing any bits.</p>
3874 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be
3875 a non-aggregate first class value, and a type to cast it to, which must also be
3876 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes of
3878 and the destination type, <tt>ty2</tt>, must be identical. If the source
3879 type is a pointer, the destination type must also be a pointer. This
3880 instruction supports bitwise conversion of vectors to integers and to vectors
3881 of other types (as long as they have the same size).</p>
3884 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3885 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
3886 this conversion. The conversion is done as if the <tt>value</tt> had been
3887 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only be
3888 converted to other pointer types with this instruction. To convert pointers to
3889 other types, use the <a href="#i_inttoptr">inttoptr</a> or
3890 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
3894 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
3895 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
3896 %Z = bitcast <2 x int> %V to i64; <i>; yields i64: %V</i>
3900 <!-- ======================================================================= -->
3901 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
3902 <div class="doc_text">
3903 <p>The instructions in this category are the "miscellaneous"
3904 instructions, which defy better classification.</p>
3907 <!-- _______________________________________________________________________ -->
3908 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
3910 <div class="doc_text">
3912 <pre> <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
3915 <p>The '<tt>icmp</tt>' instruction returns a boolean value or
3916 a vector of boolean values based on comparison
3917 of its two integer, integer vector, or pointer operands.</p>
3919 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
3920 the condition code indicating the kind of comparison to perform. It is not
3921 a value, just a keyword. The possible condition code are:
3924 <li><tt>eq</tt>: equal</li>
3925 <li><tt>ne</tt>: not equal </li>
3926 <li><tt>ugt</tt>: unsigned greater than</li>
3927 <li><tt>uge</tt>: unsigned greater or equal</li>
3928 <li><tt>ult</tt>: unsigned less than</li>
3929 <li><tt>ule</tt>: unsigned less or equal</li>
3930 <li><tt>sgt</tt>: signed greater than</li>
3931 <li><tt>sge</tt>: signed greater or equal</li>
3932 <li><tt>slt</tt>: signed less than</li>
3933 <li><tt>sle</tt>: signed less or equal</li>
3935 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
3936 <a href="#t_pointer">pointer</a>
3937 or integer <a href="#t_vector">vector</a> typed.
3938 They must also be identical types.</p>
3940 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to
3941 the condition code given as <tt>cond</tt>. The comparison performed always
3942 yields either an <a href="#t_primitive"><tt>i1</tt></a> or vector of <tt>i1</tt> result, as follows:
3945 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
3946 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3948 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
3949 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.</li>
3950 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
3951 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
3952 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
3953 <tt>true</tt> if <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
3954 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
3955 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
3956 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
3957 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
3958 <li><tt>sgt</tt>: interprets the operands as signed values and yields
3959 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
3960 <li><tt>sge</tt>: interprets the operands as signed values and yields
3961 <tt>true</tt> if <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
3962 <li><tt>slt</tt>: interprets the operands as signed values and yields
3963 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
3964 <li><tt>sle</tt>: interprets the operands as signed values and yields
3965 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
3967 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
3968 values are compared as if they were integers.</p>
3969 <p>If the operands are integer vectors, then they are compared
3970 element by element. The result is an <tt>i1</tt> vector with
3971 the same number of elements as the values being compared.
3972 Otherwise, the result is an <tt>i1</tt>.
3976 <pre> <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
3977 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
3978 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
3979 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
3980 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
3981 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
3985 <!-- _______________________________________________________________________ -->
3986 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
3988 <div class="doc_text">
3990 <pre> <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
3993 <p>The '<tt>fcmp</tt>' instruction returns a boolean value
3994 or vector of boolean values based on comparison
3995 of its operands.</p>
3997 If the operands are floating point scalars, then the result
3998 type is a boolean (<a href="#t_primitive"><tt>i1</tt></a>).
4000 <p>If the operands are floating point vectors, then the result type
4001 is a vector of boolean with the same number of elements as the
4002 operands being compared.</p>
4004 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
4005 the condition code indicating the kind of comparison to perform. It is not
4006 a value, just a keyword. The possible condition code are:</p>
4008 <li><tt>false</tt>: no comparison, always returns false</li>
4009 <li><tt>oeq</tt>: ordered and equal</li>
4010 <li><tt>ogt</tt>: ordered and greater than </li>
4011 <li><tt>oge</tt>: ordered and greater than or equal</li>
4012 <li><tt>olt</tt>: ordered and less than </li>
4013 <li><tt>ole</tt>: ordered and less than or equal</li>
4014 <li><tt>one</tt>: ordered and not equal</li>
4015 <li><tt>ord</tt>: ordered (no nans)</li>
4016 <li><tt>ueq</tt>: unordered or equal</li>
4017 <li><tt>ugt</tt>: unordered or greater than </li>
4018 <li><tt>uge</tt>: unordered or greater than or equal</li>
4019 <li><tt>ult</tt>: unordered or less than </li>
4020 <li><tt>ule</tt>: unordered or less than or equal</li>
4021 <li><tt>une</tt>: unordered or not equal</li>
4022 <li><tt>uno</tt>: unordered (either nans)</li>
4023 <li><tt>true</tt>: no comparison, always returns true</li>
4025 <p><i>Ordered</i> means that neither operand is a QNAN while
4026 <i>unordered</i> means that either operand may be a QNAN.</p>
4027 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be
4028 either a <a href="#t_floating">floating point</a> type
4029 or a <a href="#t_vector">vector</a> of floating point type.
4030 They must have identical types.</p>
4032 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4033 according to the condition code given as <tt>cond</tt>.
4034 If the operands are vectors, then the vectors are compared
4036 Each comparison performed
4037 always yields an <a href="#t_primitive">i1</a> result, as follows:</p>
4039 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
4040 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4041 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
4042 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4043 <tt>op1</tt> is greather than <tt>op2</tt>.</li>
4044 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4045 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4046 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4047 <tt>op1</tt> is less than <tt>op2</tt>.</li>
4048 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4049 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4050 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4051 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
4052 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
4053 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
4054 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
4055 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
4056 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4057 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
4058 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4059 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
4060 <tt>op1</tt> is less than <tt>op2</tt>.</li>
4061 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
4062 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4063 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
4064 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
4065 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
4066 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
4070 <pre> <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
4071 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
4072 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
4073 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
4077 <!-- _______________________________________________________________________ -->
4078 <div class="doc_subsubsection">
4079 <a name="i_vicmp">'<tt>vicmp</tt>' Instruction</a>
4081 <div class="doc_text">
4083 <pre> <result> = vicmp <cond> <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4086 <p>The '<tt>vicmp</tt>' instruction returns an integer vector value based on
4087 element-wise comparison of its two integer vector operands.</p>
4089 <p>The '<tt>vicmp</tt>' instruction takes three operands. The first operand is
4090 the condition code indicating the kind of comparison to perform. It is not
4091 a value, just a keyword. The possible condition code are:</p>
4093 <li><tt>eq</tt>: equal</li>
4094 <li><tt>ne</tt>: not equal </li>
4095 <li><tt>ugt</tt>: unsigned greater than</li>
4096 <li><tt>uge</tt>: unsigned greater or equal</li>
4097 <li><tt>ult</tt>: unsigned less than</li>
4098 <li><tt>ule</tt>: unsigned less or equal</li>
4099 <li><tt>sgt</tt>: signed greater than</li>
4100 <li><tt>sge</tt>: signed greater or equal</li>
4101 <li><tt>slt</tt>: signed less than</li>
4102 <li><tt>sle</tt>: signed less or equal</li>
4104 <p>The remaining two arguments must be <a href="#t_vector">vector</a> or
4105 <a href="#t_integer">integer</a> typed. They must also be identical types.</p>
4107 <p>The '<tt>vicmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4108 according to the condition code given as <tt>cond</tt>. The comparison yields a
4109 <a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, of
4110 identical type as the values being compared. The most significant bit in each
4111 element is 1 if the element-wise comparison evaluates to true, and is 0
4112 otherwise. All other bits of the result are undefined. The condition codes
4113 are evaluated identically to the <a href="#i_icmp">'<tt>icmp</tt>'
4114 instruction</a>.</p>
4118 <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>
4119 <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>
4123 <!-- _______________________________________________________________________ -->
4124 <div class="doc_subsubsection">
4125 <a name="i_vfcmp">'<tt>vfcmp</tt>' Instruction</a>
4127 <div class="doc_text">
4129 <pre> <result> = vfcmp <cond> <ty> <op1>, <op2></pre>
4131 <p>The '<tt>vfcmp</tt>' instruction returns an integer vector value based on
4132 element-wise comparison of its two floating point vector operands. The output
4133 elements have the same width as the input elements.</p>
4135 <p>The '<tt>vfcmp</tt>' instruction takes three operands. The first operand is
4136 the condition code indicating the kind of comparison to perform. It is not
4137 a value, just a keyword. The possible condition code are:</p>
4139 <li><tt>false</tt>: no comparison, always returns false</li>
4140 <li><tt>oeq</tt>: ordered and equal</li>
4141 <li><tt>ogt</tt>: ordered and greater than </li>
4142 <li><tt>oge</tt>: ordered and greater than or equal</li>
4143 <li><tt>olt</tt>: ordered and less than </li>
4144 <li><tt>ole</tt>: ordered and less than or equal</li>
4145 <li><tt>one</tt>: ordered and not equal</li>
4146 <li><tt>ord</tt>: ordered (no nans)</li>
4147 <li><tt>ueq</tt>: unordered or equal</li>
4148 <li><tt>ugt</tt>: unordered or greater than </li>
4149 <li><tt>uge</tt>: unordered or greater than or equal</li>
4150 <li><tt>ult</tt>: unordered or less than </li>
4151 <li><tt>ule</tt>: unordered or less than or equal</li>
4152 <li><tt>une</tt>: unordered or not equal</li>
4153 <li><tt>uno</tt>: unordered (either nans)</li>
4154 <li><tt>true</tt>: no comparison, always returns true</li>
4156 <p>The remaining two arguments must be <a href="#t_vector">vector</a> of
4157 <a href="#t_floating">floating point</a> typed. They must also be identical
4160 <p>The '<tt>vfcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4161 according to the condition code given as <tt>cond</tt>. The comparison yields a
4162 <a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, with
4163 an identical number of elements as the values being compared, and each element
4164 having identical with to the width of the floating point elements. The most
4165 significant bit in each element is 1 if the element-wise comparison evaluates to
4166 true, and is 0 otherwise. All other bits of the result are undefined. The
4167 condition codes are evaluated identically to the
4168 <a href="#i_fcmp">'<tt>fcmp</tt>' instruction</a>.</p>
4172 <i>; yields: result=<2 x i32> < i32 0, i32 -1 ></i>
4173 <result> = vfcmp oeq <2 x float> < float 4, float 0 >, < float 5, float 0 >
4175 <i>; yields: result=<2 x i64> < i64 -1, i64 0 ></i>
4176 <result> = vfcmp ult <2 x double> < double 1, double 2 >, < double 2, double 2>
4180 <!-- _______________________________________________________________________ -->
4181 <div class="doc_subsubsection">
4182 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
4185 <div class="doc_text">
4189 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
4191 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
4192 the SSA graph representing the function.</p>
4195 <p>The type of the incoming values is specified with the first type
4196 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
4197 as arguments, with one pair for each predecessor basic block of the
4198 current block. Only values of <a href="#t_firstclass">first class</a>
4199 type may be used as the value arguments to the PHI node. Only labels
4200 may be used as the label arguments.</p>
4202 <p>There must be no non-phi instructions between the start of a basic
4203 block and the PHI instructions: i.e. PHI instructions must be first in
4208 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
4209 specified by the pair corresponding to the predecessor basic block that executed
4210 just prior to the current block.</p>
4214 Loop: ; Infinite loop that counts from 0 on up...
4215 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
4216 %nextindvar = add i32 %indvar, 1
4221 <!-- _______________________________________________________________________ -->
4222 <div class="doc_subsubsection">
4223 <a name="i_select">'<tt>select</tt>' Instruction</a>
4226 <div class="doc_text">
4231 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
4233 <i>selty</i> is either i1 or {<N x i1>}
4239 The '<tt>select</tt>' instruction is used to choose one value based on a
4240 condition, without branching.
4247 The '<tt>select</tt>' instruction requires an 'i1' value or
4248 a vector of 'i1' values indicating the
4249 condition, and two values of the same <a href="#t_firstclass">first class</a>
4250 type. If the val1/val2 are vectors and
4251 the condition is a scalar, then entire vectors are selected, not
4252 individual elements.
4258 If the condition is an i1 and it evaluates to 1, the instruction returns the first
4259 value argument; otherwise, it returns the second value argument.
4262 If the condition is a vector of i1, then the value arguments must
4263 be vectors of the same size, and the selection is done element
4270 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
4275 <!-- _______________________________________________________________________ -->
4276 <div class="doc_subsubsection">
4277 <a name="i_call">'<tt>call</tt>' Instruction</a>
4280 <div class="doc_text">
4284 <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>]
4289 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
4293 <p>This instruction requires several arguments:</p>
4297 <p>The optional "tail" marker indicates whether the callee function accesses
4298 any allocas or varargs in the caller. If the "tail" marker is present, the
4299 function call is eligible for tail call optimization. Note that calls may
4300 be marked "tail" even if they do not occur before a <a
4301 href="#i_ret"><tt>ret</tt></a> instruction.</p>
4304 <p>The optional "cconv" marker indicates which <a href="#callingconv">calling
4305 convention</a> the call should use. If none is specified, the call defaults
4306 to using C calling conventions.</p>
4310 <p>The optional <a href="#paramattrs">Parameter Attributes</a> list for
4311 return values. Only '<tt>zeroext</tt>', '<tt>signext</tt>',
4312 and '<tt>inreg</tt>' attributes are valid here.</p>
4316 <p>'<tt>ty</tt>': the type of the call instruction itself which is also
4317 the type of the return value. Functions that return no value are marked
4318 <tt><a href="#t_void">void</a></tt>.</p>
4321 <p>'<tt>fnty</tt>': shall be the signature of the pointer to function
4322 value being invoked. The argument types must match the types implied by
4323 this signature. This type can be omitted if the function is not varargs
4324 and if the function type does not return a pointer to a function.</p>
4327 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
4328 be invoked. In most cases, this is a direct function invocation, but
4329 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
4330 to function value.</p>
4333 <p>'<tt>function args</tt>': argument list whose types match the
4334 function signature argument types. All arguments must be of
4335 <a href="#t_firstclass">first class</a> type. If the function signature
4336 indicates the function accepts a variable number of arguments, the extra
4337 arguments can be specified.</p>
4340 <p>The optional <a href="#fnattrs">function attributes</a> list. Only
4341 '<tt>noreturn</tt>', '<tt>nounwind</tt>', '<tt>readonly</tt>' and
4342 '<tt>readnone</tt>' attributes are valid here.</p>
4348 <p>The '<tt>call</tt>' instruction is used to cause control flow to
4349 transfer to a specified function, with its incoming arguments bound to
4350 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
4351 instruction in the called function, control flow continues with the
4352 instruction after the function call, and the return value of the
4353 function is bound to the result argument.</p>
4358 %retval = call i32 @test(i32 %argc)
4359 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42) <i>; yields i32</i>
4360 %X = tail call i32 @foo() <i>; yields i32</i>
4361 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
4362 call void %foo(i8 97 signext)
4364 %struct.A = type { i32, i8 }
4365 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
4366 %gr = extractvalue %struct.A %r, 0 <i>; yields i32</i>
4367 %gr1 = extractvalue %struct.A %r, 1 <i>; yields i8</i>
4368 %Z = call void @foo() noreturn <i>; indicates that %foo never returns normally</i>
4369 %ZZ = call zeroext i32 @bar() <i>; Return value is %zero extended</i>
4374 <!-- _______________________________________________________________________ -->
4375 <div class="doc_subsubsection">
4376 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
4379 <div class="doc_text">
4384 <resultval> = va_arg <va_list*> <arglist>, <argty>
4389 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
4390 the "variable argument" area of a function call. It is used to implement the
4391 <tt>va_arg</tt> macro in C.</p>
4395 <p>This instruction takes a <tt>va_list*</tt> value and the type of
4396 the argument. It returns a value of the specified argument type and
4397 increments the <tt>va_list</tt> to point to the next argument. The
4398 actual type of <tt>va_list</tt> is target specific.</p>
4402 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
4403 type from the specified <tt>va_list</tt> and causes the
4404 <tt>va_list</tt> to point to the next argument. For more information,
4405 see the variable argument handling <a href="#int_varargs">Intrinsic
4408 <p>It is legal for this instruction to be called in a function which does not
4409 take a variable number of arguments, for example, the <tt>vfprintf</tt>
4412 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
4413 href="#intrinsics">intrinsic function</a> because it takes a type as an
4418 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
4422 <!-- *********************************************************************** -->
4423 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
4424 <!-- *********************************************************************** -->
4426 <div class="doc_text">
4428 <p>LLVM supports the notion of an "intrinsic function". These functions have
4429 well known names and semantics and are required to follow certain restrictions.
4430 Overall, these intrinsics represent an extension mechanism for the LLVM
4431 language that does not require changing all of the transformations in LLVM when
4432 adding to the language (or the bitcode reader/writer, the parser, etc...).</p>
4434 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
4435 prefix is reserved in LLVM for intrinsic names; thus, function names may not
4436 begin with this prefix. Intrinsic functions must always be external functions:
4437 you cannot define the body of intrinsic functions. Intrinsic functions may
4438 only be used in call or invoke instructions: it is illegal to take the address
4439 of an intrinsic function. Additionally, because intrinsic functions are part
4440 of the LLVM language, it is required if any are added that they be documented
4443 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
4444 a family of functions that perform the same operation but on different data
4445 types. Because LLVM can represent over 8 million different integer types,
4446 overloading is used commonly to allow an intrinsic function to operate on any
4447 integer type. One or more of the argument types or the result type can be
4448 overloaded to accept any integer type. Argument types may also be defined as
4449 exactly matching a previous argument's type or the result type. This allows an
4450 intrinsic function which accepts multiple arguments, but needs all of them to
4451 be of the same type, to only be overloaded with respect to a single argument or
4454 <p>Overloaded intrinsics will have the names of its overloaded argument types
4455 encoded into its function name, each preceded by a period. Only those types
4456 which are overloaded result in a name suffix. Arguments whose type is matched
4457 against another type do not. For example, the <tt>llvm.ctpop</tt> function can
4458 take an integer of any width and returns an integer of exactly the same integer
4459 width. This leads to a family of functions such as
4460 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29 %val)</tt>.
4461 Only one type, the return type, is overloaded, and only one type suffix is
4462 required. Because the argument's type is matched against the return type, it
4463 does not require its own name suffix.</p>
4465 <p>To learn how to add an intrinsic function, please see the
4466 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.
4471 <!-- ======================================================================= -->
4472 <div class="doc_subsection">
4473 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
4476 <div class="doc_text">
4478 <p>Variable argument support is defined in LLVM with the <a
4479 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
4480 intrinsic functions. These functions are related to the similarly
4481 named macros defined in the <tt><stdarg.h></tt> header file.</p>
4483 <p>All of these functions operate on arguments that use a
4484 target-specific value type "<tt>va_list</tt>". The LLVM assembly
4485 language reference manual does not define what this type is, so all
4486 transformations should be prepared to handle these functions regardless of
4489 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
4490 instruction and the variable argument handling intrinsic functions are
4493 <div class="doc_code">
4495 define i32 @test(i32 %X, ...) {
4496 ; Initialize variable argument processing
4498 %ap2 = bitcast i8** %ap to i8*
4499 call void @llvm.va_start(i8* %ap2)
4501 ; Read a single integer argument
4502 %tmp = va_arg i8** %ap, i32
4504 ; Demonstrate usage of llvm.va_copy and llvm.va_end
4506 %aq2 = bitcast i8** %aq to i8*
4507 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
4508 call void @llvm.va_end(i8* %aq2)
4510 ; Stop processing of arguments.
4511 call void @llvm.va_end(i8* %ap2)
4515 declare void @llvm.va_start(i8*)
4516 declare void @llvm.va_copy(i8*, i8*)
4517 declare void @llvm.va_end(i8*)
4523 <!-- _______________________________________________________________________ -->
4524 <div class="doc_subsubsection">
4525 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
4529 <div class="doc_text">
4531 <pre> declare void %llvm.va_start(i8* <arglist>)<br></pre>
4533 <p>The '<tt>llvm.va_start</tt>' intrinsic initializes
4534 <tt>*<arglist></tt> for subsequent use by <tt><a
4535 href="#i_va_arg">va_arg</a></tt>.</p>
4539 <p>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
4543 <p>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
4544 macro available in C. In a target-dependent way, it initializes the
4545 <tt>va_list</tt> element to which the argument points, so that the next call to
4546 <tt>va_arg</tt> will produce the first variable argument passed to the function.
4547 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
4548 last argument of the function as the compiler can figure that out.</p>
4552 <!-- _______________________________________________________________________ -->
4553 <div class="doc_subsubsection">
4554 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
4557 <div class="doc_text">
4559 <pre> declare void @llvm.va_end(i8* <arglist>)<br></pre>
4562 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
4563 which has been initialized previously with <tt><a href="#int_va_start">llvm.va_start</a></tt>
4564 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
4568 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
4572 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
4573 macro available in C. In a target-dependent way, it destroys the
4574 <tt>va_list</tt> element to which the argument points. Calls to <a
4575 href="#int_va_start"><tt>llvm.va_start</tt></a> and <a href="#int_va_copy">
4576 <tt>llvm.va_copy</tt></a> must be matched exactly with calls to
4577 <tt>llvm.va_end</tt>.</p>
4581 <!-- _______________________________________________________________________ -->
4582 <div class="doc_subsubsection">
4583 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
4586 <div class="doc_text">
4591 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
4596 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
4597 from the source argument list to the destination argument list.</p>
4601 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
4602 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
4607 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
4608 macro available in C. In a target-dependent way, it copies the source
4609 <tt>va_list</tt> element into the destination <tt>va_list</tt> element. This
4610 intrinsic is necessary because the <tt><a href="#int_va_start">
4611 llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
4612 example, memory allocation.</p>
4616 <!-- ======================================================================= -->
4617 <div class="doc_subsection">
4618 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
4621 <div class="doc_text">
4624 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
4625 Collection</a> (GC) requires the implementation and generation of these
4627 These intrinsics allow identification of <a href="#int_gcroot">GC roots on the
4628 stack</a>, as well as garbage collector implementations that require <a
4629 href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a> barriers.
4630 Front-ends for type-safe garbage collected languages should generate these
4631 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
4632 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
4635 <p>The garbage collection intrinsics only operate on objects in the generic
4636 address space (address space zero).</p>
4640 <!-- _______________________________________________________________________ -->
4641 <div class="doc_subsubsection">
4642 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
4645 <div class="doc_text">
4650 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
4655 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
4656 the code generator, and allows some metadata to be associated with it.</p>
4660 <p>The first argument specifies the address of a stack object that contains the
4661 root pointer. The second pointer (which must be either a constant or a global
4662 value address) contains the meta-data to be associated with the root.</p>
4666 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
4667 location. At compile-time, the code generator generates information to allow
4668 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
4669 intrinsic may only be used in a function which <a href="#gc">specifies a GC
4675 <!-- _______________________________________________________________________ -->
4676 <div class="doc_subsubsection">
4677 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
4680 <div class="doc_text">
4685 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
4690 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
4691 locations, allowing garbage collector implementations that require read
4696 <p>The second argument is the address to read from, which should be an address
4697 allocated from the garbage collector. The first object is a pointer to the
4698 start of the referenced object, if needed by the language runtime (otherwise
4703 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
4704 instruction, but may be replaced with substantially more complex code by the
4705 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
4706 may only be used in a function which <a href="#gc">specifies a GC
4712 <!-- _______________________________________________________________________ -->
4713 <div class="doc_subsubsection">
4714 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
4717 <div class="doc_text">
4722 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
4727 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
4728 locations, allowing garbage collector implementations that require write
4729 barriers (such as generational or reference counting collectors).</p>
4733 <p>The first argument is the reference to store, the second is the start of the
4734 object to store it to, and the third is the address of the field of Obj to
4735 store to. If the runtime does not require a pointer to the object, Obj may be
4740 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
4741 instruction, but may be replaced with substantially more complex code by the
4742 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
4743 may only be used in a function which <a href="#gc">specifies a GC
4750 <!-- ======================================================================= -->
4751 <div class="doc_subsection">
4752 <a name="int_codegen">Code Generator Intrinsics</a>
4755 <div class="doc_text">
4757 These intrinsics are provided by LLVM to expose special features that may only
4758 be implemented with code generator support.
4763 <!-- _______________________________________________________________________ -->
4764 <div class="doc_subsubsection">
4765 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
4768 <div class="doc_text">
4772 declare i8 *@llvm.returnaddress(i32 <level>)
4778 The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
4779 target-specific value indicating the return address of the current function
4780 or one of its callers.
4786 The argument to this intrinsic indicates which function to return the address
4787 for. Zero indicates the calling function, one indicates its caller, etc. The
4788 argument is <b>required</b> to be a constant integer value.
4794 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
4795 the return address of the specified call frame, or zero if it cannot be
4796 identified. The value returned by this intrinsic is likely to be incorrect or 0
4797 for arguments other than zero, so it should only be used for debugging purposes.
4801 Note that calling this intrinsic does not prevent function inlining or other
4802 aggressive transformations, so the value returned may not be that of the obvious
4803 source-language caller.
4808 <!-- _______________________________________________________________________ -->
4809 <div class="doc_subsubsection">
4810 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
4813 <div class="doc_text">
4817 declare i8 *@llvm.frameaddress(i32 <level>)
4823 The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
4824 target-specific frame pointer value for the specified stack frame.
4830 The argument to this intrinsic indicates which function to return the frame
4831 pointer for. Zero indicates the calling function, one indicates its caller,
4832 etc. The argument is <b>required</b> to be a constant integer value.
4838 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
4839 the frame address of the specified call frame, or zero if it cannot be
4840 identified. The value returned by this intrinsic is likely to be incorrect or 0
4841 for arguments other than zero, so it should only be used for debugging purposes.
4845 Note that calling this intrinsic does not prevent function inlining or other
4846 aggressive transformations, so the value returned may not be that of the obvious
4847 source-language caller.
4851 <!-- _______________________________________________________________________ -->
4852 <div class="doc_subsubsection">
4853 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
4856 <div class="doc_text">
4860 declare i8 *@llvm.stacksave()
4866 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
4867 the function stack, for use with <a href="#int_stackrestore">
4868 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
4869 features like scoped automatic variable sized arrays in C99.
4875 This intrinsic returns a opaque pointer value that can be passed to <a
4876 href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
4877 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
4878 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
4879 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
4880 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
4881 that were allocated after the <tt>llvm.stacksave</tt> was executed.
4886 <!-- _______________________________________________________________________ -->
4887 <div class="doc_subsubsection">
4888 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
4891 <div class="doc_text">
4895 declare void @llvm.stackrestore(i8 * %ptr)
4901 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
4902 the function stack to the state it was in when the corresponding <a
4903 href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
4904 useful for implementing language features like scoped automatic variable sized
4911 See the description for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.
4917 <!-- _______________________________________________________________________ -->
4918 <div class="doc_subsubsection">
4919 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
4922 <div class="doc_text">
4926 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>)
4933 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
4934 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
4936 effect on the behavior of the program but can change its performance
4943 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
4944 determining if the fetch should be for a read (0) or write (1), and
4945 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
4946 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
4947 <tt>locality</tt> arguments must be constant integers.
4953 This intrinsic does not modify the behavior of the program. In particular,
4954 prefetches cannot trap and do not produce a value. On targets that support this
4955 intrinsic, the prefetch can provide hints to the processor cache for better
4961 <!-- _______________________________________________________________________ -->
4962 <div class="doc_subsubsection">
4963 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
4966 <div class="doc_text">
4970 declare void @llvm.pcmarker(i32 <id>)
4977 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
4979 code to simulators and other tools. The method is target specific, but it is
4980 expected that the marker will use exported symbols to transmit the PC of the
4982 The marker makes no guarantees that it will remain with any specific instruction
4983 after optimizations. It is possible that the presence of a marker will inhibit
4984 optimizations. The intended use is to be inserted after optimizations to allow
4985 correlations of simulation runs.
4991 <tt>id</tt> is a numerical id identifying the marker.
4997 This intrinsic does not modify the behavior of the program. Backends that do not
4998 support this intrinisic may ignore it.
5003 <!-- _______________________________________________________________________ -->
5004 <div class="doc_subsubsection">
5005 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
5008 <div class="doc_text">
5012 declare i64 @llvm.readcyclecounter( )
5019 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
5020 counter register (or similar low latency, high accuracy clocks) on those targets
5021 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
5022 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
5023 should only be used for small timings.
5029 When directly supported, reading the cycle counter should not modify any memory.
5030 Implementations are allowed to either return a application specific value or a
5031 system wide value. On backends without support, this is lowered to a constant 0.
5036 <!-- ======================================================================= -->
5037 <div class="doc_subsection">
5038 <a name="int_libc">Standard C Library Intrinsics</a>
5041 <div class="doc_text">
5043 LLVM provides intrinsics for a few important standard C library functions.
5044 These intrinsics allow source-language front-ends to pass information about the
5045 alignment of the pointer arguments to the code generator, providing opportunity
5046 for more efficient code generation.
5051 <!-- _______________________________________________________________________ -->
5052 <div class="doc_subsubsection">
5053 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
5056 <div class="doc_text">
5060 declare void @llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
5061 i32 <len>, i32 <align>)
5062 declare void @llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
5063 i64 <len>, i32 <align>)
5069 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
5070 location to the destination location.
5074 Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
5075 intrinsics do not return a value, and takes an extra alignment argument.
5081 The first argument is a pointer to the destination, the second is a pointer to
5082 the source. The third argument is an integer argument
5083 specifying the number of bytes to copy, and the fourth argument is the alignment
5084 of the source and destination locations.
5088 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5089 the caller guarantees that both the source and destination pointers are aligned
5096 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
5097 location to the destination location, which are not allowed to overlap. It
5098 copies "len" bytes of memory over. If the argument is known to be aligned to
5099 some boundary, this can be specified as the fourth argument, otherwise it should
5105 <!-- _______________________________________________________________________ -->
5106 <div class="doc_subsubsection">
5107 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
5110 <div class="doc_text">
5114 declare void @llvm.memmove.i32(i8 * <dest>, i8 * <src>,
5115 i32 <len>, i32 <align>)
5116 declare void @llvm.memmove.i64(i8 * <dest>, i8 * <src>,
5117 i64 <len>, i32 <align>)
5123 The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
5124 location to the destination location. It is similar to the
5125 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to overlap.
5129 Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
5130 intrinsics do not return a value, and takes an extra alignment argument.
5136 The first argument is a pointer to the destination, the second is a pointer to
5137 the source. The third argument is an integer argument
5138 specifying the number of bytes to copy, and the fourth argument is the alignment
5139 of the source and destination locations.
5143 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5144 the caller guarantees that the source and destination pointers are aligned to
5151 The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
5152 location to the destination location, which may overlap. It
5153 copies "len" bytes of memory over. If the argument is known to be aligned to
5154 some boundary, this can be specified as the fourth argument, otherwise it should
5160 <!-- _______________________________________________________________________ -->
5161 <div class="doc_subsubsection">
5162 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
5165 <div class="doc_text">
5169 declare void @llvm.memset.i32(i8 * <dest>, i8 <val>,
5170 i32 <len>, i32 <align>)
5171 declare void @llvm.memset.i64(i8 * <dest>, i8 <val>,
5172 i64 <len>, i32 <align>)
5178 The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
5183 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
5184 does not return a value, and takes an extra alignment argument.
5190 The first argument is a pointer to the destination to fill, the second is the
5191 byte value to fill it with, the third argument is an integer
5192 argument specifying the number of bytes to fill, and the fourth argument is the
5193 known alignment of destination location.
5197 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5198 the caller guarantees that the destination pointer is aligned to that boundary.
5204 The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
5206 destination location. If the argument is known to be aligned to some boundary,
5207 this can be specified as the fourth argument, otherwise it should be set to 0 or
5213 <!-- _______________________________________________________________________ -->
5214 <div class="doc_subsubsection">
5215 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
5218 <div class="doc_text">
5221 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
5222 floating point or vector of floating point type. Not all targets support all
5225 declare float @llvm.sqrt.f32(float %Val)
5226 declare double @llvm.sqrt.f64(double %Val)
5227 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
5228 declare fp128 @llvm.sqrt.f128(fp128 %Val)
5229 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
5235 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
5236 returning the same value as the libm '<tt>sqrt</tt>' functions would. Unlike
5237 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
5238 negative numbers other than -0.0 (which allows for better optimization, because
5239 there is no need to worry about errno being set). <tt>llvm.sqrt(-0.0)</tt> is
5240 defined to return -0.0 like IEEE sqrt.
5246 The argument and return value are floating point numbers of the same type.
5252 This function returns the sqrt of the specified operand if it is a nonnegative
5253 floating point number.
5257 <!-- _______________________________________________________________________ -->
5258 <div class="doc_subsubsection">
5259 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
5262 <div class="doc_text">
5265 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
5266 floating point or vector of floating point type. Not all targets support all
5269 declare float @llvm.powi.f32(float %Val, i32 %power)
5270 declare double @llvm.powi.f64(double %Val, i32 %power)
5271 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
5272 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
5273 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
5279 The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
5280 specified (positive or negative) power. The order of evaluation of
5281 multiplications is not defined. When a vector of floating point type is
5282 used, the second argument remains a scalar integer value.
5288 The second argument is an integer power, and the first is a value to raise to
5295 This function returns the first value raised to the second power with an
5296 unspecified sequence of rounding operations.</p>
5299 <!-- _______________________________________________________________________ -->
5300 <div class="doc_subsubsection">
5301 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
5304 <div class="doc_text">
5307 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
5308 floating point or vector of floating point type. Not all targets support all
5311 declare float @llvm.sin.f32(float %Val)
5312 declare double @llvm.sin.f64(double %Val)
5313 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
5314 declare fp128 @llvm.sin.f128(fp128 %Val)
5315 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
5321 The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.
5327 The argument and return value are floating point numbers of the same type.
5333 This function returns the sine of the specified operand, returning the
5334 same values as the libm <tt>sin</tt> functions would, and handles error
5335 conditions in the same way.</p>
5338 <!-- _______________________________________________________________________ -->
5339 <div class="doc_subsubsection">
5340 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
5343 <div class="doc_text">
5346 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
5347 floating point or vector of floating point type. Not all targets support all
5350 declare float @llvm.cos.f32(float %Val)
5351 declare double @llvm.cos.f64(double %Val)
5352 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
5353 declare fp128 @llvm.cos.f128(fp128 %Val)
5354 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
5360 The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.
5366 The argument and return value are floating point numbers of the same type.
5372 This function returns the cosine of the specified operand, returning the
5373 same values as the libm <tt>cos</tt> functions would, and handles error
5374 conditions in the same way.</p>
5377 <!-- _______________________________________________________________________ -->
5378 <div class="doc_subsubsection">
5379 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
5382 <div class="doc_text">
5385 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
5386 floating point or vector of floating point type. Not all targets support all
5389 declare float @llvm.pow.f32(float %Val, float %Power)
5390 declare double @llvm.pow.f64(double %Val, double %Power)
5391 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
5392 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
5393 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
5399 The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
5400 specified (positive or negative) power.
5406 The second argument is a floating point power, and the first is a value to
5407 raise to that power.
5413 This function returns the first value raised to the second power,
5415 same values as the libm <tt>pow</tt> functions would, and handles error
5416 conditions in the same way.</p>
5420 <!-- ======================================================================= -->
5421 <div class="doc_subsection">
5422 <a name="int_manip">Bit Manipulation Intrinsics</a>
5425 <div class="doc_text">
5427 LLVM provides intrinsics for a few important bit manipulation operations.
5428 These allow efficient code generation for some algorithms.
5433 <!-- _______________________________________________________________________ -->
5434 <div class="doc_subsubsection">
5435 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
5438 <div class="doc_text">
5441 <p>This is an overloaded intrinsic function. You can use bswap on any integer
5442 type that is an even number of bytes (i.e. BitWidth % 16 == 0).</p>
5444 declare i16 @llvm.bswap.i16(i16 <id>)
5445 declare i32 @llvm.bswap.i32(i32 <id>)
5446 declare i64 @llvm.bswap.i64(i64 <id>)
5452 The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
5453 values with an even number of bytes (positive multiple of 16 bits). These are
5454 useful for performing operations on data that is not in the target's native
5461 The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
5462 and low byte of the input i16 swapped. Similarly, the <tt>llvm.bswap.i32</tt>
5463 intrinsic returns an i32 value that has the four bytes of the input i32
5464 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the returned
5465 i32 will have its bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i48</tt>,
5466 <tt>llvm.bswap.i64</tt> and other intrinsics extend this concept to
5467 additional even-byte lengths (6 bytes, 8 bytes and more, respectively).
5472 <!-- _______________________________________________________________________ -->
5473 <div class="doc_subsubsection">
5474 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
5477 <div class="doc_text">
5480 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
5481 width. Not all targets support all bit widths however.</p>
5483 declare i8 @llvm.ctpop.i8 (i8 <src>)
5484 declare i16 @llvm.ctpop.i16(i16 <src>)
5485 declare i32 @llvm.ctpop.i32(i32 <src>)
5486 declare i64 @llvm.ctpop.i64(i64 <src>)
5487 declare i256 @llvm.ctpop.i256(i256 <src>)
5493 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
5500 The only argument is the value to be counted. The argument may be of any
5501 integer type. The return type must match the argument type.
5507 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
5511 <!-- _______________________________________________________________________ -->
5512 <div class="doc_subsubsection">
5513 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
5516 <div class="doc_text">
5519 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
5520 integer bit width. Not all targets support all bit widths however.</p>
5522 declare i8 @llvm.ctlz.i8 (i8 <src>)
5523 declare i16 @llvm.ctlz.i16(i16 <src>)
5524 declare i32 @llvm.ctlz.i32(i32 <src>)
5525 declare i64 @llvm.ctlz.i64(i64 <src>)
5526 declare i256 @llvm.ctlz.i256(i256 <src>)
5532 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
5533 leading zeros in a variable.
5539 The only argument is the value to be counted. The argument may be of any
5540 integer type. The return type must match the argument type.
5546 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
5547 in a variable. If the src == 0 then the result is the size in bits of the type
5548 of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.
5554 <!-- _______________________________________________________________________ -->
5555 <div class="doc_subsubsection">
5556 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
5559 <div class="doc_text">
5562 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
5563 integer bit width. Not all targets support all bit widths however.</p>
5565 declare i8 @llvm.cttz.i8 (i8 <src>)
5566 declare i16 @llvm.cttz.i16(i16 <src>)
5567 declare i32 @llvm.cttz.i32(i32 <src>)
5568 declare i64 @llvm.cttz.i64(i64 <src>)
5569 declare i256 @llvm.cttz.i256(i256 <src>)
5575 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
5582 The only argument is the value to be counted. The argument may be of any
5583 integer type. The return type must match the argument type.
5589 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
5590 in a variable. If the src == 0 then the result is the size in bits of the type
5591 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
5595 <!-- _______________________________________________________________________ -->
5596 <div class="doc_subsubsection">
5597 <a name="int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic</a>
5600 <div class="doc_text">
5603 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.select</tt>
5604 on any integer bit width.</p>
5606 declare i17 @llvm.part.select.i17 (i17 %val, i32 %loBit, i32 %hiBit)
5607 declare i29 @llvm.part.select.i29 (i29 %val, i32 %loBit, i32 %hiBit)
5611 <p>The '<tt>llvm.part.select</tt>' family of intrinsic functions selects a
5612 range of bits from an integer value and returns them in the same bit width as
5613 the original value.</p>
5616 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5617 any bit width but they must have the same bit width. The second and third
5618 arguments must be <tt>i32</tt> type since they specify only a bit index.</p>
5621 <p>The operation of the '<tt>llvm.part.select</tt>' intrinsic has two modes
5622 of operation: forwards and reverse. If <tt>%loBit</tt> is greater than
5623 <tt>%hiBits</tt> then the intrinsic operates in reverse mode. Otherwise it
5624 operates in forward mode.</p>
5625 <p>In forward mode, this intrinsic is the equivalent of shifting <tt>%val</tt>
5626 right by <tt>%loBit</tt> bits and then ANDing it with a mask with
5627 only the <tt>%hiBit - %loBit</tt> bits set, as follows:</p>
5629 <li>The <tt>%val</tt> is shifted right (LSHR) by the number of bits specified
5630 by <tt>%loBits</tt>. This normalizes the value to the low order bits.</li>
5631 <li>The <tt>%loBits</tt> value is subtracted from the <tt>%hiBits</tt> value
5632 to determine the number of bits to retain.</li>
5633 <li>A mask of the retained bits is created by shifting a -1 value.</li>
5634 <li>The mask is ANDed with <tt>%val</tt> to produce the result.</li>
5636 <p>In reverse mode, a similar computation is made except that the bits are
5637 returned in the reverse order. So, for example, if <tt>X</tt> has the value
5638 <tt>i16 0x0ACF (101011001111)</tt> and we apply
5639 <tt>part.select(i16 X, 8, 3)</tt> to it, we get back the value
5640 <tt>i16 0x0026 (000000100110)</tt>.</p>
5643 <div class="doc_subsubsection">
5644 <a name="int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic</a>
5647 <div class="doc_text">
5650 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.set</tt>
5651 on any integer bit width.</p>
5653 declare i17 @llvm.part.set.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
5654 declare i29 @llvm.part.set.i29.i9 (i29 %val, i9 %repl, i32 %lo, i32 %hi)
5658 <p>The '<tt>llvm.part.set</tt>' family of intrinsic functions replaces a range
5659 of bits in an integer value with another integer value. It returns the integer
5660 with the replaced bits.</p>
5663 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5664 any bit width but they must have the same bit width. <tt>%val</tt> is the value
5665 whose bits will be replaced. The second argument, <tt>%repl</tt> may be an
5666 integer of any bit width. The third and fourth arguments must be <tt>i32</tt>
5667 type since they specify only a bit index.</p>
5670 <p>The operation of the '<tt>llvm.part.set</tt>' intrinsic has two modes
5671 of operation: forwards and reverse. If <tt>%lo</tt> is greater than
5672 <tt>%hi</tt> then the intrinsic operates in reverse mode. Otherwise it
5673 operates in forward mode.</p>
5674 <p>For both modes, the <tt>%repl</tt> value is prepared for use by either
5675 truncating it down to the size of the replacement area or zero extending it
5676 up to that size.</p>
5677 <p>In forward mode, the bits between <tt>%lo</tt> and <tt>%hi</tt> (inclusive)
5678 are replaced with corresponding bits from <tt>%repl</tt>. That is the 0th bit
5679 in <tt>%repl</tt> replaces the <tt>%lo</tt>th bit in <tt>%val</tt> and etc. up
5680 to the <tt>%hi</tt>th bit.</p>
5681 <p>In reverse mode, a similar computation is made except that the bits are
5682 reversed. That is, the <tt>0</tt>th bit in <tt>%repl</tt> replaces the
5683 <tt>%hi</tt> bit in <tt>%val</tt> and etc. down to the <tt>%lo</tt>th bit.</p>
5686 llvm.part.set(0xFFFF, 0, 4, 7) -> 0xFF0F
5687 llvm.part.set(0xFFFF, 0, 7, 4) -> 0xFF0F
5688 llvm.part.set(0xFFFF, 1, 7, 4) -> 0xFF8F
5689 llvm.part.set(0xFFFF, F, 8, 3) -> 0xFFE7
5690 llvm.part.set(0xFFFF, 0, 3, 8) -> 0xFE07
5694 <!-- ======================================================================= -->
5695 <div class="doc_subsection">
5696 <a name="int_debugger">Debugger Intrinsics</a>
5699 <div class="doc_text">
5701 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
5702 are described in the <a
5703 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
5704 Debugging</a> document.
5709 <!-- ======================================================================= -->
5710 <div class="doc_subsection">
5711 <a name="int_eh">Exception Handling Intrinsics</a>
5714 <div class="doc_text">
5715 <p> The LLVM exception handling intrinsics (which all start with
5716 <tt>llvm.eh.</tt> prefix), are described in the <a
5717 href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
5718 Handling</a> document. </p>
5721 <!-- ======================================================================= -->
5722 <div class="doc_subsection">
5723 <a name="int_trampoline">Trampoline Intrinsic</a>
5726 <div class="doc_text">
5728 This intrinsic makes it possible to excise one parameter, marked with
5729 the <tt>nest</tt> attribute, from a function. The result is a callable
5730 function pointer lacking the nest parameter - the caller does not need
5731 to provide a value for it. Instead, the value to use is stored in
5732 advance in a "trampoline", a block of memory usually allocated
5733 on the stack, which also contains code to splice the nest value into the
5734 argument list. This is used to implement the GCC nested function address
5738 For example, if the function is
5739 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
5740 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as follows:</p>
5742 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
5743 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
5744 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
5745 %fp = bitcast i8* %p to i32 (i32, i32)*
5747 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
5748 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
5751 <!-- _______________________________________________________________________ -->
5752 <div class="doc_subsubsection">
5753 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
5755 <div class="doc_text">
5758 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
5762 This fills the memory pointed to by <tt>tramp</tt> with code
5763 and returns a function pointer suitable for executing it.
5767 The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
5768 pointers. The <tt>tramp</tt> argument must point to a sufficiently large
5769 and sufficiently aligned block of memory; this memory is written to by the
5770 intrinsic. Note that the size and the alignment are target-specific - LLVM
5771 currently provides no portable way of determining them, so a front-end that
5772 generates this intrinsic needs to have some target-specific knowledge.
5773 The <tt>func</tt> argument must hold a function bitcast to an <tt>i8*</tt>.
5777 The block of memory pointed to by <tt>tramp</tt> is filled with target
5778 dependent code, turning it into a function. A pointer to this function is
5779 returned, but needs to be bitcast to an
5780 <a href="#int_trampoline">appropriate function pointer type</a>
5781 before being called. The new function's signature is the same as that of
5782 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
5783 removed. At most one such <tt>nest</tt> argument is allowed, and it must be
5784 of pointer type. Calling the new function is equivalent to calling
5785 <tt>func</tt> with the same argument list, but with <tt>nval</tt> used for the
5786 missing <tt>nest</tt> argument. If, after calling
5787 <tt>llvm.init.trampoline</tt>, the memory pointed to by <tt>tramp</tt> is
5788 modified, then the effect of any later call to the returned function pointer is
5793 <!-- ======================================================================= -->
5794 <div class="doc_subsection">
5795 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
5798 <div class="doc_text">
5800 These intrinsic functions expand the "universal IR" of LLVM to represent
5801 hardware constructs for atomic operations and memory synchronization. This
5802 provides an interface to the hardware, not an interface to the programmer. It
5803 is aimed at a low enough level to allow any programming models or APIs
5804 (Application Programming Interfaces) which
5805 need atomic behaviors to map cleanly onto it. It is also modeled primarily on
5806 hardware behavior. Just as hardware provides a "universal IR" for source
5807 languages, it also provides a starting point for developing a "universal"
5808 atomic operation and synchronization IR.
5811 These do <em>not</em> form an API such as high-level threading libraries,
5812 software transaction memory systems, atomic primitives, and intrinsic
5813 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
5814 application libraries. The hardware interface provided by LLVM should allow
5815 a clean implementation of all of these APIs and parallel programming models.
5816 No one model or paradigm should be selected above others unless the hardware
5817 itself ubiquitously does so.
5822 <!-- _______________________________________________________________________ -->
5823 <div class="doc_subsubsection">
5824 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
5826 <div class="doc_text">
5829 declare void @llvm.memory.barrier( i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>,
5835 The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
5836 specific pairs of memory access types.
5840 The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
5841 The first four arguments enables a specific barrier as listed below. The fith
5842 argument specifies that the barrier applies to io or device or uncached memory.
5846 <li><tt>ll</tt>: load-load barrier</li>
5847 <li><tt>ls</tt>: load-store barrier</li>
5848 <li><tt>sl</tt>: store-load barrier</li>
5849 <li><tt>ss</tt>: store-store barrier</li>
5850 <li><tt>device</tt>: barrier applies to device and uncached memory also.</li>
5854 This intrinsic causes the system to enforce some ordering constraints upon
5855 the loads and stores of the program. This barrier does not indicate
5856 <em>when</em> any events will occur, it only enforces an <em>order</em> in
5857 which they occur. For any of the specified pairs of load and store operations
5858 (f.ex. load-load, or store-load), all of the first operations preceding the
5859 barrier will complete before any of the second operations succeeding the
5860 barrier begin. Specifically the semantics for each pairing is as follows:
5863 <li><tt>ll</tt>: All loads before the barrier must complete before any load
5864 after the barrier begins.</li>
5866 <li><tt>ls</tt>: All loads before the barrier must complete before any
5867 store after the barrier begins.</li>
5868 <li><tt>ss</tt>: All stores before the barrier must complete before any
5869 store after the barrier begins.</li>
5870 <li><tt>sl</tt>: All stores before the barrier must complete before any
5871 load after the barrier begins.</li>
5874 These semantics are applied with a logical "and" behavior when more than one
5875 is enabled in a single memory barrier intrinsic.
5878 Backends may implement stronger barriers than those requested when they do not
5879 support as fine grained a barrier as requested. Some architectures do not
5880 need all types of barriers and on such architectures, these become noops.
5887 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
5888 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
5889 <i>; guarantee the above finishes</i>
5890 store i32 8, %ptr <i>; before this begins</i>
5894 <!-- _______________________________________________________________________ -->
5895 <div class="doc_subsubsection">
5896 <a name="int_atomic_cmp_swap">'<tt>llvm.atomic.cmp.swap.*</tt>' Intrinsic</a>
5898 <div class="doc_text">
5901 This is an overloaded intrinsic. You can use <tt>llvm.atomic.cmp.swap</tt> on
5902 any integer bit width and for different address spaces. Not all targets
5903 support all bit widths however.</p>
5906 declare i8 @llvm.atomic.cmp.swap.i8.p0i8( i8* <ptr>, i8 <cmp>, i8 <val> )
5907 declare i16 @llvm.atomic.cmp.swap.i16.p0i16( i16* <ptr>, i16 <cmp>, i16 <val> )
5908 declare i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* <ptr>, i32 <cmp>, i32 <val> )
5909 declare i64 @llvm.atomic.cmp.swap.i64.p0i64( i64* <ptr>, i64 <cmp>, i64 <val> )
5914 This loads a value in memory and compares it to a given value. If they are
5915 equal, it stores a new value into the memory.
5919 The <tt>llvm.atomic.cmp.swap</tt> intrinsic takes three arguments. The result as
5920 well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
5921 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
5922 this integer type. While any bit width integer may be used, targets may only
5923 lower representations they support in hardware.
5928 This entire intrinsic must be executed atomically. It first loads the value
5929 in memory pointed to by <tt>ptr</tt> and compares it with the value
5930 <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the memory. The
5931 loaded value is yielded in all cases. This provides the equivalent of an
5932 atomic compare-and-swap operation within the SSA framework.
5940 %val1 = add i32 4, 4
5941 %result1 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 4, %val1 )
5942 <i>; yields {i32}:result1 = 4</i>
5943 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
5944 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
5946 %val2 = add i32 1, 1
5947 %result2 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 5, %val2 )
5948 <i>; yields {i32}:result2 = 8</i>
5949 %stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
5951 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
5955 <!-- _______________________________________________________________________ -->
5956 <div class="doc_subsubsection">
5957 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
5959 <div class="doc_text">
5963 This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
5964 integer bit width. Not all targets support all bit widths however.</p>
5966 declare i8 @llvm.atomic.swap.i8.p0i8( i8* <ptr>, i8 <val> )
5967 declare i16 @llvm.atomic.swap.i16.p0i16( i16* <ptr>, i16 <val> )
5968 declare i32 @llvm.atomic.swap.i32.p0i32( i32* <ptr>, i32 <val> )
5969 declare i64 @llvm.atomic.swap.i64.p0i64( i64* <ptr>, i64 <val> )
5974 This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
5975 the value from memory. It then stores the value in <tt>val</tt> in the memory
5981 The <tt>llvm.atomic.swap</tt> intrinsic takes two arguments. Both the
5982 <tt>val</tt> argument and the result must be integers of the same bit width.
5983 The first argument, <tt>ptr</tt>, must be a pointer to a value of this
5984 integer type. The targets may only lower integer representations they
5989 This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
5990 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
5991 equivalent of an atomic swap operation within the SSA framework.
5999 %val1 = add i32 4, 4
6000 %result1 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val1 )
6001 <i>; yields {i32}:result1 = 4</i>
6002 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
6003 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
6005 %val2 = add i32 1, 1
6006 %result2 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val2 )
6007 <i>; yields {i32}:result2 = 8</i>
6009 %stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
6010 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
6014 <!-- _______________________________________________________________________ -->
6015 <div class="doc_subsubsection">
6016 <a name="int_atomic_load_add">'<tt>llvm.atomic.load.add.*</tt>' Intrinsic</a>
6019 <div class="doc_text">
6022 This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.add</tt> on any
6023 integer bit width. Not all targets support all bit widths however.</p>
6025 declare i8 @llvm.atomic.load.add.i8..p0i8( i8* <ptr>, i8 <delta> )
6026 declare i16 @llvm.atomic.load.add.i16..p0i16( i16* <ptr>, i16 <delta> )
6027 declare i32 @llvm.atomic.load.add.i32..p0i32( i32* <ptr>, i32 <delta> )
6028 declare i64 @llvm.atomic.load.add.i64..p0i64( i64* <ptr>, i64 <delta> )
6033 This intrinsic adds <tt>delta</tt> to the value stored in memory at
6034 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
6039 The intrinsic takes two arguments, the first a pointer to an integer value
6040 and the second an integer value. The result is also an integer value. These
6041 integer types can have any bit width, but they must all have the same bit
6042 width. The targets may only lower integer representations they support.
6046 This intrinsic does a series of operations atomically. It first loads the
6047 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
6048 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
6055 %result1 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 4 )
6056 <i>; yields {i32}:result1 = 4</i>
6057 %result2 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 2 )
6058 <i>; yields {i32}:result2 = 8</i>
6059 %result3 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 5 )
6060 <i>; yields {i32}:result3 = 10</i>
6061 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
6065 <!-- _______________________________________________________________________ -->
6066 <div class="doc_subsubsection">
6067 <a name="int_atomic_load_sub">'<tt>llvm.atomic.load.sub.*</tt>' Intrinsic</a>
6070 <div class="doc_text">
6073 This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.sub</tt> on
6074 any integer bit width and for different address spaces. Not all targets
6075 support all bit widths however.</p>
6077 declare i8 @llvm.atomic.load.sub.i8.p0i32( i8* <ptr>, i8 <delta> )
6078 declare i16 @llvm.atomic.load.sub.i16.p0i32( i16* <ptr>, i16 <delta> )
6079 declare i32 @llvm.atomic.load.sub.i32.p0i32( i32* <ptr>, i32 <delta> )
6080 declare i64 @llvm.atomic.load.sub.i64.p0i32( i64* <ptr>, i64 <delta> )
6085 This intrinsic subtracts <tt>delta</tt> to the value stored in memory at
6086 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
6091 The intrinsic takes two arguments, the first a pointer to an integer value
6092 and the second an integer value. The result is also an integer value. These
6093 integer types can have any bit width, but they must all have the same bit
6094 width. The targets may only lower integer representations they support.
6098 This intrinsic does a series of operations atomically. It first loads the
6099 value stored at <tt>ptr</tt>. It then subtracts <tt>delta</tt>, stores the
6100 result to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
6107 %result1 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 4 )
6108 <i>; yields {i32}:result1 = 8</i>
6109 %result2 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 2 )
6110 <i>; yields {i32}:result2 = 4</i>
6111 %result3 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 5 )
6112 <i>; yields {i32}:result3 = 2</i>
6113 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = -3</i>
6117 <!-- _______________________________________________________________________ -->
6118 <div class="doc_subsubsection">
6119 <a name="int_atomic_load_and">'<tt>llvm.atomic.load.and.*</tt>' Intrinsic</a><br>
6120 <a name="int_atomic_load_nand">'<tt>llvm.atomic.load.nand.*</tt>' Intrinsic</a><br>
6121 <a name="int_atomic_load_or">'<tt>llvm.atomic.load.or.*</tt>' Intrinsic</a><br>
6122 <a name="int_atomic_load_xor">'<tt>llvm.atomic.load.xor.*</tt>' Intrinsic</a><br>
6125 <div class="doc_text">
6128 These are overloaded intrinsics. You can use <tt>llvm.atomic.load_and</tt>,
6129 <tt>llvm.atomic.load_nand</tt>, <tt>llvm.atomic.load_or</tt>, and
6130 <tt>llvm.atomic.load_xor</tt> on any integer bit width and for different
6131 address spaces. Not all targets support all bit widths however.</p>
6133 declare i8 @llvm.atomic.load.and.i8.p0i8( i8* <ptr>, i8 <delta> )
6134 declare i16 @llvm.atomic.load.and.i16.p0i16( i16* <ptr>, i16 <delta> )
6135 declare i32 @llvm.atomic.load.and.i32.p0i32( i32* <ptr>, i32 <delta> )
6136 declare i64 @llvm.atomic.load.and.i64.p0i64( i64* <ptr>, i64 <delta> )
6141 declare i8 @llvm.atomic.load.or.i8.p0i8( i8* <ptr>, i8 <delta> )
6142 declare i16 @llvm.atomic.load.or.i16.p0i16( i16* <ptr>, i16 <delta> )
6143 declare i32 @llvm.atomic.load.or.i32.p0i32( i32* <ptr>, i32 <delta> )
6144 declare i64 @llvm.atomic.load.or.i64.p0i64( i64* <ptr>, i64 <delta> )
6149 declare i8 @llvm.atomic.load.nand.i8.p0i32( i8* <ptr>, i8 <delta> )
6150 declare i16 @llvm.atomic.load.nand.i16.p0i32( i16* <ptr>, i16 <delta> )
6151 declare i32 @llvm.atomic.load.nand.i32.p0i32( i32* <ptr>, i32 <delta> )
6152 declare i64 @llvm.atomic.load.nand.i64.p0i32( i64* <ptr>, i64 <delta> )
6157 declare i8 @llvm.atomic.load.xor.i8.p0i32( i8* <ptr>, i8 <delta> )
6158 declare i16 @llvm.atomic.load.xor.i16.p0i32( i16* <ptr>, i16 <delta> )
6159 declare i32 @llvm.atomic.load.xor.i32.p0i32( i32* <ptr>, i32 <delta> )
6160 declare i64 @llvm.atomic.load.xor.i64.p0i32( i64* <ptr>, i64 <delta> )
6165 These intrinsics bitwise the operation (and, nand, or, xor) <tt>delta</tt> to
6166 the value stored in memory at <tt>ptr</tt>. It yields the original value
6172 These intrinsics take two arguments, the first a pointer to an integer value
6173 and the second an integer value. The result is also an integer value. These
6174 integer types can have any bit width, but they must all have the same bit
6175 width. The targets may only lower integer representations they support.
6179 These intrinsics does a series of operations atomically. They first load the
6180 value stored at <tt>ptr</tt>. They then do the bitwise operation
6181 <tt>delta</tt>, store the result to <tt>ptr</tt>. They yield the original
6182 value stored at <tt>ptr</tt>.
6188 store i32 0x0F0F, %ptr
6189 %result0 = call i32 @llvm.atomic.load.nand.i32.p0i32( i32* %ptr, i32 0xFF )
6190 <i>; yields {i32}:result0 = 0x0F0F</i>
6191 %result1 = call i32 @llvm.atomic.load.and.i32.p0i32( i32* %ptr, i32 0xFF )
6192 <i>; yields {i32}:result1 = 0xFFFFFFF0</i>
6193 %result2 = call i32 @llvm.atomic.load.or.i32.p0i32( i32* %ptr, i32 0F )
6194 <i>; yields {i32}:result2 = 0xF0</i>
6195 %result3 = call i32 @llvm.atomic.load.xor.i32.p0i32( i32* %ptr, i32 0F )
6196 <i>; yields {i32}:result3 = FF</i>
6197 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = F0</i>
6202 <!-- _______________________________________________________________________ -->
6203 <div class="doc_subsubsection">
6204 <a name="int_atomic_load_max">'<tt>llvm.atomic.load.max.*</tt>' Intrinsic</a><br>
6205 <a name="int_atomic_load_min">'<tt>llvm.atomic.load.min.*</tt>' Intrinsic</a><br>
6206 <a name="int_atomic_load_umax">'<tt>llvm.atomic.load.umax.*</tt>' Intrinsic</a><br>
6207 <a name="int_atomic_load_umin">'<tt>llvm.atomic.load.umin.*</tt>' Intrinsic</a><br>
6210 <div class="doc_text">
6213 These are overloaded intrinsics. You can use <tt>llvm.atomic.load_max</tt>,
6214 <tt>llvm.atomic.load_min</tt>, <tt>llvm.atomic.load_umax</tt>, and
6215 <tt>llvm.atomic.load_umin</tt> on any integer bit width and for different
6216 address spaces. Not all targets
6217 support all bit widths however.</p>
6219 declare i8 @llvm.atomic.load.max.i8.p0i8( i8* <ptr>, i8 <delta> )
6220 declare i16 @llvm.atomic.load.max.i16.p0i16( i16* <ptr>, i16 <delta> )
6221 declare i32 @llvm.atomic.load.max.i32.p0i32( i32* <ptr>, i32 <delta> )
6222 declare i64 @llvm.atomic.load.max.i64.p0i64( i64* <ptr>, i64 <delta> )
6227 declare i8 @llvm.atomic.load.min.i8.p0i8( i8* <ptr>, i8 <delta> )
6228 declare i16 @llvm.atomic.load.min.i16.p0i16( i16* <ptr>, i16 <delta> )
6229 declare i32 @llvm.atomic.load.min.i32..p0i32( i32* <ptr>, i32 <delta> )
6230 declare i64 @llvm.atomic.load.min.i64..p0i64( i64* <ptr>, i64 <delta> )
6235 declare i8 @llvm.atomic.load.umax.i8.p0i8( i8* <ptr>, i8 <delta> )
6236 declare i16 @llvm.atomic.load.umax.i16.p0i16( i16* <ptr>, i16 <delta> )
6237 declare i32 @llvm.atomic.load.umax.i32.p0i32( i32* <ptr>, i32 <delta> )
6238 declare i64 @llvm.atomic.load.umax.i64.p0i64( i64* <ptr>, i64 <delta> )
6243 declare i8 @llvm.atomic.load.umin.i8..p0i8( i8* <ptr>, i8 <delta> )
6244 declare i16 @llvm.atomic.load.umin.i16.p0i16( i16* <ptr>, i16 <delta> )
6245 declare i32 @llvm.atomic.load.umin.i32..p0i32( i32* <ptr>, i32 <delta> )
6246 declare i64 @llvm.atomic.load.umin.i64..p0i64( i64* <ptr>, i64 <delta> )
6251 These intrinsics takes the signed or unsigned minimum or maximum of
6252 <tt>delta</tt> and the value stored in memory at <tt>ptr</tt>. It yields the
6253 original value at <tt>ptr</tt>.
6258 These intrinsics take two arguments, the first a pointer to an integer value
6259 and the second an integer value. The result is also an integer value. These
6260 integer types can have any bit width, but they must all have the same bit
6261 width. The targets may only lower integer representations they support.
6265 These intrinsics does a series of operations atomically. They first load the
6266 value stored at <tt>ptr</tt>. They then do the signed or unsigned min or max
6267 <tt>delta</tt> and the value, store the result to <tt>ptr</tt>. They yield
6268 the original value stored at <tt>ptr</tt>.
6275 %result0 = call i32 @llvm.atomic.load.min.i32.p0i32( i32* %ptr, i32 -2 )
6276 <i>; yields {i32}:result0 = 7</i>
6277 %result1 = call i32 @llvm.atomic.load.max.i32.p0i32( i32* %ptr, i32 8 )
6278 <i>; yields {i32}:result1 = -2</i>
6279 %result2 = call i32 @llvm.atomic.load.umin.i32.p0i32( i32* %ptr, i32 10 )
6280 <i>; yields {i32}:result2 = 8</i>
6281 %result3 = call i32 @llvm.atomic.load.umax.i32.p0i32( i32* %ptr, i32 30 )
6282 <i>; yields {i32}:result3 = 8</i>
6283 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 30</i>
6287 <!-- ======================================================================= -->
6288 <div class="doc_subsection">
6289 <a name="int_general">General Intrinsics</a>
6292 <div class="doc_text">
6293 <p> This class of intrinsics is designed to be generic and has
6294 no specific purpose. </p>
6297 <!-- _______________________________________________________________________ -->
6298 <div class="doc_subsubsection">
6299 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
6302 <div class="doc_text">
6306 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
6312 The '<tt>llvm.var.annotation</tt>' intrinsic
6318 The first argument is a pointer to a value, the second is a pointer to a
6319 global string, the third is a pointer to a global string which is the source
6320 file name, and the last argument is the line number.
6326 This intrinsic allows annotation of local variables with arbitrary strings.
6327 This can be useful for special purpose optimizations that want to look for these
6328 annotations. These have no other defined use, they are ignored by code
6329 generation and optimization.
6333 <!-- _______________________________________________________________________ -->
6334 <div class="doc_subsubsection">
6335 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
6338 <div class="doc_text">
6341 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
6342 any integer bit width.
6345 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
6346 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
6347 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
6348 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
6349 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
6355 The '<tt>llvm.annotation</tt>' intrinsic.
6361 The first argument is an integer value (result of some expression),
6362 the second is a pointer to a global string, the third is a pointer to a global
6363 string which is the source file name, and the last argument is the line number.
6364 It returns the value of the first argument.
6370 This intrinsic allows annotations to be put on arbitrary expressions
6371 with arbitrary strings. This can be useful for special purpose optimizations
6372 that want to look for these annotations. These have no other defined use, they
6373 are ignored by code generation and optimization.
6377 <!-- _______________________________________________________________________ -->
6378 <div class="doc_subsubsection">
6379 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
6382 <div class="doc_text">
6386 declare void @llvm.trap()
6392 The '<tt>llvm.trap</tt>' intrinsic
6404 This intrinsics is lowered to the target dependent trap instruction. If the
6405 target does not have a trap instruction, this intrinsic will be lowered to the
6406 call of the abort() function.
6410 <!-- *********************************************************************** -->
6413 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
6414 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
6415 <a href="http://validator.w3.org/check/referer"><img
6416 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
6418 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
6419 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
6420 Last modified: $Date$