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="#gc">Garbage Collector Names</a></li>
30 <li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
31 <li><a href="#datalayout">Data Layout</a></li>
34 <li><a href="#typesystem">Type System</a>
36 <li><a href="#t_classifications">Type Classifications</a></li>
37 <li><a href="#t_primitive">Primitive Types</a>
39 <li><a href="#t_floating">Floating Point Types</a></li>
40 <li><a href="#t_void">Void Type</a></li>
41 <li><a href="#t_label">Label Type</a></li>
44 <li><a href="#t_derived">Derived Types</a>
46 <li><a href="#t_integer">Integer Type</a></li>
47 <li><a href="#t_array">Array Type</a></li>
48 <li><a href="#t_function">Function Type</a></li>
49 <li><a href="#t_pointer">Pointer Type</a></li>
50 <li><a href="#t_struct">Structure Type</a></li>
51 <li><a href="#t_pstruct">Packed Structure Type</a></li>
52 <li><a href="#t_vector">Vector Type</a></li>
53 <li><a href="#t_opaque">Opaque Type</a></li>
58 <li><a href="#constants">Constants</a>
60 <li><a href="#simpleconstants">Simple Constants</a>
61 <li><a href="#aggregateconstants">Aggregate Constants</a>
62 <li><a href="#globalconstants">Global Variable and Function Addresses</a>
63 <li><a href="#undefvalues">Undefined Values</a>
64 <li><a href="#constantexprs">Constant Expressions</a>
67 <li><a href="#othervalues">Other Values</a>
69 <li><a href="#inlineasm">Inline Assembler Expressions</a>
72 <li><a href="#instref">Instruction Reference</a>
74 <li><a href="#terminators">Terminator Instructions</a>
76 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
77 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
78 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
79 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
80 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
81 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
84 <li><a href="#binaryops">Binary Operations</a>
86 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
87 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
88 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
89 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
90 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
91 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
92 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
93 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
94 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
97 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
99 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
100 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
101 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
102 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
103 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
104 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
107 <li><a href="#vectorops">Vector Operations</a>
109 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
110 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
111 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
114 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
116 <li><a href="#i_malloc">'<tt>malloc</tt>' Instruction</a></li>
117 <li><a href="#i_free">'<tt>free</tt>' Instruction</a></li>
118 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
119 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
120 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
121 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
124 <li><a href="#convertops">Conversion Operations</a>
126 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
127 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
128 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
129 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
130 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
131 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
132 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
133 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
134 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
135 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
136 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
137 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
139 <li><a href="#otherops">Other Operations</a>
141 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
142 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
143 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
144 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
145 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
146 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
147 <li><a href="#i_getresult">'<tt>getresult</tt>' Instruction</a></li>
152 <li><a href="#intrinsics">Intrinsic Functions</a>
154 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
156 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
157 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
158 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
161 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
163 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
164 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
165 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
168 <li><a href="#int_codegen">Code Generator Intrinsics</a>
170 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
171 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
172 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
173 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
174 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
175 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
176 <li><a href="#int_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
179 <li><a href="#int_libc">Standard C Library Intrinsics</a>
181 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
182 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
183 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
184 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
185 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
186 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
187 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
188 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
191 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
193 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
194 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
195 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
196 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
197 <li><a href="#int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic </a></li>
198 <li><a href="#int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic </a></li>
201 <li><a href="#int_debugger">Debugger intrinsics</a></li>
202 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
203 <li><a href="#int_trampoline">Trampoline Intrinsic</a>
205 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
208 <li><a href="#int_atomics">Atomic intrinsics</a>
210 <li><a href="#int_memory_barrier"><tt>llvm.memory_barrier</tt></a></li>
211 <li><a href="#int_atomic_lcs"><tt>llvm.atomic.lcs</tt></a></li>
212 <li><a href="#int_atomic_las"><tt>llvm.atomic.las</tt></a></li>
213 <li><a href="#int_atomic_swap"><tt>llvm.atomic.swap</tt></a></li>
216 <li><a href="#int_general">General intrinsics</a>
218 <li><a href="#int_var_annotation">
219 <tt>llvm.var.annotation</tt>' Intrinsic</a></li>
220 <li><a href="#int_annotation">
221 <tt>llvm.annotation.*</tt>' Intrinsic</a></li>
222 <li><a href="#int_trap">
223 <tt>llvm.trap</tt>' Intrinsic</a></li>
230 <div class="doc_author">
231 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
232 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
235 <!-- *********************************************************************** -->
236 <div class="doc_section"> <a name="abstract">Abstract </a></div>
237 <!-- *********************************************************************** -->
239 <div class="doc_text">
240 <p>This document is a reference manual for the LLVM assembly language.
241 LLVM is an SSA based representation that provides type safety,
242 low-level operations, flexibility, and the capability of representing
243 'all' high-level languages cleanly. It is the common code
244 representation used throughout all phases of the LLVM compilation
248 <!-- *********************************************************************** -->
249 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
250 <!-- *********************************************************************** -->
252 <div class="doc_text">
254 <p>The LLVM code representation is designed to be used in three
255 different forms: as an in-memory compiler IR, as an on-disk bitcode
256 representation (suitable for fast loading by a Just-In-Time compiler),
257 and as a human readable assembly language representation. This allows
258 LLVM to provide a powerful intermediate representation for efficient
259 compiler transformations and analysis, while providing a natural means
260 to debug and visualize the transformations. The three different forms
261 of LLVM are all equivalent. This document describes the human readable
262 representation and notation.</p>
264 <p>The LLVM representation aims to be light-weight and low-level
265 while being expressive, typed, and extensible at the same time. It
266 aims to be a "universal IR" of sorts, by being at a low enough level
267 that high-level ideas may be cleanly mapped to it (similar to how
268 microprocessors are "universal IR's", allowing many source languages to
269 be mapped to them). By providing type information, LLVM can be used as
270 the target of optimizations: for example, through pointer analysis, it
271 can be proven that a C automatic variable is never accessed outside of
272 the current function... allowing it to be promoted to a simple SSA
273 value instead of a memory location.</p>
277 <!-- _______________________________________________________________________ -->
278 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
280 <div class="doc_text">
282 <p>It is important to note that this document describes 'well formed'
283 LLVM assembly language. There is a difference between what the parser
284 accepts and what is considered 'well formed'. For example, the
285 following instruction is syntactically okay, but not well formed:</p>
287 <div class="doc_code">
289 %x = <a href="#i_add">add</a> i32 1, %x
293 <p>...because the definition of <tt>%x</tt> does not dominate all of
294 its uses. The LLVM infrastructure provides a verification pass that may
295 be used to verify that an LLVM module is well formed. This pass is
296 automatically run by the parser after parsing input assembly and by
297 the optimizer before it outputs bitcode. The violations pointed out
298 by the verifier pass indicate bugs in transformation passes or input to
302 <!-- Describe the typesetting conventions here. -->
304 <!-- *********************************************************************** -->
305 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
306 <!-- *********************************************************************** -->
308 <div class="doc_text">
310 <p>LLVM identifiers come in two basic types: global and local. Global
311 identifiers (functions, global variables) begin with the @ character. Local
312 identifiers (register names, types) begin with the % character. Additionally,
313 there are three different formats for identifiers, for different purposes:
316 <li>Named values are represented as a string of characters with their prefix.
317 For example, %foo, @DivisionByZero, %a.really.long.identifier. The actual
318 regular expression used is '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
319 Identifiers which require other characters in their names can be surrounded
320 with quotes. In this way, anything except a <tt>"</tt> character can
321 be used in a named value.</li>
323 <li>Unnamed values are represented as an unsigned numeric value with their
324 prefix. For example, %12, @2, %44.</li>
326 <li>Constants, which are described in a <a href="#constants">section about
327 constants</a>, below.</li>
330 <p>LLVM requires that values start with a prefix for two reasons: Compilers
331 don't need to worry about name clashes with reserved words, and the set of
332 reserved words may be expanded in the future without penalty. Additionally,
333 unnamed identifiers allow a compiler to quickly come up with a temporary
334 variable without having to avoid symbol table conflicts.</p>
336 <p>Reserved words in LLVM are very similar to reserved words in other
337 languages. There are keywords for different opcodes
338 ('<tt><a href="#i_add">add</a></tt>',
339 '<tt><a href="#i_bitcast">bitcast</a></tt>',
340 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
341 href="#t_void">void</a></tt>', '<tt><a href="#t_primitive">i32</a></tt>', etc...),
342 and others. These reserved words cannot conflict with variable names, because
343 none of them start with a prefix character ('%' or '@').</p>
345 <p>Here is an example of LLVM code to multiply the integer variable
346 '<tt>%X</tt>' by 8:</p>
350 <div class="doc_code">
352 %result = <a href="#i_mul">mul</a> i32 %X, 8
356 <p>After strength reduction:</p>
358 <div class="doc_code">
360 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
364 <p>And the hard way:</p>
366 <div class="doc_code">
368 <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
369 <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
370 %result = <a href="#i_add">add</a> i32 %1, %1
374 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
375 important lexical features of LLVM:</p>
379 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
382 <li>Unnamed temporaries are created when the result of a computation is not
383 assigned to a named value.</li>
385 <li>Unnamed temporaries are numbered sequentially</li>
389 <p>...and it also shows a convention that we follow in this document. When
390 demonstrating instructions, we will follow an instruction with a comment that
391 defines the type and name of value produced. Comments are shown in italic
396 <!-- *********************************************************************** -->
397 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
398 <!-- *********************************************************************** -->
400 <!-- ======================================================================= -->
401 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
404 <div class="doc_text">
406 <p>LLVM programs are composed of "Module"s, each of which is a
407 translation unit of the input programs. Each module consists of
408 functions, global variables, and symbol table entries. Modules may be
409 combined together with the LLVM linker, which merges function (and
410 global variable) definitions, resolves forward declarations, and merges
411 symbol table entries. Here is an example of the "hello world" module:</p>
413 <div class="doc_code">
414 <pre><i>; Declare the string constant as a global constant...</i>
415 <a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a
416 href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
418 <i>; External declaration of the puts function</i>
419 <a href="#functionstructure">declare</a> i32 @puts(i8 *) <i>; i32(i8 *)* </i>
421 <i>; Definition of main function</i>
422 define i32 @main() { <i>; i32()* </i>
423 <i>; Convert [13x i8 ]* to i8 *...</i>
425 href="#i_getelementptr">getelementptr</a> [13 x i8 ]* @.LC0, i64 0, i64 0 <i>; i8 *</i>
427 <i>; Call puts function to write out the string to stdout...</i>
429 href="#i_call">call</a> i32 @puts(i8 * %cast210) <i>; i32</i>
431 href="#i_ret">ret</a> i32 0<br>}<br>
435 <p>This example is made up of a <a href="#globalvars">global variable</a>
436 named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
437 function, and a <a href="#functionstructure">function definition</a>
438 for "<tt>main</tt>".</p>
440 <p>In general, a module is made up of a list of global values,
441 where both functions and global variables are global values. Global values are
442 represented by a pointer to a memory location (in this case, a pointer to an
443 array of char, and a pointer to a function), and have one of the following <a
444 href="#linkage">linkage types</a>.</p>
448 <!-- ======================================================================= -->
449 <div class="doc_subsection">
450 <a name="linkage">Linkage Types</a>
453 <div class="doc_text">
456 All Global Variables and Functions have one of the following types of linkage:
461 <dt><tt><b><a name="linkage_internal">internal</a></b></tt> </dt>
463 <dd>Global values with internal linkage are only directly accessible by
464 objects in the current module. In particular, linking code into a module with
465 an internal global value may cause the internal to be renamed as necessary to
466 avoid collisions. Because the symbol is internal to the module, all
467 references can be updated. This corresponds to the notion of the
468 '<tt>static</tt>' keyword in C.
471 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
473 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
474 the same name when linkage occurs. This is typically used to implement
475 inline functions, templates, or other code which must be generated in each
476 translation unit that uses it. Unreferenced <tt>linkonce</tt> globals are
477 allowed to be discarded.
480 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
482 <dd>"<tt>weak</tt>" linkage is exactly the same as <tt>linkonce</tt> linkage,
483 except that unreferenced <tt>weak</tt> globals may not be discarded. This is
484 used for globals that may be emitted in multiple translation units, but that
485 are not guaranteed to be emitted into every translation unit that uses them.
486 One example of this are common globals in C, such as "<tt>int X;</tt>" at
490 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
492 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
493 pointer to array type. When two global variables with appending linkage are
494 linked together, the two global arrays are appended together. This is the
495 LLVM, typesafe, equivalent of having the system linker append together
496 "sections" with identical names when .o files are linked.
499 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt>: </dt>
500 <dd>The semantics of this linkage follow the ELF model: the symbol is weak
501 until linked, if not linked, the symbol becomes null instead of being an
505 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
507 <dd>If none of the above identifiers are used, the global is externally
508 visible, meaning that it participates in linkage and can be used to resolve
509 external symbol references.
514 The next two types of linkage are targeted for Microsoft Windows platform
515 only. They are designed to support importing (exporting) symbols from (to)
520 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt>: </dt>
522 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
523 or variable via a global pointer to a pointer that is set up by the DLL
524 exporting the symbol. On Microsoft Windows targets, the pointer name is
525 formed by combining <code>_imp__</code> and the function or variable name.
528 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt>: </dt>
530 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
531 pointer to a pointer in a DLL, so that it can be referenced with the
532 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
533 name is formed by combining <code>_imp__</code> and the function or variable
539 <p><a name="linkage_external"></a>For example, since the "<tt>.LC0</tt>"
540 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
541 variable and was linked with this one, one of the two would be renamed,
542 preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
543 external (i.e., lacking any linkage declarations), they are accessible
544 outside of the current module.</p>
545 <p>It is illegal for a function <i>declaration</i>
546 to have any linkage type other than "externally visible", <tt>dllimport</tt>,
547 or <tt>extern_weak</tt>.</p>
548 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt> and <tt>weak</tt>
552 <!-- ======================================================================= -->
553 <div class="doc_subsection">
554 <a name="callingconv">Calling Conventions</a>
557 <div class="doc_text">
559 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
560 and <a href="#i_invoke">invokes</a> can all have an optional calling convention
561 specified for the call. The calling convention of any pair of dynamic
562 caller/callee must match, or the behavior of the program is undefined. The
563 following calling conventions are supported by LLVM, and more may be added in
567 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
569 <dd>This calling convention (the default if no other calling convention is
570 specified) matches the target C calling conventions. This calling convention
571 supports varargs function calls and tolerates some mismatch in the declared
572 prototype and implemented declaration of the function (as does normal C).
575 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
577 <dd>This calling convention attempts to make calls as fast as possible
578 (e.g. by passing things in registers). This calling convention allows the
579 target to use whatever tricks it wants to produce fast code for the target,
580 without having to conform to an externally specified ABI. Implementations of
581 this convention should allow arbitrary tail call optimization to be supported.
582 This calling convention does not support varargs and requires the prototype of
583 all callees to exactly match the prototype of the function definition.
586 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
588 <dd>This calling convention attempts to make code in the caller as efficient
589 as possible under the assumption that the call is not commonly executed. As
590 such, these calls often preserve all registers so that the call does not break
591 any live ranges in the caller side. This calling convention does not support
592 varargs and requires the prototype of all callees to exactly match the
593 prototype of the function definition.
596 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
598 <dd>Any calling convention may be specified by number, allowing
599 target-specific calling conventions to be used. Target specific calling
600 conventions start at 64.
604 <p>More calling conventions can be added/defined on an as-needed basis, to
605 support pascal conventions or any other well-known target-independent
610 <!-- ======================================================================= -->
611 <div class="doc_subsection">
612 <a name="visibility">Visibility Styles</a>
615 <div class="doc_text">
618 All Global Variables and Functions have one of the following visibility styles:
622 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
624 <dd>On ELF, default visibility means that the declaration is visible to other
625 modules and, in shared libraries, means that the declared entity may be
626 overridden. On Darwin, default visibility means that the declaration is
627 visible to other modules. Default visibility corresponds to "external
628 linkage" in the language.
631 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
633 <dd>Two declarations of an object with hidden visibility refer to the same
634 object if they are in the same shared object. Usually, hidden visibility
635 indicates that the symbol will not be placed into the dynamic symbol table,
636 so no other module (executable or shared library) can reference it
640 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
642 <dd>On ELF, protected visibility indicates that the symbol will be placed in
643 the dynamic symbol table, but that references within the defining module will
644 bind to the local symbol. That is, the symbol cannot be overridden by another
651 <!-- ======================================================================= -->
652 <div class="doc_subsection">
653 <a name="globalvars">Global Variables</a>
656 <div class="doc_text">
658 <p>Global variables define regions of memory allocated at compilation time
659 instead of run-time. Global variables may optionally be initialized, may have
660 an explicit section to be placed in, and may have an optional explicit alignment
661 specified. A variable may be defined as "thread_local", which means that it
662 will not be shared by threads (each thread will have a separated copy of the
663 variable). A variable may be defined as a global "constant," which indicates
664 that the contents of the variable will <b>never</b> be modified (enabling better
665 optimization, allowing the global data to be placed in the read-only section of
666 an executable, etc). Note that variables that need runtime initialization
667 cannot be marked "constant" as there is a store to the variable.</p>
670 LLVM explicitly allows <em>declarations</em> of global variables to be marked
671 constant, even if the final definition of the global is not. This capability
672 can be used to enable slightly better optimization of the program, but requires
673 the language definition to guarantee that optimizations based on the
674 'constantness' are valid for the translation units that do not include the
678 <p>As SSA values, global variables define pointer values that are in
679 scope (i.e. they dominate) all basic blocks in the program. Global
680 variables always define a pointer to their "content" type because they
681 describe a region of memory, and all memory objects in LLVM are
682 accessed through pointers.</p>
684 <p>A global variable may be declared to reside in a target-specifc numbered
685 address space. For targets that support them, address spaces may affect how
686 optimizations are performed and/or what target instructions are used to access
687 the variable. The default address space is zero. The address space qualifier
688 must precede any other attributes.</p>
690 <p>LLVM allows an explicit section to be specified for globals. If the target
691 supports it, it will emit globals to the section specified.</p>
693 <p>An explicit alignment may be specified for a global. If not present, or if
694 the alignment is set to zero, the alignment of the global is set by the target
695 to whatever it feels convenient. If an explicit alignment is specified, the
696 global is forced to have at least that much alignment. All alignments must be
699 <p>For example, the following defines a global in a numbered address space with
700 an initializer, section, and alignment:</p>
702 <div class="doc_code">
704 @G = constant float 1.0 addrspace(5), section "foo", align 4
711 <!-- ======================================================================= -->
712 <div class="doc_subsection">
713 <a name="functionstructure">Functions</a>
716 <div class="doc_text">
718 <p>LLVM function definitions consist of the "<tt>define</tt>" keyord,
719 an optional <a href="#linkage">linkage type</a>, an optional
720 <a href="#visibility">visibility style</a>, an optional
721 <a href="#callingconv">calling convention</a>, a return type, an optional
722 <a href="#paramattrs">parameter attribute</a> for the return type, a function
723 name, a (possibly empty) argument list (each with optional
724 <a href="#paramattrs">parameter attributes</a>), an optional section, an
725 optional alignment, an optional <a href="#gc">garbage collector name</a>, an
726 opening curly brace, a list of basic blocks, and a closing curly brace.
728 LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
729 optional <a href="#linkage">linkage type</a>, an optional
730 <a href="#visibility">visibility style</a>, an optional
731 <a href="#callingconv">calling convention</a>, a return type, an optional
732 <a href="#paramattrs">parameter attribute</a> for the return type, a function
733 name, a possibly empty list of arguments, an optional alignment, and an optional
734 <a href="#gc">garbage collector name</a>.</p>
736 <p>A function definition contains a list of basic blocks, forming the CFG for
737 the function. Each basic block may optionally start with a label (giving the
738 basic block a symbol table entry), contains a list of instructions, and ends
739 with a <a href="#terminators">terminator</a> instruction (such as a branch or
740 function return).</p>
742 <p>The first basic block in a function is special in two ways: it is immediately
743 executed on entrance to the function, and it is not allowed to have predecessor
744 basic blocks (i.e. there can not be any branches to the entry block of a
745 function). Because the block can have no predecessors, it also cannot have any
746 <a href="#i_phi">PHI nodes</a>.</p>
748 <p>LLVM allows an explicit section to be specified for functions. If the target
749 supports it, it will emit functions to the section specified.</p>
751 <p>An explicit alignment may be specified for a function. If not present, or if
752 the alignment is set to zero, the alignment of the function is set by the target
753 to whatever it feels convenient. If an explicit alignment is specified, the
754 function is forced to have at least that much alignment. All alignments must be
760 <!-- ======================================================================= -->
761 <div class="doc_subsection">
762 <a name="aliasstructure">Aliases</a>
764 <div class="doc_text">
765 <p>Aliases act as "second name" for the aliasee value (which can be either
766 function, global variable, another alias or bitcast of global value). Aliases
767 may have an optional <a href="#linkage">linkage type</a>, and an
768 optional <a href="#visibility">visibility style</a>.</p>
772 <div class="doc_code">
774 @<Name> = [Linkage] [Visibility] alias <AliaseeTy> @<Aliasee>
782 <!-- ======================================================================= -->
783 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
784 <div class="doc_text">
785 <p>The return type and each parameter of a function type may have a set of
786 <i>parameter attributes</i> associated with them. Parameter attributes are
787 used to communicate additional information about the result or parameters of
788 a function. Parameter attributes are considered to be part of the function,
789 not of the function type, so functions with different parameter attributes
790 can have the same function type.</p>
792 <p>Parameter attributes are simple keywords that follow the type specified. If
793 multiple parameter attributes are needed, they are space separated. For
796 <div class="doc_code">
798 declare i32 @printf(i8* noalias , ...) nounwind
799 declare i32 @atoi(i8*) nounwind readonly
803 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
804 <tt>readonly</tt>) come immediately after the argument list.</p>
806 <p>Currently, only the following parameter attributes are defined:</p>
808 <dt><tt>zeroext</tt></dt>
809 <dd>This indicates that the parameter should be zero extended just before
810 a call to this function.</dd>
812 <dt><tt>signext</tt></dt>
813 <dd>This indicates that the parameter should be sign extended just before
814 a call to this function.</dd>
816 <dt><tt>inreg</tt></dt>
817 <dd>This indicates that the parameter should be placed in register (if
818 possible) during assembling function call. Support for this attribute is
821 <dt><tt>byval</tt></dt>
822 <dd>This indicates that the pointer parameter should really be passed by
823 value to the function. The attribute implies that a hidden copy of the
824 pointee is made between the caller and the callee, so the callee is unable
825 to modify the value in the callee. This attribute is only valid on llvm
826 pointer arguments. It is generally used to pass structs and arrays by
827 value, but is also valid on scalars (even though this is silly).</dd>
829 <dt><tt>sret</tt></dt>
830 <dd>This indicates that the pointer parameter specifies the address of a
831 structure that is the return value of the function in the source program.
832 Loads and stores to the structure are assumed not to trap.
833 May only be applied to the first parameter.</dd>
835 <dt><tt>noalias</tt></dt>
836 <dd>This indicates that the parameter does not alias any global or any other
837 parameter. The caller is responsible for ensuring that this is the case,
838 usually by placing the value in a stack allocation.</dd>
840 <dt><tt>noreturn</tt></dt>
841 <dd>This function attribute indicates that the function never returns. This
842 indicates to LLVM that every call to this function should be treated as if
843 an <tt>unreachable</tt> instruction immediately followed the call.</dd>
845 <dt><tt>nounwind</tt></dt>
846 <dd>This function attribute indicates that no exceptions unwind out of the
847 function. Usually this is because the function makes no use of exceptions,
848 but it may also be that the function catches any exceptions thrown when
851 <dt><tt>nest</tt></dt>
852 <dd>This indicates that the parameter can be excised using the
853 <a href="#int_trampoline">trampoline intrinsics</a>.</dd>
854 <dt><tt>readonly</tt></dt>
855 <dd>This function attribute indicates that the function has no side-effects
856 except for producing a return value or throwing an exception. The value
857 returned must only depend on the function arguments and/or global variables.
858 It may use values obtained by dereferencing pointers.</dd>
859 <dt><tt>readnone</tt></dt>
860 <dd>A <tt>readnone</tt> function has the same restrictions as a <tt>readonly</tt>
861 function, but in addition it is not allowed to dereference any pointer arguments
867 <!-- ======================================================================= -->
868 <div class="doc_subsection">
869 <a name="gc">Garbage Collector Names</a>
872 <div class="doc_text">
873 <p>Each function may specify a garbage collector name, which is simply a
876 <div class="doc_code"><pre
877 >define void @f() gc "name" { ...</pre></div>
879 <p>The compiler declares the supported values of <i>name</i>. Specifying a
880 collector which will cause the compiler to alter its output in order to support
881 the named garbage collection algorithm.</p>
884 <!-- ======================================================================= -->
885 <div class="doc_subsection">
886 <a name="moduleasm">Module-Level Inline Assembly</a>
889 <div class="doc_text">
891 Modules may contain "module-level inline asm" blocks, which corresponds to the
892 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
893 LLVM and treated as a single unit, but may be separated in the .ll file if
894 desired. The syntax is very simple:
897 <div class="doc_code">
899 module asm "inline asm code goes here"
900 module asm "more can go here"
904 <p>The strings can contain any character by escaping non-printable characters.
905 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
910 The inline asm code is simply printed to the machine code .s file when
911 assembly code is generated.
915 <!-- ======================================================================= -->
916 <div class="doc_subsection">
917 <a name="datalayout">Data Layout</a>
920 <div class="doc_text">
921 <p>A module may specify a target specific data layout string that specifies how
922 data is to be laid out in memory. The syntax for the data layout is simply:</p>
923 <pre> target datalayout = "<i>layout specification</i>"</pre>
924 <p>The <i>layout specification</i> consists of a list of specifications
925 separated by the minus sign character ('-'). Each specification starts with a
926 letter and may include other information after the letter to define some
927 aspect of the data layout. The specifications accepted are as follows: </p>
930 <dd>Specifies that the target lays out data in big-endian form. That is, the
931 bits with the most significance have the lowest address location.</dd>
933 <dd>Specifies that hte target lays out data in little-endian form. That is,
934 the bits with the least significance have the lowest address location.</dd>
935 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
936 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
937 <i>preferred</i> alignments. All sizes are in bits. Specifying the <i>pref</i>
938 alignment is optional. If omitted, the preceding <tt>:</tt> should be omitted
940 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
941 <dd>This specifies the alignment for an integer type of a given bit
942 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
943 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
944 <dd>This specifies the alignment for a vector type of a given bit
946 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
947 <dd>This specifies the alignment for a floating point type of a given bit
948 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
950 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
951 <dd>This specifies the alignment for an aggregate type of a given bit
954 <p>When constructing the data layout for a given target, LLVM starts with a
955 default set of specifications which are then (possibly) overriden by the
956 specifications in the <tt>datalayout</tt> keyword. The default specifications
957 are given in this list:</p>
959 <li><tt>E</tt> - big endian</li>
960 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
961 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
962 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
963 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
964 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
965 <li><tt>i64:32:64</tt> - i64 has abi alignment of 32-bits but preferred
966 alignment of 64-bits</li>
967 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
968 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
969 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
970 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
971 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
973 <p>When llvm is determining the alignment for a given type, it uses the
976 <li>If the type sought is an exact match for one of the specifications, that
977 specification is used.</li>
978 <li>If no match is found, and the type sought is an integer type, then the
979 smallest integer type that is larger than the bitwidth of the sought type is
980 used. If none of the specifications are larger than the bitwidth then the the
981 largest integer type is used. For example, given the default specifications
982 above, the i7 type will use the alignment of i8 (next largest) while both
983 i65 and i256 will use the alignment of i64 (largest specified).</li>
984 <li>If no match is found, and the type sought is a vector type, then the
985 largest vector type that is smaller than the sought vector type will be used
986 as a fall back. This happens because <128 x double> can be implemented in
987 terms of 64 <2 x double>, for example.</li>
991 <!-- *********************************************************************** -->
992 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
993 <!-- *********************************************************************** -->
995 <div class="doc_text">
997 <p>The LLVM type system is one of the most important features of the
998 intermediate representation. Being typed enables a number of
999 optimizations to be performed on the IR directly, without having to do
1000 extra analyses on the side before the transformation. A strong type
1001 system makes it easier to read the generated code and enables novel
1002 analyses and transformations that are not feasible to perform on normal
1003 three address code representations.</p>
1007 <!-- ======================================================================= -->
1008 <div class="doc_subsection"> <a name="t_classifications">Type
1009 Classifications</a> </div>
1010 <div class="doc_text">
1011 <p>The types fall into a few useful
1012 classifications:</p>
1014 <table border="1" cellspacing="0" cellpadding="4">
1016 <tr><th>Classification</th><th>Types</th></tr>
1018 <td><a href="#t_integer">integer</a></td>
1019 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1022 <td><a href="#t_floating">floating point</a></td>
1023 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1026 <td><a name="t_firstclass">first class</a></td>
1027 <td><a href="#t_integer">integer</a>,
1028 <a href="#t_floating">floating point</a>,
1029 <a href="#t_pointer">pointer</a>,
1030 <a href="#t_vector">vector</a>
1034 <td><a href="#t_primitive">primitive</a></td>
1035 <td><a href="#t_label">label</a>,
1036 <a href="#t_void">void</a>,
1037 <a href="#t_integer">integer</a>,
1038 <a href="#t_floating">floating point</a>.</td>
1041 <td><a href="#t_derived">derived</a></td>
1042 <td><a href="#t_integer">integer</a>,
1043 <a href="#t_array">array</a>,
1044 <a href="#t_function">function</a>,
1045 <a href="#t_pointer">pointer</a>,
1046 <a href="#t_struct">structure</a>,
1047 <a href="#t_pstruct">packed structure</a>,
1048 <a href="#t_vector">vector</a>,
1049 <a href="#t_opaque">opaque</a>.
1054 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
1055 most important. Values of these types are the only ones which can be
1056 produced by instructions, passed as arguments, or used as operands to
1057 instructions. This means that all structures and arrays must be
1058 manipulated either by pointer or by component.</p>
1061 <!-- ======================================================================= -->
1062 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
1064 <div class="doc_text">
1065 <p>The primitive types are the fundamental building blocks of the LLVM
1070 <!-- _______________________________________________________________________ -->
1071 <div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1073 <div class="doc_text">
1076 <tr><th>Type</th><th>Description</th></tr>
1077 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1078 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1079 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1080 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1081 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1086 <!-- _______________________________________________________________________ -->
1087 <div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1089 <div class="doc_text">
1091 <p>The void type does not represent any value and has no size.</p>
1100 <!-- _______________________________________________________________________ -->
1101 <div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1103 <div class="doc_text">
1105 <p>The label type represents code labels.</p>
1115 <!-- ======================================================================= -->
1116 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1118 <div class="doc_text">
1120 <p>The real power in LLVM comes from the derived types in the system.
1121 This is what allows a programmer to represent arrays, functions,
1122 pointers, and other useful types. Note that these derived types may be
1123 recursive: For example, it is possible to have a two dimensional array.</p>
1127 <!-- _______________________________________________________________________ -->
1128 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1130 <div class="doc_text">
1133 <p>The integer type is a very simple derived type that simply specifies an
1134 arbitrary bit width for the integer type desired. Any bit width from 1 bit to
1135 2^23-1 (about 8 million) can be specified.</p>
1143 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1147 <table class="layout">
1150 <td><tt>i1</tt></td>
1151 <td>a single-bit integer.</td>
1153 <td><tt>i32</tt></td>
1154 <td>a 32-bit integer.</td>
1156 <td><tt>i1942652</tt></td>
1157 <td>a really big integer of over 1 million bits.</td>
1163 <!-- _______________________________________________________________________ -->
1164 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1166 <div class="doc_text">
1170 <p>The array type is a very simple derived type that arranges elements
1171 sequentially in memory. The array type requires a size (number of
1172 elements) and an underlying data type.</p>
1177 [<# elements> x <elementtype>]
1180 <p>The number of elements is a constant integer value; elementtype may
1181 be any type with a size.</p>
1184 <table class="layout">
1186 <td class="left"><tt>[40 x i32]</tt></td>
1187 <td class="left">Array of 40 32-bit integer values.</td>
1190 <td class="left"><tt>[41 x i32]</tt></td>
1191 <td class="left">Array of 41 32-bit integer values.</td>
1194 <td class="left"><tt>[4 x i8]</tt></td>
1195 <td class="left">Array of 4 8-bit integer values.</td>
1198 <p>Here are some examples of multidimensional arrays:</p>
1199 <table class="layout">
1201 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1202 <td class="left">3x4 array of 32-bit integer values.</td>
1205 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1206 <td class="left">12x10 array of single precision floating point values.</td>
1209 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1210 <td class="left">2x3x4 array of 16-bit integer values.</td>
1214 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1215 length array. Normally, accesses past the end of an array are undefined in
1216 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
1217 As a special case, however, zero length arrays are recognized to be variable
1218 length. This allows implementation of 'pascal style arrays' with the LLVM
1219 type "{ i32, [0 x float]}", for example.</p>
1223 <!-- _______________________________________________________________________ -->
1224 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1225 <div class="doc_text">
1229 <p>The function type can be thought of as a function signature. It
1230 consists of a return type and a list of formal parameter types. The
1231 return type of a function type is a scalar type, a void type, or a struct type.
1232 If the return type is a struct type then all struct elements must be of first
1233 class types, and the struct must have at least one element.</p>
1238 <returntype list> (<parameter list>)
1241 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1242 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1243 which indicates that the function takes a variable number of arguments.
1244 Variable argument functions can access their arguments with the <a
1245 href="#int_varargs">variable argument handling intrinsic</a> functions.
1246 '<tt><returntype list></tt>' is a comma-separated list of
1247 <a href="#t_firstclass">first class</a> type specifiers.</p>
1250 <table class="layout">
1252 <td class="left"><tt>i32 (i32)</tt></td>
1253 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1255 </tr><tr class="layout">
1256 <td class="left"><tt>float (i16 signext, i32 *) *
1258 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1259 an <tt>i16</tt> that should be sign extended and a
1260 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1263 </tr><tr class="layout">
1264 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1265 <td class="left">A vararg function that takes at least one
1266 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1267 which returns an integer. This is the signature for <tt>printf</tt> in
1270 </tr><tr class="layout">
1271 <td class="left"><tt>{i32, i32} (i32)</tt></td>
1272 <td class="left">A function taking an <tt>i32></tt>, returning two
1273 <tt> i32 </tt> values as an aggregate of type <tt>{ i32, i32 }</tt>
1279 <!-- _______________________________________________________________________ -->
1280 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1281 <div class="doc_text">
1283 <p>The structure type is used to represent a collection of data members
1284 together in memory. The packing of the field types is defined to match
1285 the ABI of the underlying processor. The elements of a structure may
1286 be any type that has a size.</p>
1287 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1288 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1289 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1292 <pre> { <type list> }<br></pre>
1294 <table class="layout">
1296 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1297 <td class="left">A triple of three <tt>i32</tt> values</td>
1298 </tr><tr class="layout">
1299 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1300 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1301 second element is a <a href="#t_pointer">pointer</a> to a
1302 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1303 an <tt>i32</tt>.</td>
1308 <!-- _______________________________________________________________________ -->
1309 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1311 <div class="doc_text">
1313 <p>The packed structure type is used to represent a collection of data members
1314 together in memory. There is no padding between fields. Further, the alignment
1315 of a packed structure is 1 byte. The elements of a packed structure may
1316 be any type that has a size.</p>
1317 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1318 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1319 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1322 <pre> < { <type list> } > <br></pre>
1324 <table class="layout">
1326 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1327 <td class="left">A triple of three <tt>i32</tt> values</td>
1328 </tr><tr class="layout">
1329 <td class="left"><tt>< { float, i32 (i32)* } ></tt></td>
1330 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1331 second element is a <a href="#t_pointer">pointer</a> to a
1332 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1333 an <tt>i32</tt>.</td>
1338 <!-- _______________________________________________________________________ -->
1339 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1340 <div class="doc_text">
1342 <p>As in many languages, the pointer type represents a pointer or
1343 reference to another object, which must live in memory. Pointer types may have
1344 an optional address space attribute defining the target-specific numbered
1345 address space where the pointed-to object resides. The default address space is
1348 <pre> <type> *<br></pre>
1350 <table class="layout">
1352 <td class="left"><tt>[4x i32]*</tt></td>
1353 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1354 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1357 <td class="left"><tt>i32 (i32 *) *</tt></td>
1358 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
1359 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1363 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1364 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1365 that resides in address space #5.</td>
1370 <!-- _______________________________________________________________________ -->
1371 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1372 <div class="doc_text">
1376 <p>A vector type is a simple derived type that represents a vector
1377 of elements. Vector types are used when multiple primitive data
1378 are operated in parallel using a single instruction (SIMD).
1379 A vector type requires a size (number of
1380 elements) and an underlying primitive data type. Vectors must have a power
1381 of two length (1, 2, 4, 8, 16 ...). Vector types are
1382 considered <a href="#t_firstclass">first class</a>.</p>
1387 < <# elements> x <elementtype> >
1390 <p>The number of elements is a constant integer value; elementtype may
1391 be any integer or floating point type.</p>
1395 <table class="layout">
1397 <td class="left"><tt><4 x i32></tt></td>
1398 <td class="left">Vector of 4 32-bit integer values.</td>
1401 <td class="left"><tt><8 x float></tt></td>
1402 <td class="left">Vector of 8 32-bit floating-point values.</td>
1405 <td class="left"><tt><2 x i64></tt></td>
1406 <td class="left">Vector of 2 64-bit integer values.</td>
1411 <!-- _______________________________________________________________________ -->
1412 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1413 <div class="doc_text">
1417 <p>Opaque types are used to represent unknown types in the system. This
1418 corresponds (for example) to the C notion of a forward declared structure type.
1419 In LLVM, opaque types can eventually be resolved to any type (not just a
1420 structure type).</p>
1430 <table class="layout">
1432 <td class="left"><tt>opaque</tt></td>
1433 <td class="left">An opaque type.</td>
1439 <!-- *********************************************************************** -->
1440 <div class="doc_section"> <a name="constants">Constants</a> </div>
1441 <!-- *********************************************************************** -->
1443 <div class="doc_text">
1445 <p>LLVM has several different basic types of constants. This section describes
1446 them all and their syntax.</p>
1450 <!-- ======================================================================= -->
1451 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1453 <div class="doc_text">
1456 <dt><b>Boolean constants</b></dt>
1458 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1459 constants of the <tt><a href="#t_primitive">i1</a></tt> type.
1462 <dt><b>Integer constants</b></dt>
1464 <dd>Standard integers (such as '4') are constants of the <a
1465 href="#t_integer">integer</a> type. Negative numbers may be used with
1469 <dt><b>Floating point constants</b></dt>
1471 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1472 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1473 notation (see below). The assembler requires the exact decimal value of
1474 a floating-point constant. For example, the assembler accepts 1.25 but
1475 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
1476 constants must have a <a href="#t_floating">floating point</a> type. </dd>
1478 <dt><b>Null pointer constants</b></dt>
1480 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1481 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1485 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1486 of floating point constants. For example, the form '<tt>double
1487 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1488 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1489 (and the only time that they are generated by the disassembler) is when a
1490 floating point constant must be emitted but it cannot be represented as a
1491 decimal floating point number. For example, NaN's, infinities, and other
1492 special values are represented in their IEEE hexadecimal format so that
1493 assembly and disassembly do not cause any bits to change in the constants.</p>
1497 <!-- ======================================================================= -->
1498 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1501 <div class="doc_text">
1502 <p>Aggregate constants arise from aggregation of simple constants
1503 and smaller aggregate constants.</p>
1506 <dt><b>Structure constants</b></dt>
1508 <dd>Structure constants are represented with notation similar to structure
1509 type definitions (a comma separated list of elements, surrounded by braces
1510 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
1511 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>". Structure constants
1512 must have <a href="#t_struct">structure type</a>, and the number and
1513 types of elements must match those specified by the type.
1516 <dt><b>Array constants</b></dt>
1518 <dd>Array constants are represented with notation similar to array type
1519 definitions (a comma separated list of elements, surrounded by square brackets
1520 (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74 ]</tt>". Array
1521 constants must have <a href="#t_array">array type</a>, and the number and
1522 types of elements must match those specified by the type.
1525 <dt><b>Vector constants</b></dt>
1527 <dd>Vector constants are represented with notation similar to vector type
1528 definitions (a comma separated list of elements, surrounded by
1529 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32 42,
1530 i32 11, i32 74, i32 100 ></tt>". Vector constants must have <a
1531 href="#t_vector">vector type</a>, and the number and types of elements must
1532 match those specified by the type.
1535 <dt><b>Zero initialization</b></dt>
1537 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1538 value to zero of <em>any</em> type, including scalar and aggregate types.
1539 This is often used to avoid having to print large zero initializers (e.g. for
1540 large arrays) and is always exactly equivalent to using explicit zero
1547 <!-- ======================================================================= -->
1548 <div class="doc_subsection">
1549 <a name="globalconstants">Global Variable and Function Addresses</a>
1552 <div class="doc_text">
1554 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1555 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1556 constants. These constants are explicitly referenced when the <a
1557 href="#identifiers">identifier for the global</a> is used and always have <a
1558 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1561 <div class="doc_code">
1565 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1571 <!-- ======================================================================= -->
1572 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1573 <div class="doc_text">
1574 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1575 no specific value. Undefined values may be of any type and be used anywhere
1576 a constant is permitted.</p>
1578 <p>Undefined values indicate to the compiler that the program is well defined
1579 no matter what value is used, giving the compiler more freedom to optimize.
1583 <!-- ======================================================================= -->
1584 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1587 <div class="doc_text">
1589 <p>Constant expressions are used to allow expressions involving other constants
1590 to be used as constants. Constant expressions may be of any <a
1591 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1592 that does not have side effects (e.g. load and call are not supported). The
1593 following is the syntax for constant expressions:</p>
1596 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1597 <dd>Truncate a constant to another type. The bit size of CST must be larger
1598 than the bit size of TYPE. Both types must be integers.</dd>
1600 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1601 <dd>Zero extend a constant to another type. The bit size of CST must be
1602 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1604 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1605 <dd>Sign extend a constant to another type. The bit size of CST must be
1606 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1608 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1609 <dd>Truncate a floating point constant to another floating point type. The
1610 size of CST must be larger than the size of TYPE. Both types must be
1611 floating point.</dd>
1613 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1614 <dd>Floating point extend a constant to another type. The size of CST must be
1615 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1617 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
1618 <dd>Convert a floating point constant to the corresponding unsigned integer
1619 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1620 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1621 of the same number of elements. If the value won't fit in the integer type,
1622 the results are undefined.</dd>
1624 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
1625 <dd>Convert a floating point constant to the corresponding signed integer
1626 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1627 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1628 of the same number of elements. If the value won't fit in the integer type,
1629 the results are undefined.</dd>
1631 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
1632 <dd>Convert an unsigned integer constant to the corresponding floating point
1633 constant. TYPE must be a scalar or vector floating point type. CST must be of
1634 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1635 of the same number of elements. If the value won't fit in the floating point
1636 type, the results are undefined.</dd>
1638 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
1639 <dd>Convert a signed integer constant to the corresponding floating point
1640 constant. TYPE must be a scalar or vector floating point type. CST must be of
1641 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1642 of the same number of elements. If the value won't fit in the floating point
1643 type, the results are undefined.</dd>
1645 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
1646 <dd>Convert a pointer typed constant to the corresponding integer constant
1647 TYPE must be an integer type. CST must be of pointer type. The CST value is
1648 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1650 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
1651 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1652 pointer type. CST must be of integer type. The CST value is zero extended,
1653 truncated, or unchanged to make it fit in a pointer size. This one is
1654 <i>really</i> dangerous!</dd>
1656 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
1657 <dd>Convert a constant, CST, to another TYPE. The size of CST and TYPE must be
1658 identical (same number of bits). The conversion is done as if the CST value
1659 was stored to memory and read back as TYPE. In other words, no bits change
1660 with this operator, just the type. This can be used for conversion of
1661 vector types to any other type, as long as they have the same bit width. For
1662 pointers it is only valid to cast to another pointer type.
1665 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1667 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1668 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1669 instruction, the index list may have zero or more indexes, which are required
1670 to make sense for the type of "CSTPTR".</dd>
1672 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1674 <dd>Perform the <a href="#i_select">select operation</a> on
1677 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
1678 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
1680 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
1681 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
1683 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1685 <dd>Perform the <a href="#i_extractelement">extractelement
1686 operation</a> on constants.
1688 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1690 <dd>Perform the <a href="#i_insertelement">insertelement
1691 operation</a> on constants.</dd>
1694 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
1696 <dd>Perform the <a href="#i_shufflevector">shufflevector
1697 operation</a> on constants.</dd>
1699 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1701 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1702 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1703 binary</a> operations. The constraints on operands are the same as those for
1704 the corresponding instruction (e.g. no bitwise operations on floating point
1705 values are allowed).</dd>
1709 <!-- *********************************************************************** -->
1710 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1711 <!-- *********************************************************************** -->
1713 <!-- ======================================================================= -->
1714 <div class="doc_subsection">
1715 <a name="inlineasm">Inline Assembler Expressions</a>
1718 <div class="doc_text">
1721 LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
1722 Module-Level Inline Assembly</a>) through the use of a special value. This
1723 value represents the inline assembler as a string (containing the instructions
1724 to emit), a list of operand constraints (stored as a string), and a flag that
1725 indicates whether or not the inline asm expression has side effects. An example
1726 inline assembler expression is:
1729 <div class="doc_code">
1731 i32 (i32) asm "bswap $0", "=r,r"
1736 Inline assembler expressions may <b>only</b> be used as the callee operand of
1737 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
1740 <div class="doc_code">
1742 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
1747 Inline asms with side effects not visible in the constraint list must be marked
1748 as having side effects. This is done through the use of the
1749 '<tt>sideeffect</tt>' keyword, like so:
1752 <div class="doc_code">
1754 call void asm sideeffect "eieio", ""()
1758 <p>TODO: The format of the asm and constraints string still need to be
1759 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1760 need to be documented).
1765 <!-- *********************************************************************** -->
1766 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1767 <!-- *********************************************************************** -->
1769 <div class="doc_text">
1771 <p>The LLVM instruction set consists of several different
1772 classifications of instructions: <a href="#terminators">terminator
1773 instructions</a>, <a href="#binaryops">binary instructions</a>,
1774 <a href="#bitwiseops">bitwise binary instructions</a>, <a
1775 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1776 instructions</a>.</p>
1780 <!-- ======================================================================= -->
1781 <div class="doc_subsection"> <a name="terminators">Terminator
1782 Instructions</a> </div>
1784 <div class="doc_text">
1786 <p>As mentioned <a href="#functionstructure">previously</a>, every
1787 basic block in a program ends with a "Terminator" instruction, which
1788 indicates which block should be executed after the current block is
1789 finished. These terminator instructions typically yield a '<tt>void</tt>'
1790 value: they produce control flow, not values (the one exception being
1791 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1792 <p>There are six different terminator instructions: the '<a
1793 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1794 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1795 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1796 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1797 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1801 <!-- _______________________________________________________________________ -->
1802 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1803 Instruction</a> </div>
1804 <div class="doc_text">
1806 <pre> ret <type> <value> <i>; Return a value from a non-void function</i>
1807 ret void <i>; Return from void function</i>
1808 ret <type> <value>, <type> <value> <i>; Return two values from a non-void function </i>
1813 <p>The '<tt>ret</tt>' instruction is used to return control flow (and a
1814 value) from a function back to the caller.</p>
1815 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
1816 returns value(s) and then causes control flow, and one that just causes
1817 control flow to occur.</p>
1821 <p>The '<tt>ret</tt>' instruction may return zero, one or multiple values.
1822 The type of each return value must be a '<a href="#t_firstclass">first
1823 class</a>' type. Note that a function is not <a href="#wellformed">well
1824 formed</a> if there exists a '<tt>ret</tt>' instruction inside of the
1825 function that returns values that do not match the return type of the
1830 <p>When the '<tt>ret</tt>' instruction is executed, control flow
1831 returns back to the calling function's context. If the caller is a "<a
1832 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1833 the instruction after the call. If the caller was an "<a
1834 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1835 at the beginning of the "normal" destination block. If the instruction
1836 returns a value, that value shall set the call or invoke instruction's
1837 return value. If the instruction returns multiple values then these
1838 values can only be accessed through a '<a href="#i_getresult"><tt>getresult</tt>
1839 </a>' instruction.</p>
1844 ret i32 5 <i>; Return an integer value of 5</i>
1845 ret void <i>; Return from a void function</i>
1846 ret i32 4, i8 2 <i>; Return two values 4 and 2 </i>
1849 <!-- _______________________________________________________________________ -->
1850 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1851 <div class="doc_text">
1853 <pre> br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1856 <p>The '<tt>br</tt>' instruction is used to cause control flow to
1857 transfer to a different basic block in the current function. There are
1858 two forms of this instruction, corresponding to a conditional branch
1859 and an unconditional branch.</p>
1861 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1862 single '<tt>i1</tt>' value and two '<tt>label</tt>' values. The
1863 unconditional form of the '<tt>br</tt>' instruction takes a single
1864 '<tt>label</tt>' value as a target.</p>
1866 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
1867 argument is evaluated. If the value is <tt>true</tt>, control flows
1868 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1869 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1871 <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
1872 href="#i_ret">ret</a> i32 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> i32 0<br></pre>
1874 <!-- _______________________________________________________________________ -->
1875 <div class="doc_subsubsection">
1876 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
1879 <div class="doc_text">
1883 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
1888 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
1889 several different places. It is a generalization of the '<tt>br</tt>'
1890 instruction, allowing a branch to occur to one of many possible
1896 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
1897 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
1898 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
1899 table is not allowed to contain duplicate constant entries.</p>
1903 <p>The <tt>switch</tt> instruction specifies a table of values and
1904 destinations. When the '<tt>switch</tt>' instruction is executed, this
1905 table is searched for the given value. If the value is found, control flow is
1906 transfered to the corresponding destination; otherwise, control flow is
1907 transfered to the default destination.</p>
1909 <h5>Implementation:</h5>
1911 <p>Depending on properties of the target machine and the particular
1912 <tt>switch</tt> instruction, this instruction may be code generated in different
1913 ways. For example, it could be generated as a series of chained conditional
1914 branches or with a lookup table.</p>
1919 <i>; Emulate a conditional br instruction</i>
1920 %Val = <a href="#i_zext">zext</a> i1 %value to i32
1921 switch i32 %Val, label %truedest [i32 0, label %falsedest ]
1923 <i>; Emulate an unconditional br instruction</i>
1924 switch i32 0, label %dest [ ]
1926 <i>; Implement a jump table:</i>
1927 switch i32 %val, label %otherwise [ i32 0, label %onzero
1929 i32 2, label %ontwo ]
1933 <!-- _______________________________________________________________________ -->
1934 <div class="doc_subsubsection">
1935 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
1938 <div class="doc_text">
1943 <result> = invoke [<a href="#callingconv">cconv</a>] <ptr to function ty> <function ptr val>(<function args>)
1944 to label <normal label> unwind label <exception label>
1949 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
1950 function, with the possibility of control flow transfer to either the
1951 '<tt>normal</tt>' label or the
1952 '<tt>exception</tt>' label. If the callee function returns with the
1953 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
1954 "normal" label. If the callee (or any indirect callees) returns with the "<a
1955 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
1956 continued at the dynamically nearest "exception" label. If the callee function
1957 returns multiple values then individual return values are only accessible through
1958 a '<tt><a href="#i_getresult">getresult</a></tt>' instruction.</p>
1962 <p>This instruction requires several arguments:</p>
1966 The optional "cconv" marker indicates which <a href="#callingconv">calling
1967 convention</a> the call should use. If none is specified, the call defaults
1968 to using C calling conventions.
1970 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
1971 function value being invoked. In most cases, this is a direct function
1972 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
1973 an arbitrary pointer to function value.
1976 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
1977 function to be invoked. </li>
1979 <li>'<tt>function args</tt>': argument list whose types match the function
1980 signature argument types. If the function signature indicates the function
1981 accepts a variable number of arguments, the extra arguments can be
1984 <li>'<tt>normal label</tt>': the label reached when the called function
1985 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
1987 <li>'<tt>exception label</tt>': the label reached when a callee returns with
1988 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
1994 <p>This instruction is designed to operate as a standard '<tt><a
1995 href="#i_call">call</a></tt>' instruction in most regards. The primary
1996 difference is that it establishes an association with a label, which is used by
1997 the runtime library to unwind the stack.</p>
1999 <p>This instruction is used in languages with destructors to ensure that proper
2000 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
2001 exception. Additionally, this is important for implementation of
2002 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
2006 %retval = invoke i32 @Test(i32 15) to label %Continue
2007 unwind label %TestCleanup <i>; {i32}:retval set</i>
2008 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
2009 unwind label %TestCleanup <i>; {i32}:retval set</i>
2014 <!-- _______________________________________________________________________ -->
2016 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
2017 Instruction</a> </div>
2019 <div class="doc_text">
2028 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
2029 at the first callee in the dynamic call stack which used an <a
2030 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
2031 primarily used to implement exception handling.</p>
2035 <p>The '<tt>unwind</tt>' instruction causes execution of the current function to
2036 immediately halt. The dynamic call stack is then searched for the first <a
2037 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
2038 execution continues at the "exceptional" destination block specified by the
2039 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
2040 dynamic call chain, undefined behavior results.</p>
2043 <!-- _______________________________________________________________________ -->
2045 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
2046 Instruction</a> </div>
2048 <div class="doc_text">
2057 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
2058 instruction is used to inform the optimizer that a particular portion of the
2059 code is not reachable. This can be used to indicate that the code after a
2060 no-return function cannot be reached, and other facts.</p>
2064 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
2069 <!-- ======================================================================= -->
2070 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
2071 <div class="doc_text">
2072 <p>Binary operators are used to do most of the computation in a
2073 program. They require two operands of the same type, execute an operation on them, and
2074 produce a single value. The operands might represent
2075 multiple data, as is the case with the <a href="#t_vector">vector</a> data type.
2076 The result value has the same type as its operands.</p>
2077 <p>There are several different binary operators:</p>
2079 <!-- _______________________________________________________________________ -->
2080 <div class="doc_subsubsection"> <a name="i_add">'<tt>add</tt>'
2081 Instruction</a> </div>
2082 <div class="doc_text">
2084 <pre> <result> = add <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2087 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
2089 <p>The two arguments to the '<tt>add</tt>' instruction must be either <a
2090 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a> values.
2091 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2092 Both arguments must have identical types.</p>
2094 <p>The value produced is the integer or floating point sum of the two
2096 <p>If an integer sum has unsigned overflow, the result returned is the
2097 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2099 <p>Because LLVM integers use a two's complement representation, this
2100 instruction is appropriate for both signed and unsigned integers.</p>
2102 <pre> <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
2105 <!-- _______________________________________________________________________ -->
2106 <div class="doc_subsubsection"> <a name="i_sub">'<tt>sub</tt>'
2107 Instruction</a> </div>
2108 <div class="doc_text">
2110 <pre> <result> = sub <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2113 <p>The '<tt>sub</tt>' instruction returns the difference of its two
2115 <p>Note that the '<tt>sub</tt>' instruction is used to represent the '<tt>neg</tt>'
2116 instruction present in most other intermediate representations.</p>
2118 <p>The two arguments to the '<tt>sub</tt>' instruction must be either <a
2119 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
2121 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2122 Both arguments must have identical types.</p>
2124 <p>The value produced is the integer or floating point difference of
2125 the two operands.</p>
2126 <p>If an integer difference has unsigned overflow, the result returned is the
2127 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2129 <p>Because LLVM integers use a two's complement representation, this
2130 instruction is appropriate for both signed and unsigned integers.</p>
2133 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
2134 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
2137 <!-- _______________________________________________________________________ -->
2138 <div class="doc_subsubsection"> <a name="i_mul">'<tt>mul</tt>'
2139 Instruction</a> </div>
2140 <div class="doc_text">
2142 <pre> <result> = mul <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2145 <p>The '<tt>mul</tt>' instruction returns the product of its two
2148 <p>The two arguments to the '<tt>mul</tt>' instruction must be either <a
2149 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
2151 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2152 Both arguments must have identical types.</p>
2154 <p>The value produced is the integer or floating point product of the
2156 <p>If the result of an integer multiplication has unsigned overflow,
2157 the result returned is the mathematical result modulo
2158 2<sup>n</sup>, where n is the bit width of the result.</p>
2159 <p>Because LLVM integers use a two's complement representation, and the
2160 result is the same width as the operands, this instruction returns the
2161 correct result for both signed and unsigned integers. If a full product
2162 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands
2163 should be sign-extended or zero-extended as appropriate to the
2164 width of the full product.</p>
2166 <pre> <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
2169 <!-- _______________________________________________________________________ -->
2170 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
2172 <div class="doc_text">
2174 <pre> <result> = udiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2177 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two
2180 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
2181 <a href="#t_integer">integer</a> values. Both arguments must have identical
2182 types. This instruction can also take <a href="#t_vector">vector</a> versions
2183 of the values in which case the elements must be integers.</p>
2185 <p>The value produced is the unsigned integer quotient of the two operands.</p>
2186 <p>Note that unsigned integer division and signed integer division are distinct
2187 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
2188 <p>Division by zero leads to undefined behavior.</p>
2190 <pre> <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2193 <!-- _______________________________________________________________________ -->
2194 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
2196 <div class="doc_text">
2198 <pre> <result> = sdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2201 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
2204 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
2205 <a href="#t_integer">integer</a> values. Both arguments must have identical
2206 types. This instruction can also take <a href="#t_vector">vector</a> versions
2207 of the values in which case the elements must be integers.</p>
2209 <p>The value produced is the signed integer quotient of the two operands rounded towards zero.</p>
2210 <p>Note that signed integer division and unsigned integer division are distinct
2211 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
2212 <p>Division by zero leads to undefined behavior. Overflow also leads to
2213 undefined behavior; this is a rare case, but can occur, for example,
2214 by doing a 32-bit division of -2147483648 by -1.</p>
2216 <pre> <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2219 <!-- _______________________________________________________________________ -->
2220 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
2221 Instruction</a> </div>
2222 <div class="doc_text">
2224 <pre> <result> = fdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2227 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
2230 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
2231 <a href="#t_floating">floating point</a> values. Both arguments must have
2232 identical types. This instruction can also take <a href="#t_vector">vector</a>
2233 versions of floating point values.</p>
2235 <p>The value produced is the floating point quotient of the two operands.</p>
2237 <pre> <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
2240 <!-- _______________________________________________________________________ -->
2241 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
2243 <div class="doc_text">
2245 <pre> <result> = urem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2248 <p>The '<tt>urem</tt>' instruction returns the remainder from the
2249 unsigned division of its two arguments.</p>
2251 <p>The two arguments to the '<tt>urem</tt>' instruction must be
2252 <a href="#t_integer">integer</a> values. Both arguments must have identical
2253 types. This instruction can also take <a href="#t_vector">vector</a> versions
2254 of the values in which case the elements must be integers.</p>
2256 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
2257 This instruction always performs an unsigned division to get the remainder.</p>
2258 <p>Note that unsigned integer remainder and signed integer remainder are
2259 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
2260 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
2262 <pre> <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2266 <!-- _______________________________________________________________________ -->
2267 <div class="doc_subsubsection"> <a name="i_srem">'<tt>srem</tt>'
2268 Instruction</a> </div>
2269 <div class="doc_text">
2271 <pre> <result> = srem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2274 <p>The '<tt>srem</tt>' instruction returns the remainder from the
2275 signed division of its two operands. This instruction can also take
2276 <a href="#t_vector">vector</a> versions of the values in which case
2277 the elements must be integers.</p>
2280 <p>The two arguments to the '<tt>srem</tt>' instruction must be
2281 <a href="#t_integer">integer</a> values. Both arguments must have identical
2284 <p>This instruction returns the <i>remainder</i> of a division (where the result
2285 has the same sign as the dividend, <tt>var1</tt>), not the <i>modulo</i>
2286 operator (where the result has the same sign as the divisor, <tt>var2</tt>) of
2287 a value. For more information about the difference, see <a
2288 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
2289 Math Forum</a>. For a table of how this is implemented in various languages,
2290 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
2291 Wikipedia: modulo operation</a>.</p>
2292 <p>Note that signed integer remainder and unsigned integer remainder are
2293 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
2294 <p>Taking the remainder of a division by zero leads to undefined behavior.
2295 Overflow also leads to undefined behavior; this is a rare case, but can occur,
2296 for example, by taking the remainder of a 32-bit division of -2147483648 by -1.
2297 (The remainder doesn't actually overflow, but this rule lets srem be
2298 implemented using instructions that return both the result of the division
2299 and the remainder.)</p>
2301 <pre> <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2305 <!-- _______________________________________________________________________ -->
2306 <div class="doc_subsubsection"> <a name="i_frem">'<tt>frem</tt>'
2307 Instruction</a> </div>
2308 <div class="doc_text">
2310 <pre> <result> = frem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2313 <p>The '<tt>frem</tt>' instruction returns the remainder from the
2314 division of its two operands.</p>
2316 <p>The two arguments to the '<tt>frem</tt>' instruction must be
2317 <a href="#t_floating">floating point</a> values. Both arguments must have
2318 identical types. This instruction can also take <a href="#t_vector">vector</a>
2319 versions of floating point values.</p>
2321 <p>This instruction returns the <i>remainder</i> of a division.
2322 The remainder has the same sign as the dividend.</p>
2324 <pre> <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
2328 <!-- ======================================================================= -->
2329 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
2330 Operations</a> </div>
2331 <div class="doc_text">
2332 <p>Bitwise binary operators are used to do various forms of
2333 bit-twiddling in a program. They are generally very efficient
2334 instructions and can commonly be strength reduced from other
2335 instructions. They require two operands of the same type, execute an operation on them,
2336 and produce a single value. The resulting value is the same type as its operands.</p>
2339 <!-- _______________________________________________________________________ -->
2340 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
2341 Instruction</a> </div>
2342 <div class="doc_text">
2344 <pre> <result> = shl <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2349 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2350 the left a specified number of bits.</p>
2354 <p>Both arguments to the '<tt>shl</tt>' instruction must be the same <a
2355 href="#t_integer">integer</a> type. '<tt>var2</tt>' is treated as an
2360 <p>The value produced is <tt>var1</tt> * 2<sup><tt>var2</tt></sup> mod 2<sup>n</sup>,
2361 where n is the width of the result. If <tt>var2</tt> is (statically or dynamically) negative or
2362 equal to or larger than the number of bits in <tt>var1</tt>, the result is undefined.</p>
2364 <h5>Example:</h5><pre>
2365 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
2366 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
2367 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
2368 <result> = shl i32 1, 32 <i>; undefined</i>
2371 <!-- _______________________________________________________________________ -->
2372 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
2373 Instruction</a> </div>
2374 <div class="doc_text">
2376 <pre> <result> = lshr <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2380 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
2381 operand shifted to the right a specified number of bits with zero fill.</p>
2384 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
2385 <a href="#t_integer">integer</a> type. '<tt>var2</tt>' is treated as an
2390 <p>This instruction always performs a logical shift right operation. The most
2391 significant bits of the result will be filled with zero bits after the
2392 shift. If <tt>var2</tt> is (statically or dynamically) equal to or larger than
2393 the number of bits in <tt>var1</tt>, the result is undefined.</p>
2397 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
2398 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
2399 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
2400 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
2401 <result> = lshr i32 1, 32 <i>; undefined</i>
2405 <!-- _______________________________________________________________________ -->
2406 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
2407 Instruction</a> </div>
2408 <div class="doc_text">
2411 <pre> <result> = ashr <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2415 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
2416 operand shifted to the right a specified number of bits with sign extension.</p>
2419 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
2420 <a href="#t_integer">integer</a> type. '<tt>var2</tt>' is treated as an
2424 <p>This instruction always performs an arithmetic shift right operation,
2425 The most significant bits of the result will be filled with the sign bit
2426 of <tt>var1</tt>. If <tt>var2</tt> is (statically or dynamically) equal to or
2427 larger than the number of bits in <tt>var1</tt>, the result is undefined.
2432 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
2433 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
2434 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
2435 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
2436 <result> = ashr i32 1, 32 <i>; undefined</i>
2440 <!-- _______________________________________________________________________ -->
2441 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
2442 Instruction</a> </div>
2443 <div class="doc_text">
2445 <pre> <result> = and <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2448 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
2449 its two operands.</p>
2451 <p>The two arguments to the '<tt>and</tt>' instruction must be <a
2452 href="#t_integer">integer</a> values. Both arguments must have
2453 identical types.</p>
2455 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
2457 <div style="align: center">
2458 <table border="1" cellspacing="0" cellpadding="4">
2489 <pre> <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
2490 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
2491 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
2494 <!-- _______________________________________________________________________ -->
2495 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
2496 <div class="doc_text">
2498 <pre> <result> = or <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2501 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
2502 or of its two operands.</p>
2504 <p>The two arguments to the '<tt>or</tt>' instruction must be <a
2505 href="#t_integer">integer</a> values. Both arguments must have
2506 identical types.</p>
2508 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
2510 <div style="align: center">
2511 <table border="1" cellspacing="0" cellpadding="4">
2542 <pre> <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
2543 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
2544 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
2547 <!-- _______________________________________________________________________ -->
2548 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
2549 Instruction</a> </div>
2550 <div class="doc_text">
2552 <pre> <result> = xor <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2555 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
2556 or of its two operands. The <tt>xor</tt> is used to implement the
2557 "one's complement" operation, which is the "~" operator in C.</p>
2559 <p>The two arguments to the '<tt>xor</tt>' instruction must be <a
2560 href="#t_integer">integer</a> values. Both arguments must have
2561 identical types.</p>
2563 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
2565 <div style="align: center">
2566 <table border="1" cellspacing="0" cellpadding="4">
2598 <pre> <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
2599 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
2600 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
2601 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
2605 <!-- ======================================================================= -->
2606 <div class="doc_subsection">
2607 <a name="vectorops">Vector Operations</a>
2610 <div class="doc_text">
2612 <p>LLVM supports several instructions to represent vector operations in a
2613 target-independent manner. These instructions cover the element-access and
2614 vector-specific operations needed to process vectors effectively. While LLVM
2615 does directly support these vector operations, many sophisticated algorithms
2616 will want to use target-specific intrinsics to take full advantage of a specific
2621 <!-- _______________________________________________________________________ -->
2622 <div class="doc_subsubsection">
2623 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2626 <div class="doc_text">
2631 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
2637 The '<tt>extractelement</tt>' instruction extracts a single scalar
2638 element from a vector at a specified index.
2645 The first operand of an '<tt>extractelement</tt>' instruction is a
2646 value of <a href="#t_vector">vector</a> type. The second operand is
2647 an index indicating the position from which to extract the element.
2648 The index may be a variable.</p>
2653 The result is a scalar of the same type as the element type of
2654 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2655 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2656 results are undefined.
2662 %result = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
2667 <!-- _______________________________________________________________________ -->
2668 <div class="doc_subsubsection">
2669 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
2672 <div class="doc_text">
2677 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
2683 The '<tt>insertelement</tt>' instruction inserts a scalar
2684 element into a vector at a specified index.
2691 The first operand of an '<tt>insertelement</tt>' instruction is a
2692 value of <a href="#t_vector">vector</a> type. The second operand is a
2693 scalar value whose type must equal the element type of the first
2694 operand. The third operand is an index indicating the position at
2695 which to insert the value. The index may be a variable.</p>
2700 The result is a vector of the same type as <tt>val</tt>. Its
2701 element values are those of <tt>val</tt> except at position
2702 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
2703 exceeds the length of <tt>val</tt>, the results are undefined.
2709 %result = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
2713 <!-- _______________________________________________________________________ -->
2714 <div class="doc_subsubsection">
2715 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
2718 <div class="doc_text">
2723 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <n x i32> <mask> <i>; yields <n x <ty>></i>
2729 The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
2730 from two input vectors, returning a vector of the same type.
2736 The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
2737 with types that match each other and types that match the result of the
2738 instruction. The third argument is a shuffle mask, which has the same number
2739 of elements as the other vector type, but whose element type is always 'i32'.
2743 The shuffle mask operand is required to be a constant vector with either
2744 constant integer or undef values.
2750 The elements of the two input vectors are numbered from left to right across
2751 both of the vectors. The shuffle mask operand specifies, for each element of
2752 the result vector, which element of the two input registers the result element
2753 gets. The element selector may be undef (meaning "don't care") and the second
2754 operand may be undef if performing a shuffle from only one vector.
2760 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
2761 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
2762 %result = shufflevector <4 x i32> %v1, <4 x i32> undef,
2763 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
2768 <!-- ======================================================================= -->
2769 <div class="doc_subsection">
2770 <a name="memoryops">Memory Access and Addressing Operations</a>
2773 <div class="doc_text">
2775 <p>A key design point of an SSA-based representation is how it
2776 represents memory. In LLVM, no memory locations are in SSA form, which
2777 makes things very simple. This section describes how to read, write,
2778 allocate, and free memory in LLVM.</p>
2782 <!-- _______________________________________________________________________ -->
2783 <div class="doc_subsubsection">
2784 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
2787 <div class="doc_text">
2792 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2797 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
2798 heap and returns a pointer to it. The object is always allocated in the generic
2799 address space (address space zero).</p>
2803 <p>The '<tt>malloc</tt>' instruction allocates
2804 <tt>sizeof(<type>)*NumElements</tt>
2805 bytes of memory from the operating system and returns a pointer of the
2806 appropriate type to the program. If "NumElements" is specified, it is the
2807 number of elements allocated, otherwise "NumElements" is defaulted to be one.
2808 If a constant alignment is specified, the value result of the allocation is guaranteed to
2809 be aligned to at least that boundary. If not specified, or if zero, the target can
2810 choose to align the allocation on any convenient boundary.</p>
2812 <p>'<tt>type</tt>' must be a sized type.</p>
2816 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
2817 a pointer is returned. The result of a zero byte allocattion is undefined. The
2818 result is null if there is insufficient memory available.</p>
2823 %array = malloc [4 x i8 ] <i>; yields {[%4 x i8]*}:array</i>
2825 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
2826 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
2827 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
2828 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
2829 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
2833 <!-- _______________________________________________________________________ -->
2834 <div class="doc_subsubsection">
2835 <a name="i_free">'<tt>free</tt>' Instruction</a>
2838 <div class="doc_text">
2843 free <type> <value> <i>; yields {void}</i>
2848 <p>The '<tt>free</tt>' instruction returns memory back to the unused
2849 memory heap to be reallocated in the future.</p>
2853 <p>'<tt>value</tt>' shall be a pointer value that points to a value
2854 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
2859 <p>Access to the memory pointed to by the pointer is no longer defined
2860 after this instruction executes. If the pointer is null, the operation
2866 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
2867 free [4 x i8]* %array
2871 <!-- _______________________________________________________________________ -->
2872 <div class="doc_subsubsection">
2873 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
2876 <div class="doc_text">
2881 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2886 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
2887 currently executing function, to be automatically released when this function
2888 returns to its caller. The object is always allocated in the generic address
2889 space (address space zero).</p>
2893 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
2894 bytes of memory on the runtime stack, returning a pointer of the
2895 appropriate type to the program. If "NumElements" is specified, it is the
2896 number of elements allocated, otherwise "NumElements" is defaulted to be one.
2897 If a constant alignment is specified, the value result of the allocation is guaranteed
2898 to be aligned to at least that boundary. If not specified, or if zero, the target
2899 can choose to align the allocation on any convenient boundary.</p>
2901 <p>'<tt>type</tt>' may be any sized type.</p>
2905 <p>Memory is allocated; a pointer is returned. The operation is undefiend if
2906 there is insufficient stack space for the allocation. '<tt>alloca</tt>'d
2907 memory is automatically released when the function returns. The '<tt>alloca</tt>'
2908 instruction is commonly used to represent automatic variables that must
2909 have an address available. When the function returns (either with the <tt><a
2910 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
2911 instructions), the memory is reclaimed. Allocating zero bytes
2912 is legal, but the result is undefined.</p>
2917 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
2918 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
2919 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
2920 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
2924 <!-- _______________________________________________________________________ -->
2925 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
2926 Instruction</a> </div>
2927 <div class="doc_text">
2929 <pre> <result> = load <ty>* <pointer>[, align <alignment>]<br> <result> = volatile load <ty>* <pointer>[, align <alignment>]<br></pre>
2931 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
2933 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
2934 address from which to load. The pointer must point to a <a
2935 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
2936 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
2937 the number or order of execution of this <tt>load</tt> with other
2938 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
2941 The optional constant "align" argument specifies the alignment of the operation
2942 (that is, the alignment of the memory address). A value of 0 or an
2943 omitted "align" argument means that the operation has the preferential
2944 alignment for the target. It is the responsibility of the code emitter
2945 to ensure that the alignment information is correct. Overestimating
2946 the alignment results in an undefined behavior. Underestimating the
2947 alignment may produce less efficient code. An alignment of 1 is always
2951 <p>The location of memory pointed to is loaded.</p>
2953 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
2955 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
2956 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
2959 <!-- _______________________________________________________________________ -->
2960 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
2961 Instruction</a> </div>
2962 <div class="doc_text">
2964 <pre> store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
2965 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
2968 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
2970 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
2971 to store and an address at which to store it. The type of the '<tt><pointer></tt>'
2972 operand must be a pointer to the <a href="#t_firstclass">first class</a> type
2973 of the '<tt><value></tt>'
2974 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
2975 optimizer is not allowed to modify the number or order of execution of
2976 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
2977 href="#i_store">store</a></tt> instructions.</p>
2979 The optional constant "align" argument specifies the alignment of the operation
2980 (that is, the alignment of the memory address). A value of 0 or an
2981 omitted "align" argument means that the operation has the preferential
2982 alignment for the target. It is the responsibility of the code emitter
2983 to ensure that the alignment information is correct. Overestimating
2984 the alignment results in an undefined behavior. Underestimating the
2985 alignment may produce less efficient code. An alignment of 1 is always
2989 <p>The contents of memory are updated to contain '<tt><value></tt>'
2990 at the location specified by the '<tt><pointer></tt>' operand.</p>
2992 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
2993 store i32 3, i32* %ptr <i>; yields {void}</i>
2994 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
2998 <!-- _______________________________________________________________________ -->
2999 <div class="doc_subsubsection">
3000 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
3003 <div class="doc_text">
3006 <result> = getelementptr <ty>* <ptrval>{, <ty> <idx>}*
3012 The '<tt>getelementptr</tt>' instruction is used to get the address of a
3013 subelement of an aggregate data structure.</p>
3017 <p>This instruction takes a list of integer operands that indicate what
3018 elements of the aggregate object to index to. The actual types of the arguments
3019 provided depend on the type of the first pointer argument. The
3020 '<tt>getelementptr</tt>' instruction is used to index down through the type
3021 levels of a structure or to a specific index in an array. When indexing into a
3022 structure, only <tt>i32</tt> integer constants are allowed. When indexing
3023 into an array or pointer, only integers of 32 or 64 bits are allowed; 32-bit
3024 values will be sign extended to 64-bits if required.</p>
3026 <p>For example, let's consider a C code fragment and how it gets
3027 compiled to LLVM:</p>
3029 <div class="doc_code">
3042 int *foo(struct ST *s) {
3043 return &s[1].Z.B[5][13];
3048 <p>The LLVM code generated by the GCC frontend is:</p>
3050 <div class="doc_code">
3052 %RT = type { i8 , [10 x [20 x i32]], i8 }
3053 %ST = type { i32, double, %RT }
3055 define i32* %foo(%ST* %s) {
3057 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
3065 <p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
3066 on the pointer type that is being indexed into. <a href="#t_pointer">Pointer</a>
3067 and <a href="#t_array">array</a> types can use a 32-bit or 64-bit
3068 <a href="#t_integer">integer</a> type but the value will always be sign extended
3069 to 64-bits. <a href="#t_struct">Structure</a> and <a href="#t_pstruct">packed
3070 structure</a> types require <tt>i32</tt> <b>constants</b>.</p>
3072 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
3073 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
3074 }</tt>' type, a structure. The second index indexes into the third element of
3075 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
3076 i8 }</tt>' type, another structure. The third index indexes into the second
3077 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
3078 array. The two dimensions of the array are subscripted into, yielding an
3079 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
3080 to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
3082 <p>Note that it is perfectly legal to index partially through a
3083 structure, returning a pointer to an inner element. Because of this,
3084 the LLVM code for the given testcase is equivalent to:</p>
3087 define i32* %foo(%ST* %s) {
3088 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
3089 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
3090 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
3091 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
3092 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
3097 <p>Note that it is undefined to access an array out of bounds: array and
3098 pointer indexes must always be within the defined bounds of the array type.
3099 The one exception for this rule is zero length arrays. These arrays are
3100 defined to be accessible as variable length arrays, which requires access
3101 beyond the zero'th element.</p>
3103 <p>The getelementptr instruction is often confusing. For some more insight
3104 into how it works, see <a href="GetElementPtr.html">the getelementptr
3110 <i>; yields [12 x i8]*:aptr</i>
3111 %aptr = getelementptr {i32, [12 x i8]}* %sptr, i64 0, i32 1
3115 <!-- ======================================================================= -->
3116 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
3118 <div class="doc_text">
3119 <p>The instructions in this category are the conversion instructions (casting)
3120 which all take a single operand and a type. They perform various bit conversions
3124 <!-- _______________________________________________________________________ -->
3125 <div class="doc_subsubsection">
3126 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
3128 <div class="doc_text">
3132 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
3137 The '<tt>trunc</tt>' instruction truncates its operand to the type <tt>ty2</tt>.
3142 The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
3143 be an <a href="#t_integer">integer</a> type, and a type that specifies the size
3144 and type of the result, which must be an <a href="#t_integer">integer</a>
3145 type. The bit size of <tt>value</tt> must be larger than the bit size of
3146 <tt>ty2</tt>. Equal sized types are not allowed.</p>
3150 The '<tt>trunc</tt>' instruction truncates the high order bits in <tt>value</tt>
3151 and converts the remaining bits to <tt>ty2</tt>. Since the source size must be
3152 larger than the destination size, <tt>trunc</tt> cannot be a <i>no-op cast</i>.
3153 It will always truncate bits.</p>
3157 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
3158 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
3159 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
3163 <!-- _______________________________________________________________________ -->
3164 <div class="doc_subsubsection">
3165 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
3167 <div class="doc_text">
3171 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
3175 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
3180 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
3181 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3182 also be of <a href="#t_integer">integer</a> type. The bit size of the
3183 <tt>value</tt> must be smaller than the bit size of the destination type,
3187 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
3188 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
3190 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
3194 %X = zext i32 257 to i64 <i>; yields i64:257</i>
3195 %Y = zext i1 true to i32 <i>; yields i32:1</i>
3199 <!-- _______________________________________________________________________ -->
3200 <div class="doc_subsubsection">
3201 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
3203 <div class="doc_text">
3207 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
3211 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
3215 The '<tt>sext</tt>' instruction takes a value to cast, which must be of
3216 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3217 also be of <a href="#t_integer">integer</a> type. The bit size of the
3218 <tt>value</tt> must be smaller than the bit size of the destination type,
3223 The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
3224 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size of
3225 the type <tt>ty2</tt>.</p>
3227 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
3231 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
3232 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
3236 <!-- _______________________________________________________________________ -->
3237 <div class="doc_subsubsection">
3238 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
3241 <div class="doc_text">
3246 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
3250 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
3255 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
3256 point</a> value to cast and a <a href="#t_floating">floating point</a> type to
3257 cast it to. The size of <tt>value</tt> must be larger than the size of
3258 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
3259 <i>no-op cast</i>.</p>
3262 <p> The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
3263 <a href="#t_floating">floating point</a> type to a smaller
3264 <a href="#t_floating">floating point</a> type. If the value cannot fit within
3265 the destination type, <tt>ty2</tt>, then the results are undefined.</p>
3269 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
3270 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
3274 <!-- _______________________________________________________________________ -->
3275 <div class="doc_subsubsection">
3276 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
3278 <div class="doc_text">
3282 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
3286 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
3287 floating point value.</p>
3290 <p>The '<tt>fpext</tt>' instruction takes a
3291 <a href="#t_floating">floating point</a> <tt>value</tt> to cast,
3292 and a <a href="#t_floating">floating point</a> type to cast it to. The source
3293 type must be smaller than the destination type.</p>
3296 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
3297 <a href="#t_floating">floating point</a> type to a larger
3298 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
3299 used to make a <i>no-op cast</i> because it always changes bits. Use
3300 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
3304 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
3305 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
3309 <!-- _______________________________________________________________________ -->
3310 <div class="doc_subsubsection">
3311 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
3313 <div class="doc_text">
3317 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
3321 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
3322 unsigned integer equivalent of type <tt>ty2</tt>.
3326 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
3327 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3328 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3329 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3330 vector integer type with the same number of elements as <tt>ty</tt></p>
3333 <p> The '<tt>fptoui</tt>' instruction converts its
3334 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3335 towards zero) unsigned integer value. If the value cannot fit in <tt>ty2</tt>,
3336 the results are undefined.</p>
3340 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
3341 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
3342 %X = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
3346 <!-- _______________________________________________________________________ -->
3347 <div class="doc_subsubsection">
3348 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
3350 <div class="doc_text">
3354 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
3358 <p>The '<tt>fptosi</tt>' instruction converts
3359 <a href="#t_floating">floating point</a> <tt>value</tt> to type <tt>ty2</tt>.
3363 <p> The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
3364 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3365 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3366 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3367 vector integer type with the same number of elements as <tt>ty</tt></p>
3370 <p>The '<tt>fptosi</tt>' instruction converts its
3371 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3372 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
3373 the results are undefined.</p>
3377 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
3378 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
3379 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
3383 <!-- _______________________________________________________________________ -->
3384 <div class="doc_subsubsection">
3385 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
3387 <div class="doc_text">
3391 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3395 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
3396 integer and converts that value to the <tt>ty2</tt> type.</p>
3399 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
3400 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3401 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3402 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3403 floating point type with the same number of elements as <tt>ty</tt></p>
3406 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
3407 integer quantity and converts it to the corresponding floating point value. If
3408 the value cannot fit in the floating point value, the results are undefined.</p>
3412 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
3413 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
3417 <!-- _______________________________________________________________________ -->
3418 <div class="doc_subsubsection">
3419 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
3421 <div class="doc_text">
3425 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3429 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed
3430 integer and converts that value to the <tt>ty2</tt> type.</p>
3433 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
3434 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3435 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3436 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3437 floating point type with the same number of elements as <tt>ty</tt></p>
3440 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed
3441 integer quantity and converts it to the corresponding floating point value. If
3442 the value cannot fit in the floating point value, the results are undefined.</p>
3446 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
3447 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
3451 <!-- _______________________________________________________________________ -->
3452 <div class="doc_subsubsection">
3453 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
3455 <div class="doc_text">
3459 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
3463 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
3464 the integer type <tt>ty2</tt>.</p>
3467 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
3468 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
3469 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.
3472 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
3473 <tt>ty2</tt> by interpreting the pointer value as an integer and either
3474 truncating or zero extending that value to the size of the integer type. If
3475 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
3476 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
3477 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
3482 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
3483 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
3487 <!-- _______________________________________________________________________ -->
3488 <div class="doc_subsubsection">
3489 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
3491 <div class="doc_text">
3495 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
3499 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to
3500 a pointer type, <tt>ty2</tt>.</p>
3503 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
3504 value to cast, and a type to cast it to, which must be a
3505 <a href="#t_pointer">pointer</a> type.
3508 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
3509 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
3510 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
3511 size of a pointer then a truncation is done. If <tt>value</tt> is smaller than
3512 the size of a pointer then a zero extension is done. If they are the same size,
3513 nothing is done (<i>no-op cast</i>).</p>
3517 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
3518 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
3519 %Y = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
3523 <!-- _______________________________________________________________________ -->
3524 <div class="doc_subsubsection">
3525 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
3527 <div class="doc_text">
3531 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
3535 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3536 <tt>ty2</tt> without changing any bits.</p>
3539 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be
3540 a first class value, and a type to cast it to, which must also be a <a
3541 href="#t_firstclass">first class</a> type. The bit sizes of <tt>value</tt>
3542 and the destination type, <tt>ty2</tt>, must be identical. If the source
3543 type is a pointer, the destination type must also be a pointer.</p>
3546 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3547 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
3548 this conversion. The conversion is done as if the <tt>value</tt> had been
3549 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only be
3550 converted to other pointer types with this instruction. To convert pointers to
3551 other types, use the <a href="#i_inttoptr">inttoptr</a> or
3552 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
3556 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
3557 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
3558 %Z = bitcast <2xint> %V to i64; <i>; yields i64: %V</i>
3562 <!-- ======================================================================= -->
3563 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
3564 <div class="doc_text">
3565 <p>The instructions in this category are the "miscellaneous"
3566 instructions, which defy better classification.</p>
3569 <!-- _______________________________________________________________________ -->
3570 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
3572 <div class="doc_text">
3574 <pre> <result> = icmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3577 <p>The '<tt>icmp</tt>' instruction returns a boolean value based on comparison
3578 of its two integer or pointer operands.</p>
3580 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
3581 the condition code indicating the kind of comparison to perform. It is not
3582 a value, just a keyword. The possible condition code are:
3584 <li><tt>eq</tt>: equal</li>
3585 <li><tt>ne</tt>: not equal </li>
3586 <li><tt>ugt</tt>: unsigned greater than</li>
3587 <li><tt>uge</tt>: unsigned greater or equal</li>
3588 <li><tt>ult</tt>: unsigned less than</li>
3589 <li><tt>ule</tt>: unsigned less or equal</li>
3590 <li><tt>sgt</tt>: signed greater than</li>
3591 <li><tt>sge</tt>: signed greater or equal</li>
3592 <li><tt>slt</tt>: signed less than</li>
3593 <li><tt>sle</tt>: signed less or equal</li>
3595 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
3596 <a href="#t_pointer">pointer</a> typed. They must also be identical types.</p>
3598 <p>The '<tt>icmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3599 the condition code given as <tt>cond</tt>. The comparison performed always
3600 yields a <a href="#t_primitive">i1</a> result, as follows:
3602 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
3603 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3605 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
3606 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3607 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
3608 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3609 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
3610 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3611 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
3612 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3613 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
3614 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3615 <li><tt>sgt</tt>: interprets the operands as signed values and yields
3616 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3617 <li><tt>sge</tt>: interprets the operands as signed values and yields
3618 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3619 <li><tt>slt</tt>: interprets the operands as signed values and yields
3620 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3621 <li><tt>sle</tt>: interprets the operands as signed values and yields
3622 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3624 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
3625 values are compared as if they were integers.</p>
3628 <pre> <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
3629 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
3630 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
3631 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
3632 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
3633 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
3637 <!-- _______________________________________________________________________ -->
3638 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
3640 <div class="doc_text">
3642 <pre> <result> = fcmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3645 <p>The '<tt>fcmp</tt>' instruction returns a boolean value based on comparison
3646 of its floating point operands.</p>
3648 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
3649 the condition code indicating the kind of comparison to perform. It is not
3650 a value, just a keyword. The possible condition code are:
3652 <li><tt>false</tt>: no comparison, always returns false</li>
3653 <li><tt>oeq</tt>: ordered and equal</li>
3654 <li><tt>ogt</tt>: ordered and greater than </li>
3655 <li><tt>oge</tt>: ordered and greater than or equal</li>
3656 <li><tt>olt</tt>: ordered and less than </li>
3657 <li><tt>ole</tt>: ordered and less than or equal</li>
3658 <li><tt>one</tt>: ordered and not equal</li>
3659 <li><tt>ord</tt>: ordered (no nans)</li>
3660 <li><tt>ueq</tt>: unordered or equal</li>
3661 <li><tt>ugt</tt>: unordered or greater than </li>
3662 <li><tt>uge</tt>: unordered or greater than or equal</li>
3663 <li><tt>ult</tt>: unordered or less than </li>
3664 <li><tt>ule</tt>: unordered or less than or equal</li>
3665 <li><tt>une</tt>: unordered or not equal</li>
3666 <li><tt>uno</tt>: unordered (either nans)</li>
3667 <li><tt>true</tt>: no comparison, always returns true</li>
3669 <p><i>Ordered</i> means that neither operand is a QNAN while
3670 <i>unordered</i> means that either operand may be a QNAN.</p>
3671 <p>The <tt>val1</tt> and <tt>val2</tt> arguments must be
3672 <a href="#t_floating">floating point</a> typed. They must have identical
3675 <p>The '<tt>fcmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3676 the condition code given as <tt>cond</tt>. The comparison performed always
3677 yields a <a href="#t_primitive">i1</a> result, as follows:
3679 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
3680 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3681 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3682 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3683 <tt>var1</tt> is greather than <tt>var2</tt>.</li>
3684 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3685 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3686 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3687 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3688 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3689 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3690 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3691 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3692 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
3693 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
3694 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3695 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
3696 <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3697 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
3698 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3699 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
3700 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3701 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
3702 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3703 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
3704 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3705 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
3706 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
3710 <pre> <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
3711 <result> = icmp one float 4.0, 5.0 <i>; yields: result=true</i>
3712 <result> = icmp olt float 4.0, 5.0 <i>; yields: result=true</i>
3713 <result> = icmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
3717 <!-- _______________________________________________________________________ -->
3718 <div class="doc_subsubsection"> <a name="i_phi">'<tt>phi</tt>'
3719 Instruction</a> </div>
3720 <div class="doc_text">
3722 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
3724 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
3725 the SSA graph representing the function.</p>
3727 <p>The type of the incoming values is specified with the first type
3728 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
3729 as arguments, with one pair for each predecessor basic block of the
3730 current block. Only values of <a href="#t_firstclass">first class</a>
3731 type may be used as the value arguments to the PHI node. Only labels
3732 may be used as the label arguments.</p>
3733 <p>There must be no non-phi instructions between the start of a basic
3734 block and the PHI instructions: i.e. PHI instructions must be first in
3737 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
3738 specified by the pair corresponding to the predecessor basic block that executed
3739 just prior to the current block.</p>
3741 <pre>Loop: ; Infinite loop that counts from 0 on up...<br> %indvar = phi i32 [ 0, %LoopHeader ], [ %nextindvar, %Loop ]<br> %nextindvar = add i32 %indvar, 1<br> br label %Loop<br></pre>
3744 <!-- _______________________________________________________________________ -->
3745 <div class="doc_subsubsection">
3746 <a name="i_select">'<tt>select</tt>' Instruction</a>
3749 <div class="doc_text">
3754 <result> = select i1 <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
3760 The '<tt>select</tt>' instruction is used to choose one value based on a
3761 condition, without branching.
3768 The '<tt>select</tt>' instruction requires a boolean value indicating the condition, and two values of the same <a href="#t_firstclass">first class</a> type.
3774 If the boolean condition evaluates to true, the instruction returns the first
3775 value argument; otherwise, it returns the second value argument.
3781 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
3786 <!-- _______________________________________________________________________ -->
3787 <div class="doc_subsubsection">
3788 <a name="i_call">'<tt>call</tt>' Instruction</a>
3791 <div class="doc_text">
3795 <result> = [tail] call [<a href="#callingconv">cconv</a>] <ty> [<fnty>*] <fnptrval>(<param list>)
3800 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
3804 <p>This instruction requires several arguments:</p>
3808 <p>The optional "tail" marker indicates whether the callee function accesses
3809 any allocas or varargs in the caller. If the "tail" marker is present, the
3810 function call is eligible for tail call optimization. Note that calls may
3811 be marked "tail" even if they do not occur before a <a
3812 href="#i_ret"><tt>ret</tt></a> instruction.
3815 <p>The optional "cconv" marker indicates which <a href="#callingconv">calling
3816 convention</a> the call should use. If none is specified, the call defaults
3817 to using C calling conventions.
3820 <p>'<tt>ty</tt>': the type of the call instruction itself which is also
3821 the type of the return value. Functions that return no value are marked
3822 <tt><a href="#t_void">void</a></tt>.</p>
3825 <p>'<tt>fnty</tt>': shall be the signature of the pointer to function
3826 value being invoked. The argument types must match the types implied by
3827 this signature. This type can be omitted if the function is not varargs
3828 and if the function type does not return a pointer to a function.</p>
3831 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
3832 be invoked. In most cases, this is a direct function invocation, but
3833 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
3834 to function value.</p>
3837 <p>'<tt>function args</tt>': argument list whose types match the
3838 function signature argument types. All arguments must be of
3839 <a href="#t_firstclass">first class</a> type. If the function signature
3840 indicates the function accepts a variable number of arguments, the extra
3841 arguments can be specified.</p>
3847 <p>The '<tt>call</tt>' instruction is used to cause control flow to
3848 transfer to a specified function, with its incoming arguments bound to
3849 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
3850 instruction in the called function, control flow continues with the
3851 instruction after the function call, and the return value of the
3852 function is bound to the result argument. If the callee returns multiple
3853 values then the return values of the function are only accessible through
3854 the '<tt><a href="#i_getresult">getresult</a></tt>' instruction.</p>
3859 %retval = call i32 @test(i32 %argc)
3860 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42) <i>; yields i32</i>
3861 %X = tail call i32 @foo() <i>; yields i32</i>
3862 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
3863 call void %foo(i8 97 signext)
3865 %struct.A = type { i32, i8 }
3866 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
3867 %gr = getresult %struct.A %r, 0 <i>; yields i32</i>
3868 %gr1 = getresult %struct.A %r, 1 <i>; yields i8</i>
3873 <!-- _______________________________________________________________________ -->
3874 <div class="doc_subsubsection">
3875 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
3878 <div class="doc_text">
3883 <resultval> = va_arg <va_list*> <arglist>, <argty>
3888 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
3889 the "variable argument" area of a function call. It is used to implement the
3890 <tt>va_arg</tt> macro in C.</p>
3894 <p>This instruction takes a <tt>va_list*</tt> value and the type of
3895 the argument. It returns a value of the specified argument type and
3896 increments the <tt>va_list</tt> to point to the next argument. The
3897 actual type of <tt>va_list</tt> is target specific.</p>
3901 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
3902 type from the specified <tt>va_list</tt> and causes the
3903 <tt>va_list</tt> to point to the next argument. For more information,
3904 see the variable argument handling <a href="#int_varargs">Intrinsic
3907 <p>It is legal for this instruction to be called in a function which does not
3908 take a variable number of arguments, for example, the <tt>vfprintf</tt>
3911 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
3912 href="#intrinsics">intrinsic function</a> because it takes a type as an
3917 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
3921 <!-- _______________________________________________________________________ -->
3922 <div class="doc_subsubsection">
3923 <a name="i_getresult">'<tt>getresult</tt>' Instruction</a>
3926 <div class="doc_text">
3930 <resultval> = getresult <type> <retval>, <index>
3935 <p> The '<tt>getresult</tt>' instruction is used to extract individual values
3936 from a '<tt><a href="#i_call">call</a></tt>'
3937 or '<tt><a href="#i_invoke">invoke</a></tt>' instruction that returns multiple
3942 <p>The '<tt>getresult</tt>' instruction takes a call or invoke value as its
3943 first argument, or an undef value. The value must have <a
3944 href="#t_struct">structure type</a>. The second argument is a constant
3945 unsigned index value which must be in range for the number of values returned
3950 <p>The '<tt>getresult</tt>' instruction extracts the element identified by
3951 '<tt>index</tt>' from the aggregate value.</p>
3956 %struct.A = type { i32, i8 }
3958 %r = call %struct.A @foo()
3959 %gr = getresult %struct.A %r, 0 <i>; yields i32:%gr</i>
3960 %gr1 = getresult %struct.A %r, 1 <i>; yields i8:%gr1</i>
3967 <!-- *********************************************************************** -->
3968 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
3969 <!-- *********************************************************************** -->
3971 <div class="doc_text">
3973 <p>LLVM supports the notion of an "intrinsic function". These functions have
3974 well known names and semantics and are required to follow certain restrictions.
3975 Overall, these intrinsics represent an extension mechanism for the LLVM
3976 language that does not require changing all of the transformations in LLVM when
3977 adding to the language (or the bitcode reader/writer, the parser, etc...).</p>
3979 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
3980 prefix is reserved in LLVM for intrinsic names; thus, function names may not
3981 begin with this prefix. Intrinsic functions must always be external functions:
3982 you cannot define the body of intrinsic functions. Intrinsic functions may
3983 only be used in call or invoke instructions: it is illegal to take the address
3984 of an intrinsic function. Additionally, because intrinsic functions are part
3985 of the LLVM language, it is required if any are added that they be documented
3988 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
3989 a family of functions that perform the same operation but on different data
3990 types. Because LLVM can represent over 8 million different integer types,
3991 overloading is used commonly to allow an intrinsic function to operate on any
3992 integer type. One or more of the argument types or the result type can be
3993 overloaded to accept any integer type. Argument types may also be defined as
3994 exactly matching a previous argument's type or the result type. This allows an
3995 intrinsic function which accepts multiple arguments, but needs all of them to
3996 be of the same type, to only be overloaded with respect to a single argument or
3999 <p>Overloaded intrinsics will have the names of its overloaded argument types
4000 encoded into its function name, each preceded by a period. Only those types
4001 which are overloaded result in a name suffix. Arguments whose type is matched
4002 against another type do not. For example, the <tt>llvm.ctpop</tt> function can
4003 take an integer of any width and returns an integer of exactly the same integer
4004 width. This leads to a family of functions such as
4005 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29 %val)</tt>.
4006 Only one type, the return type, is overloaded, and only one type suffix is
4007 required. Because the argument's type is matched against the return type, it
4008 does not require its own name suffix.</p>
4010 <p>To learn how to add an intrinsic function, please see the
4011 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.
4016 <!-- ======================================================================= -->
4017 <div class="doc_subsection">
4018 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
4021 <div class="doc_text">
4023 <p>Variable argument support is defined in LLVM with the <a
4024 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
4025 intrinsic functions. These functions are related to the similarly
4026 named macros defined in the <tt><stdarg.h></tt> header file.</p>
4028 <p>All of these functions operate on arguments that use a
4029 target-specific value type "<tt>va_list</tt>". The LLVM assembly
4030 language reference manual does not define what this type is, so all
4031 transformations should be prepared to handle these functions regardless of
4034 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
4035 instruction and the variable argument handling intrinsic functions are
4038 <div class="doc_code">
4040 define i32 @test(i32 %X, ...) {
4041 ; Initialize variable argument processing
4043 %ap2 = bitcast i8** %ap to i8*
4044 call void @llvm.va_start(i8* %ap2)
4046 ; Read a single integer argument
4047 %tmp = va_arg i8** %ap, i32
4049 ; Demonstrate usage of llvm.va_copy and llvm.va_end
4051 %aq2 = bitcast i8** %aq to i8*
4052 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
4053 call void @llvm.va_end(i8* %aq2)
4055 ; Stop processing of arguments.
4056 call void @llvm.va_end(i8* %ap2)
4060 declare void @llvm.va_start(i8*)
4061 declare void @llvm.va_copy(i8*, i8*)
4062 declare void @llvm.va_end(i8*)
4068 <!-- _______________________________________________________________________ -->
4069 <div class="doc_subsubsection">
4070 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
4074 <div class="doc_text">
4076 <pre> declare void %llvm.va_start(i8* <arglist>)<br></pre>
4078 <P>The '<tt>llvm.va_start</tt>' intrinsic initializes
4079 <tt>*<arglist></tt> for subsequent use by <tt><a
4080 href="#i_va_arg">va_arg</a></tt>.</p>
4084 <P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
4088 <P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
4089 macro available in C. In a target-dependent way, it initializes the
4090 <tt>va_list</tt> element to which the argument points, so that the next call to
4091 <tt>va_arg</tt> will produce the first variable argument passed to the function.
4092 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
4093 last argument of the function as the compiler can figure that out.</p>
4097 <!-- _______________________________________________________________________ -->
4098 <div class="doc_subsubsection">
4099 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
4102 <div class="doc_text">
4104 <pre> declare void @llvm.va_end(i8* <arglist>)<br></pre>
4107 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
4108 which has been initialized previously with <tt><a href="#int_va_start">llvm.va_start</a></tt>
4109 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
4113 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
4117 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
4118 macro available in C. In a target-dependent way, it destroys the
4119 <tt>va_list</tt> element to which the argument points. Calls to <a
4120 href="#int_va_start"><tt>llvm.va_start</tt></a> and <a href="#int_va_copy">
4121 <tt>llvm.va_copy</tt></a> must be matched exactly with calls to
4122 <tt>llvm.va_end</tt>.</p>
4126 <!-- _______________________________________________________________________ -->
4127 <div class="doc_subsubsection">
4128 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
4131 <div class="doc_text">
4136 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
4141 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
4142 from the source argument list to the destination argument list.</p>
4146 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
4147 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
4152 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
4153 macro available in C. In a target-dependent way, it copies the source
4154 <tt>va_list</tt> element into the destination <tt>va_list</tt> element. This
4155 intrinsic is necessary because the <tt><a href="#int_va_start">
4156 llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
4157 example, memory allocation.</p>
4161 <!-- ======================================================================= -->
4162 <div class="doc_subsection">
4163 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
4166 <div class="doc_text">
4169 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
4170 Collection</a> requires the implementation and generation of these intrinsics.
4171 These intrinsics allow identification of <a href="#int_gcroot">GC roots on the
4172 stack</a>, as well as garbage collector implementations that require <a
4173 href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a> barriers.
4174 Front-ends for type-safe garbage collected languages should generate these
4175 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
4176 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
4179 <p>The garbage collection intrinsics only operate on objects in the generic
4180 address space (address space zero).</p>
4184 <!-- _______________________________________________________________________ -->
4185 <div class="doc_subsubsection">
4186 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
4189 <div class="doc_text">
4194 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
4199 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
4200 the code generator, and allows some metadata to be associated with it.</p>
4204 <p>The first argument specifies the address of a stack object that contains the
4205 root pointer. The second pointer (which must be either a constant or a global
4206 value address) contains the meta-data to be associated with the root.</p>
4210 <p>At runtime, a call to this intrinsic stores a null pointer into the "ptrloc"
4211 location. At compile-time, the code generator generates information to allow
4212 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
4213 intrinsic may only be used in a function which <a href="#gc">specifies a GC
4219 <!-- _______________________________________________________________________ -->
4220 <div class="doc_subsubsection">
4221 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
4224 <div class="doc_text">
4229 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
4234 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
4235 locations, allowing garbage collector implementations that require read
4240 <p>The second argument is the address to read from, which should be an address
4241 allocated from the garbage collector. The first object is a pointer to the
4242 start of the referenced object, if needed by the language runtime (otherwise
4247 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
4248 instruction, but may be replaced with substantially more complex code by the
4249 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
4250 may only be used in a function which <a href="#gc">specifies a GC
4256 <!-- _______________________________________________________________________ -->
4257 <div class="doc_subsubsection">
4258 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
4261 <div class="doc_text">
4266 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
4271 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
4272 locations, allowing garbage collector implementations that require write
4273 barriers (such as generational or reference counting collectors).</p>
4277 <p>The first argument is the reference to store, the second is the start of the
4278 object to store it to, and the third is the address of the field of Obj to
4279 store to. If the runtime does not require a pointer to the object, Obj may be
4284 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
4285 instruction, but may be replaced with substantially more complex code by the
4286 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
4287 may only be used in a function which <a href="#gc">specifies a GC
4294 <!-- ======================================================================= -->
4295 <div class="doc_subsection">
4296 <a name="int_codegen">Code Generator Intrinsics</a>
4299 <div class="doc_text">
4301 These intrinsics are provided by LLVM to expose special features that may only
4302 be implemented with code generator support.
4307 <!-- _______________________________________________________________________ -->
4308 <div class="doc_subsubsection">
4309 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
4312 <div class="doc_text">
4316 declare i8 *@llvm.returnaddress(i32 <level>)
4322 The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
4323 target-specific value indicating the return address of the current function
4324 or one of its callers.
4330 The argument to this intrinsic indicates which function to return the address
4331 for. Zero indicates the calling function, one indicates its caller, etc. The
4332 argument is <b>required</b> to be a constant integer value.
4338 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
4339 the return address of the specified call frame, or zero if it cannot be
4340 identified. The value returned by this intrinsic is likely to be incorrect or 0
4341 for arguments other than zero, so it should only be used for debugging purposes.
4345 Note that calling this intrinsic does not prevent function inlining or other
4346 aggressive transformations, so the value returned may not be that of the obvious
4347 source-language caller.
4352 <!-- _______________________________________________________________________ -->
4353 <div class="doc_subsubsection">
4354 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
4357 <div class="doc_text">
4361 declare i8 *@llvm.frameaddress(i32 <level>)
4367 The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
4368 target-specific frame pointer value for the specified stack frame.
4374 The argument to this intrinsic indicates which function to return the frame
4375 pointer for. Zero indicates the calling function, one indicates its caller,
4376 etc. The argument is <b>required</b> to be a constant integer value.
4382 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
4383 the frame address of the specified call frame, or zero if it cannot be
4384 identified. The value returned by this intrinsic is likely to be incorrect or 0
4385 for arguments other than zero, so it should only be used for debugging purposes.
4389 Note that calling this intrinsic does not prevent function inlining or other
4390 aggressive transformations, so the value returned may not be that of the obvious
4391 source-language caller.
4395 <!-- _______________________________________________________________________ -->
4396 <div class="doc_subsubsection">
4397 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
4400 <div class="doc_text">
4404 declare i8 *@llvm.stacksave()
4410 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
4411 the function stack, for use with <a href="#int_stackrestore">
4412 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
4413 features like scoped automatic variable sized arrays in C99.
4419 This intrinsic returns a opaque pointer value that can be passed to <a
4420 href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
4421 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
4422 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
4423 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
4424 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
4425 that were allocated after the <tt>llvm.stacksave</tt> was executed.
4430 <!-- _______________________________________________________________________ -->
4431 <div class="doc_subsubsection">
4432 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
4435 <div class="doc_text">
4439 declare void @llvm.stackrestore(i8 * %ptr)
4445 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
4446 the function stack to the state it was in when the corresponding <a
4447 href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
4448 useful for implementing language features like scoped automatic variable sized
4455 See the description for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.
4461 <!-- _______________________________________________________________________ -->
4462 <div class="doc_subsubsection">
4463 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
4466 <div class="doc_text">
4470 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>)
4477 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
4478 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
4480 effect on the behavior of the program but can change its performance
4487 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
4488 determining if the fetch should be for a read (0) or write (1), and
4489 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
4490 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
4491 <tt>locality</tt> arguments must be constant integers.
4497 This intrinsic does not modify the behavior of the program. In particular,
4498 prefetches cannot trap and do not produce a value. On targets that support this
4499 intrinsic, the prefetch can provide hints to the processor cache for better
4505 <!-- _______________________________________________________________________ -->
4506 <div class="doc_subsubsection">
4507 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
4510 <div class="doc_text">
4514 declare void @llvm.pcmarker(i32 <id>)
4521 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
4523 code to simulators and other tools. The method is target specific, but it is
4524 expected that the marker will use exported symbols to transmit the PC of the marker.
4525 The marker makes no guarantees that it will remain with any specific instruction
4526 after optimizations. It is possible that the presence of a marker will inhibit
4527 optimizations. The intended use is to be inserted after optimizations to allow
4528 correlations of simulation runs.
4534 <tt>id</tt> is a numerical id identifying the marker.
4540 This intrinsic does not modify the behavior of the program. Backends that do not
4541 support this intrinisic may ignore it.
4546 <!-- _______________________________________________________________________ -->
4547 <div class="doc_subsubsection">
4548 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
4551 <div class="doc_text">
4555 declare i64 @llvm.readcyclecounter( )
4562 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
4563 counter register (or similar low latency, high accuracy clocks) on those targets
4564 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
4565 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
4566 should only be used for small timings.
4572 When directly supported, reading the cycle counter should not modify any memory.
4573 Implementations are allowed to either return a application specific value or a
4574 system wide value. On backends without support, this is lowered to a constant 0.
4579 <!-- ======================================================================= -->
4580 <div class="doc_subsection">
4581 <a name="int_libc">Standard C Library Intrinsics</a>
4584 <div class="doc_text">
4586 LLVM provides intrinsics for a few important standard C library functions.
4587 These intrinsics allow source-language front-ends to pass information about the
4588 alignment of the pointer arguments to the code generator, providing opportunity
4589 for more efficient code generation.
4594 <!-- _______________________________________________________________________ -->
4595 <div class="doc_subsubsection">
4596 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
4599 <div class="doc_text">
4603 declare void @llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
4604 i32 <len>, i32 <align>)
4605 declare void @llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
4606 i64 <len>, i32 <align>)
4612 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4613 location to the destination location.
4617 Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
4618 intrinsics do not return a value, and takes an extra alignment argument.
4624 The first argument is a pointer to the destination, the second is a pointer to
4625 the source. The third argument is an integer argument
4626 specifying the number of bytes to copy, and the fourth argument is the alignment
4627 of the source and destination locations.
4631 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4632 the caller guarantees that both the source and destination pointers are aligned
4639 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4640 location to the destination location, which are not allowed to overlap. It
4641 copies "len" bytes of memory over. If the argument is known to be aligned to
4642 some boundary, this can be specified as the fourth argument, otherwise it should
4648 <!-- _______________________________________________________________________ -->
4649 <div class="doc_subsubsection">
4650 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
4653 <div class="doc_text">
4657 declare void @llvm.memmove.i32(i8 * <dest>, i8 * <src>,
4658 i32 <len>, i32 <align>)
4659 declare void @llvm.memmove.i64(i8 * <dest>, i8 * <src>,
4660 i64 <len>, i32 <align>)
4666 The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
4667 location to the destination location. It is similar to the
4668 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to overlap.
4672 Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
4673 intrinsics do not return a value, and takes an extra alignment argument.
4679 The first argument is a pointer to the destination, the second is a pointer to
4680 the source. The third argument is an integer argument
4681 specifying the number of bytes to copy, and the fourth argument is the alignment
4682 of the source and destination locations.
4686 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4687 the caller guarantees that the source and destination pointers are aligned to
4694 The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
4695 location to the destination location, which may overlap. It
4696 copies "len" bytes of memory over. If the argument is known to be aligned to
4697 some boundary, this can be specified as the fourth argument, otherwise it should
4703 <!-- _______________________________________________________________________ -->
4704 <div class="doc_subsubsection">
4705 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
4708 <div class="doc_text">
4712 declare void @llvm.memset.i32(i8 * <dest>, i8 <val>,
4713 i32 <len>, i32 <align>)
4714 declare void @llvm.memset.i64(i8 * <dest>, i8 <val>,
4715 i64 <len>, i32 <align>)
4721 The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
4726 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
4727 does not return a value, and takes an extra alignment argument.
4733 The first argument is a pointer to the destination to fill, the second is the
4734 byte value to fill it with, the third argument is an integer
4735 argument specifying the number of bytes to fill, and the fourth argument is the
4736 known alignment of destination location.
4740 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4741 the caller guarantees that the destination pointer is aligned to that boundary.
4747 The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
4749 destination location. If the argument is known to be aligned to some boundary,
4750 this can be specified as the fourth argument, otherwise it should be set to 0 or
4756 <!-- _______________________________________________________________________ -->
4757 <div class="doc_subsubsection">
4758 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
4761 <div class="doc_text">
4764 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
4765 floating point or vector of floating point type. Not all targets support all
4768 declare float @llvm.sqrt.f32(float %Val)
4769 declare double @llvm.sqrt.f64(double %Val)
4770 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
4771 declare fp128 @llvm.sqrt.f128(fp128 %Val)
4772 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
4778 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
4779 returning the same value as the libm '<tt>sqrt</tt>' functions would. Unlike
4780 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
4781 negative numbers other than -0.0 (which allows for better optimization, because
4782 there is no need to worry about errno being set). <tt>llvm.sqrt(-0.0)</tt> is
4783 defined to return -0.0 like IEEE sqrt.
4789 The argument and return value are floating point numbers of the same type.
4795 This function returns the sqrt of the specified operand if it is a nonnegative
4796 floating point number.
4800 <!-- _______________________________________________________________________ -->
4801 <div class="doc_subsubsection">
4802 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
4805 <div class="doc_text">
4808 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
4809 floating point or vector of floating point type. Not all targets support all
4812 declare float @llvm.powi.f32(float %Val, i32 %power)
4813 declare double @llvm.powi.f64(double %Val, i32 %power)
4814 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
4815 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
4816 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
4822 The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
4823 specified (positive or negative) power. The order of evaluation of
4824 multiplications is not defined. When a vector of floating point type is
4825 used, the second argument remains a scalar integer value.
4831 The second argument is an integer power, and the first is a value to raise to
4838 This function returns the first value raised to the second power with an
4839 unspecified sequence of rounding operations.</p>
4842 <!-- _______________________________________________________________________ -->
4843 <div class="doc_subsubsection">
4844 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
4847 <div class="doc_text">
4850 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
4851 floating point or vector of floating point type. Not all targets support all
4854 declare float @llvm.sin.f32(float %Val)
4855 declare double @llvm.sin.f64(double %Val)
4856 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
4857 declare fp128 @llvm.sin.f128(fp128 %Val)
4858 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
4864 The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.
4870 The argument and return value are floating point numbers of the same type.
4876 This function returns the sine of the specified operand, returning the
4877 same values as the libm <tt>sin</tt> functions would, and handles error
4878 conditions in the same way.</p>
4881 <!-- _______________________________________________________________________ -->
4882 <div class="doc_subsubsection">
4883 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
4886 <div class="doc_text">
4889 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
4890 floating point or vector of floating point type. Not all targets support all
4893 declare float @llvm.cos.f32(float %Val)
4894 declare double @llvm.cos.f64(double %Val)
4895 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
4896 declare fp128 @llvm.cos.f128(fp128 %Val)
4897 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
4903 The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.
4909 The argument and return value are floating point numbers of the same type.
4915 This function returns the cosine of the specified operand, returning the
4916 same values as the libm <tt>cos</tt> functions would, and handles error
4917 conditions in the same way.</p>
4920 <!-- _______________________________________________________________________ -->
4921 <div class="doc_subsubsection">
4922 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
4925 <div class="doc_text">
4928 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
4929 floating point or vector of floating point type. Not all targets support all
4932 declare float @llvm.pow.f32(float %Val, float %Power)
4933 declare double @llvm.pow.f64(double %Val, double %Power)
4934 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
4935 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
4936 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
4942 The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
4943 specified (positive or negative) power.
4949 The second argument is a floating point power, and the first is a value to
4950 raise to that power.
4956 This function returns the first value raised to the second power,
4958 same values as the libm <tt>pow</tt> functions would, and handles error
4959 conditions in the same way.</p>
4963 <!-- ======================================================================= -->
4964 <div class="doc_subsection">
4965 <a name="int_manip">Bit Manipulation Intrinsics</a>
4968 <div class="doc_text">
4970 LLVM provides intrinsics for a few important bit manipulation operations.
4971 These allow efficient code generation for some algorithms.
4976 <!-- _______________________________________________________________________ -->
4977 <div class="doc_subsubsection">
4978 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
4981 <div class="doc_text">
4984 <p>This is an overloaded intrinsic function. You can use bswap on any integer
4985 type that is an even number of bytes (i.e. BitWidth % 16 == 0).
4987 declare i16 @llvm.bswap.i16(i16 <id>)
4988 declare i32 @llvm.bswap.i32(i32 <id>)
4989 declare i64 @llvm.bswap.i64(i64 <id>)
4995 The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
4996 values with an even number of bytes (positive multiple of 16 bits). These are
4997 useful for performing operations on data that is not in the target's native
5004 The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
5005 and low byte of the input i16 swapped. Similarly, the <tt>llvm.bswap.i32</tt>
5006 intrinsic returns an i32 value that has the four bytes of the input i32
5007 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the returned
5008 i32 will have its bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i48</tt>,
5009 <tt>llvm.bswap.i64</tt> and other intrinsics extend this concept to
5010 additional even-byte lengths (6 bytes, 8 bytes and more, respectively).
5015 <!-- _______________________________________________________________________ -->
5016 <div class="doc_subsubsection">
5017 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
5020 <div class="doc_text">
5023 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
5024 width. Not all targets support all bit widths however.
5026 declare i8 @llvm.ctpop.i8 (i8 <src>)
5027 declare i16 @llvm.ctpop.i16(i16 <src>)
5028 declare i32 @llvm.ctpop.i32(i32 <src>)
5029 declare i64 @llvm.ctpop.i64(i64 <src>)
5030 declare i256 @llvm.ctpop.i256(i256 <src>)
5036 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
5043 The only argument is the value to be counted. The argument may be of any
5044 integer type. The return type must match the argument type.
5050 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
5054 <!-- _______________________________________________________________________ -->
5055 <div class="doc_subsubsection">
5056 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
5059 <div class="doc_text">
5062 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
5063 integer bit width. Not all targets support all bit widths however.
5065 declare i8 @llvm.ctlz.i8 (i8 <src>)
5066 declare i16 @llvm.ctlz.i16(i16 <src>)
5067 declare i32 @llvm.ctlz.i32(i32 <src>)
5068 declare i64 @llvm.ctlz.i64(i64 <src>)
5069 declare i256 @llvm.ctlz.i256(i256 <src>)
5075 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
5076 leading zeros in a variable.
5082 The only argument is the value to be counted. The argument may be of any
5083 integer type. The return type must match the argument type.
5089 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
5090 in a variable. If the src == 0 then the result is the size in bits of the type
5091 of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.
5097 <!-- _______________________________________________________________________ -->
5098 <div class="doc_subsubsection">
5099 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
5102 <div class="doc_text">
5105 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
5106 integer bit width. Not all targets support all bit widths however.
5108 declare i8 @llvm.cttz.i8 (i8 <src>)
5109 declare i16 @llvm.cttz.i16(i16 <src>)
5110 declare i32 @llvm.cttz.i32(i32 <src>)
5111 declare i64 @llvm.cttz.i64(i64 <src>)
5112 declare i256 @llvm.cttz.i256(i256 <src>)
5118 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
5125 The only argument is the value to be counted. The argument may be of any
5126 integer type. The return type must match the argument type.
5132 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
5133 in a variable. If the src == 0 then the result is the size in bits of the type
5134 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
5138 <!-- _______________________________________________________________________ -->
5139 <div class="doc_subsubsection">
5140 <a name="int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic</a>
5143 <div class="doc_text">
5146 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.select</tt>
5147 on any integer bit width.
5149 declare i17 @llvm.part.select.i17 (i17 %val, i32 %loBit, i32 %hiBit)
5150 declare i29 @llvm.part.select.i29 (i29 %val, i32 %loBit, i32 %hiBit)
5154 <p>The '<tt>llvm.part.select</tt>' family of intrinsic functions selects a
5155 range of bits from an integer value and returns them in the same bit width as
5156 the original value.</p>
5159 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5160 any bit width but they must have the same bit width. The second and third
5161 arguments must be <tt>i32</tt> type since they specify only a bit index.</p>
5164 <p>The operation of the '<tt>llvm.part.select</tt>' intrinsic has two modes
5165 of operation: forwards and reverse. If <tt>%loBit</tt> is greater than
5166 <tt>%hiBits</tt> then the intrinsic operates in reverse mode. Otherwise it
5167 operates in forward mode.</p>
5168 <p>In forward mode, this intrinsic is the equivalent of shifting <tt>%val</tt>
5169 right by <tt>%loBit</tt> bits and then ANDing it with a mask with
5170 only the <tt>%hiBit - %loBit</tt> bits set, as follows:</p>
5172 <li>The <tt>%val</tt> is shifted right (LSHR) by the number of bits specified
5173 by <tt>%loBits</tt>. This normalizes the value to the low order bits.</li>
5174 <li>The <tt>%loBits</tt> value is subtracted from the <tt>%hiBits</tt> value
5175 to determine the number of bits to retain.</li>
5176 <li>A mask of the retained bits is created by shifting a -1 value.</li>
5177 <li>The mask is ANDed with <tt>%val</tt> to produce the result.
5179 <p>In reverse mode, a similar computation is made except that the bits are
5180 returned in the reverse order. So, for example, if <tt>X</tt> has the value
5181 <tt>i16 0x0ACF (101011001111)</tt> and we apply
5182 <tt>part.select(i16 X, 8, 3)</tt> to it, we get back the value
5183 <tt>i16 0x0026 (000000100110)</tt>.</p>
5186 <div class="doc_subsubsection">
5187 <a name="int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic</a>
5190 <div class="doc_text">
5193 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.set</tt>
5194 on any integer bit width.
5196 declare i17 @llvm.part.set.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
5197 declare i29 @llvm.part.set.i29.i9 (i29 %val, i9 %repl, i32 %lo, i32 %hi)
5201 <p>The '<tt>llvm.part.set</tt>' family of intrinsic functions replaces a range
5202 of bits in an integer value with another integer value. It returns the integer
5203 with the replaced bits.</p>
5206 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5207 any bit width but they must have the same bit width. <tt>%val</tt> is the value
5208 whose bits will be replaced. The second argument, <tt>%repl</tt> may be an
5209 integer of any bit width. The third and fourth arguments must be <tt>i32</tt>
5210 type since they specify only a bit index.</p>
5213 <p>The operation of the '<tt>llvm.part.set</tt>' intrinsic has two modes
5214 of operation: forwards and reverse. If <tt>%lo</tt> is greater than
5215 <tt>%hi</tt> then the intrinsic operates in reverse mode. Otherwise it
5216 operates in forward mode.</p>
5217 <p>For both modes, the <tt>%repl</tt> value is prepared for use by either
5218 truncating it down to the size of the replacement area or zero extending it
5219 up to that size.</p>
5220 <p>In forward mode, the bits between <tt>%lo</tt> and <tt>%hi</tt> (inclusive)
5221 are replaced with corresponding bits from <tt>%repl</tt>. That is the 0th bit
5222 in <tt>%repl</tt> replaces the <tt>%lo</tt>th bit in <tt>%val</tt> and etc. up
5223 to the <tt>%hi</tt>th bit.
5224 <p>In reverse mode, a similar computation is made except that the bits are
5225 reversed. That is, the <tt>0</tt>th bit in <tt>%repl</tt> replaces the
5226 <tt>%hi</tt> bit in <tt>%val</tt> and etc. down to the <tt>%lo</tt>th bit.
5229 llvm.part.set(0xFFFF, 0, 4, 7) -> 0xFF0F
5230 llvm.part.set(0xFFFF, 0, 7, 4) -> 0xFF0F
5231 llvm.part.set(0xFFFF, 1, 7, 4) -> 0xFF8F
5232 llvm.part.set(0xFFFF, F, 8, 3) -> 0xFFE7
5233 llvm.part.set(0xFFFF, 0, 3, 8) -> 0xFE07
5237 <!-- ======================================================================= -->
5238 <div class="doc_subsection">
5239 <a name="int_debugger">Debugger Intrinsics</a>
5242 <div class="doc_text">
5244 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
5245 are described in the <a
5246 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
5247 Debugging</a> document.
5252 <!-- ======================================================================= -->
5253 <div class="doc_subsection">
5254 <a name="int_eh">Exception Handling Intrinsics</a>
5257 <div class="doc_text">
5258 <p> The LLVM exception handling intrinsics (which all start with
5259 <tt>llvm.eh.</tt> prefix), are described in the <a
5260 href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
5261 Handling</a> document. </p>
5264 <!-- ======================================================================= -->
5265 <div class="doc_subsection">
5266 <a name="int_trampoline">Trampoline Intrinsic</a>
5269 <div class="doc_text">
5271 This intrinsic makes it possible to excise one parameter, marked with
5272 the <tt>nest</tt> attribute, from a function. The result is a callable
5273 function pointer lacking the nest parameter - the caller does not need
5274 to provide a value for it. Instead, the value to use is stored in
5275 advance in a "trampoline", a block of memory usually allocated
5276 on the stack, which also contains code to splice the nest value into the
5277 argument list. This is used to implement the GCC nested function address
5281 For example, if the function is
5282 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
5283 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as follows:</p>
5285 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
5286 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
5287 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
5288 %fp = bitcast i8* %p to i32 (i32, i32)*
5290 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
5291 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
5294 <!-- _______________________________________________________________________ -->
5295 <div class="doc_subsubsection">
5296 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
5298 <div class="doc_text">
5301 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
5305 This fills the memory pointed to by <tt>tramp</tt> with code
5306 and returns a function pointer suitable for executing it.
5310 The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
5311 pointers. The <tt>tramp</tt> argument must point to a sufficiently large
5312 and sufficiently aligned block of memory; this memory is written to by the
5313 intrinsic. Note that the size and the alignment are target-specific - LLVM
5314 currently provides no portable way of determining them, so a front-end that
5315 generates this intrinsic needs to have some target-specific knowledge.
5316 The <tt>func</tt> argument must hold a function bitcast to an <tt>i8*</tt>.
5320 The block of memory pointed to by <tt>tramp</tt> is filled with target
5321 dependent code, turning it into a function. A pointer to this function is
5322 returned, but needs to be bitcast to an
5323 <a href="#int_trampoline">appropriate function pointer type</a>
5324 before being called. The new function's signature is the same as that of
5325 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
5326 removed. At most one such <tt>nest</tt> argument is allowed, and it must be
5327 of pointer type. Calling the new function is equivalent to calling
5328 <tt>func</tt> with the same argument list, but with <tt>nval</tt> used for the
5329 missing <tt>nest</tt> argument. If, after calling
5330 <tt>llvm.init.trampoline</tt>, the memory pointed to by <tt>tramp</tt> is
5331 modified, then the effect of any later call to the returned function pointer is
5336 <!-- ======================================================================= -->
5337 <div class="doc_subsection">
5338 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
5341 <div class="doc_text">
5343 These intrinsic functions expand the "universal IR" of LLVM to represent
5344 hardware constructs for atomic operations and memory synchronization. This
5345 provides an interface to the hardware, not an interface to the programmer. It
5346 is aimed at a low enough level to allow any programming models or APIs which
5347 need atomic behaviors to map cleanly onto it. It is also modeled primarily on
5348 hardware behavior. Just as hardware provides a "universal IR" for source
5349 languages, it also provides a starting point for developing a "universal"
5350 atomic operation and synchronization IR.
5353 These do <em>not</em> form an API such as high-level threading libraries,
5354 software transaction memory systems, atomic primitives, and intrinsic
5355 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
5356 application libraries. The hardware interface provided by LLVM should allow
5357 a clean implementation of all of these APIs and parallel programming models.
5358 No one model or paradigm should be selected above others unless the hardware
5359 itself ubiquitously does so.
5364 <!-- _______________________________________________________________________ -->
5365 <div class="doc_subsubsection">
5366 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
5368 <div class="doc_text">
5371 declare void @llvm.memory.barrier( i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>,
5377 The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
5378 specific pairs of memory access types.
5382 The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
5383 The first four arguments enables a specific barrier as listed below. The fith
5384 argument specifies that the barrier applies to io or device or uncached memory.
5388 <li><tt>ll</tt>: load-load barrier</li>
5389 <li><tt>ls</tt>: load-store barrier</li>
5390 <li><tt>sl</tt>: store-load barrier</li>
5391 <li><tt>ss</tt>: store-store barrier</li>
5392 <li><tt>device</tt>: barrier applies to device and uncached memory also.
5396 This intrinsic causes the system to enforce some ordering constraints upon
5397 the loads and stores of the program. This barrier does not indicate
5398 <em>when</em> any events will occur, it only enforces an <em>order</em> in
5399 which they occur. For any of the specified pairs of load and store operations
5400 (f.ex. load-load, or store-load), all of the first operations preceding the
5401 barrier will complete before any of the second operations succeeding the
5402 barrier begin. Specifically the semantics for each pairing is as follows:
5405 <li><tt>ll</tt>: All loads before the barrier must complete before any load
5406 after the barrier begins.</li>
5408 <li><tt>ls</tt>: All loads before the barrier must complete before any
5409 store after the barrier begins.</li>
5410 <li><tt>ss</tt>: All stores before the barrier must complete before any
5411 store after the barrier begins.</li>
5412 <li><tt>sl</tt>: All stores before the barrier must complete before any
5413 load after the barrier begins.</li>
5416 These semantics are applied with a logical "and" behavior when more than one
5417 is enabled in a single memory barrier intrinsic.
5420 Backends may implement stronger barriers than those requested when they do not
5421 support as fine grained a barrier as requested. Some architectures do not
5422 need all types of barriers and on such architectures, these become noops.
5429 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
5430 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
5431 <i>; guarantee the above finishes</i>
5432 store i32 8, %ptr <i>; before this begins</i>
5436 <!-- _______________________________________________________________________ -->
5437 <div class="doc_subsubsection">
5438 <a name="int_atomic_lcs">'<tt>llvm.atomic.lcs.*</tt>' Intrinsic</a>
5440 <div class="doc_text">
5443 This is an overloaded intrinsic. You can use <tt>llvm.atomic.lcs</tt> on any
5444 integer bit width. Not all targets support all bit widths however.</p>
5447 declare i8 @llvm.atomic.lcs.i8( i8* <ptr>, i8 <cmp>, i8 <val> )
5448 declare i16 @llvm.atomic.lcs.i16( i16* <ptr>, i16 <cmp>, i16 <val> )
5449 declare i32 @llvm.atomic.lcs.i32( i32* <ptr>, i32 <cmp>, i32 <val> )
5450 declare i64 @llvm.atomic.lcs.i64( i64* <ptr>, i64 <cmp>, i64 <val> )
5455 This loads a value in memory and compares it to a given value. If they are
5456 equal, it stores a new value into the memory.
5460 The <tt>llvm.atomic.lcs</tt> intrinsic takes three arguments. The result as
5461 well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
5462 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
5463 this integer type. While any bit width integer may be used, targets may only
5464 lower representations they support in hardware.
5469 This entire intrinsic must be executed atomically. It first loads the value
5470 in memory pointed to by <tt>ptr</tt> and compares it with the value
5471 <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the memory. The
5472 loaded value is yielded in all cases. This provides the equivalent of an
5473 atomic compare-and-swap operation within the SSA framework.
5481 %val1 = add i32 4, 4
5482 %result1 = call i32 @llvm.atomic.lcs.i32( i32* %ptr, i32 4, %val1 )
5483 <i>; yields {i32}:result1 = 4</i>
5484 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
5485 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
5487 %val2 = add i32 1, 1
5488 %result2 = call i32 @llvm.atomic.lcs.i32( i32* %ptr, i32 5, %val2 )
5489 <i>; yields {i32}:result2 = 8</i>
5490 %stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
5492 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
5496 <!-- _______________________________________________________________________ -->
5497 <div class="doc_subsubsection">
5498 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
5500 <div class="doc_text">
5504 This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
5505 integer bit width. Not all targets support all bit widths however.</p>
5507 declare i8 @llvm.atomic.swap.i8( i8* <ptr>, i8 <val> )
5508 declare i16 @llvm.atomic.swap.i16( i16* <ptr>, i16 <val> )
5509 declare i32 @llvm.atomic.swap.i32( i32* <ptr>, i32 <val> )
5510 declare i64 @llvm.atomic.swap.i64( i64* <ptr>, i64 <val> )
5515 This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
5516 the value from memory. It then stores the value in <tt>val</tt> in the memory
5522 The <tt>llvm.atomic.ls</tt> intrinsic takes two arguments. Both the
5523 <tt>val</tt> argument and the result must be integers of the same bit width.
5524 The first argument, <tt>ptr</tt>, must be a pointer to a value of this
5525 integer type. The targets may only lower integer representations they
5530 This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
5531 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
5532 equivalent of an atomic swap operation within the SSA framework.
5540 %val1 = add i32 4, 4
5541 %result1 = call i32 @llvm.atomic.swap.i32( i32* %ptr, i32 %val1 )
5542 <i>; yields {i32}:result1 = 4</i>
5543 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
5544 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
5546 %val2 = add i32 1, 1
5547 %result2 = call i32 @llvm.atomic.swap.i32( i32* %ptr, i32 %val2 )
5548 <i>; yields {i32}:result2 = 8</i>
5550 %stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
5551 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
5555 <!-- _______________________________________________________________________ -->
5556 <div class="doc_subsubsection">
5557 <a name="int_atomic_las">'<tt>llvm.atomic.las.*</tt>' Intrinsic</a>
5560 <div class="doc_text">
5563 This is an overloaded intrinsic. You can use <tt>llvm.atomic.las</tt> on any
5564 integer bit width. Not all targets support all bit widths however.</p>
5566 declare i8 @llvm.atomic.las.i8.( i8* <ptr>, i8 <delta> )
5567 declare i16 @llvm.atomic.las.i16.( i16* <ptr>, i16 <delta> )
5568 declare i32 @llvm.atomic.las.i32.( i32* <ptr>, i32 <delta> )
5569 declare i64 @llvm.atomic.las.i64.( i64* <ptr>, i64 <delta> )
5574 This intrinsic adds <tt>delta</tt> to the value stored in memory at
5575 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
5580 The intrinsic takes two arguments, the first a pointer to an integer value
5581 and the second an integer value. The result is also an integer value. These
5582 integer types can have any bit width, but they must all have the same bit
5583 width. The targets may only lower integer representations they support.
5587 This intrinsic does a series of operations atomically. It first loads the
5588 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
5589 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
5596 %result1 = call i32 @llvm.atomic.las.i32( i32* %ptr, i32 4 )
5597 <i>; yields {i32}:result1 = 4</i>
5598 %result2 = call i32 @llvm.atomic.las.i32( i32* %ptr, i32 2 )
5599 <i>; yields {i32}:result2 = 8</i>
5600 %result3 = call i32 @llvm.atomic.las.i32( i32* %ptr, i32 5 )
5601 <i>; yields {i32}:result3 = 10</i>
5602 %memval = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
5607 <!-- ======================================================================= -->
5608 <div class="doc_subsection">
5609 <a name="int_general">General Intrinsics</a>
5612 <div class="doc_text">
5613 <p> This class of intrinsics is designed to be generic and has
5614 no specific purpose. </p>
5617 <!-- _______________________________________________________________________ -->
5618 <div class="doc_subsubsection">
5619 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
5622 <div class="doc_text">
5626 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
5632 The '<tt>llvm.var.annotation</tt>' intrinsic
5638 The first argument is a pointer to a value, the second is a pointer to a
5639 global string, the third is a pointer to a global string which is the source
5640 file name, and the last argument is the line number.
5646 This intrinsic allows annotation of local variables with arbitrary strings.
5647 This can be useful for special purpose optimizations that want to look for these
5648 annotations. These have no other defined use, they are ignored by code
5649 generation and optimization.
5653 <!-- _______________________________________________________________________ -->
5654 <div class="doc_subsubsection">
5655 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
5658 <div class="doc_text">
5661 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
5662 any integer bit width.
5665 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
5666 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
5667 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
5668 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
5669 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
5675 The '<tt>llvm.annotation</tt>' intrinsic.
5681 The first argument is an integer value (result of some expression),
5682 the second is a pointer to a global string, the third is a pointer to a global
5683 string which is the source file name, and the last argument is the line number.
5684 It returns the value of the first argument.
5690 This intrinsic allows annotations to be put on arbitrary expressions
5691 with arbitrary strings. This can be useful for special purpose optimizations
5692 that want to look for these annotations. These have no other defined use, they
5693 are ignored by code generation and optimization.
5696 <!-- _______________________________________________________________________ -->
5697 <div class="doc_subsubsection">
5698 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
5701 <div class="doc_text">
5705 declare void @llvm.trap()
5711 The '<tt>llvm.trap</tt>' intrinsic
5723 This intrinsics is lowered to the target dependent trap instruction. If the
5724 target does not have a trap instruction, this intrinsic will be lowered to the
5725 call of the abort() function.
5729 <!-- *********************************************************************** -->
5732 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
5733 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
5734 <a href="http://validator.w3.org/check/referer"><img
5735 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
5737 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
5738 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
5739 Last modified: $Date$