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>
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>
62 <li><a href="#aggregateconstants">Aggregate Constants</a>
63 <li><a href="#globalconstants">Global Variable and Function Addresses</a>
64 <li><a href="#undefvalues">Undefined Values</a>
65 <li><a href="#constantexprs">Constant Expressions</a>
68 <li><a href="#othervalues">Other Values</a>
70 <li><a href="#inlineasm">Inline Assembler Expressions</a>
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>
146 <li><a href="#otherops">Other Operations</a>
148 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
149 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
150 <li><a href="#i_vicmp">'<tt>vicmp</tt>' Instruction</a></li>
151 <li><a href="#i_vfcmp">'<tt>vfcmp</tt>' Instruction</a></li>
152 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
153 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
154 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
155 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
156 <li><a href="#i_getresult">'<tt>getresult</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:
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. In this way, anything except a <tt>"</tt> character can
339 be used in a named value.</li>
341 <li>Unnamed values are represented as an unsigned numeric value with their
342 prefix. For example, %12, @2, %44.</li>
344 <li>Constants, which are described in a <a href="#constants">section about
345 constants</a>, below.</li>
348 <p>LLVM requires that values start with a prefix for two reasons: Compilers
349 don't need to worry about name clashes with reserved words, and the set of
350 reserved words may be expanded in the future without penalty. Additionally,
351 unnamed identifiers allow a compiler to quickly come up with a temporary
352 variable without having to avoid symbol table conflicts.</p>
354 <p>Reserved words in LLVM are very similar to reserved words in other
355 languages. There are keywords for different opcodes
356 ('<tt><a href="#i_add">add</a></tt>',
357 '<tt><a href="#i_bitcast">bitcast</a></tt>',
358 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
359 href="#t_void">void</a></tt>', '<tt><a href="#t_primitive">i32</a></tt>', etc...),
360 and others. These reserved words cannot conflict with variable names, because
361 none of them start with a prefix character ('%' or '@').</p>
363 <p>Here is an example of LLVM code to multiply the integer variable
364 '<tt>%X</tt>' by 8:</p>
368 <div class="doc_code">
370 %result = <a href="#i_mul">mul</a> i32 %X, 8
374 <p>After strength reduction:</p>
376 <div class="doc_code">
378 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
382 <p>And the hard way:</p>
384 <div class="doc_code">
386 <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
387 <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
388 %result = <a href="#i_add">add</a> i32 %1, %1
392 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
393 important lexical features of LLVM:</p>
397 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
400 <li>Unnamed temporaries are created when the result of a computation is not
401 assigned to a named value.</li>
403 <li>Unnamed temporaries are numbered sequentially</li>
407 <p>...and it also shows a convention that we follow in this document. When
408 demonstrating instructions, we will follow an instruction with a comment that
409 defines the type and name of value produced. Comments are shown in italic
414 <!-- *********************************************************************** -->
415 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
416 <!-- *********************************************************************** -->
418 <!-- ======================================================================= -->
419 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
422 <div class="doc_text">
424 <p>LLVM programs are composed of "Module"s, each of which is a
425 translation unit of the input programs. Each module consists of
426 functions, global variables, and symbol table entries. Modules may be
427 combined together with the LLVM linker, which merges function (and
428 global variable) definitions, resolves forward declarations, and merges
429 symbol table entries. Here is an example of the "hello world" module:</p>
431 <div class="doc_code">
432 <pre><i>; Declare the string constant as a global constant...</i>
433 <a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a
434 href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
436 <i>; External declaration of the puts function</i>
437 <a href="#functionstructure">declare</a> i32 @puts(i8 *) <i>; i32(i8 *)* </i>
439 <i>; Definition of main function</i>
440 define i32 @main() { <i>; i32()* </i>
441 <i>; Convert [13x i8 ]* to i8 *...</i>
443 href="#i_getelementptr">getelementptr</a> [13 x i8 ]* @.LC0, i64 0, i64 0 <i>; i8 *</i>
445 <i>; Call puts function to write out the string to stdout...</i>
447 href="#i_call">call</a> i32 @puts(i8 * %cast210) <i>; i32</i>
449 href="#i_ret">ret</a> i32 0<br>}<br>
453 <p>This example is made up of a <a href="#globalvars">global variable</a>
454 named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
455 function, and a <a href="#functionstructure">function definition</a>
456 for "<tt>main</tt>".</p>
458 <p>In general, a module is made up of a list of global values,
459 where both functions and global variables are global values. Global values are
460 represented by a pointer to a memory location (in this case, a pointer to an
461 array of char, and a pointer to a function), and have one of the following <a
462 href="#linkage">linkage types</a>.</p>
466 <!-- ======================================================================= -->
467 <div class="doc_subsection">
468 <a name="linkage">Linkage Types</a>
471 <div class="doc_text">
474 All Global Variables and Functions have one of the following types of linkage:
479 <dt><tt><b><a name="linkage_internal">internal</a></b></tt>: </dt>
481 <dd>Global values with internal linkage are only directly accessible by
482 objects in the current module. In particular, linking code into a module with
483 an internal global value may cause the internal to be renamed as necessary to
484 avoid collisions. Because the symbol is internal to the module, all
485 references can be updated. This corresponds to the notion of the
486 '<tt>static</tt>' keyword in C.
489 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
491 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
492 the same name when linkage occurs. This is typically used to implement
493 inline functions, templates, or other code which must be generated in each
494 translation unit that uses it. Unreferenced <tt>linkonce</tt> globals are
495 allowed to be discarded.
498 <dt><tt><b><a name="linkage_common">common</a></b></tt>: </dt>
500 <dd>"<tt>common</tt>" linkage is exactly the same as <tt>linkonce</tt>
501 linkage, except that unreferenced <tt>common</tt> globals may not be
502 discarded. This is used for globals that may be emitted in multiple
503 translation units, but that are not guaranteed to be emitted into every
504 translation unit that uses them. One example of this is tentative
505 definitions in C, such as "<tt>int X;</tt>" at global scope.
508 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
510 <dd>"<tt>weak</tt>" linkage is the same as <tt>common</tt> linkage, except
511 that some targets may choose to emit different assembly sequences for them
512 for target-dependent reasons. This is used for globals that are declared
513 "weak" in C source code.
516 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
518 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
519 pointer to array type. When two global variables with appending linkage are
520 linked together, the two global arrays are appended together. This is the
521 LLVM, typesafe, equivalent of having the system linker append together
522 "sections" with identical names when .o files are linked.
525 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt>: </dt>
526 <dd>The semantics of this linkage follow the ELF object file model: the
527 symbol is weak until linked, if not linked, the symbol becomes null instead
528 of being an undefined reference.
531 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
533 <dd>If none of the above identifiers are used, the global is externally
534 visible, meaning that it participates in linkage and can be used to resolve
535 external symbol references.
540 The next two types of linkage are targeted for Microsoft Windows platform
541 only. They are designed to support importing (exporting) symbols from (to)
542 DLLs (Dynamic Link Libraries).
546 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt>: </dt>
548 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
549 or variable via a global pointer to a pointer that is set up by the DLL
550 exporting the symbol. On Microsoft Windows targets, the pointer name is
551 formed by combining <code>_imp__</code> and the function or variable name.
554 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt>: </dt>
556 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
557 pointer to a pointer in a DLL, so that it can be referenced with the
558 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
559 name is formed by combining <code>_imp__</code> and the function or variable
565 <p><a name="linkage_external"></a>For example, since the "<tt>.LC0</tt>"
566 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
567 variable and was linked with this one, one of the two would be renamed,
568 preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
569 external (i.e., lacking any linkage declarations), they are accessible
570 outside of the current module.</p>
571 <p>It is illegal for a function <i>declaration</i>
572 to have any linkage type other than "externally visible", <tt>dllimport</tt>,
573 or <tt>extern_weak</tt>.</p>
574 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt> and <tt>weak</tt>
578 <!-- ======================================================================= -->
579 <div class="doc_subsection">
580 <a name="callingconv">Calling Conventions</a>
583 <div class="doc_text">
585 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
586 and <a href="#i_invoke">invokes</a> can all have an optional calling convention
587 specified for the call. The calling convention of any pair of dynamic
588 caller/callee must match, or the behavior of the program is undefined. The
589 following calling conventions are supported by LLVM, and more may be added in
593 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
595 <dd>This calling convention (the default if no other calling convention is
596 specified) matches the target C calling conventions. This calling convention
597 supports varargs function calls and tolerates some mismatch in the declared
598 prototype and implemented declaration of the function (as does normal C).
601 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
603 <dd>This calling convention attempts to make calls as fast as possible
604 (e.g. by passing things in registers). This calling convention allows the
605 target to use whatever tricks it wants to produce fast code for the target,
606 without having to conform to an externally specified ABI (Application Binary
607 Interface). Implementations of this convention should allow arbitrary
608 <a href="CodeGenerator.html#tailcallopt">tail call optimization</a> to be
609 supported. This calling convention does not support varargs and requires the
610 prototype of all callees to exactly match the prototype of the function
614 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
616 <dd>This calling convention attempts to make code in the caller as efficient
617 as possible under the assumption that the call is not commonly executed. As
618 such, these calls often preserve all registers so that the call does not break
619 any live ranges in the caller side. This calling convention does not support
620 varargs and requires the prototype of all callees to exactly match the
621 prototype of the function definition.
624 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
626 <dd>Any calling convention may be specified by number, allowing
627 target-specific calling conventions to be used. Target specific calling
628 conventions start at 64.
632 <p>More calling conventions can be added/defined on an as-needed basis, to
633 support pascal conventions or any other well-known target-independent
638 <!-- ======================================================================= -->
639 <div class="doc_subsection">
640 <a name="visibility">Visibility Styles</a>
643 <div class="doc_text">
646 All Global Variables and Functions have one of the following visibility styles:
650 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
652 <dd>On targets that use the ELF object file format, default visibility means
653 that the declaration is visible to other
654 modules and, in shared libraries, means that the declared entity may be
655 overridden. On Darwin, default visibility means that the declaration is
656 visible to other modules. Default visibility corresponds to "external
657 linkage" in the language.
660 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
662 <dd>Two declarations of an object with hidden visibility refer to the same
663 object if they are in the same shared object. Usually, hidden visibility
664 indicates that the symbol will not be placed into the dynamic symbol table,
665 so no other module (executable or shared library) can reference it
669 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
671 <dd>On ELF, protected visibility indicates that the symbol will be placed in
672 the dynamic symbol table, but that references within the defining module will
673 bind to the local symbol. That is, the symbol cannot be overridden by another
680 <!-- ======================================================================= -->
681 <div class="doc_subsection">
682 <a name="globalvars">Global Variables</a>
685 <div class="doc_text">
687 <p>Global variables define regions of memory allocated at compilation time
688 instead of run-time. Global variables may optionally be initialized, may have
689 an explicit section to be placed in, and may have an optional explicit alignment
690 specified. A variable may be defined as "thread_local", which means that it
691 will not be shared by threads (each thread will have a separated copy of the
692 variable). A variable may be defined as a global "constant," which indicates
693 that the contents of the variable will <b>never</b> be modified (enabling better
694 optimization, allowing the global data to be placed in the read-only section of
695 an executable, etc). Note that variables that need runtime initialization
696 cannot be marked "constant" as there is a store to the variable.</p>
699 LLVM explicitly allows <em>declarations</em> of global variables to be marked
700 constant, even if the final definition of the global is not. This capability
701 can be used to enable slightly better optimization of the program, but requires
702 the language definition to guarantee that optimizations based on the
703 'constantness' are valid for the translation units that do not include the
707 <p>As SSA values, global variables define pointer values that are in
708 scope (i.e. they dominate) all basic blocks in the program. Global
709 variables always define a pointer to their "content" type because they
710 describe a region of memory, and all memory objects in LLVM are
711 accessed through pointers.</p>
713 <p>A global variable may be declared to reside in a target-specifc numbered
714 address space. For targets that support them, address spaces may affect how
715 optimizations are performed and/or what target instructions are used to access
716 the variable. The default address space is zero. The address space qualifier
717 must precede any other attributes.</p>
719 <p>LLVM allows an explicit section to be specified for globals. If the target
720 supports it, it will emit globals to the section specified.</p>
722 <p>An explicit alignment may be specified for a global. If not present, or if
723 the alignment is set to zero, the alignment of the global is set by the target
724 to whatever it feels convenient. If an explicit alignment is specified, the
725 global is forced to have at least that much alignment. All alignments must be
728 <p>For example, the following defines a global in a numbered address space with
729 an initializer, section, and alignment:</p>
731 <div class="doc_code">
733 @G = constant float 1.0 addrspace(5), section "foo", align 4
740 <!-- ======================================================================= -->
741 <div class="doc_subsection">
742 <a name="functionstructure">Functions</a>
745 <div class="doc_text">
747 <p>LLVM function definitions consist of the "<tt>define</tt>" keyord,
748 an optional <a href="#linkage">linkage type</a>, an optional
749 <a href="#visibility">visibility style</a>, an optional
750 <a href="#callingconv">calling convention</a>, a return type, an optional
751 <a href="#paramattrs">parameter attribute</a> for the return type, a function
752 name, a (possibly empty) argument list (each with optional
753 <a href="#paramattrs">parameter attributes</a>), an optional section, an
754 optional alignment, an optional <a href="#gc">garbage collector name</a>,
755 an opening curly brace, a list of basic blocks, and a closing curly brace.
757 LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
758 optional <a href="#linkage">linkage type</a>, an optional
759 <a href="#visibility">visibility style</a>, an optional
760 <a href="#callingconv">calling convention</a>, a return type, an optional
761 <a href="#paramattrs">parameter attribute</a> for the return type, a function
762 name, a possibly empty list of arguments, an optional alignment, and an optional
763 <a href="#gc">garbage collector name</a>.</p>
765 <p>A function definition contains a list of basic blocks, forming the CFG
766 (Control Flow Graph) for
767 the function. Each basic block may optionally start with a label (giving the
768 basic block a symbol table entry), contains a list of instructions, and ends
769 with a <a href="#terminators">terminator</a> instruction (such as a branch or
770 function return).</p>
772 <p>The first basic block in a function is special in two ways: it is immediately
773 executed on entrance to the function, and it is not allowed to have predecessor
774 basic blocks (i.e. there can not be any branches to the entry block of a
775 function). Because the block can have no predecessors, it also cannot have any
776 <a href="#i_phi">PHI nodes</a>.</p>
778 <p>LLVM allows an explicit section to be specified for functions. If the target
779 supports it, it will emit functions to the section specified.</p>
781 <p>An explicit alignment may be specified for a function. If not present, or if
782 the alignment is set to zero, the alignment of the function is set by the target
783 to whatever it feels convenient. If an explicit alignment is specified, the
784 function is forced to have at least that much alignment. All alignments must be
790 <!-- ======================================================================= -->
791 <div class="doc_subsection">
792 <a name="aliasstructure">Aliases</a>
794 <div class="doc_text">
795 <p>Aliases act as "second name" for the aliasee value (which can be either
796 function, global variable, another alias or bitcast of global value). Aliases
797 may have an optional <a href="#linkage">linkage type</a>, and an
798 optional <a href="#visibility">visibility style</a>.</p>
802 <div class="doc_code">
804 @<Name> = alias [Linkage] [Visibility] <AliaseeTy> @<Aliasee>
812 <!-- ======================================================================= -->
813 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
814 <div class="doc_text">
815 <p>The return type and each parameter of a function type may have a set of
816 <i>parameter attributes</i> associated with them. Parameter attributes are
817 used to communicate additional information about the result or parameters of
818 a function. Parameter attributes are considered to be part of the function,
819 not of the function type, so functions with different parameter attributes
820 can have the same function type.</p>
822 <p>Parameter attributes are simple keywords that follow the type specified. If
823 multiple parameter attributes are needed, they are space separated. For
826 <div class="doc_code">
828 declare i32 @printf(i8* noalias , ...)
829 declare i32 @atoi(i8 zeroext*)
833 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
834 <tt>readonly</tt>) come immediately after the argument list.</p>
836 <p>Currently, only the following parameter attributes are defined:</p>
838 <dt><tt>zeroext</tt></dt>
839 <dd>This indicates that the parameter should be zero extended just before
840 a call to this function.</dd>
842 <dt><tt>signext</tt></dt>
843 <dd>This indicates that the parameter should be sign extended just before
844 a call to this function.</dd>
846 <dt><tt>inreg</tt></dt>
847 <dd>This indicates that this parameter or return value should be treated
848 in a special target-dependent fashion during while emitting code for a
849 function call or return (usually, by putting it in a register as opposed
850 to memory; in some places it is used to distinguish between two different
851 kinds of registers). Use of this attribute is target-specific</dd>
853 <dt><tt>byval</tt></dt>
854 <dd>This indicates that the pointer parameter should really be passed by
855 value to the function. The attribute implies that a hidden copy of the
856 pointee is made between the caller and the callee, so the callee is unable
857 to modify the value in the callee. This attribute is only valid on LLVM
858 pointer arguments. It is generally used to pass structs and arrays by
859 value, but is also valid on scalars (even though this is silly).</dd>
861 <dt><tt>sret</tt></dt>
862 <dd>This indicates that the pointer parameter specifies the address of a
863 structure that is the return value of the function in the source program.
864 Loads and stores to the structure are assumed not to trap.
865 May only be applied to the first parameter.</dd>
867 <dt><tt>noalias</tt></dt>
868 <dd>This indicates that the parameter does not alias any global or any other
869 parameter. The caller is responsible for ensuring that this is the case,
870 usually by placing the value in a stack allocation.</dd>
872 <dt><tt>nest</tt></dt>
873 <dd>This indicates that the pointer parameter can be excised using the
874 <a href="#int_trampoline">trampoline intrinsics</a>.</dd>
879 <!-- ======================================================================= -->
880 <div class="doc_subsection">
881 <a name="gc">Garbage Collector Names</a>
884 <div class="doc_text">
885 <p>Each function may specify a garbage collector name, which is simply a
888 <div class="doc_code"><pre
889 >define void @f() gc "name" { ...</pre></div>
891 <p>The compiler declares the supported values of <i>name</i>. Specifying a
892 collector which will cause the compiler to alter its output in order to support
893 the named garbage collection algorithm.</p>
896 <!-- ======================================================================= -->
897 <div class="doc_subsection">
898 <a name="fnattrs">Function Attributes</a>
901 <div class="doc_text">
903 <p>Function attributes are set to communicate additional information about
904 a function. Function attributes are considered to be part of the function,
905 not of the function type, so functions with different parameter attributes
906 can have the same function type.</p>
908 <p>Function attributes are simple keywords that follow the type specified. If
909 multiple attributes are needed, they are space separated. For
912 <div class="doc_code">
914 define void @f() noinline { ... }
915 define void @f() alwaysinline { ... }
916 define void @f() alwaysinline optsize { ... }
917 define void @f() optsize
922 <dt><tt>alwaysinline</tt></dt>
923 <dd>This attribute indicates that the inliner should attempt to inline this
924 function into callers whenever possible, ignoring any active inlining size
925 threshold for this caller.</dd>
927 <dt><tt>noinline</tt></dt>
928 <dd>This attribute indicates that the inliner should never inline this function
929 in any situation. This attribute may not be used together with
930 <tt>alwaysinline</tt> attribute.</dd>
932 <dt><tt>optsize</tt></dt>
933 <dd>This attribute suggests that optimization passes and code generator passes
934 make choices that keep the code size of this function low, and otherwise do
935 optimizations specifically to reduce code size.</dd>
937 <dt><tt>noreturn</tt></dt>
938 <dd>This function attribute indicates that the function never returns normally.
939 This produces undefined behavior at runtime if the function ever does
940 dynamically return.</dd>
942 <dt><tt>nounwind</tt></dt>
943 <dd>This function attribute indicates that the function never returns with an
944 unwind or exceptional control flow. If the function does unwind, its runtime
945 behavior is undefined.</dd>
947 <dt><tt>readnone</tt></dt>
948 <dd>This attribute indicates that the function computes its result (or its
949 thrown exception) based strictly on its arguments. It does not read any global
950 mutable state (e.g. memory, control registers, etc) visible to caller functions.
951 Furthermore, <tt>readnone</tt> functions never change any state visible to their
954 <dt><tt>readonly</tt></dt>
955 <dd>This function attribute indicates that the function has no side-effects on
956 the calling function, but that it depends on state (memory state, control
957 register state, etc) that may be set in the caller. A readonly function always
958 returns the same value (or throws the same exception) whenever it is called with
959 a particular set of arguments and global state.</dd>
965 <!-- ======================================================================= -->
966 <div class="doc_subsection">
967 <a name="moduleasm">Module-Level Inline Assembly</a>
970 <div class="doc_text">
972 Modules may contain "module-level inline asm" blocks, which corresponds to the
973 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
974 LLVM and treated as a single unit, but may be separated in the .ll file if
975 desired. The syntax is very simple:
978 <div class="doc_code">
980 module asm "inline asm code goes here"
981 module asm "more can go here"
985 <p>The strings can contain any character by escaping non-printable characters.
986 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
991 The inline asm code is simply printed to the machine code .s file when
992 assembly code is generated.
996 <!-- ======================================================================= -->
997 <div class="doc_subsection">
998 <a name="datalayout">Data Layout</a>
1001 <div class="doc_text">
1002 <p>A module may specify a target specific data layout string that specifies how
1003 data is to be laid out in memory. The syntax for the data layout is simply:</p>
1004 <pre> target datalayout = "<i>layout specification</i>"</pre>
1005 <p>The <i>layout specification</i> consists of a list of specifications
1006 separated by the minus sign character ('-'). Each specification starts with a
1007 letter and may include other information after the letter to define some
1008 aspect of the data layout. The specifications accepted are as follows: </p>
1011 <dd>Specifies that the target lays out data in big-endian form. That is, the
1012 bits with the most significance have the lowest address location.</dd>
1014 <dd>Specifies that the target lays out data in little-endian form. That is,
1015 the bits with the least significance have the lowest address location.</dd>
1016 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1017 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
1018 <i>preferred</i> alignments. All sizes are in bits. Specifying the <i>pref</i>
1019 alignment is optional. If omitted, the preceding <tt>:</tt> should be omitted
1021 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1022 <dd>This specifies the alignment for an integer type of a given bit
1023 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
1024 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1025 <dd>This specifies the alignment for a vector type of a given bit
1027 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1028 <dd>This specifies the alignment for a floating point type of a given bit
1029 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
1031 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
1032 <dd>This specifies the alignment for an aggregate type of a given bit
1035 <p>When constructing the data layout for a given target, LLVM starts with a
1036 default set of specifications which are then (possibly) overriden by the
1037 specifications in the <tt>datalayout</tt> keyword. The default specifications
1038 are given in this list:</p>
1040 <li><tt>E</tt> - big endian</li>
1041 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
1042 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
1043 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
1044 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
1045 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
1046 <li><tt>i64:32:64</tt> - i64 has ABI alignment of 32-bits but preferred
1047 alignment of 64-bits</li>
1048 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
1049 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
1050 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
1051 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
1052 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
1054 <p>When LLVM is determining the alignment for a given type, it uses the
1057 <li>If the type sought is an exact match for one of the specifications, that
1058 specification is used.</li>
1059 <li>If no match is found, and the type sought is an integer type, then the
1060 smallest integer type that is larger than the bitwidth of the sought type is
1061 used. If none of the specifications are larger than the bitwidth then the the
1062 largest integer type is used. For example, given the default specifications
1063 above, the i7 type will use the alignment of i8 (next largest) while both
1064 i65 and i256 will use the alignment of i64 (largest specified).</li>
1065 <li>If no match is found, and the type sought is a vector type, then the
1066 largest vector type that is smaller than the sought vector type will be used
1067 as a fall back. This happens because <128 x double> can be implemented in
1068 terms of 64 <2 x double>, for example.</li>
1072 <!-- *********************************************************************** -->
1073 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
1074 <!-- *********************************************************************** -->
1076 <div class="doc_text">
1078 <p>The LLVM type system is one of the most important features of the
1079 intermediate representation. Being typed enables a number of
1080 optimizations to be performed on the intermediate representation directly,
1081 without having to do
1082 extra analyses on the side before the transformation. A strong type
1083 system makes it easier to read the generated code and enables novel
1084 analyses and transformations that are not feasible to perform on normal
1085 three address code representations.</p>
1089 <!-- ======================================================================= -->
1090 <div class="doc_subsection"> <a name="t_classifications">Type
1091 Classifications</a> </div>
1092 <div class="doc_text">
1093 <p>The types fall into a few useful
1094 classifications:</p>
1096 <table border="1" cellspacing="0" cellpadding="4">
1098 <tr><th>Classification</th><th>Types</th></tr>
1100 <td><a href="#t_integer">integer</a></td>
1101 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1104 <td><a href="#t_floating">floating point</a></td>
1105 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1108 <td><a name="t_firstclass">first class</a></td>
1109 <td><a href="#t_integer">integer</a>,
1110 <a href="#t_floating">floating point</a>,
1111 <a href="#t_pointer">pointer</a>,
1112 <a href="#t_vector">vector</a>,
1113 <a href="#t_struct">structure</a>,
1114 <a href="#t_array">array</a>,
1115 <a href="#t_label">label</a>.
1119 <td><a href="#t_primitive">primitive</a></td>
1120 <td><a href="#t_label">label</a>,
1121 <a href="#t_void">void</a>,
1122 <a href="#t_floating">floating point</a>.</td>
1125 <td><a href="#t_derived">derived</a></td>
1126 <td><a href="#t_integer">integer</a>,
1127 <a href="#t_array">array</a>,
1128 <a href="#t_function">function</a>,
1129 <a href="#t_pointer">pointer</a>,
1130 <a href="#t_struct">structure</a>,
1131 <a href="#t_pstruct">packed structure</a>,
1132 <a href="#t_vector">vector</a>,
1133 <a href="#t_opaque">opaque</a>.
1138 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
1139 most important. Values of these types are the only ones which can be
1140 produced by instructions, passed as arguments, or used as operands to
1144 <!-- ======================================================================= -->
1145 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
1147 <div class="doc_text">
1148 <p>The primitive types are the fundamental building blocks of the LLVM
1153 <!-- _______________________________________________________________________ -->
1154 <div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1156 <div class="doc_text">
1159 <tr><th>Type</th><th>Description</th></tr>
1160 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1161 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1162 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1163 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1164 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1169 <!-- _______________________________________________________________________ -->
1170 <div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1172 <div class="doc_text">
1174 <p>The void type does not represent any value and has no size.</p>
1183 <!-- _______________________________________________________________________ -->
1184 <div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1186 <div class="doc_text">
1188 <p>The label type represents code labels.</p>
1198 <!-- ======================================================================= -->
1199 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1201 <div class="doc_text">
1203 <p>The real power in LLVM comes from the derived types in the system.
1204 This is what allows a programmer to represent arrays, functions,
1205 pointers, and other useful types. Note that these derived types may be
1206 recursive: For example, it is possible to have a two dimensional array.</p>
1210 <!-- _______________________________________________________________________ -->
1211 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1213 <div class="doc_text">
1216 <p>The integer type is a very simple derived type that simply specifies an
1217 arbitrary bit width for the integer type desired. Any bit width from 1 bit to
1218 2^23-1 (about 8 million) can be specified.</p>
1226 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1230 <table class="layout">
1233 <td><tt>i1</tt></td>
1234 <td>a single-bit integer.</td>
1236 <td><tt>i32</tt></td>
1237 <td>a 32-bit integer.</td>
1239 <td><tt>i1942652</tt></td>
1240 <td>a really big integer of over 1 million bits.</td>
1246 <!-- _______________________________________________________________________ -->
1247 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1249 <div class="doc_text">
1253 <p>The array type is a very simple derived type that arranges elements
1254 sequentially in memory. The array type requires a size (number of
1255 elements) and an underlying data type.</p>
1260 [<# elements> x <elementtype>]
1263 <p>The number of elements is a constant integer value; elementtype may
1264 be any type with a size.</p>
1267 <table class="layout">
1269 <td class="left"><tt>[40 x i32]</tt></td>
1270 <td class="left">Array of 40 32-bit integer values.</td>
1273 <td class="left"><tt>[41 x i32]</tt></td>
1274 <td class="left">Array of 41 32-bit integer values.</td>
1277 <td class="left"><tt>[4 x i8]</tt></td>
1278 <td class="left">Array of 4 8-bit integer values.</td>
1281 <p>Here are some examples of multidimensional arrays:</p>
1282 <table class="layout">
1284 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1285 <td class="left">3x4 array of 32-bit integer values.</td>
1288 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1289 <td class="left">12x10 array of single precision floating point values.</td>
1292 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1293 <td class="left">2x3x4 array of 16-bit integer values.</td>
1297 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1298 length array. Normally, accesses past the end of an array are undefined in
1299 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
1300 As a special case, however, zero length arrays are recognized to be variable
1301 length. This allows implementation of 'pascal style arrays' with the LLVM
1302 type "{ i32, [0 x float]}", for example.</p>
1306 <!-- _______________________________________________________________________ -->
1307 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1308 <div class="doc_text">
1312 <p>The function type can be thought of as a function signature. It
1313 consists of a return type and a list of formal parameter types. The
1314 return type of a function type is a scalar type, a void type, or a struct type.
1315 If the return type is a struct type then all struct elements must be of first
1316 class types, and the struct must have at least one element.</p>
1321 <returntype list> (<parameter list>)
1324 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1325 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1326 which indicates that the function takes a variable number of arguments.
1327 Variable argument functions can access their arguments with the <a
1328 href="#int_varargs">variable argument handling intrinsic</a> functions.
1329 '<tt><returntype list></tt>' is a comma-separated list of
1330 <a href="#t_firstclass">first class</a> type specifiers.</p>
1333 <table class="layout">
1335 <td class="left"><tt>i32 (i32)</tt></td>
1336 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1338 </tr><tr class="layout">
1339 <td class="left"><tt>float (i16 signext, i32 *) *
1341 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1342 an <tt>i16</tt> that should be sign extended and a
1343 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1346 </tr><tr class="layout">
1347 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1348 <td class="left">A vararg function that takes at least one
1349 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1350 which returns an integer. This is the signature for <tt>printf</tt> in
1353 </tr><tr class="layout">
1354 <td class="left"><tt>{i32, i32} (i32)</tt></td>
1355 <td class="left">A function taking an <tt>i32></tt>, returning two
1356 <tt> i32 </tt> values as an aggregate of type <tt>{ i32, i32 }</tt>
1362 <!-- _______________________________________________________________________ -->
1363 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1364 <div class="doc_text">
1366 <p>The structure type is used to represent a collection of data members
1367 together in memory. The packing of the field types is defined to match
1368 the ABI of the underlying processor. The elements of a structure may
1369 be any type that has a size.</p>
1370 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1371 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1372 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1375 <pre> { <type list> }<br></pre>
1377 <table class="layout">
1379 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1380 <td class="left">A triple of three <tt>i32</tt> values</td>
1381 </tr><tr class="layout">
1382 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1383 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1384 second element is a <a href="#t_pointer">pointer</a> to a
1385 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1386 an <tt>i32</tt>.</td>
1391 <!-- _______________________________________________________________________ -->
1392 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1394 <div class="doc_text">
1396 <p>The packed structure type is used to represent a collection of data members
1397 together in memory. There is no padding between fields. Further, the alignment
1398 of a packed structure is 1 byte. The elements of a packed structure may
1399 be any type that has a size.</p>
1400 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1401 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1402 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1405 <pre> < { <type list> } > <br></pre>
1407 <table class="layout">
1409 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1410 <td class="left">A triple of three <tt>i32</tt> values</td>
1411 </tr><tr class="layout">
1413 <tt>< { float, i32 (i32)* } ></tt></td>
1414 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1415 second element is a <a href="#t_pointer">pointer</a> to a
1416 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1417 an <tt>i32</tt>.</td>
1422 <!-- _______________________________________________________________________ -->
1423 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1424 <div class="doc_text">
1426 <p>As in many languages, the pointer type represents a pointer or
1427 reference to another object, which must live in memory. Pointer types may have
1428 an optional address space attribute defining the target-specific numbered
1429 address space where the pointed-to object resides. The default address space is
1432 <pre> <type> *<br></pre>
1434 <table class="layout">
1436 <td class="left"><tt>[4x i32]*</tt></td>
1437 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1438 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1441 <td class="left"><tt>i32 (i32 *) *</tt></td>
1442 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
1443 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1447 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1448 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1449 that resides in address space #5.</td>
1454 <!-- _______________________________________________________________________ -->
1455 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1456 <div class="doc_text">
1460 <p>A vector type is a simple derived type that represents a vector
1461 of elements. Vector types are used when multiple primitive data
1462 are operated in parallel using a single instruction (SIMD).
1463 A vector type requires a size (number of
1464 elements) and an underlying primitive data type. Vectors must have a power
1465 of two length (1, 2, 4, 8, 16 ...). Vector types are
1466 considered <a href="#t_firstclass">first class</a>.</p>
1471 < <# elements> x <elementtype> >
1474 <p>The number of elements is a constant integer value; elementtype may
1475 be any integer or floating point type.</p>
1479 <table class="layout">
1481 <td class="left"><tt><4 x i32></tt></td>
1482 <td class="left">Vector of 4 32-bit integer values.</td>
1485 <td class="left"><tt><8 x float></tt></td>
1486 <td class="left">Vector of 8 32-bit floating-point values.</td>
1489 <td class="left"><tt><2 x i64></tt></td>
1490 <td class="left">Vector of 2 64-bit integer values.</td>
1495 <!-- _______________________________________________________________________ -->
1496 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1497 <div class="doc_text">
1501 <p>Opaque types are used to represent unknown types in the system. This
1502 corresponds (for example) to the C notion of a forward declared structure type.
1503 In LLVM, opaque types can eventually be resolved to any type (not just a
1504 structure type).</p>
1514 <table class="layout">
1516 <td class="left"><tt>opaque</tt></td>
1517 <td class="left">An opaque type.</td>
1523 <!-- *********************************************************************** -->
1524 <div class="doc_section"> <a name="constants">Constants</a> </div>
1525 <!-- *********************************************************************** -->
1527 <div class="doc_text">
1529 <p>LLVM has several different basic types of constants. This section describes
1530 them all and their syntax.</p>
1534 <!-- ======================================================================= -->
1535 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1537 <div class="doc_text">
1540 <dt><b>Boolean constants</b></dt>
1542 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1543 constants of the <tt><a href="#t_primitive">i1</a></tt> type.
1546 <dt><b>Integer constants</b></dt>
1548 <dd>Standard integers (such as '4') are constants of the <a
1549 href="#t_integer">integer</a> type. Negative numbers may be used with
1553 <dt><b>Floating point constants</b></dt>
1555 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1556 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1557 notation (see below). The assembler requires the exact decimal value of
1558 a floating-point constant. For example, the assembler accepts 1.25 but
1559 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
1560 constants must have a <a href="#t_floating">floating point</a> type. </dd>
1562 <dt><b>Null pointer constants</b></dt>
1564 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1565 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1569 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1570 of floating point constants. For example, the form '<tt>double
1571 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1572 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1573 (and the only time that they are generated by the disassembler) is when a
1574 floating point constant must be emitted but it cannot be represented as a
1575 decimal floating point number. For example, NaN's, infinities, and other
1576 special values are represented in their IEEE hexadecimal format so that
1577 assembly and disassembly do not cause any bits to change in the constants.</p>
1581 <!-- ======================================================================= -->
1582 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1585 <div class="doc_text">
1586 <p>Aggregate constants arise from aggregation of simple constants
1587 and smaller aggregate constants.</p>
1590 <dt><b>Structure constants</b></dt>
1592 <dd>Structure constants are represented with notation similar to structure
1593 type definitions (a comma separated list of elements, surrounded by braces
1594 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
1595 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>". Structure constants
1596 must have <a href="#t_struct">structure type</a>, and the number and
1597 types of elements must match those specified by the type.
1600 <dt><b>Array constants</b></dt>
1602 <dd>Array constants are represented with notation similar to array type
1603 definitions (a comma separated list of elements, surrounded by square brackets
1604 (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74 ]</tt>". Array
1605 constants must have <a href="#t_array">array type</a>, and the number and
1606 types of elements must match those specified by the type.
1609 <dt><b>Vector constants</b></dt>
1611 <dd>Vector constants are represented with notation similar to vector type
1612 definitions (a comma separated list of elements, surrounded by
1613 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32 42,
1614 i32 11, i32 74, i32 100 ></tt>". Vector constants must have <a
1615 href="#t_vector">vector type</a>, and the number and types of elements must
1616 match those specified by the type.
1619 <dt><b>Zero initialization</b></dt>
1621 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1622 value to zero of <em>any</em> type, including scalar and aggregate types.
1623 This is often used to avoid having to print large zero initializers (e.g. for
1624 large arrays) and is always exactly equivalent to using explicit zero
1631 <!-- ======================================================================= -->
1632 <div class="doc_subsection">
1633 <a name="globalconstants">Global Variable and Function Addresses</a>
1636 <div class="doc_text">
1638 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1639 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1640 constants. These constants are explicitly referenced when the <a
1641 href="#identifiers">identifier for the global</a> is used and always have <a
1642 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1645 <div class="doc_code">
1649 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1655 <!-- ======================================================================= -->
1656 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1657 <div class="doc_text">
1658 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1659 no specific value. Undefined values may be of any type and be used anywhere
1660 a constant is permitted.</p>
1662 <p>Undefined values indicate to the compiler that the program is well defined
1663 no matter what value is used, giving the compiler more freedom to optimize.
1667 <!-- ======================================================================= -->
1668 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1671 <div class="doc_text">
1673 <p>Constant expressions are used to allow expressions involving other constants
1674 to be used as constants. Constant expressions may be of any <a
1675 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1676 that does not have side effects (e.g. load and call are not supported). The
1677 following is the syntax for constant expressions:</p>
1680 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1681 <dd>Truncate a constant to another type. The bit size of CST must be larger
1682 than the bit size of TYPE. Both types must be integers.</dd>
1684 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1685 <dd>Zero extend a constant to another type. The bit size of CST must be
1686 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1688 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1689 <dd>Sign extend a constant to another type. The bit size of CST must be
1690 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1692 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1693 <dd>Truncate a floating point constant to another floating point type. The
1694 size of CST must be larger than the size of TYPE. Both types must be
1695 floating point.</dd>
1697 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1698 <dd>Floating point extend a constant to another type. The size of CST must be
1699 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1701 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
1702 <dd>Convert a floating point constant to the corresponding unsigned integer
1703 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1704 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1705 of the same number of elements. If the value won't fit in the integer type,
1706 the results are undefined.</dd>
1708 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
1709 <dd>Convert a floating point constant to the corresponding signed integer
1710 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1711 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1712 of the same number of elements. If the value won't fit in the integer type,
1713 the results are undefined.</dd>
1715 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
1716 <dd>Convert an unsigned integer constant to the corresponding floating point
1717 constant. TYPE must be a scalar or vector floating point type. CST must be of
1718 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1719 of the same number of elements. If the value won't fit in the floating point
1720 type, the results are undefined.</dd>
1722 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
1723 <dd>Convert a signed integer constant to the corresponding floating point
1724 constant. TYPE must be a scalar or vector floating point type. CST must be of
1725 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1726 of the same number of elements. If the value won't fit in the floating point
1727 type, the results are undefined.</dd>
1729 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
1730 <dd>Convert a pointer typed constant to the corresponding integer constant
1731 TYPE must be an integer type. CST must be of pointer type. The CST value is
1732 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1734 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
1735 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1736 pointer type. CST must be of integer type. The CST value is zero extended,
1737 truncated, or unchanged to make it fit in a pointer size. This one is
1738 <i>really</i> dangerous!</dd>
1740 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
1741 <dd>Convert a constant, CST, to another TYPE. The size of CST and TYPE must be
1742 identical (same number of bits). The conversion is done as if the CST value
1743 was stored to memory and read back as TYPE. In other words, no bits change
1744 with this operator, just the type. This can be used for conversion of
1745 vector types to any other type, as long as they have the same bit width. For
1746 pointers it is only valid to cast to another pointer type. It is not valid
1747 to bitcast to or from an aggregate type.
1750 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1752 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1753 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1754 instruction, the index list may have zero or more indexes, which are required
1755 to make sense for the type of "CSTPTR".</dd>
1757 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1759 <dd>Perform the <a href="#i_select">select operation</a> on
1762 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
1763 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
1765 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
1766 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
1768 <dt><b><tt>vicmp COND ( VAL1, VAL2 )</tt></b></dt>
1769 <dd>Performs the <a href="#i_vicmp">vicmp operation</a> on constants.</dd>
1771 <dt><b><tt>vfcmp COND ( VAL1, VAL2 )</tt></b></dt>
1772 <dd>Performs the <a href="#i_vfcmp">vfcmp operation</a> on constants.</dd>
1774 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1776 <dd>Perform the <a href="#i_extractelement">extractelement
1777 operation</a> on constants.
1779 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1781 <dd>Perform the <a href="#i_insertelement">insertelement
1782 operation</a> on constants.</dd>
1785 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
1787 <dd>Perform the <a href="#i_shufflevector">shufflevector
1788 operation</a> on constants.</dd>
1790 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1792 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1793 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1794 binary</a> operations. The constraints on operands are the same as those for
1795 the corresponding instruction (e.g. no bitwise operations on floating point
1796 values are allowed).</dd>
1800 <!-- *********************************************************************** -->
1801 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1802 <!-- *********************************************************************** -->
1804 <!-- ======================================================================= -->
1805 <div class="doc_subsection">
1806 <a name="inlineasm">Inline Assembler Expressions</a>
1809 <div class="doc_text">
1812 LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
1813 Module-Level Inline Assembly</a>) through the use of a special value. This
1814 value represents the inline assembler as a string (containing the instructions
1815 to emit), a list of operand constraints (stored as a string), and a flag that
1816 indicates whether or not the inline asm expression has side effects. An example
1817 inline assembler expression is:
1820 <div class="doc_code">
1822 i32 (i32) asm "bswap $0", "=r,r"
1827 Inline assembler expressions may <b>only</b> be used as the callee operand of
1828 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
1831 <div class="doc_code">
1833 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
1838 Inline asms with side effects not visible in the constraint list must be marked
1839 as having side effects. This is done through the use of the
1840 '<tt>sideeffect</tt>' keyword, like so:
1843 <div class="doc_code">
1845 call void asm sideeffect "eieio", ""()
1849 <p>TODO: The format of the asm and constraints string still need to be
1850 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1851 need to be documented).
1856 <!-- *********************************************************************** -->
1857 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1858 <!-- *********************************************************************** -->
1860 <div class="doc_text">
1862 <p>The LLVM instruction set consists of several different
1863 classifications of instructions: <a href="#terminators">terminator
1864 instructions</a>, <a href="#binaryops">binary instructions</a>,
1865 <a href="#bitwiseops">bitwise binary instructions</a>, <a
1866 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1867 instructions</a>.</p>
1871 <!-- ======================================================================= -->
1872 <div class="doc_subsection"> <a name="terminators">Terminator
1873 Instructions</a> </div>
1875 <div class="doc_text">
1877 <p>As mentioned <a href="#functionstructure">previously</a>, every
1878 basic block in a program ends with a "Terminator" instruction, which
1879 indicates which block should be executed after the current block is
1880 finished. These terminator instructions typically yield a '<tt>void</tt>'
1881 value: they produce control flow, not values (the one exception being
1882 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1883 <p>There are six different terminator instructions: the '<a
1884 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1885 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1886 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1887 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1888 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1892 <!-- _______________________________________________________________________ -->
1893 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1894 Instruction</a> </div>
1895 <div class="doc_text">
1897 <pre> ret <type> <value> <i>; Return a value from a non-void function</i>
1898 ret void <i>; Return from void function</i>
1899 ret <type> <value>, <type> <value> <i>; Return two values from a non-void function </i>
1904 <p>The '<tt>ret</tt>' instruction is used to return control flow (and a
1905 value) from a function back to the caller.</p>
1906 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
1907 returns value(s) and then causes control flow, and one that just causes
1908 control flow to occur.</p>
1912 <p>The '<tt>ret</tt>' instruction may return zero, one or multiple values.
1913 The type of each return value must be a '<a href="#t_firstclass">first
1914 class</a>' type. Note that a function is not <a href="#wellformed">well
1915 formed</a> if there exists a '<tt>ret</tt>' instruction inside of the
1916 function that returns values that do not match the return type of the
1921 <p>When the '<tt>ret</tt>' instruction is executed, control flow
1922 returns back to the calling function's context. If the caller is a "<a
1923 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1924 the instruction after the call. If the caller was an "<a
1925 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1926 at the beginning of the "normal" destination block. If the instruction
1927 returns a value, that value shall set the call or invoke instruction's
1928 return value. If the instruction returns multiple values then these
1929 values can only be accessed through a '<a href="#i_getresult"><tt>getresult</tt>
1930 </a>' instruction.</p>
1935 ret i32 5 <i>; Return an integer value of 5</i>
1936 ret void <i>; Return from a void function</i>
1937 ret i32 4, i8 2 <i>; Return two values 4 and 2 </i>
1940 <!-- _______________________________________________________________________ -->
1941 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1942 <div class="doc_text">
1944 <pre> br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1947 <p>The '<tt>br</tt>' instruction is used to cause control flow to
1948 transfer to a different basic block in the current function. There are
1949 two forms of this instruction, corresponding to a conditional branch
1950 and an unconditional branch.</p>
1952 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1953 single '<tt>i1</tt>' value and two '<tt>label</tt>' values. The
1954 unconditional form of the '<tt>br</tt>' instruction takes a single
1955 '<tt>label</tt>' value as a target.</p>
1957 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
1958 argument is evaluated. If the value is <tt>true</tt>, control flows
1959 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1960 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1962 <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
1963 href="#i_ret">ret</a> i32 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> i32 0<br></pre>
1965 <!-- _______________________________________________________________________ -->
1966 <div class="doc_subsubsection">
1967 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
1970 <div class="doc_text">
1974 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
1979 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
1980 several different places. It is a generalization of the '<tt>br</tt>'
1981 instruction, allowing a branch to occur to one of many possible
1987 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
1988 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
1989 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
1990 table is not allowed to contain duplicate constant entries.</p>
1994 <p>The <tt>switch</tt> instruction specifies a table of values and
1995 destinations. When the '<tt>switch</tt>' instruction is executed, this
1996 table is searched for the given value. If the value is found, control flow is
1997 transfered to the corresponding destination; otherwise, control flow is
1998 transfered to the default destination.</p>
2000 <h5>Implementation:</h5>
2002 <p>Depending on properties of the target machine and the particular
2003 <tt>switch</tt> instruction, this instruction may be code generated in different
2004 ways. For example, it could be generated as a series of chained conditional
2005 branches or with a lookup table.</p>
2010 <i>; Emulate a conditional br instruction</i>
2011 %Val = <a href="#i_zext">zext</a> i1 %value to i32
2012 switch i32 %Val, label %truedest [i32 0, label %falsedest ]
2014 <i>; Emulate an unconditional br instruction</i>
2015 switch i32 0, label %dest [ ]
2017 <i>; Implement a jump table:</i>
2018 switch i32 %val, label %otherwise [ i32 0, label %onzero
2020 i32 2, label %ontwo ]
2024 <!-- _______________________________________________________________________ -->
2025 <div class="doc_subsubsection">
2026 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
2029 <div class="doc_text">
2034 <result> = invoke [<a href="#callingconv">cconv</a>] <ptr to function ty> <function ptr val>(<function args>)
2035 to label <normal label> unwind label <exception label>
2040 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
2041 function, with the possibility of control flow transfer to either the
2042 '<tt>normal</tt>' label or the
2043 '<tt>exception</tt>' label. If the callee function returns with the
2044 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
2045 "normal" label. If the callee (or any indirect callees) returns with the "<a
2046 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
2047 continued at the dynamically nearest "exception" label. If the callee function
2048 returns multiple values then individual return values are only accessible through
2049 a '<tt><a href="#i_getresult">getresult</a></tt>' instruction.</p>
2053 <p>This instruction requires several arguments:</p>
2057 The optional "cconv" marker indicates which <a href="#callingconv">calling
2058 convention</a> the call should use. If none is specified, the call defaults
2059 to using C calling conventions.
2061 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
2062 function value being invoked. In most cases, this is a direct function
2063 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
2064 an arbitrary pointer to function value.
2067 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
2068 function to be invoked. </li>
2070 <li>'<tt>function args</tt>': argument list whose types match the function
2071 signature argument types. If the function signature indicates the function
2072 accepts a variable number of arguments, the extra arguments can be
2075 <li>'<tt>normal label</tt>': the label reached when the called function
2076 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
2078 <li>'<tt>exception label</tt>': the label reached when a callee returns with
2079 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
2085 <p>This instruction is designed to operate as a standard '<tt><a
2086 href="#i_call">call</a></tt>' instruction in most regards. The primary
2087 difference is that it establishes an association with a label, which is used by
2088 the runtime library to unwind the stack.</p>
2090 <p>This instruction is used in languages with destructors to ensure that proper
2091 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
2092 exception. Additionally, this is important for implementation of
2093 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
2097 %retval = invoke i32 @Test(i32 15) to label %Continue
2098 unwind label %TestCleanup <i>; {i32}:retval set</i>
2099 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
2100 unwind label %TestCleanup <i>; {i32}:retval set</i>
2105 <!-- _______________________________________________________________________ -->
2107 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
2108 Instruction</a> </div>
2110 <div class="doc_text">
2119 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
2120 at the first callee in the dynamic call stack which used an <a
2121 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
2122 primarily used to implement exception handling.</p>
2126 <p>The '<tt>unwind</tt>' instruction causes execution of the current function to
2127 immediately halt. The dynamic call stack is then searched for the first <a
2128 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
2129 execution continues at the "exceptional" destination block specified by the
2130 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
2131 dynamic call chain, undefined behavior results.</p>
2134 <!-- _______________________________________________________________________ -->
2136 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
2137 Instruction</a> </div>
2139 <div class="doc_text">
2148 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
2149 instruction is used to inform the optimizer that a particular portion of the
2150 code is not reachable. This can be used to indicate that the code after a
2151 no-return function cannot be reached, and other facts.</p>
2155 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
2160 <!-- ======================================================================= -->
2161 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
2162 <div class="doc_text">
2163 <p>Binary operators are used to do most of the computation in a
2164 program. They require two operands of the same type, execute an operation on them, and
2165 produce a single value. The operands might represent
2166 multiple data, as is the case with the <a href="#t_vector">vector</a> data type.
2167 The result value has the same type as its operands.</p>
2168 <p>There are several different binary operators:</p>
2170 <!-- _______________________________________________________________________ -->
2171 <div class="doc_subsubsection">
2172 <a name="i_add">'<tt>add</tt>' Instruction</a>
2175 <div class="doc_text">
2180 <result> = add <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2185 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
2189 <p>The two arguments to the '<tt>add</tt>' instruction must be <a
2190 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>, or
2191 <a href="#t_vector">vector</a> values. Both arguments must have identical
2196 <p>The value produced is the integer or floating point sum of the two
2199 <p>If an integer sum has unsigned overflow, the result returned is the
2200 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2203 <p>Because LLVM integers use a two's complement representation, this
2204 instruction is appropriate for both signed and unsigned integers.</p>
2209 <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
2212 <!-- _______________________________________________________________________ -->
2213 <div class="doc_subsubsection">
2214 <a name="i_sub">'<tt>sub</tt>' Instruction</a>
2217 <div class="doc_text">
2222 <result> = sub <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2227 <p>The '<tt>sub</tt>' instruction returns the difference of its two
2230 <p>Note that the '<tt>sub</tt>' instruction is used to represent the
2231 '<tt>neg</tt>' instruction present in most other intermediate
2232 representations.</p>
2236 <p>The two arguments to the '<tt>sub</tt>' instruction must be <a
2237 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>,
2238 or <a href="#t_vector">vector</a> values. Both arguments must have identical
2243 <p>The value produced is the integer or floating point difference of
2244 the two operands.</p>
2246 <p>If an integer difference has unsigned overflow, the result returned is the
2247 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2250 <p>Because LLVM integers use a two's complement representation, this
2251 instruction is appropriate for both signed and unsigned integers.</p>
2255 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
2256 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
2260 <!-- _______________________________________________________________________ -->
2261 <div class="doc_subsubsection">
2262 <a name="i_mul">'<tt>mul</tt>' Instruction</a>
2265 <div class="doc_text">
2268 <pre> <result> = mul <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2271 <p>The '<tt>mul</tt>' instruction returns the product of its two
2276 <p>The two arguments to the '<tt>mul</tt>' instruction must be <a
2277 href="#t_integer">integer</a>, <a href="#t_floating">floating point</a>,
2278 or <a href="#t_vector">vector</a> values. Both arguments must have identical
2283 <p>The value produced is the integer or floating point product of the
2286 <p>If the result of an integer multiplication has unsigned overflow,
2287 the result returned is the mathematical result modulo
2288 2<sup>n</sup>, where n is the bit width of the result.</p>
2289 <p>Because LLVM integers use a two's complement representation, and the
2290 result is the same width as the operands, this instruction returns the
2291 correct result for both signed and unsigned integers. If a full product
2292 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands
2293 should be sign-extended or zero-extended as appropriate to the
2294 width of the full product.</p>
2296 <pre> <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
2300 <!-- _______________________________________________________________________ -->
2301 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
2303 <div class="doc_text">
2305 <pre> <result> = udiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2308 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two
2313 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
2314 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2315 values. Both arguments must have identical types.</p>
2319 <p>The value produced is the unsigned integer quotient of the two operands.</p>
2320 <p>Note that unsigned integer division and signed integer division are distinct
2321 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
2322 <p>Division by zero leads to undefined behavior.</p>
2324 <pre> <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2327 <!-- _______________________________________________________________________ -->
2328 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
2330 <div class="doc_text">
2333 <result> = sdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2338 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
2343 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
2344 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2345 values. Both arguments must have identical types.</p>
2348 <p>The value produced is the signed integer quotient of the two operands rounded towards zero.</p>
2349 <p>Note that signed integer division and unsigned integer division are distinct
2350 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
2351 <p>Division by zero leads to undefined behavior. Overflow also leads to
2352 undefined behavior; this is a rare case, but can occur, for example,
2353 by doing a 32-bit division of -2147483648 by -1.</p>
2355 <pre> <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2358 <!-- _______________________________________________________________________ -->
2359 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
2360 Instruction</a> </div>
2361 <div class="doc_text">
2364 <result> = fdiv <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2368 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
2373 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
2374 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a>
2375 of floating point values. Both arguments must have identical types.</p>
2379 <p>The value produced is the floating point quotient of the two operands.</p>
2384 <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
2388 <!-- _______________________________________________________________________ -->
2389 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
2391 <div class="doc_text">
2393 <pre> <result> = urem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2396 <p>The '<tt>urem</tt>' instruction returns the remainder from the
2397 unsigned division of its two arguments.</p>
2399 <p>The two arguments to the '<tt>urem</tt>' instruction must be
2400 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2401 values. Both arguments must have identical types.</p>
2403 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
2404 This instruction always performs an unsigned division to get the remainder.</p>
2405 <p>Note that unsigned integer remainder and signed integer remainder are
2406 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
2407 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
2409 <pre> <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2413 <!-- _______________________________________________________________________ -->
2414 <div class="doc_subsubsection">
2415 <a name="i_srem">'<tt>srem</tt>' Instruction</a>
2418 <div class="doc_text">
2423 <result> = srem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2428 <p>The '<tt>srem</tt>' instruction returns the remainder from the
2429 signed division of its two operands. This instruction can also take
2430 <a href="#t_vector">vector</a> versions of the values in which case
2431 the elements must be integers.</p>
2435 <p>The two arguments to the '<tt>srem</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>
2441 <p>This instruction returns the <i>remainder</i> of a division (where the result
2442 has the same sign as the dividend, <tt>op1</tt>), not the <i>modulo</i>
2443 operator (where the result has the same sign as the divisor, <tt>op2</tt>) of
2444 a value. For more information about the difference, see <a
2445 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
2446 Math Forum</a>. For a table of how this is implemented in various languages,
2447 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
2448 Wikipedia: modulo operation</a>.</p>
2449 <p>Note that signed integer remainder and unsigned integer remainder are
2450 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
2451 <p>Taking the remainder of a division by zero leads to undefined behavior.
2452 Overflow also leads to undefined behavior; this is a rare case, but can occur,
2453 for example, by taking the remainder of a 32-bit division of -2147483648 by -1.
2454 (The remainder doesn't actually overflow, but this rule lets srem be
2455 implemented using instructions that return both the result of the division
2456 and the remainder.)</p>
2458 <pre> <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2462 <!-- _______________________________________________________________________ -->
2463 <div class="doc_subsubsection">
2464 <a name="i_frem">'<tt>frem</tt>' Instruction</a> </div>
2466 <div class="doc_text">
2469 <pre> <result> = frem <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2472 <p>The '<tt>frem</tt>' instruction returns the remainder from the
2473 division of its two operands.</p>
2475 <p>The two arguments to the '<tt>frem</tt>' instruction must be
2476 <a href="#t_floating">floating point</a> or <a href="#t_vector">vector</a>
2477 of floating point values. Both arguments must have identical types.</p>
2481 <p>This instruction returns the <i>remainder</i> of a division.
2482 The remainder has the same sign as the dividend.</p>
2487 <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
2491 <!-- ======================================================================= -->
2492 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
2493 Operations</a> </div>
2494 <div class="doc_text">
2495 <p>Bitwise binary operators are used to do various forms of
2496 bit-twiddling in a program. They are generally very efficient
2497 instructions and can commonly be strength reduced from other
2498 instructions. They require two operands of the same type, execute an operation on them,
2499 and produce a single value. The resulting value is the same type as its operands.</p>
2502 <!-- _______________________________________________________________________ -->
2503 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
2504 Instruction</a> </div>
2505 <div class="doc_text">
2507 <pre> <result> = shl <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2512 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2513 the left a specified number of bits.</p>
2517 <p>Both arguments to the '<tt>shl</tt>' instruction must be the same <a
2518 href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2519 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
2523 <p>The value produced is <tt>op1</tt> * 2<sup><tt>op2</tt></sup> mod 2<sup>n</sup>,
2524 where n is the width of the result. If <tt>op2</tt> is (statically or dynamically) negative or
2525 equal to or larger than the number of bits in <tt>op1</tt>, the result is undefined.</p>
2527 <h5>Example:</h5><pre>
2528 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
2529 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
2530 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
2531 <result> = shl i32 1, 32 <i>; undefined</i>
2534 <!-- _______________________________________________________________________ -->
2535 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
2536 Instruction</a> </div>
2537 <div class="doc_text">
2539 <pre> <result> = lshr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2543 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
2544 operand shifted to the right a specified number of bits with zero fill.</p>
2547 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
2548 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2549 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
2553 <p>This instruction always performs a logical shift right operation. The most
2554 significant bits of the result will be filled with zero bits after the
2555 shift. If <tt>op2</tt> is (statically or dynamically) equal to or larger than
2556 the number of bits in <tt>op1</tt>, the result is undefined.</p>
2560 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
2561 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
2562 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
2563 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
2564 <result> = lshr i32 1, 32 <i>; undefined</i>
2568 <!-- _______________________________________________________________________ -->
2569 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
2570 Instruction</a> </div>
2571 <div class="doc_text">
2574 <pre> <result> = ashr <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2578 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
2579 operand shifted to the right a specified number of bits with sign extension.</p>
2582 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
2583 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2584 type. '<tt>op2</tt>' is treated as an unsigned value.</p>
2587 <p>This instruction always performs an arithmetic shift right operation,
2588 The most significant bits of the result will be filled with the sign bit
2589 of <tt>op1</tt>. If <tt>op2</tt> is (statically or dynamically) equal to or
2590 larger than the number of bits in <tt>op1</tt>, the result is undefined.
2595 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
2596 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
2597 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
2598 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
2599 <result> = ashr i32 1, 32 <i>; undefined</i>
2603 <!-- _______________________________________________________________________ -->
2604 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
2605 Instruction</a> </div>
2607 <div class="doc_text">
2612 <result> = and <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2617 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
2618 its two operands.</p>
2622 <p>The two arguments to the '<tt>and</tt>' instruction must be
2623 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2624 values. Both arguments must have identical types.</p>
2627 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
2630 <table border="1" cellspacing="0" cellpadding="4">
2662 <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
2663 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
2664 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
2667 <!-- _______________________________________________________________________ -->
2668 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
2669 <div class="doc_text">
2671 <pre> <result> = or <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2674 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
2675 or of its two operands.</p>
2678 <p>The two arguments to the '<tt>or</tt>' instruction must be
2679 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2680 values. Both arguments must have identical types.</p>
2682 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
2685 <table border="1" cellspacing="0" cellpadding="4">
2716 <pre> <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
2717 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
2718 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
2721 <!-- _______________________________________________________________________ -->
2722 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
2723 Instruction</a> </div>
2724 <div class="doc_text">
2726 <pre> <result> = xor <ty> <op1>, <op2> <i>; yields {ty}:result</i>
2729 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
2730 or of its two operands. The <tt>xor</tt> is used to implement the
2731 "one's complement" operation, which is the "~" operator in C.</p>
2733 <p>The two arguments to the '<tt>xor</tt>' instruction must be
2734 <a href="#t_integer">integer</a> or <a href="#t_vector">vector</a> of integer
2735 values. Both arguments must have identical types.</p>
2739 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
2742 <table border="1" cellspacing="0" cellpadding="4">
2774 <pre> <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
2775 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
2776 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
2777 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
2781 <!-- ======================================================================= -->
2782 <div class="doc_subsection">
2783 <a name="vectorops">Vector Operations</a>
2786 <div class="doc_text">
2788 <p>LLVM supports several instructions to represent vector operations in a
2789 target-independent manner. These instructions cover the element-access and
2790 vector-specific operations needed to process vectors effectively. While LLVM
2791 does directly support these vector operations, many sophisticated algorithms
2792 will want to use target-specific intrinsics to take full advantage of a specific
2797 <!-- _______________________________________________________________________ -->
2798 <div class="doc_subsubsection">
2799 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2802 <div class="doc_text">
2807 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
2813 The '<tt>extractelement</tt>' instruction extracts a single scalar
2814 element from a vector at a specified index.
2821 The first operand of an '<tt>extractelement</tt>' instruction is a
2822 value of <a href="#t_vector">vector</a> type. The second operand is
2823 an index indicating the position from which to extract the element.
2824 The index may be a variable.</p>
2829 The result is a scalar of the same type as the element type of
2830 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2831 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2832 results are undefined.
2838 %result = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
2843 <!-- _______________________________________________________________________ -->
2844 <div class="doc_subsubsection">
2845 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
2848 <div class="doc_text">
2853 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
2859 The '<tt>insertelement</tt>' instruction inserts a scalar
2860 element into a vector at a specified index.
2867 The first operand of an '<tt>insertelement</tt>' instruction is a
2868 value of <a href="#t_vector">vector</a> type. The second operand is a
2869 scalar value whose type must equal the element type of the first
2870 operand. The third operand is an index indicating the position at
2871 which to insert the value. The index may be a variable.</p>
2876 The result is a vector of the same type as <tt>val</tt>. Its
2877 element values are those of <tt>val</tt> except at position
2878 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
2879 exceeds the length of <tt>val</tt>, the results are undefined.
2885 %result = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
2889 <!-- _______________________________________________________________________ -->
2890 <div class="doc_subsubsection">
2891 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
2894 <div class="doc_text">
2899 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <n x i32> <mask> <i>; yields <n x <ty>></i>
2905 The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
2906 from two input vectors, returning a vector of the same type.
2912 The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
2913 with types that match each other and types that match the result of the
2914 instruction. The third argument is a shuffle mask, which has the same number
2915 of elements as the other vector type, but whose element type is always 'i32'.
2919 The shuffle mask operand is required to be a constant vector with either
2920 constant integer or undef values.
2926 The elements of the two input vectors are numbered from left to right across
2927 both of the vectors. The shuffle mask operand specifies, for each element of
2928 the result vector, which element of the two input registers the result element
2929 gets. The element selector may be undef (meaning "don't care") and the second
2930 operand may be undef if performing a shuffle from only one vector.
2936 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
2937 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
2938 %result = shufflevector <4 x i32> %v1, <4 x i32> undef,
2939 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
2944 <!-- ======================================================================= -->
2945 <div class="doc_subsection">
2946 <a name="aggregateops">Aggregate Operations</a>
2949 <div class="doc_text">
2951 <p>LLVM supports several instructions for working with aggregate values.
2956 <!-- _______________________________________________________________________ -->
2957 <div class="doc_subsubsection">
2958 <a name="i_extractvalue">'<tt>extractvalue</tt>' Instruction</a>
2961 <div class="doc_text">
2966 <result> = extractvalue <aggregate type> <val>, <idx>{, <idx>}*
2972 The '<tt>extractvalue</tt>' instruction extracts the value of a struct field
2973 or array element from an aggregate value.
2980 The first operand of an '<tt>extractvalue</tt>' instruction is a
2981 value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a>
2982 type. The operands are constant indices to specify which value to extract
2983 in a similar manner as indices in a
2984 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
2990 The result is the value at the position in the aggregate specified by
2997 %result = extractvalue {i32, float} %agg, 0 <i>; yields i32</i>
3002 <!-- _______________________________________________________________________ -->
3003 <div class="doc_subsubsection">
3004 <a name="i_insertvalue">'<tt>insertvalue</tt>' Instruction</a>
3007 <div class="doc_text">
3012 <result> = insertvalue <aggregate type> <val>, <ty> <val>, <idx> <i>; yields <n x <ty>></i>
3018 The '<tt>insertvalue</tt>' instruction inserts a value
3019 into a struct field or array element in an aggregate.
3026 The first operand of an '<tt>insertvalue</tt>' instruction is a
3027 value of <a href="#t_struct">struct</a> or <a href="#t_array">array</a> type.
3028 The second operand is a first-class value to insert.
3029 The following operands are constant indices
3030 indicating the position at which to insert the value in a similar manner as
3032 '<tt><a href="#i_getelementptr">getelementptr</a></tt>' instruction.
3033 The value to insert must have the same type as the value identified
3039 The result is an aggregate of the same type as <tt>val</tt>. Its
3040 value is that of <tt>val</tt> except that the value at the position
3041 specified by the indices is that of <tt>elt</tt>.
3047 %result = insertvalue {i32, float} %agg, i32 1, 0 <i>; yields {i32, float}</i>
3052 <!-- ======================================================================= -->
3053 <div class="doc_subsection">
3054 <a name="memoryops">Memory Access and Addressing Operations</a>
3057 <div class="doc_text">
3059 <p>A key design point of an SSA-based representation is how it
3060 represents memory. In LLVM, no memory locations are in SSA form, which
3061 makes things very simple. This section describes how to read, write,
3062 allocate, and free memory in LLVM.</p>
3066 <!-- _______________________________________________________________________ -->
3067 <div class="doc_subsubsection">
3068 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
3071 <div class="doc_text">
3076 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3081 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
3082 heap and returns a pointer to it. The object is always allocated in the generic
3083 address space (address space zero).</p>
3087 <p>The '<tt>malloc</tt>' instruction allocates
3088 <tt>sizeof(<type>)*NumElements</tt>
3089 bytes of memory from the operating system and returns a pointer of the
3090 appropriate type to the program. If "NumElements" is specified, it is the
3091 number of elements allocated, otherwise "NumElements" is defaulted to be one.
3092 If a constant alignment is specified, the value result of the allocation is guaranteed to
3093 be aligned to at least that boundary. If not specified, or if zero, the target can
3094 choose to align the allocation on any convenient boundary.</p>
3096 <p>'<tt>type</tt>' must be a sized type.</p>
3100 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
3101 a pointer is returned. The result of a zero byte allocattion is undefined. The
3102 result is null if there is insufficient memory available.</p>
3107 %array = malloc [4 x i8 ] <i>; yields {[%4 x i8]*}:array</i>
3109 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
3110 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
3111 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
3112 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
3113 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
3117 <!-- _______________________________________________________________________ -->
3118 <div class="doc_subsubsection">
3119 <a name="i_free">'<tt>free</tt>' Instruction</a>
3122 <div class="doc_text">
3127 free <type> <value> <i>; yields {void}</i>
3132 <p>The '<tt>free</tt>' instruction returns memory back to the unused
3133 memory heap to be reallocated in the future.</p>
3137 <p>'<tt>value</tt>' shall be a pointer value that points to a value
3138 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
3143 <p>Access to the memory pointed to by the pointer is no longer defined
3144 after this instruction executes. If the pointer is null, the operation
3150 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
3151 free [4 x i8]* %array
3155 <!-- _______________________________________________________________________ -->
3156 <div class="doc_subsubsection">
3157 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
3160 <div class="doc_text">
3165 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
3170 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
3171 currently executing function, to be automatically released when this function
3172 returns to its caller. The object is always allocated in the generic address
3173 space (address space zero).</p>
3177 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
3178 bytes of memory on the runtime stack, returning a pointer of the
3179 appropriate type to the program. If "NumElements" is specified, it is the
3180 number of elements allocated, otherwise "NumElements" is defaulted to be one.
3181 If a constant alignment is specified, the value result of the allocation is guaranteed
3182 to be aligned to at least that boundary. If not specified, or if zero, the target
3183 can choose to align the allocation on any convenient boundary.</p>
3185 <p>'<tt>type</tt>' may be any sized type.</p>
3189 <p>Memory is allocated; a pointer is returned. The operation is undefiend if
3190 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
3191 memory is automatically released when the function returns. The '<tt>alloca</tt>'
3192 instruction is commonly used to represent automatic variables that must
3193 have an address available. When the function returns (either with the <tt><a
3194 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
3195 instructions), the memory is reclaimed. Allocating zero bytes
3196 is legal, but the result is undefined.</p>
3201 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
3202 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
3203 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
3204 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
3208 <!-- _______________________________________________________________________ -->
3209 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
3210 Instruction</a> </div>
3211 <div class="doc_text">
3213 <pre> <result> = load <ty>* <pointer>[, align <alignment>]<br> <result> = volatile load <ty>* <pointer>[, align <alignment>]<br></pre>
3215 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
3217 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
3218 address from which to load. The pointer must point to a <a
3219 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
3220 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
3221 the number or order of execution of this <tt>load</tt> with other
3222 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
3225 The optional constant "align" argument specifies the alignment of the operation
3226 (that is, the alignment of the memory address). A value of 0 or an
3227 omitted "align" argument means that the operation has the preferential
3228 alignment for the target. It is the responsibility of the code emitter
3229 to ensure that the alignment information is correct. Overestimating
3230 the alignment results in an undefined behavior. Underestimating the
3231 alignment may produce less efficient code. An alignment of 1 is always
3235 <p>The location of memory pointed to is loaded.</p>
3237 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3239 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
3240 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
3243 <!-- _______________________________________________________________________ -->
3244 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
3245 Instruction</a> </div>
3246 <div class="doc_text">
3248 <pre> store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3249 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
3252 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
3254 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
3255 to store and an address at which to store it. The type of the '<tt><pointer></tt>'
3256 operand must be a pointer to the <a href="#t_firstclass">first class</a> type
3257 of the '<tt><value></tt>'
3258 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
3259 optimizer is not allowed to modify the number or order of execution of
3260 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
3261 href="#i_store">store</a></tt> instructions.</p>
3263 The optional constant "align" argument specifies the alignment of the operation
3264 (that is, the alignment of the memory address). A value of 0 or an
3265 omitted "align" argument means that the operation has the preferential
3266 alignment for the target. It is the responsibility of the code emitter
3267 to ensure that the alignment information is correct. Overestimating
3268 the alignment results in an undefined behavior. Underestimating the
3269 alignment may produce less efficient code. An alignment of 1 is always
3273 <p>The contents of memory are updated to contain '<tt><value></tt>'
3274 at the location specified by the '<tt><pointer></tt>' operand.</p>
3276 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
3277 store i32 3, i32* %ptr <i>; yields {void}</i>
3278 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
3282 <!-- _______________________________________________________________________ -->
3283 <div class="doc_subsubsection">
3284 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
3287 <div class="doc_text">
3290 <result> = getelementptr <ty>* <ptrval>{, <ty> <idx>}*
3296 The '<tt>getelementptr</tt>' instruction is used to get the address of a
3297 subelement of an aggregate data structure.</p>
3301 <p>This instruction takes a list of integer operands that indicate what
3302 elements of the aggregate object to index to. The actual types of the arguments
3303 provided depend on the type of the first pointer argument. The
3304 '<tt>getelementptr</tt>' instruction is used to index down through the type
3305 levels of a structure or to a specific index in an array. When indexing into a
3306 structure, only <tt>i32</tt> integer constants are allowed. When indexing
3307 into an array or pointer, only integers of 32 or 64 bits are allowed; 32-bit
3308 values will be sign extended to 64-bits if required.</p>
3310 <p>For example, let's consider a C code fragment and how it gets
3311 compiled to LLVM:</p>
3313 <div class="doc_code">
3326 int *foo(struct ST *s) {
3327 return &s[1].Z.B[5][13];
3332 <p>The LLVM code generated by the GCC frontend is:</p>
3334 <div class="doc_code">
3336 %RT = type { i8 , [10 x [20 x i32]], i8 }
3337 %ST = type { i32, double, %RT }
3339 define i32* %foo(%ST* %s) {
3341 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
3349 <p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
3350 on the pointer type that is being indexed into. <a href="#t_pointer">Pointer</a>
3351 and <a href="#t_array">array</a> types can use a 32-bit or 64-bit
3352 <a href="#t_integer">integer</a> type but the value will always be sign extended
3353 to 64-bits. <a href="#t_struct">Structure</a> and <a href="#t_pstruct">packed
3354 structure</a> types require <tt>i32</tt> <b>constants</b>.</p>
3356 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
3357 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
3358 }</tt>' type, a structure. The second index indexes into the third element of
3359 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
3360 i8 }</tt>' type, another structure. The third index indexes into the second
3361 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
3362 array. The two dimensions of the array are subscripted into, yielding an
3363 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
3364 to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
3366 <p>Note that it is perfectly legal to index partially through a
3367 structure, returning a pointer to an inner element. Because of this,
3368 the LLVM code for the given testcase is equivalent to:</p>
3371 define i32* %foo(%ST* %s) {
3372 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
3373 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
3374 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
3375 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
3376 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
3381 <p>Note that it is undefined to access an array out of bounds: array and
3382 pointer indexes must always be within the defined bounds of the array type.
3383 The one exception for this rule is zero length arrays. These arrays are
3384 defined to be accessible as variable length arrays, which requires access
3385 beyond the zero'th element.</p>
3387 <p>The getelementptr instruction is often confusing. For some more insight
3388 into how it works, see <a href="GetElementPtr.html">the getelementptr
3394 <i>; yields [12 x i8]*:aptr</i>
3395 %aptr = getelementptr {i32, [12 x i8]}* %sptr, i64 0, i32 1
3399 <!-- ======================================================================= -->
3400 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
3402 <div class="doc_text">
3403 <p>The instructions in this category are the conversion instructions (casting)
3404 which all take a single operand and a type. They perform various bit conversions
3408 <!-- _______________________________________________________________________ -->
3409 <div class="doc_subsubsection">
3410 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
3412 <div class="doc_text">
3416 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
3421 The '<tt>trunc</tt>' instruction truncates its operand to the type <tt>ty2</tt>.
3426 The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
3427 be an <a href="#t_integer">integer</a> type, and a type that specifies the size
3428 and type of the result, which must be an <a href="#t_integer">integer</a>
3429 type. The bit size of <tt>value</tt> must be larger than the bit size of
3430 <tt>ty2</tt>. Equal sized types are not allowed.</p>
3434 The '<tt>trunc</tt>' instruction truncates the high order bits in <tt>value</tt>
3435 and converts the remaining bits to <tt>ty2</tt>. Since the source size must be
3436 larger than the destination size, <tt>trunc</tt> cannot be a <i>no-op cast</i>.
3437 It will always truncate bits.</p>
3441 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
3442 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
3443 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
3447 <!-- _______________________________________________________________________ -->
3448 <div class="doc_subsubsection">
3449 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
3451 <div class="doc_text">
3455 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
3459 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
3464 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
3465 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3466 also be of <a href="#t_integer">integer</a> type. The bit size of the
3467 <tt>value</tt> must be smaller than the bit size of the destination type,
3471 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
3472 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
3474 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
3478 %X = zext i32 257 to i64 <i>; yields i64:257</i>
3479 %Y = zext i1 true to i32 <i>; yields i32:1</i>
3483 <!-- _______________________________________________________________________ -->
3484 <div class="doc_subsubsection">
3485 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
3487 <div class="doc_text">
3491 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
3495 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
3499 The '<tt>sext</tt>' instruction takes a value to cast, which must be of
3500 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3501 also be of <a href="#t_integer">integer</a> type. The bit size of the
3502 <tt>value</tt> must be smaller than the bit size of the destination type,
3507 The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
3508 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size of
3509 the type <tt>ty2</tt>.</p>
3511 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
3515 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
3516 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
3520 <!-- _______________________________________________________________________ -->
3521 <div class="doc_subsubsection">
3522 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
3525 <div class="doc_text">
3530 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
3534 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
3539 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
3540 point</a> value to cast and a <a href="#t_floating">floating point</a> type to
3541 cast it to. The size of <tt>value</tt> must be larger than the size of
3542 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
3543 <i>no-op cast</i>.</p>
3546 <p> The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
3547 <a href="#t_floating">floating point</a> type to a smaller
3548 <a href="#t_floating">floating point</a> type. If the value cannot fit within
3549 the destination type, <tt>ty2</tt>, then the results are undefined.</p>
3553 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
3554 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
3558 <!-- _______________________________________________________________________ -->
3559 <div class="doc_subsubsection">
3560 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
3562 <div class="doc_text">
3566 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
3570 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
3571 floating point value.</p>
3574 <p>The '<tt>fpext</tt>' instruction takes a
3575 <a href="#t_floating">floating point</a> <tt>value</tt> to cast,
3576 and a <a href="#t_floating">floating point</a> type to cast it to. The source
3577 type must be smaller than the destination type.</p>
3580 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
3581 <a href="#t_floating">floating point</a> type to a larger
3582 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
3583 used to make a <i>no-op cast</i> because it always changes bits. Use
3584 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
3588 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
3589 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
3593 <!-- _______________________________________________________________________ -->
3594 <div class="doc_subsubsection">
3595 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
3597 <div class="doc_text">
3601 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
3605 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
3606 unsigned integer equivalent of type <tt>ty2</tt>.
3610 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
3611 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3612 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3613 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3614 vector integer type with the same number of elements as <tt>ty</tt></p>
3617 <p> The '<tt>fptoui</tt>' instruction converts its
3618 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3619 towards zero) unsigned integer value. If the value cannot fit in <tt>ty2</tt>,
3620 the results are undefined.</p>
3624 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
3625 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
3626 %X = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
3630 <!-- _______________________________________________________________________ -->
3631 <div class="doc_subsubsection">
3632 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
3634 <div class="doc_text">
3638 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
3642 <p>The '<tt>fptosi</tt>' instruction converts
3643 <a href="#t_floating">floating point</a> <tt>value</tt> to type <tt>ty2</tt>.
3647 <p> The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
3648 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3649 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3650 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3651 vector integer type with the same number of elements as <tt>ty</tt></p>
3654 <p>The '<tt>fptosi</tt>' instruction converts its
3655 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3656 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
3657 the results are undefined.</p>
3661 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
3662 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
3663 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
3667 <!-- _______________________________________________________________________ -->
3668 <div class="doc_subsubsection">
3669 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
3671 <div class="doc_text">
3675 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3679 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
3680 integer and converts that value to the <tt>ty2</tt> type.</p>
3683 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
3684 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3685 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3686 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3687 floating point type with the same number of elements as <tt>ty</tt></p>
3690 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
3691 integer quantity and converts it to the corresponding floating point value. If
3692 the value cannot fit in the floating point value, the results are undefined.</p>
3696 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
3697 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
3701 <!-- _______________________________________________________________________ -->
3702 <div class="doc_subsubsection">
3703 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
3705 <div class="doc_text">
3709 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3713 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed
3714 integer and converts that value to the <tt>ty2</tt> type.</p>
3717 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
3718 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3719 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3720 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3721 floating point type with the same number of elements as <tt>ty</tt></p>
3724 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed
3725 integer quantity and converts it to the corresponding floating point value. If
3726 the value cannot fit in the floating point value, the results are undefined.</p>
3730 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
3731 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
3735 <!-- _______________________________________________________________________ -->
3736 <div class="doc_subsubsection">
3737 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
3739 <div class="doc_text">
3743 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
3747 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
3748 the integer type <tt>ty2</tt>.</p>
3751 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
3752 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
3753 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.
3756 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
3757 <tt>ty2</tt> by interpreting the pointer value as an integer and either
3758 truncating or zero extending that value to the size of the integer type. If
3759 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
3760 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
3761 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
3766 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
3767 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
3771 <!-- _______________________________________________________________________ -->
3772 <div class="doc_subsubsection">
3773 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
3775 <div class="doc_text">
3779 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
3783 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to
3784 a pointer type, <tt>ty2</tt>.</p>
3787 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
3788 value to cast, and a type to cast it to, which must be a
3789 <a href="#t_pointer">pointer</a> type.
3792 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
3793 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
3794 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
3795 size of a pointer then a truncation is done. If <tt>value</tt> is smaller than
3796 the size of a pointer then a zero extension is done. If they are the same size,
3797 nothing is done (<i>no-op cast</i>).</p>
3801 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
3802 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
3803 %Y = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
3807 <!-- _______________________________________________________________________ -->
3808 <div class="doc_subsubsection">
3809 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
3811 <div class="doc_text">
3815 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
3820 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3821 <tt>ty2</tt> without changing any bits.</p>
3825 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be
3826 a non-aggregate first class value, and a type to cast it to, which must also be
3827 a non-aggregate <a href="#t_firstclass">first class</a> type. The bit sizes of
3829 and the destination type, <tt>ty2</tt>, must be identical. If the source
3830 type is a pointer, the destination type must also be a pointer. This
3831 instruction supports bitwise conversion of vectors to integers and to vectors
3832 of other types (as long as they have the same size).</p>
3835 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3836 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
3837 this conversion. The conversion is done as if the <tt>value</tt> had been
3838 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only be
3839 converted to other pointer types with this instruction. To convert pointers to
3840 other types, use the <a href="#i_inttoptr">inttoptr</a> or
3841 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
3845 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
3846 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
3847 %Z = bitcast <2xint> %V to i64; <i>; yields i64: %V</i>
3851 <!-- ======================================================================= -->
3852 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
3853 <div class="doc_text">
3854 <p>The instructions in this category are the "miscellaneous"
3855 instructions, which defy better classification.</p>
3858 <!-- _______________________________________________________________________ -->
3859 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
3861 <div class="doc_text">
3863 <pre> <result> = icmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
3866 <p>The '<tt>icmp</tt>' instruction returns a boolean value or
3867 a vector of boolean values based on comparison
3868 of its two integer, integer vector, or pointer operands.</p>
3870 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
3871 the condition code indicating the kind of comparison to perform. It is not
3872 a value, just a keyword. The possible condition code are:
3874 <li><tt>eq</tt>: equal</li>
3875 <li><tt>ne</tt>: not equal </li>
3876 <li><tt>ugt</tt>: unsigned greater than</li>
3877 <li><tt>uge</tt>: unsigned greater or equal</li>
3878 <li><tt>ult</tt>: unsigned less than</li>
3879 <li><tt>ule</tt>: unsigned less or equal</li>
3880 <li><tt>sgt</tt>: signed greater than</li>
3881 <li><tt>sge</tt>: signed greater or equal</li>
3882 <li><tt>slt</tt>: signed less than</li>
3883 <li><tt>sle</tt>: signed less or equal</li>
3885 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
3886 <a href="#t_pointer">pointer</a>
3887 or integer <a href="#t_vector">vector</a> typed.
3888 They must also be identical types.</p>
3890 <p>The '<tt>icmp</tt>' compares <tt>op1</tt> and <tt>op2</tt> according to
3891 the condition code given as <tt>cond</tt>. The comparison performed always
3892 yields either an <a href="#t_primitive"><tt>i1</tt></a> or vector of <tt>i1</tt> result, as follows:
3894 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
3895 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3897 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
3898 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3899 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
3900 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
3901 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
3902 <tt>true</tt> if <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
3903 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
3904 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
3905 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
3906 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
3907 <li><tt>sgt</tt>: interprets the operands as signed values and yields
3908 <tt>true</tt> if <tt>op1</tt> is greater than <tt>op2</tt>.</li>
3909 <li><tt>sge</tt>: interprets the operands as signed values and yields
3910 <tt>true</tt> if <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
3911 <li><tt>slt</tt>: interprets the operands as signed values and yields
3912 <tt>true</tt> if <tt>op1</tt> is less than <tt>op2</tt>.</li>
3913 <li><tt>sle</tt>: interprets the operands as signed values and yields
3914 <tt>true</tt> if <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
3916 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
3917 values are compared as if they were integers.</p>
3918 <p>If the operands are integer vectors, then they are compared
3919 element by element. The result is an <tt>i1</tt> vector with
3920 the same number of elements as the values being compared.
3921 Otherwise, the result is an <tt>i1</tt>.
3925 <pre> <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
3926 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
3927 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
3928 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
3929 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
3930 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
3934 <!-- _______________________________________________________________________ -->
3935 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
3937 <div class="doc_text">
3939 <pre> <result> = fcmp <cond> <ty> <op1>, <op2> <i>; yields {i1} or {<N x i1>}:result</i>
3942 <p>The '<tt>fcmp</tt>' instruction returns a boolean value
3943 or vector of boolean values based on comparison
3946 If the operands are floating point scalars, then the result
3947 type is a boolean (<a href="#t_primitive"><tt>i1</tt></a>).
3949 <p>If the operands are floating point vectors, then the result type
3950 is a vector of boolean with the same number of elements as the
3951 operands being compared.</p>
3953 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
3954 the condition code indicating the kind of comparison to perform. It is not
3955 a value, just a keyword. The possible condition code are:
3957 <li><tt>false</tt>: no comparison, always returns false</li>
3958 <li><tt>oeq</tt>: ordered and equal</li>
3959 <li><tt>ogt</tt>: ordered and greater than </li>
3960 <li><tt>oge</tt>: ordered and greater than or equal</li>
3961 <li><tt>olt</tt>: ordered and less than </li>
3962 <li><tt>ole</tt>: ordered and less than or equal</li>
3963 <li><tt>one</tt>: ordered and not equal</li>
3964 <li><tt>ord</tt>: ordered (no nans)</li>
3965 <li><tt>ueq</tt>: unordered or equal</li>
3966 <li><tt>ugt</tt>: unordered or greater than </li>
3967 <li><tt>uge</tt>: unordered or greater than or equal</li>
3968 <li><tt>ult</tt>: unordered or less than </li>
3969 <li><tt>ule</tt>: unordered or less than or equal</li>
3970 <li><tt>une</tt>: unordered or not equal</li>
3971 <li><tt>uno</tt>: unordered (either nans)</li>
3972 <li><tt>true</tt>: no comparison, always returns true</li>
3974 <p><i>Ordered</i> means that neither operand is a QNAN while
3975 <i>unordered</i> means that either operand may be a QNAN.</p>
3976 <p>Each of <tt>val1</tt> and <tt>val2</tt> arguments must be
3977 either a <a href="#t_floating">floating point</a> type
3978 or a <a href="#t_vector">vector</a> of floating point type.
3979 They must have identical types.</p>
3981 <p>The '<tt>fcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
3982 according to the condition code given as <tt>cond</tt>.
3983 If the operands are vectors, then the vectors are compared
3985 Each comparison performed
3986 always yields an <a href="#t_primitive">i1</a> result, as follows:
3988 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
3989 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3990 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
3991 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3992 <tt>op1</tt> is greather than <tt>op2</tt>.</li>
3993 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3994 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
3995 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3996 <tt>op1</tt> is less than <tt>op2</tt>.</li>
3997 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3998 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
3999 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
4000 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
4001 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
4002 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
4003 <tt>op1</tt> is equal to <tt>op2</tt>.</li>
4004 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
4005 <tt>op1</tt> is greater than <tt>op2</tt>.</li>
4006 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
4007 <tt>op1</tt> is greater than or equal to <tt>op2</tt>.</li>
4008 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
4009 <tt>op1</tt> is less than <tt>op2</tt>.</li>
4010 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
4011 <tt>op1</tt> is less than or equal to <tt>op2</tt>.</li>
4012 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
4013 <tt>op1</tt> is not equal to <tt>op2</tt>.</li>
4014 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
4015 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
4019 <pre> <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
4020 <result> = fcmp one float 4.0, 5.0 <i>; yields: result=true</i>
4021 <result> = fcmp olt float 4.0, 5.0 <i>; yields: result=true</i>
4022 <result> = fcmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
4026 <!-- _______________________________________________________________________ -->
4027 <div class="doc_subsubsection">
4028 <a name="i_vicmp">'<tt>vicmp</tt>' Instruction</a>
4030 <div class="doc_text">
4032 <pre> <result> = vicmp <cond> <ty> <op1>, <op2> <i>; yields {ty}:result</i>
4035 <p>The '<tt>vicmp</tt>' instruction returns an integer vector value based on
4036 element-wise comparison of its two integer vector operands.</p>
4038 <p>The '<tt>vicmp</tt>' instruction takes three operands. The first operand is
4039 the condition code indicating the kind of comparison to perform. It is not
4040 a value, just a keyword. The possible condition code are:
4042 <li><tt>eq</tt>: equal</li>
4043 <li><tt>ne</tt>: not equal </li>
4044 <li><tt>ugt</tt>: unsigned greater than</li>
4045 <li><tt>uge</tt>: unsigned greater or equal</li>
4046 <li><tt>ult</tt>: unsigned less than</li>
4047 <li><tt>ule</tt>: unsigned less or equal</li>
4048 <li><tt>sgt</tt>: signed greater than</li>
4049 <li><tt>sge</tt>: signed greater or equal</li>
4050 <li><tt>slt</tt>: signed less than</li>
4051 <li><tt>sle</tt>: signed less or equal</li>
4053 <p>The remaining two arguments must be <a href="#t_vector">vector</a> or
4054 <a href="#t_integer">integer</a> typed. They must also be identical types.</p>
4056 <p>The '<tt>vicmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4057 according to the condition code given as <tt>cond</tt>. The comparison yields a
4058 <a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, of
4059 identical type as the values being compared. The most significant bit in each
4060 element is 1 if the element-wise comparison evaluates to true, and is 0
4061 otherwise. All other bits of the result are undefined. The condition codes
4062 are evaluated identically to the <a href="#i_icmp">'<tt>icmp</tt>'
4067 <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>
4068 <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>
4072 <!-- _______________________________________________________________________ -->
4073 <div class="doc_subsubsection">
4074 <a name="i_vfcmp">'<tt>vfcmp</tt>' Instruction</a>
4076 <div class="doc_text">
4078 <pre> <result> = vfcmp <cond> <ty> <op1>, <op2></pre>
4080 <p>The '<tt>vfcmp</tt>' instruction returns an integer vector value based on
4081 element-wise comparison of its two floating point vector operands. The output
4082 elements have the same width as the input elements.</p>
4084 <p>The '<tt>vfcmp</tt>' instruction takes three operands. The first operand is
4085 the condition code indicating the kind of comparison to perform. It is not
4086 a value, just a keyword. The possible condition code are:
4088 <li><tt>false</tt>: no comparison, always returns false</li>
4089 <li><tt>oeq</tt>: ordered and equal</li>
4090 <li><tt>ogt</tt>: ordered and greater than </li>
4091 <li><tt>oge</tt>: ordered and greater than or equal</li>
4092 <li><tt>olt</tt>: ordered and less than </li>
4093 <li><tt>ole</tt>: ordered and less than or equal</li>
4094 <li><tt>one</tt>: ordered and not equal</li>
4095 <li><tt>ord</tt>: ordered (no nans)</li>
4096 <li><tt>ueq</tt>: unordered or equal</li>
4097 <li><tt>ugt</tt>: unordered or greater than </li>
4098 <li><tt>uge</tt>: unordered or greater than or equal</li>
4099 <li><tt>ult</tt>: unordered or less than </li>
4100 <li><tt>ule</tt>: unordered or less than or equal</li>
4101 <li><tt>une</tt>: unordered or not equal</li>
4102 <li><tt>uno</tt>: unordered (either nans)</li>
4103 <li><tt>true</tt>: no comparison, always returns true</li>
4105 <p>The remaining two arguments must be <a href="#t_vector">vector</a> of
4106 <a href="#t_floating">floating point</a> typed. They must also be identical
4109 <p>The '<tt>vfcmp</tt>' instruction compares <tt>op1</tt> and <tt>op2</tt>
4110 according to the condition code given as <tt>cond</tt>. The comparison yields a
4111 <a href="#t_vector">vector</a> of <a href="#t_integer">integer</a> result, with
4112 an identical number of elements as the values being compared, and each element
4113 having identical with to the width of the floating point elements. The most
4114 significant bit in each element is 1 if the element-wise comparison evaluates to
4115 true, and is 0 otherwise. All other bits of the result are undefined. The
4116 condition codes are evaluated identically to the
4117 <a href="#i_fcmp">'<tt>fcmp</tt>' instruction</a>.
4121 <result> = vfcmp oeq <2 x float> < float 4, float 0 >, < float 5, float 0 > <i>; yields: result=<2 x i32> < i32 0, i32 -1 ></i>
4122 <result> = vfcmp ult <2 x double> < double 1, double 2 >, < double 2, double 2> <i>; yields: result=<2 x i64> < i64 -1, i64 0 ></i>
4126 <!-- _______________________________________________________________________ -->
4127 <div class="doc_subsubsection">
4128 <a name="i_phi">'<tt>phi</tt>' Instruction</a>
4131 <div class="doc_text">
4135 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
4137 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
4138 the SSA graph representing the function.</p>
4141 <p>The type of the incoming values is specified with the first type
4142 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
4143 as arguments, with one pair for each predecessor basic block of the
4144 current block. Only values of <a href="#t_firstclass">first class</a>
4145 type may be used as the value arguments to the PHI node. Only labels
4146 may be used as the label arguments.</p>
4148 <p>There must be no non-phi instructions between the start of a basic
4149 block and the PHI instructions: i.e. PHI instructions must be first in
4154 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
4155 specified by the pair corresponding to the predecessor basic block that executed
4156 just prior to the current block.</p>
4160 Loop: ; Infinite loop that counts from 0 on up...
4161 %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]
4162 %nextindvar = add i32 %indvar, 1
4167 <!-- _______________________________________________________________________ -->
4168 <div class="doc_subsubsection">
4169 <a name="i_select">'<tt>select</tt>' Instruction</a>
4172 <div class="doc_text">
4177 <result> = select <i>selty</i> <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
4179 <i>selty</i> is either i1 or {<N x i1>}
4185 The '<tt>select</tt>' instruction is used to choose one value based on a
4186 condition, without branching.
4193 The '<tt>select</tt>' instruction requires an 'i1' value or
4194 a vector of 'i1' values indicating the
4195 condition, and two values of the same <a href="#t_firstclass">first class</a>
4196 type. If the val1/val2 are vectors and
4197 the condition is a scalar, then entire vectors are selected, not
4198 individual elements.
4204 If the condition is an i1 and it evaluates to 1, the instruction returns the first
4205 value argument; otherwise, it returns the second value argument.
4208 If the condition is a vector of i1, then the value arguments must
4209 be vectors of the same size, and the selection is done element
4216 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
4221 <!-- _______________________________________________________________________ -->
4222 <div class="doc_subsubsection">
4223 <a name="i_call">'<tt>call</tt>' Instruction</a>
4226 <div class="doc_text">
4230 <result> = [tail] call [<a href="#callingconv">cconv</a>] <ty> [<fnty>*] <fnptrval>(<param list>)
4235 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
4239 <p>This instruction requires several arguments:</p>
4243 <p>The optional "tail" marker indicates whether the callee function accesses
4244 any allocas or varargs in the caller. If the "tail" marker is present, the
4245 function call is eligible for tail call optimization. Note that calls may
4246 be marked "tail" even if they do not occur before a <a
4247 href="#i_ret"><tt>ret</tt></a> instruction.
4250 <p>The optional "cconv" marker indicates which <a href="#callingconv">calling
4251 convention</a> the call should use. If none is specified, the call defaults
4252 to using C calling conventions.
4255 <p>'<tt>ty</tt>': the type of the call instruction itself which is also
4256 the type of the return value. Functions that return no value are marked
4257 <tt><a href="#t_void">void</a></tt>.</p>
4260 <p>'<tt>fnty</tt>': shall be the signature of the pointer to function
4261 value being invoked. The argument types must match the types implied by
4262 this signature. This type can be omitted if the function is not varargs
4263 and if the function type does not return a pointer to a function.</p>
4266 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
4267 be invoked. In most cases, this is a direct function invocation, but
4268 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
4269 to function value.</p>
4272 <p>'<tt>function args</tt>': argument list whose types match the
4273 function signature argument types. All arguments must be of
4274 <a href="#t_firstclass">first class</a> type. If the function signature
4275 indicates the function accepts a variable number of arguments, the extra
4276 arguments can be specified.</p>
4282 <p>The '<tt>call</tt>' instruction is used to cause control flow to
4283 transfer to a specified function, with its incoming arguments bound to
4284 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
4285 instruction in the called function, control flow continues with the
4286 instruction after the function call, and the return value of the
4287 function is bound to the result argument. If the callee returns multiple
4288 values then the return values of the function are only accessible through
4289 the '<tt><a href="#i_getresult">getresult</a></tt>' instruction.</p>
4294 %retval = call i32 @test(i32 %argc)
4295 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42) <i>; yields i32</i>
4296 %X = tail call i32 @foo() <i>; yields i32</i>
4297 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
4298 call void %foo(i8 97 signext)
4300 %struct.A = type { i32, i8 }
4301 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
4302 %gr = getresult %struct.A %r, 0 <i>; yields i32</i>
4303 %gr1 = getresult %struct.A %r, 1 <i>; yields i8</i>
4308 <!-- _______________________________________________________________________ -->
4309 <div class="doc_subsubsection">
4310 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
4313 <div class="doc_text">
4318 <resultval> = va_arg <va_list*> <arglist>, <argty>
4323 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
4324 the "variable argument" area of a function call. It is used to implement the
4325 <tt>va_arg</tt> macro in C.</p>
4329 <p>This instruction takes a <tt>va_list*</tt> value and the type of
4330 the argument. It returns a value of the specified argument type and
4331 increments the <tt>va_list</tt> to point to the next argument. The
4332 actual type of <tt>va_list</tt> is target specific.</p>
4336 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
4337 type from the specified <tt>va_list</tt> and causes the
4338 <tt>va_list</tt> to point to the next argument. For more information,
4339 see the variable argument handling <a href="#int_varargs">Intrinsic
4342 <p>It is legal for this instruction to be called in a function which does not
4343 take a variable number of arguments, for example, the <tt>vfprintf</tt>
4346 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
4347 href="#intrinsics">intrinsic function</a> because it takes a type as an
4352 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
4356 <!-- _______________________________________________________________________ -->
4357 <div class="doc_subsubsection">
4358 <a name="i_getresult">'<tt>getresult</tt>' Instruction</a>
4361 <div class="doc_text">
4365 <resultval> = getresult <type> <retval>, <index>
4370 <p> The '<tt>getresult</tt>' instruction is used to extract individual values
4371 from a '<tt><a href="#i_call">call</a></tt>'
4372 or '<tt><a href="#i_invoke">invoke</a></tt>' instruction that returns multiple
4377 <p>The '<tt>getresult</tt>' instruction takes a call or invoke value as its
4378 first argument, or an undef value. The value must have <a
4379 href="#t_struct">structure type</a>. The second argument is a constant
4380 unsigned index value which must be in range for the number of values returned
4385 <p>The '<tt>getresult</tt>' instruction extracts the element identified by
4386 '<tt>index</tt>' from the aggregate value.</p>
4391 %struct.A = type { i32, i8 }
4393 %r = call %struct.A @foo()
4394 %gr = getresult %struct.A %r, 0 <i>; yields i32:%gr</i>
4395 %gr1 = getresult %struct.A %r, 1 <i>; yields i8:%gr1</i>
4402 <!-- *********************************************************************** -->
4403 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
4404 <!-- *********************************************************************** -->
4406 <div class="doc_text">
4408 <p>LLVM supports the notion of an "intrinsic function". These functions have
4409 well known names and semantics and are required to follow certain restrictions.
4410 Overall, these intrinsics represent an extension mechanism for the LLVM
4411 language that does not require changing all of the transformations in LLVM when
4412 adding to the language (or the bitcode reader/writer, the parser, etc...).</p>
4414 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
4415 prefix is reserved in LLVM for intrinsic names; thus, function names may not
4416 begin with this prefix. Intrinsic functions must always be external functions:
4417 you cannot define the body of intrinsic functions. Intrinsic functions may
4418 only be used in call or invoke instructions: it is illegal to take the address
4419 of an intrinsic function. Additionally, because intrinsic functions are part
4420 of the LLVM language, it is required if any are added that they be documented
4423 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
4424 a family of functions that perform the same operation but on different data
4425 types. Because LLVM can represent over 8 million different integer types,
4426 overloading is used commonly to allow an intrinsic function to operate on any
4427 integer type. One or more of the argument types or the result type can be
4428 overloaded to accept any integer type. Argument types may also be defined as
4429 exactly matching a previous argument's type or the result type. This allows an
4430 intrinsic function which accepts multiple arguments, but needs all of them to
4431 be of the same type, to only be overloaded with respect to a single argument or
4434 <p>Overloaded intrinsics will have the names of its overloaded argument types
4435 encoded into its function name, each preceded by a period. Only those types
4436 which are overloaded result in a name suffix. Arguments whose type is matched
4437 against another type do not. For example, the <tt>llvm.ctpop</tt> function can
4438 take an integer of any width and returns an integer of exactly the same integer
4439 width. This leads to a family of functions such as
4440 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29 %val)</tt>.
4441 Only one type, the return type, is overloaded, and only one type suffix is
4442 required. Because the argument's type is matched against the return type, it
4443 does not require its own name suffix.</p>
4445 <p>To learn how to add an intrinsic function, please see the
4446 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.
4451 <!-- ======================================================================= -->
4452 <div class="doc_subsection">
4453 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
4456 <div class="doc_text">
4458 <p>Variable argument support is defined in LLVM with the <a
4459 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
4460 intrinsic functions. These functions are related to the similarly
4461 named macros defined in the <tt><stdarg.h></tt> header file.</p>
4463 <p>All of these functions operate on arguments that use a
4464 target-specific value type "<tt>va_list</tt>". The LLVM assembly
4465 language reference manual does not define what this type is, so all
4466 transformations should be prepared to handle these functions regardless of
4469 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
4470 instruction and the variable argument handling intrinsic functions are
4473 <div class="doc_code">
4475 define i32 @test(i32 %X, ...) {
4476 ; Initialize variable argument processing
4478 %ap2 = bitcast i8** %ap to i8*
4479 call void @llvm.va_start(i8* %ap2)
4481 ; Read a single integer argument
4482 %tmp = va_arg i8** %ap, i32
4484 ; Demonstrate usage of llvm.va_copy and llvm.va_end
4486 %aq2 = bitcast i8** %aq to i8*
4487 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
4488 call void @llvm.va_end(i8* %aq2)
4490 ; Stop processing of arguments.
4491 call void @llvm.va_end(i8* %ap2)
4495 declare void @llvm.va_start(i8*)
4496 declare void @llvm.va_copy(i8*, i8*)
4497 declare void @llvm.va_end(i8*)
4503 <!-- _______________________________________________________________________ -->
4504 <div class="doc_subsubsection">
4505 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
4509 <div class="doc_text">
4511 <pre> declare void %llvm.va_start(i8* <arglist>)<br></pre>
4513 <P>The '<tt>llvm.va_start</tt>' intrinsic initializes
4514 <tt>*<arglist></tt> for subsequent use by <tt><a
4515 href="#i_va_arg">va_arg</a></tt>.</p>
4519 <P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
4523 <P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
4524 macro available in C. In a target-dependent way, it initializes the
4525 <tt>va_list</tt> element to which the argument points, so that the next call to
4526 <tt>va_arg</tt> will produce the first variable argument passed to the function.
4527 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
4528 last argument of the function as the compiler can figure that out.</p>
4532 <!-- _______________________________________________________________________ -->
4533 <div class="doc_subsubsection">
4534 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
4537 <div class="doc_text">
4539 <pre> declare void @llvm.va_end(i8* <arglist>)<br></pre>
4542 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
4543 which has been initialized previously with <tt><a href="#int_va_start">llvm.va_start</a></tt>
4544 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
4548 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
4552 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
4553 macro available in C. In a target-dependent way, it destroys the
4554 <tt>va_list</tt> element to which the argument points. Calls to <a
4555 href="#int_va_start"><tt>llvm.va_start</tt></a> and <a href="#int_va_copy">
4556 <tt>llvm.va_copy</tt></a> must be matched exactly with calls to
4557 <tt>llvm.va_end</tt>.</p>
4561 <!-- _______________________________________________________________________ -->
4562 <div class="doc_subsubsection">
4563 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
4566 <div class="doc_text">
4571 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
4576 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
4577 from the source argument list to the destination argument list.</p>
4581 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
4582 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
4587 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
4588 macro available in C. In a target-dependent way, it copies the source
4589 <tt>va_list</tt> element into the destination <tt>va_list</tt> element. This
4590 intrinsic is necessary because the <tt><a href="#int_va_start">
4591 llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
4592 example, memory allocation.</p>
4596 <!-- ======================================================================= -->
4597 <div class="doc_subsection">
4598 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
4601 <div class="doc_text">
4604 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
4605 Collection</a> (GC) requires the implementation and generation of these
4607 These intrinsics allow identification of <a href="#int_gcroot">GC roots on the
4608 stack</a>, as well as garbage collector implementations that require <a
4609 href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a> barriers.
4610 Front-ends for type-safe garbage collected languages should generate these
4611 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
4612 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
4615 <p>The garbage collection intrinsics only operate on objects in the generic
4616 address space (address space zero).</p>
4620 <!-- _______________________________________________________________________ -->
4621 <div class="doc_subsubsection">
4622 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
4625 <div class="doc_text">
4630 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
4635 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
4636 the code generator, and allows some metadata to be associated with it.</p>
4640 <p>The first argument specifies the address of a stack object that contains the
4641 root pointer. The second pointer (which must be either a constant or a global
4642 value address) contains the meta-data to be associated with the root.</p>
4646 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
4647 location. At compile-time, the code generator generates information to allow
4648 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
4649 intrinsic may only be used in a function which <a href="#gc">specifies a GC
4655 <!-- _______________________________________________________________________ -->
4656 <div class="doc_subsubsection">
4657 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
4660 <div class="doc_text">
4665 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
4670 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
4671 locations, allowing garbage collector implementations that require read
4676 <p>The second argument is the address to read from, which should be an address
4677 allocated from the garbage collector. The first object is a pointer to the
4678 start of the referenced object, if needed by the language runtime (otherwise
4683 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
4684 instruction, but may be replaced with substantially more complex code by the
4685 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
4686 may only be used in a function which <a href="#gc">specifies a GC
4692 <!-- _______________________________________________________________________ -->
4693 <div class="doc_subsubsection">
4694 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
4697 <div class="doc_text">
4702 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
4707 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
4708 locations, allowing garbage collector implementations that require write
4709 barriers (such as generational or reference counting collectors).</p>
4713 <p>The first argument is the reference to store, the second is the start of the
4714 object to store it to, and the third is the address of the field of Obj to
4715 store to. If the runtime does not require a pointer to the object, Obj may be
4720 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
4721 instruction, but may be replaced with substantially more complex code by the
4722 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
4723 may only be used in a function which <a href="#gc">specifies a GC
4730 <!-- ======================================================================= -->
4731 <div class="doc_subsection">
4732 <a name="int_codegen">Code Generator Intrinsics</a>
4735 <div class="doc_text">
4737 These intrinsics are provided by LLVM to expose special features that may only
4738 be implemented with code generator support.
4743 <!-- _______________________________________________________________________ -->
4744 <div class="doc_subsubsection">
4745 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
4748 <div class="doc_text">
4752 declare i8 *@llvm.returnaddress(i32 <level>)
4758 The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
4759 target-specific value indicating the return address of the current function
4760 or one of its callers.
4766 The argument to this intrinsic indicates which function to return the address
4767 for. Zero indicates the calling function, one indicates its caller, etc. The
4768 argument is <b>required</b> to be a constant integer value.
4774 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
4775 the return address of the specified call frame, or zero if it cannot be
4776 identified. The value returned by this intrinsic is likely to be incorrect or 0
4777 for arguments other than zero, so it should only be used for debugging purposes.
4781 Note that calling this intrinsic does not prevent function inlining or other
4782 aggressive transformations, so the value returned may not be that of the obvious
4783 source-language caller.
4788 <!-- _______________________________________________________________________ -->
4789 <div class="doc_subsubsection">
4790 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
4793 <div class="doc_text">
4797 declare i8 *@llvm.frameaddress(i32 <level>)
4803 The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
4804 target-specific frame pointer value for the specified stack frame.
4810 The argument to this intrinsic indicates which function to return the frame
4811 pointer for. Zero indicates the calling function, one indicates its caller,
4812 etc. The argument is <b>required</b> to be a constant integer value.
4818 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
4819 the frame address of the specified call frame, or zero if it cannot be
4820 identified. The value returned by this intrinsic is likely to be incorrect or 0
4821 for arguments other than zero, so it should only be used for debugging purposes.
4825 Note that calling this intrinsic does not prevent function inlining or other
4826 aggressive transformations, so the value returned may not be that of the obvious
4827 source-language caller.
4831 <!-- _______________________________________________________________________ -->
4832 <div class="doc_subsubsection">
4833 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
4836 <div class="doc_text">
4840 declare i8 *@llvm.stacksave()
4846 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
4847 the function stack, for use with <a href="#int_stackrestore">
4848 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
4849 features like scoped automatic variable sized arrays in C99.
4855 This intrinsic returns a opaque pointer value that can be passed to <a
4856 href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
4857 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
4858 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
4859 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
4860 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
4861 that were allocated after the <tt>llvm.stacksave</tt> was executed.
4866 <!-- _______________________________________________________________________ -->
4867 <div class="doc_subsubsection">
4868 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
4871 <div class="doc_text">
4875 declare void @llvm.stackrestore(i8 * %ptr)
4881 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
4882 the function stack to the state it was in when the corresponding <a
4883 href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
4884 useful for implementing language features like scoped automatic variable sized
4891 See the description for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.
4897 <!-- _______________________________________________________________________ -->
4898 <div class="doc_subsubsection">
4899 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
4902 <div class="doc_text">
4906 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>)
4913 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
4914 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
4916 effect on the behavior of the program but can change its performance
4923 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
4924 determining if the fetch should be for a read (0) or write (1), and
4925 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
4926 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
4927 <tt>locality</tt> arguments must be constant integers.
4933 This intrinsic does not modify the behavior of the program. In particular,
4934 prefetches cannot trap and do not produce a value. On targets that support this
4935 intrinsic, the prefetch can provide hints to the processor cache for better
4941 <!-- _______________________________________________________________________ -->
4942 <div class="doc_subsubsection">
4943 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
4946 <div class="doc_text">
4950 declare void @llvm.pcmarker(i32 <id>)
4957 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
4959 code to simulators and other tools. The method is target specific, but it is
4960 expected that the marker will use exported symbols to transmit the PC of the
4962 The marker makes no guarantees that it will remain with any specific instruction
4963 after optimizations. It is possible that the presence of a marker will inhibit
4964 optimizations. The intended use is to be inserted after optimizations to allow
4965 correlations of simulation runs.
4971 <tt>id</tt> is a numerical id identifying the marker.
4977 This intrinsic does not modify the behavior of the program. Backends that do not
4978 support this intrinisic may ignore it.
4983 <!-- _______________________________________________________________________ -->
4984 <div class="doc_subsubsection">
4985 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
4988 <div class="doc_text">
4992 declare i64 @llvm.readcyclecounter( )
4999 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
5000 counter register (or similar low latency, high accuracy clocks) on those targets
5001 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
5002 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
5003 should only be used for small timings.
5009 When directly supported, reading the cycle counter should not modify any memory.
5010 Implementations are allowed to either return a application specific value or a
5011 system wide value. On backends without support, this is lowered to a constant 0.
5016 <!-- ======================================================================= -->
5017 <div class="doc_subsection">
5018 <a name="int_libc">Standard C Library Intrinsics</a>
5021 <div class="doc_text">
5023 LLVM provides intrinsics for a few important standard C library functions.
5024 These intrinsics allow source-language front-ends to pass information about the
5025 alignment of the pointer arguments to the code generator, providing opportunity
5026 for more efficient code generation.
5031 <!-- _______________________________________________________________________ -->
5032 <div class="doc_subsubsection">
5033 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
5036 <div class="doc_text">
5040 declare void @llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
5041 i32 <len>, i32 <align>)
5042 declare void @llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
5043 i64 <len>, i32 <align>)
5049 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
5050 location to the destination location.
5054 Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
5055 intrinsics do not return a value, and takes an extra alignment argument.
5061 The first argument is a pointer to the destination, the second is a pointer to
5062 the source. The third argument is an integer argument
5063 specifying the number of bytes to copy, and the fourth argument is the alignment
5064 of the source and destination locations.
5068 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5069 the caller guarantees that both the source and destination pointers are aligned
5076 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
5077 location to the destination location, which are not allowed to overlap. It
5078 copies "len" bytes of memory over. If the argument is known to be aligned to
5079 some boundary, this can be specified as the fourth argument, otherwise it should
5085 <!-- _______________________________________________________________________ -->
5086 <div class="doc_subsubsection">
5087 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
5090 <div class="doc_text">
5094 declare void @llvm.memmove.i32(i8 * <dest>, i8 * <src>,
5095 i32 <len>, i32 <align>)
5096 declare void @llvm.memmove.i64(i8 * <dest>, i8 * <src>,
5097 i64 <len>, i32 <align>)
5103 The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
5104 location to the destination location. It is similar to the
5105 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to overlap.
5109 Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
5110 intrinsics do not return a value, and takes an extra alignment argument.
5116 The first argument is a pointer to the destination, the second is a pointer to
5117 the source. The third argument is an integer argument
5118 specifying the number of bytes to copy, and the fourth argument is the alignment
5119 of the source and destination locations.
5123 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5124 the caller guarantees that the source and destination pointers are aligned to
5131 The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
5132 location to the destination location, which may overlap. It
5133 copies "len" bytes of memory over. If the argument is known to be aligned to
5134 some boundary, this can be specified as the fourth argument, otherwise it should
5140 <!-- _______________________________________________________________________ -->
5141 <div class="doc_subsubsection">
5142 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
5145 <div class="doc_text">
5149 declare void @llvm.memset.i32(i8 * <dest>, i8 <val>,
5150 i32 <len>, i32 <align>)
5151 declare void @llvm.memset.i64(i8 * <dest>, i8 <val>,
5152 i64 <len>, i32 <align>)
5158 The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
5163 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
5164 does not return a value, and takes an extra alignment argument.
5170 The first argument is a pointer to the destination to fill, the second is the
5171 byte value to fill it with, the third argument is an integer
5172 argument specifying the number of bytes to fill, and the fourth argument is the
5173 known alignment of destination location.
5177 If the call to this intrinisic has an alignment value that is not 0 or 1, then
5178 the caller guarantees that the destination pointer is aligned to that boundary.
5184 The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
5186 destination location. If the argument is known to be aligned to some boundary,
5187 this can be specified as the fourth argument, otherwise it should be set to 0 or
5193 <!-- _______________________________________________________________________ -->
5194 <div class="doc_subsubsection">
5195 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
5198 <div class="doc_text">
5201 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
5202 floating point or vector of floating point type. Not all targets support all
5205 declare float @llvm.sqrt.f32(float %Val)
5206 declare double @llvm.sqrt.f64(double %Val)
5207 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
5208 declare fp128 @llvm.sqrt.f128(fp128 %Val)
5209 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
5215 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
5216 returning the same value as the libm '<tt>sqrt</tt>' functions would. Unlike
5217 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
5218 negative numbers other than -0.0 (which allows for better optimization, because
5219 there is no need to worry about errno being set). <tt>llvm.sqrt(-0.0)</tt> is
5220 defined to return -0.0 like IEEE sqrt.
5226 The argument and return value are floating point numbers of the same type.
5232 This function returns the sqrt of the specified operand if it is a nonnegative
5233 floating point number.
5237 <!-- _______________________________________________________________________ -->
5238 <div class="doc_subsubsection">
5239 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
5242 <div class="doc_text">
5245 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
5246 floating point or vector of floating point type. Not all targets support all
5249 declare float @llvm.powi.f32(float %Val, i32 %power)
5250 declare double @llvm.powi.f64(double %Val, i32 %power)
5251 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
5252 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
5253 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
5259 The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
5260 specified (positive or negative) power. The order of evaluation of
5261 multiplications is not defined. When a vector of floating point type is
5262 used, the second argument remains a scalar integer value.
5268 The second argument is an integer power, and the first is a value to raise to
5275 This function returns the first value raised to the second power with an
5276 unspecified sequence of rounding operations.</p>
5279 <!-- _______________________________________________________________________ -->
5280 <div class="doc_subsubsection">
5281 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
5284 <div class="doc_text">
5287 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
5288 floating point or vector of floating point type. Not all targets support all
5291 declare float @llvm.sin.f32(float %Val)
5292 declare double @llvm.sin.f64(double %Val)
5293 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
5294 declare fp128 @llvm.sin.f128(fp128 %Val)
5295 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
5301 The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.
5307 The argument and return value are floating point numbers of the same type.
5313 This function returns the sine of the specified operand, returning the
5314 same values as the libm <tt>sin</tt> functions would, and handles error
5315 conditions in the same way.</p>
5318 <!-- _______________________________________________________________________ -->
5319 <div class="doc_subsubsection">
5320 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
5323 <div class="doc_text">
5326 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
5327 floating point or vector of floating point type. Not all targets support all
5330 declare float @llvm.cos.f32(float %Val)
5331 declare double @llvm.cos.f64(double %Val)
5332 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
5333 declare fp128 @llvm.cos.f128(fp128 %Val)
5334 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
5340 The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.
5346 The argument and return value are floating point numbers of the same type.
5352 This function returns the cosine of the specified operand, returning the
5353 same values as the libm <tt>cos</tt> functions would, and handles error
5354 conditions in the same way.</p>
5357 <!-- _______________________________________________________________________ -->
5358 <div class="doc_subsubsection">
5359 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
5362 <div class="doc_text">
5365 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
5366 floating point or vector of floating point type. Not all targets support all
5369 declare float @llvm.pow.f32(float %Val, float %Power)
5370 declare double @llvm.pow.f64(double %Val, double %Power)
5371 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
5372 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
5373 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
5379 The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
5380 specified (positive or negative) power.
5386 The second argument is a floating point power, and the first is a value to
5387 raise to that power.
5393 This function returns the first value raised to the second power,
5395 same values as the libm <tt>pow</tt> functions would, and handles error
5396 conditions in the same way.</p>
5400 <!-- ======================================================================= -->
5401 <div class="doc_subsection">
5402 <a name="int_manip">Bit Manipulation Intrinsics</a>
5405 <div class="doc_text">
5407 LLVM provides intrinsics for a few important bit manipulation operations.
5408 These allow efficient code generation for some algorithms.
5413 <!-- _______________________________________________________________________ -->
5414 <div class="doc_subsubsection">
5415 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
5418 <div class="doc_text">
5421 <p>This is an overloaded intrinsic function. You can use bswap on any integer
5422 type that is an even number of bytes (i.e. BitWidth % 16 == 0).
5424 declare i16 @llvm.bswap.i16(i16 <id>)
5425 declare i32 @llvm.bswap.i32(i32 <id>)
5426 declare i64 @llvm.bswap.i64(i64 <id>)
5432 The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
5433 values with an even number of bytes (positive multiple of 16 bits). These are
5434 useful for performing operations on data that is not in the target's native
5441 The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
5442 and low byte of the input i16 swapped. Similarly, the <tt>llvm.bswap.i32</tt>
5443 intrinsic returns an i32 value that has the four bytes of the input i32
5444 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the returned
5445 i32 will have its bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i48</tt>,
5446 <tt>llvm.bswap.i64</tt> and other intrinsics extend this concept to
5447 additional even-byte lengths (6 bytes, 8 bytes and more, respectively).
5452 <!-- _______________________________________________________________________ -->
5453 <div class="doc_subsubsection">
5454 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
5457 <div class="doc_text">
5460 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
5461 width. Not all targets support all bit widths however.
5463 declare i8 @llvm.ctpop.i8 (i8 <src>)
5464 declare i16 @llvm.ctpop.i16(i16 <src>)
5465 declare i32 @llvm.ctpop.i32(i32 <src>)
5466 declare i64 @llvm.ctpop.i64(i64 <src>)
5467 declare i256 @llvm.ctpop.i256(i256 <src>)
5473 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
5480 The only argument is the value to be counted. The argument may be of any
5481 integer type. The return type must match the argument type.
5487 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
5491 <!-- _______________________________________________________________________ -->
5492 <div class="doc_subsubsection">
5493 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
5496 <div class="doc_text">
5499 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
5500 integer bit width. Not all targets support all bit widths however.
5502 declare i8 @llvm.ctlz.i8 (i8 <src>)
5503 declare i16 @llvm.ctlz.i16(i16 <src>)
5504 declare i32 @llvm.ctlz.i32(i32 <src>)
5505 declare i64 @llvm.ctlz.i64(i64 <src>)
5506 declare i256 @llvm.ctlz.i256(i256 <src>)
5512 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
5513 leading zeros in a variable.
5519 The only argument is the value to be counted. The argument may be of any
5520 integer type. The return type must match the argument type.
5526 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
5527 in a variable. If the src == 0 then the result is the size in bits of the type
5528 of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.
5534 <!-- _______________________________________________________________________ -->
5535 <div class="doc_subsubsection">
5536 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
5539 <div class="doc_text">
5542 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
5543 integer bit width. Not all targets support all bit widths however.
5545 declare i8 @llvm.cttz.i8 (i8 <src>)
5546 declare i16 @llvm.cttz.i16(i16 <src>)
5547 declare i32 @llvm.cttz.i32(i32 <src>)
5548 declare i64 @llvm.cttz.i64(i64 <src>)
5549 declare i256 @llvm.cttz.i256(i256 <src>)
5555 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
5562 The only argument is the value to be counted. The argument may be of any
5563 integer type. The return type must match the argument type.
5569 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
5570 in a variable. If the src == 0 then the result is the size in bits of the type
5571 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
5575 <!-- _______________________________________________________________________ -->
5576 <div class="doc_subsubsection">
5577 <a name="int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic</a>
5580 <div class="doc_text">
5583 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.select</tt>
5584 on any integer bit width.
5586 declare i17 @llvm.part.select.i17 (i17 %val, i32 %loBit, i32 %hiBit)
5587 declare i29 @llvm.part.select.i29 (i29 %val, i32 %loBit, i32 %hiBit)
5591 <p>The '<tt>llvm.part.select</tt>' family of intrinsic functions selects a
5592 range of bits from an integer value and returns them in the same bit width as
5593 the original value.</p>
5596 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5597 any bit width but they must have the same bit width. The second and third
5598 arguments must be <tt>i32</tt> type since they specify only a bit index.</p>
5601 <p>The operation of the '<tt>llvm.part.select</tt>' intrinsic has two modes
5602 of operation: forwards and reverse. If <tt>%loBit</tt> is greater than
5603 <tt>%hiBits</tt> then the intrinsic operates in reverse mode. Otherwise it
5604 operates in forward mode.</p>
5605 <p>In forward mode, this intrinsic is the equivalent of shifting <tt>%val</tt>
5606 right by <tt>%loBit</tt> bits and then ANDing it with a mask with
5607 only the <tt>%hiBit - %loBit</tt> bits set, as follows:</p>
5609 <li>The <tt>%val</tt> is shifted right (LSHR) by the number of bits specified
5610 by <tt>%loBits</tt>. This normalizes the value to the low order bits.</li>
5611 <li>The <tt>%loBits</tt> value is subtracted from the <tt>%hiBits</tt> value
5612 to determine the number of bits to retain.</li>
5613 <li>A mask of the retained bits is created by shifting a -1 value.</li>
5614 <li>The mask is ANDed with <tt>%val</tt> to produce the result.
5616 <p>In reverse mode, a similar computation is made except that the bits are
5617 returned in the reverse order. So, for example, if <tt>X</tt> has the value
5618 <tt>i16 0x0ACF (101011001111)</tt> and we apply
5619 <tt>part.select(i16 X, 8, 3)</tt> to it, we get back the value
5620 <tt>i16 0x0026 (000000100110)</tt>.</p>
5623 <div class="doc_subsubsection">
5624 <a name="int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic</a>
5627 <div class="doc_text">
5630 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.set</tt>
5631 on any integer bit width.
5633 declare i17 @llvm.part.set.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
5634 declare i29 @llvm.part.set.i29.i9 (i29 %val, i9 %repl, i32 %lo, i32 %hi)
5638 <p>The '<tt>llvm.part.set</tt>' family of intrinsic functions replaces a range
5639 of bits in an integer value with another integer value. It returns the integer
5640 with the replaced bits.</p>
5643 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5644 any bit width but they must have the same bit width. <tt>%val</tt> is the value
5645 whose bits will be replaced. The second argument, <tt>%repl</tt> may be an
5646 integer of any bit width. The third and fourth arguments must be <tt>i32</tt>
5647 type since they specify only a bit index.</p>
5650 <p>The operation of the '<tt>llvm.part.set</tt>' intrinsic has two modes
5651 of operation: forwards and reverse. If <tt>%lo</tt> is greater than
5652 <tt>%hi</tt> then the intrinsic operates in reverse mode. Otherwise it
5653 operates in forward mode.</p>
5654 <p>For both modes, the <tt>%repl</tt> value is prepared for use by either
5655 truncating it down to the size of the replacement area or zero extending it
5656 up to that size.</p>
5657 <p>In forward mode, the bits between <tt>%lo</tt> and <tt>%hi</tt> (inclusive)
5658 are replaced with corresponding bits from <tt>%repl</tt>. That is the 0th bit
5659 in <tt>%repl</tt> replaces the <tt>%lo</tt>th bit in <tt>%val</tt> and etc. up
5660 to the <tt>%hi</tt>th bit.
5661 <p>In reverse mode, a similar computation is made except that the bits are
5662 reversed. That is, the <tt>0</tt>th bit in <tt>%repl</tt> replaces the
5663 <tt>%hi</tt> bit in <tt>%val</tt> and etc. down to the <tt>%lo</tt>th bit.
5666 llvm.part.set(0xFFFF, 0, 4, 7) -> 0xFF0F
5667 llvm.part.set(0xFFFF, 0, 7, 4) -> 0xFF0F
5668 llvm.part.set(0xFFFF, 1, 7, 4) -> 0xFF8F
5669 llvm.part.set(0xFFFF, F, 8, 3) -> 0xFFE7
5670 llvm.part.set(0xFFFF, 0, 3, 8) -> 0xFE07
5674 <!-- ======================================================================= -->
5675 <div class="doc_subsection">
5676 <a name="int_debugger">Debugger Intrinsics</a>
5679 <div class="doc_text">
5681 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
5682 are described in the <a
5683 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
5684 Debugging</a> document.
5689 <!-- ======================================================================= -->
5690 <div class="doc_subsection">
5691 <a name="int_eh">Exception Handling Intrinsics</a>
5694 <div class="doc_text">
5695 <p> The LLVM exception handling intrinsics (which all start with
5696 <tt>llvm.eh.</tt> prefix), are described in the <a
5697 href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
5698 Handling</a> document. </p>
5701 <!-- ======================================================================= -->
5702 <div class="doc_subsection">
5703 <a name="int_trampoline">Trampoline Intrinsic</a>
5706 <div class="doc_text">
5708 This intrinsic makes it possible to excise one parameter, marked with
5709 the <tt>nest</tt> attribute, from a function. The result is a callable
5710 function pointer lacking the nest parameter - the caller does not need
5711 to provide a value for it. Instead, the value to use is stored in
5712 advance in a "trampoline", a block of memory usually allocated
5713 on the stack, which also contains code to splice the nest value into the
5714 argument list. This is used to implement the GCC nested function address
5718 For example, if the function is
5719 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
5720 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as follows:</p>
5722 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
5723 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
5724 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
5725 %fp = bitcast i8* %p to i32 (i32, i32)*
5727 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
5728 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
5731 <!-- _______________________________________________________________________ -->
5732 <div class="doc_subsubsection">
5733 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
5735 <div class="doc_text">
5738 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
5742 This fills the memory pointed to by <tt>tramp</tt> with code
5743 and returns a function pointer suitable for executing it.
5747 The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
5748 pointers. The <tt>tramp</tt> argument must point to a sufficiently large
5749 and sufficiently aligned block of memory; this memory is written to by the
5750 intrinsic. Note that the size and the alignment are target-specific - LLVM
5751 currently provides no portable way of determining them, so a front-end that
5752 generates this intrinsic needs to have some target-specific knowledge.
5753 The <tt>func</tt> argument must hold a function bitcast to an <tt>i8*</tt>.
5757 The block of memory pointed to by <tt>tramp</tt> is filled with target
5758 dependent code, turning it into a function. A pointer to this function is
5759 returned, but needs to be bitcast to an
5760 <a href="#int_trampoline">appropriate function pointer type</a>
5761 before being called. The new function's signature is the same as that of
5762 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
5763 removed. At most one such <tt>nest</tt> argument is allowed, and it must be
5764 of pointer type. Calling the new function is equivalent to calling
5765 <tt>func</tt> with the same argument list, but with <tt>nval</tt> used for the
5766 missing <tt>nest</tt> argument. If, after calling
5767 <tt>llvm.init.trampoline</tt>, the memory pointed to by <tt>tramp</tt> is
5768 modified, then the effect of any later call to the returned function pointer is
5773 <!-- ======================================================================= -->
5774 <div class="doc_subsection">
5775 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
5778 <div class="doc_text">
5780 These intrinsic functions expand the "universal IR" of LLVM to represent
5781 hardware constructs for atomic operations and memory synchronization. This
5782 provides an interface to the hardware, not an interface to the programmer. It
5783 is aimed at a low enough level to allow any programming models or APIs
5784 (Application Programming Interfaces) which
5785 need atomic behaviors to map cleanly onto it. It is also modeled primarily on
5786 hardware behavior. Just as hardware provides a "universal IR" for source
5787 languages, it also provides a starting point for developing a "universal"
5788 atomic operation and synchronization IR.
5791 These do <em>not</em> form an API such as high-level threading libraries,
5792 software transaction memory systems, atomic primitives, and intrinsic
5793 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
5794 application libraries. The hardware interface provided by LLVM should allow
5795 a clean implementation of all of these APIs and parallel programming models.
5796 No one model or paradigm should be selected above others unless the hardware
5797 itself ubiquitously does so.
5802 <!-- _______________________________________________________________________ -->
5803 <div class="doc_subsubsection">
5804 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
5806 <div class="doc_text">
5809 declare void @llvm.memory.barrier( i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>,
5815 The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
5816 specific pairs of memory access types.
5820 The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
5821 The first four arguments enables a specific barrier as listed below. The fith
5822 argument specifies that the barrier applies to io or device or uncached memory.
5826 <li><tt>ll</tt>: load-load barrier</li>
5827 <li><tt>ls</tt>: load-store barrier</li>
5828 <li><tt>sl</tt>: store-load barrier</li>
5829 <li><tt>ss</tt>: store-store barrier</li>
5830 <li><tt>device</tt>: barrier applies to device and uncached memory also.
5834 This intrinsic causes the system to enforce some ordering constraints upon
5835 the loads and stores of the program. This barrier does not indicate
5836 <em>when</em> any events will occur, it only enforces an <em>order</em> in
5837 which they occur. For any of the specified pairs of load and store operations
5838 (f.ex. load-load, or store-load), all of the first operations preceding the
5839 barrier will complete before any of the second operations succeeding the
5840 barrier begin. Specifically the semantics for each pairing is as follows:
5843 <li><tt>ll</tt>: All loads before the barrier must complete before any load
5844 after the barrier begins.</li>
5846 <li><tt>ls</tt>: All loads before the barrier must complete before any
5847 store after the barrier begins.</li>
5848 <li><tt>ss</tt>: All stores before the barrier must complete before any
5849 store after the barrier begins.</li>
5850 <li><tt>sl</tt>: All stores before the barrier must complete before any
5851 load after the barrier begins.</li>
5854 These semantics are applied with a logical "and" behavior when more than one
5855 is enabled in a single memory barrier intrinsic.
5858 Backends may implement stronger barriers than those requested when they do not
5859 support as fine grained a barrier as requested. Some architectures do not
5860 need all types of barriers and on such architectures, these become noops.
5867 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
5868 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
5869 <i>; guarantee the above finishes</i>
5870 store i32 8, %ptr <i>; before this begins</i>
5874 <!-- _______________________________________________________________________ -->
5875 <div class="doc_subsubsection">
5876 <a name="int_atomic_cmp_swap">'<tt>llvm.atomic.cmp.swap.*</tt>' Intrinsic</a>
5878 <div class="doc_text">
5881 This is an overloaded intrinsic. You can use <tt>llvm.atomic.cmp.swap</tt> on
5882 any integer bit width and for different address spaces. Not all targets
5883 support all bit widths however.</p>
5886 declare i8 @llvm.atomic.cmp.swap.i8.p0i8( i8* <ptr>, i8 <cmp>, i8 <val> )
5887 declare i16 @llvm.atomic.cmp.swap.i16.p0i16( i16* <ptr>, i16 <cmp>, i16 <val> )
5888 declare i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* <ptr>, i32 <cmp>, i32 <val> )
5889 declare i64 @llvm.atomic.cmp.swap.i64.p0i64( i64* <ptr>, i64 <cmp>, i64 <val> )
5894 This loads a value in memory and compares it to a given value. If they are
5895 equal, it stores a new value into the memory.
5899 The <tt>llvm.atomic.cmp.swap</tt> intrinsic takes three arguments. The result as
5900 well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
5901 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
5902 this integer type. While any bit width integer may be used, targets may only
5903 lower representations they support in hardware.
5908 This entire intrinsic must be executed atomically. It first loads the value
5909 in memory pointed to by <tt>ptr</tt> and compares it with the value
5910 <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the memory. The
5911 loaded value is yielded in all cases. This provides the equivalent of an
5912 atomic compare-and-swap operation within the SSA framework.
5920 %val1 = add i32 4, 4
5921 %result1 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 4, %val1 )
5922 <i>; yields {i32}:result1 = 4</i>
5923 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
5924 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
5926 %val2 = add i32 1, 1
5927 %result2 = call i32 @llvm.atomic.cmp.swap.i32.p0i32( i32* %ptr, i32 5, %val2 )
5928 <i>; yields {i32}:result2 = 8</i>
5929 %stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
5931 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
5935 <!-- _______________________________________________________________________ -->
5936 <div class="doc_subsubsection">
5937 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
5939 <div class="doc_text">
5943 This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
5944 integer bit width. Not all targets support all bit widths however.</p>
5946 declare i8 @llvm.atomic.swap.i8.p0i8( i8* <ptr>, i8 <val> )
5947 declare i16 @llvm.atomic.swap.i16.p0i16( i16* <ptr>, i16 <val> )
5948 declare i32 @llvm.atomic.swap.i32.p0i32( i32* <ptr>, i32 <val> )
5949 declare i64 @llvm.atomic.swap.i64.p0i64( i64* <ptr>, i64 <val> )
5954 This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
5955 the value from memory. It then stores the value in <tt>val</tt> in the memory
5961 The <tt>llvm.atomic.swap</tt> intrinsic takes two arguments. Both the
5962 <tt>val</tt> argument and the result must be integers of the same bit width.
5963 The first argument, <tt>ptr</tt>, must be a pointer to a value of this
5964 integer type. The targets may only lower integer representations they
5969 This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
5970 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
5971 equivalent of an atomic swap operation within the SSA framework.
5979 %val1 = add i32 4, 4
5980 %result1 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val1 )
5981 <i>; yields {i32}:result1 = 4</i>
5982 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
5983 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
5985 %val2 = add i32 1, 1
5986 %result2 = call i32 @llvm.atomic.swap.i32.p0i32( i32* %ptr, i32 %val2 )
5987 <i>; yields {i32}:result2 = 8</i>
5989 %stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
5990 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
5994 <!-- _______________________________________________________________________ -->
5995 <div class="doc_subsubsection">
5996 <a name="int_atomic_load_add">'<tt>llvm.atomic.load.add.*</tt>' Intrinsic</a>
5999 <div class="doc_text">
6002 This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.add</tt> on any
6003 integer bit width. Not all targets support all bit widths however.</p>
6005 declare i8 @llvm.atomic.load.add.i8..p0i8( i8* <ptr>, i8 <delta> )
6006 declare i16 @llvm.atomic.load.add.i16..p0i16( i16* <ptr>, i16 <delta> )
6007 declare i32 @llvm.atomic.load.add.i32..p0i32( i32* <ptr>, i32 <delta> )
6008 declare i64 @llvm.atomic.load.add.i64..p0i64( i64* <ptr>, i64 <delta> )
6013 This intrinsic adds <tt>delta</tt> to the value stored in memory at
6014 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
6019 The intrinsic takes two arguments, the first a pointer to an integer value
6020 and the second an integer value. The result is also an integer value. These
6021 integer types can have any bit width, but they must all have the same bit
6022 width. The targets may only lower integer representations they support.
6026 This intrinsic does a series of operations atomically. It first loads the
6027 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
6028 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
6035 %result1 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 4 )
6036 <i>; yields {i32}:result1 = 4</i>
6037 %result2 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 2 )
6038 <i>; yields {i32}:result2 = 8</i>
6039 %result3 = call i32 @llvm.atomic.load.add.i32.p0i32( i32* %ptr, i32 5 )
6040 <i>; yields {i32}:result3 = 10</i>
6041 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
6045 <!-- _______________________________________________________________________ -->
6046 <div class="doc_subsubsection">
6047 <a name="int_atomic_load_sub">'<tt>llvm.atomic.load.sub.*</tt>' Intrinsic</a>
6050 <div class="doc_text">
6053 This is an overloaded intrinsic. You can use <tt>llvm.atomic.load.sub</tt> on
6054 any integer bit width and for different address spaces. Not all targets
6055 support all bit widths however.</p>
6057 declare i8 @llvm.atomic.load.sub.i8.p0i32( i8* <ptr>, i8 <delta> )
6058 declare i16 @llvm.atomic.load.sub.i16.p0i32( i16* <ptr>, i16 <delta> )
6059 declare i32 @llvm.atomic.load.sub.i32.p0i32( i32* <ptr>, i32 <delta> )
6060 declare i64 @llvm.atomic.load.sub.i64.p0i32( i64* <ptr>, i64 <delta> )
6065 This intrinsic subtracts <tt>delta</tt> to the value stored in memory at
6066 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
6071 The intrinsic takes two arguments, the first a pointer to an integer value
6072 and the second an integer value. The result is also an integer value. These
6073 integer types can have any bit width, but they must all have the same bit
6074 width. The targets may only lower integer representations they support.
6078 This intrinsic does a series of operations atomically. It first loads the
6079 value stored at <tt>ptr</tt>. It then subtracts <tt>delta</tt>, stores the
6080 result to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
6087 %result1 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 4 )
6088 <i>; yields {i32}:result1 = 8</i>
6089 %result2 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 2 )
6090 <i>; yields {i32}:result2 = 4</i>
6091 %result3 = call i32 @llvm.atomic.load.sub.i32.p0i32( i32* %ptr, i32 5 )
6092 <i>; yields {i32}:result3 = 2</i>
6093 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = -3</i>
6097 <!-- _______________________________________________________________________ -->
6098 <div class="doc_subsubsection">
6099 <a name="int_atomic_load_and">'<tt>llvm.atomic.load.and.*</tt>' Intrinsic</a><br>
6100 <a name="int_atomic_load_nand">'<tt>llvm.atomic.load.nand.*</tt>' Intrinsic</a><br>
6101 <a name="int_atomic_load_or">'<tt>llvm.atomic.load.or.*</tt>' Intrinsic</a><br>
6102 <a name="int_atomic_load_xor">'<tt>llvm.atomic.load.xor.*</tt>' Intrinsic</a><br>
6105 <div class="doc_text">
6108 These are overloaded intrinsics. You can use <tt>llvm.atomic.load_and</tt>,
6109 <tt>llvm.atomic.load_nand</tt>, <tt>llvm.atomic.load_or</tt>, and
6110 <tt>llvm.atomic.load_xor</tt> on any integer bit width and for different
6111 address spaces. Not all targets support all bit widths however.</p>
6113 declare i8 @llvm.atomic.load.and.i8.p0i8( i8* <ptr>, i8 <delta> )
6114 declare i16 @llvm.atomic.load.and.i16.p0i16( i16* <ptr>, i16 <delta> )
6115 declare i32 @llvm.atomic.load.and.i32.p0i32( i32* <ptr>, i32 <delta> )
6116 declare i64 @llvm.atomic.load.and.i64.p0i64( i64* <ptr>, i64 <delta> )
6121 declare i8 @llvm.atomic.load.or.i8.p0i8( i8* <ptr>, i8 <delta> )
6122 declare i16 @llvm.atomic.load.or.i16.p0i16( i16* <ptr>, i16 <delta> )
6123 declare i32 @llvm.atomic.load.or.i32.p0i32( i32* <ptr>, i32 <delta> )
6124 declare i64 @llvm.atomic.load.or.i64.p0i64( i64* <ptr>, i64 <delta> )
6129 declare i8 @llvm.atomic.load.nand.i8.p0i32( i8* <ptr>, i8 <delta> )
6130 declare i16 @llvm.atomic.load.nand.i16.p0i32( i16* <ptr>, i16 <delta> )
6131 declare i32 @llvm.atomic.load.nand.i32.p0i32( i32* <ptr>, i32 <delta> )
6132 declare i64 @llvm.atomic.load.nand.i64.p0i32( i64* <ptr>, i64 <delta> )
6137 declare i8 @llvm.atomic.load.xor.i8.p0i32( i8* <ptr>, i8 <delta> )
6138 declare i16 @llvm.atomic.load.xor.i16.p0i32( i16* <ptr>, i16 <delta> )
6139 declare i32 @llvm.atomic.load.xor.i32.p0i32( i32* <ptr>, i32 <delta> )
6140 declare i64 @llvm.atomic.load.xor.i64.p0i32( i64* <ptr>, i64 <delta> )
6145 These intrinsics bitwise the operation (and, nand, or, xor) <tt>delta</tt> to
6146 the value stored in memory at <tt>ptr</tt>. It yields the original value
6152 These intrinsics take two arguments, the first a pointer to an integer value
6153 and the second an integer value. The result is also an integer value. These
6154 integer types can have any bit width, but they must all have the same bit
6155 width. The targets may only lower integer representations they support.
6159 These intrinsics does a series of operations atomically. They first load the
6160 value stored at <tt>ptr</tt>. They then do the bitwise operation
6161 <tt>delta</tt>, store the result to <tt>ptr</tt>. They yield the original
6162 value stored at <tt>ptr</tt>.
6168 store i32 0x0F0F, %ptr
6169 %result0 = call i32 @llvm.atomic.load.nand.i32.p0i32( i32* %ptr, i32 0xFF )
6170 <i>; yields {i32}:result0 = 0x0F0F</i>
6171 %result1 = call i32 @llvm.atomic.load.and.i32.p0i32( i32* %ptr, i32 0xFF )
6172 <i>; yields {i32}:result1 = 0xFFFFFFF0</i>
6173 %result2 = call i32 @llvm.atomic.load.or.i32.p0i32( i32* %ptr, i32 0F )
6174 <i>; yields {i32}:result2 = 0xF0</i>
6175 %result3 = call i32 @llvm.atomic.load.xor.i32.p0i32( i32* %ptr, i32 0F )
6176 <i>; yields {i32}:result3 = FF</i>
6177 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = F0</i>
6182 <!-- _______________________________________________________________________ -->
6183 <div class="doc_subsubsection">
6184 <a name="int_atomic_load_max">'<tt>llvm.atomic.load.max.*</tt>' Intrinsic</a><br>
6185 <a name="int_atomic_load_min">'<tt>llvm.atomic.load.min.*</tt>' Intrinsic</a><br>
6186 <a name="int_atomic_load_umax">'<tt>llvm.atomic.load.umax.*</tt>' Intrinsic</a><br>
6187 <a name="int_atomic_load_umin">'<tt>llvm.atomic.load.umin.*</tt>' Intrinsic</a><br>
6190 <div class="doc_text">
6193 These are overloaded intrinsics. You can use <tt>llvm.atomic.load_max</tt>,
6194 <tt>llvm.atomic.load_min</tt>, <tt>llvm.atomic.load_umax</tt>, and
6195 <tt>llvm.atomic.load_umin</tt> on any integer bit width and for different
6196 address spaces. Not all targets
6197 support all bit widths however.</p>
6199 declare i8 @llvm.atomic.load.max.i8.p0i8( i8* <ptr>, i8 <delta> )
6200 declare i16 @llvm.atomic.load.max.i16.p0i16( i16* <ptr>, i16 <delta> )
6201 declare i32 @llvm.atomic.load.max.i32.p0i32( i32* <ptr>, i32 <delta> )
6202 declare i64 @llvm.atomic.load.max.i64.p0i64( i64* <ptr>, i64 <delta> )
6207 declare i8 @llvm.atomic.load.min.i8.p0i8( i8* <ptr>, i8 <delta> )
6208 declare i16 @llvm.atomic.load.min.i16.p0i16( i16* <ptr>, i16 <delta> )
6209 declare i32 @llvm.atomic.load.min.i32..p0i32( i32* <ptr>, i32 <delta> )
6210 declare i64 @llvm.atomic.load.min.i64..p0i64( i64* <ptr>, i64 <delta> )
6215 declare i8 @llvm.atomic.load.umax.i8.p0i8( i8* <ptr>, i8 <delta> )
6216 declare i16 @llvm.atomic.load.umax.i16.p0i16( i16* <ptr>, i16 <delta> )
6217 declare i32 @llvm.atomic.load.umax.i32.p0i32( i32* <ptr>, i32 <delta> )
6218 declare i64 @llvm.atomic.load.umax.i64.p0i64( i64* <ptr>, i64 <delta> )
6223 declare i8 @llvm.atomic.load.umin.i8..p0i8( i8* <ptr>, i8 <delta> )
6224 declare i16 @llvm.atomic.load.umin.i16.p0i16( i16* <ptr>, i16 <delta> )
6225 declare i32 @llvm.atomic.load.umin.i32..p0i32( i32* <ptr>, i32 <delta> )
6226 declare i64 @llvm.atomic.load.umin.i64..p0i64( i64* <ptr>, i64 <delta> )
6231 These intrinsics takes the signed or unsigned minimum or maximum of
6232 <tt>delta</tt> and the value stored in memory at <tt>ptr</tt>. It yields the
6233 original value at <tt>ptr</tt>.
6238 These intrinsics take two arguments, the first a pointer to an integer value
6239 and the second an integer value. The result is also an integer value. These
6240 integer types can have any bit width, but they must all have the same bit
6241 width. The targets may only lower integer representations they support.
6245 These intrinsics does a series of operations atomically. They first load the
6246 value stored at <tt>ptr</tt>. They then do the signed or unsigned min or max
6247 <tt>delta</tt> and the value, store the result to <tt>ptr</tt>. They yield
6248 the original value stored at <tt>ptr</tt>.
6255 %result0 = call i32 @llvm.atomic.load.min.i32.p0i32( i32* %ptr, i32 -2 )
6256 <i>; yields {i32}:result0 = 7</i>
6257 %result1 = call i32 @llvm.atomic.load.max.i32.p0i32( i32* %ptr, i32 8 )
6258 <i>; yields {i32}:result1 = -2</i>
6259 %result2 = call i32 @llvm.atomic.load.umin.i32.p0i32( i32* %ptr, i32 10 )
6260 <i>; yields {i32}:result2 = 8</i>
6261 %result3 = call i32 @llvm.atomic.load.umax.i32.p0i32( i32* %ptr, i32 30 )
6262 <i>; yields {i32}:result3 = 8</i>
6263 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 30</i>
6267 <!-- ======================================================================= -->
6268 <div class="doc_subsection">
6269 <a name="int_general">General Intrinsics</a>
6272 <div class="doc_text">
6273 <p> This class of intrinsics is designed to be generic and has
6274 no specific purpose. </p>
6277 <!-- _______________________________________________________________________ -->
6278 <div class="doc_subsubsection">
6279 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
6282 <div class="doc_text">
6286 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
6292 The '<tt>llvm.var.annotation</tt>' intrinsic
6298 The first argument is a pointer to a value, the second is a pointer to a
6299 global string, the third is a pointer to a global string which is the source
6300 file name, and the last argument is the line number.
6306 This intrinsic allows annotation of local variables with arbitrary strings.
6307 This can be useful for special purpose optimizations that want to look for these
6308 annotations. These have no other defined use, they are ignored by code
6309 generation and optimization.
6313 <!-- _______________________________________________________________________ -->
6314 <div class="doc_subsubsection">
6315 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
6318 <div class="doc_text">
6321 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
6322 any integer bit width.
6325 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
6326 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
6327 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
6328 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
6329 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
6335 The '<tt>llvm.annotation</tt>' intrinsic.
6341 The first argument is an integer value (result of some expression),
6342 the second is a pointer to a global string, the third is a pointer to a global
6343 string which is the source file name, and the last argument is the line number.
6344 It returns the value of the first argument.
6350 This intrinsic allows annotations to be put on arbitrary expressions
6351 with arbitrary strings. This can be useful for special purpose optimizations
6352 that want to look for these annotations. These have no other defined use, they
6353 are ignored by code generation and optimization.
6356 <!-- _______________________________________________________________________ -->
6357 <div class="doc_subsubsection">
6358 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
6361 <div class="doc_text">
6365 declare void @llvm.trap()
6371 The '<tt>llvm.trap</tt>' intrinsic
6383 This intrinsics is lowered to the target dependent trap instruction. If the
6384 target does not have a trap instruction, this intrinsic will be lowered to the
6385 call of the abort() function.
6389 <!-- *********************************************************************** -->
6392 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
6393 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
6394 <a href="http://validator.w3.org/check/referer"><img
6395 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
6397 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
6398 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
6399 Last modified: $Date$