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="#moduleasm">Module-Level Inline Assembly</a></li>
30 <li><a href="#datalayout">Data Layout</a></li>
33 <li><a href="#typesystem">Type System</a>
35 <li><a href="#t_primitive">Primitive Types</a>
37 <li><a href="#t_classifications">Type Classifications</a></li>
40 <li><a href="#t_derived">Derived Types</a>
42 <li><a href="#t_array">Array Type</a></li>
43 <li><a href="#t_function">Function Type</a></li>
44 <li><a href="#t_pointer">Pointer Type</a></li>
45 <li><a href="#t_struct">Structure Type</a></li>
46 <li><a href="#t_pstruct">Packed Structure Type</a></li>
47 <li><a href="#t_vector">Vector Type</a></li>
48 <li><a href="#t_opaque">Opaque Type</a></li>
53 <li><a href="#constants">Constants</a>
55 <li><a href="#simpleconstants">Simple Constants</a>
56 <li><a href="#aggregateconstants">Aggregate Constants</a>
57 <li><a href="#globalconstants">Global Variable and Function Addresses</a>
58 <li><a href="#undefvalues">Undefined Values</a>
59 <li><a href="#constantexprs">Constant Expressions</a>
62 <li><a href="#othervalues">Other Values</a>
64 <li><a href="#inlineasm">Inline Assembler Expressions</a>
67 <li><a href="#instref">Instruction Reference</a>
69 <li><a href="#terminators">Terminator Instructions</a>
71 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
72 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
73 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
74 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
75 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
76 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
79 <li><a href="#binaryops">Binary Operations</a>
81 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
82 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
83 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
84 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
85 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
86 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
87 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
88 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
89 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
92 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
94 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
95 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
96 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
97 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
98 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
99 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
102 <li><a href="#vectorops">Vector Operations</a>
104 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
105 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
106 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
109 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
111 <li><a href="#i_malloc">'<tt>malloc</tt>' Instruction</a></li>
112 <li><a href="#i_free">'<tt>free</tt>' Instruction</a></li>
113 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
114 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
115 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
116 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
119 <li><a href="#convertops">Conversion Operations</a>
121 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
122 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
123 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
124 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
125 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
126 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
127 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
128 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
129 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
130 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
131 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
132 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
134 <li><a href="#otherops">Other Operations</a>
136 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
137 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
138 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
139 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
140 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
141 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
146 <li><a href="#intrinsics">Intrinsic Functions</a>
148 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
150 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
151 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
152 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
155 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
157 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
158 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
159 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
162 <li><a href="#int_codegen">Code Generator Intrinsics</a>
164 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
165 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
166 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
167 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
168 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
169 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
170 <li><a href="#int_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
173 <li><a href="#int_libc">Standard C Library Intrinsics</a>
175 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
176 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
177 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
178 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
179 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
180 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
181 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
182 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
185 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
187 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
188 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
189 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
190 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
191 <li><a href="#int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic </a></li>
192 <li><a href="#int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic </a></li>
195 <li><a href="#int_debugger">Debugger intrinsics</a></li>
196 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
197 <li><a href="#int_trampoline">Trampoline Intrinsic</a>
199 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
202 <li><a href="#int_general">General intrinsics</a>
204 <li><a href="#int_var_annotation">
205 <tt>llvm.var.annotation</tt>' Intrinsic</a></li>
208 <li><a href="#int_annotation">
209 <tt>llvm.annotation.*</tt>' Intrinsic</a></li>
216 <div class="doc_author">
217 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
218 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
221 <!-- *********************************************************************** -->
222 <div class="doc_section"> <a name="abstract">Abstract </a></div>
223 <!-- *********************************************************************** -->
225 <div class="doc_text">
226 <p>This document is a reference manual for the LLVM assembly language.
227 LLVM is an SSA based representation that provides type safety,
228 low-level operations, flexibility, and the capability of representing
229 'all' high-level languages cleanly. It is the common code
230 representation used throughout all phases of the LLVM compilation
234 <!-- *********************************************************************** -->
235 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
236 <!-- *********************************************************************** -->
238 <div class="doc_text">
240 <p>The LLVM code representation is designed to be used in three
241 different forms: as an in-memory compiler IR, as an on-disk bitcode
242 representation (suitable for fast loading by a Just-In-Time compiler),
243 and as a human readable assembly language representation. This allows
244 LLVM to provide a powerful intermediate representation for efficient
245 compiler transformations and analysis, while providing a natural means
246 to debug and visualize the transformations. The three different forms
247 of LLVM are all equivalent. This document describes the human readable
248 representation and notation.</p>
250 <p>The LLVM representation aims to be light-weight and low-level
251 while being expressive, typed, and extensible at the same time. It
252 aims to be a "universal IR" of sorts, by being at a low enough level
253 that high-level ideas may be cleanly mapped to it (similar to how
254 microprocessors are "universal IR's", allowing many source languages to
255 be mapped to them). By providing type information, LLVM can be used as
256 the target of optimizations: for example, through pointer analysis, it
257 can be proven that a C automatic variable is never accessed outside of
258 the current function... allowing it to be promoted to a simple SSA
259 value instead of a memory location.</p>
263 <!-- _______________________________________________________________________ -->
264 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
266 <div class="doc_text">
268 <p>It is important to note that this document describes 'well formed'
269 LLVM assembly language. There is a difference between what the parser
270 accepts and what is considered 'well formed'. For example, the
271 following instruction is syntactically okay, but not well formed:</p>
273 <div class="doc_code">
275 %x = <a href="#i_add">add</a> i32 1, %x
279 <p>...because the definition of <tt>%x</tt> does not dominate all of
280 its uses. The LLVM infrastructure provides a verification pass that may
281 be used to verify that an LLVM module is well formed. This pass is
282 automatically run by the parser after parsing input assembly and by
283 the optimizer before it outputs bitcode. The violations pointed out
284 by the verifier pass indicate bugs in transformation passes or input to
288 <!-- Describe the typesetting conventions here. -->
290 <!-- *********************************************************************** -->
291 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
292 <!-- *********************************************************************** -->
294 <div class="doc_text">
296 <p>LLVM identifiers come in two basic types: global and local. Global
297 identifiers (functions, global variables) begin with the @ character. Local
298 identifiers (register names, types) begin with the % character. Additionally,
299 there are three different formats for identifiers, for different purposes:
302 <li>Named values are represented as a string of characters with their prefix.
303 For example, %foo, @DivisionByZero, %a.really.long.identifier. The actual
304 regular expression used is '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
305 Identifiers which require other characters in their names can be surrounded
306 with quotes. In this way, anything except a <tt>"</tt> character can
307 be used in a named value.</li>
309 <li>Unnamed values are represented as an unsigned numeric value with their
310 prefix. For example, %12, @2, %44.</li>
312 <li>Constants, which are described in a <a href="#constants">section about
313 constants</a>, below.</li>
316 <p>LLVM requires that values start with a prefix for two reasons: Compilers
317 don't need to worry about name clashes with reserved words, and the set of
318 reserved words may be expanded in the future without penalty. Additionally,
319 unnamed identifiers allow a compiler to quickly come up with a temporary
320 variable without having to avoid symbol table conflicts.</p>
322 <p>Reserved words in LLVM are very similar to reserved words in other
323 languages. There are keywords for different opcodes
324 ('<tt><a href="#i_add">add</a></tt>',
325 '<tt><a href="#i_bitcast">bitcast</a></tt>',
326 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
327 href="#t_void">void</a></tt>', '<tt><a href="#t_primitive">i32</a></tt>', etc...),
328 and others. These reserved words cannot conflict with variable names, because
329 none of them start with a prefix character ('%' or '@').</p>
331 <p>Here is an example of LLVM code to multiply the integer variable
332 '<tt>%X</tt>' by 8:</p>
336 <div class="doc_code">
338 %result = <a href="#i_mul">mul</a> i32 %X, 8
342 <p>After strength reduction:</p>
344 <div class="doc_code">
346 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
350 <p>And the hard way:</p>
352 <div class="doc_code">
354 <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
355 <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
356 %result = <a href="#i_add">add</a> i32 %1, %1
360 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
361 important lexical features of LLVM:</p>
365 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
368 <li>Unnamed temporaries are created when the result of a computation is not
369 assigned to a named value.</li>
371 <li>Unnamed temporaries are numbered sequentially</li>
375 <p>...and it also shows a convention that we follow in this document. When
376 demonstrating instructions, we will follow an instruction with a comment that
377 defines the type and name of value produced. Comments are shown in italic
382 <!-- *********************************************************************** -->
383 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
384 <!-- *********************************************************************** -->
386 <!-- ======================================================================= -->
387 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
390 <div class="doc_text">
392 <p>LLVM programs are composed of "Module"s, each of which is a
393 translation unit of the input programs. Each module consists of
394 functions, global variables, and symbol table entries. Modules may be
395 combined together with the LLVM linker, which merges function (and
396 global variable) definitions, resolves forward declarations, and merges
397 symbol table entries. Here is an example of the "hello world" module:</p>
399 <div class="doc_code">
400 <pre><i>; Declare the string constant as a global constant...</i>
401 <a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a
402 href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
404 <i>; External declaration of the puts function</i>
405 <a href="#functionstructure">declare</a> i32 @puts(i8 *) <i>; i32(i8 *)* </i>
407 <i>; Definition of main function</i>
408 define i32 @main() { <i>; i32()* </i>
409 <i>; Convert [13x i8 ]* to i8 *...</i>
411 href="#i_getelementptr">getelementptr</a> [13 x i8 ]* @.LC0, i64 0, i64 0 <i>; i8 *</i>
413 <i>; Call puts function to write out the string to stdout...</i>
415 href="#i_call">call</a> i32 @puts(i8 * %cast210) <i>; i32</i>
417 href="#i_ret">ret</a> i32 0<br>}<br>
421 <p>This example is made up of a <a href="#globalvars">global variable</a>
422 named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
423 function, and a <a href="#functionstructure">function definition</a>
424 for "<tt>main</tt>".</p>
426 <p>In general, a module is made up of a list of global values,
427 where both functions and global variables are global values. Global values are
428 represented by a pointer to a memory location (in this case, a pointer to an
429 array of char, and a pointer to a function), and have one of the following <a
430 href="#linkage">linkage types</a>.</p>
434 <!-- ======================================================================= -->
435 <div class="doc_subsection">
436 <a name="linkage">Linkage Types</a>
439 <div class="doc_text">
442 All Global Variables and Functions have one of the following types of linkage:
447 <dt><tt><b><a name="linkage_internal">internal</a></b></tt> </dt>
449 <dd>Global values with internal linkage are only directly accessible by
450 objects in the current module. In particular, linking code into a module with
451 an internal global value may cause the internal to be renamed as necessary to
452 avoid collisions. Because the symbol is internal to the module, all
453 references can be updated. This corresponds to the notion of the
454 '<tt>static</tt>' keyword in C.
457 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
459 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
460 the same name when linkage occurs. This is typically used to implement
461 inline functions, templates, or other code which must be generated in each
462 translation unit that uses it. Unreferenced <tt>linkonce</tt> globals are
463 allowed to be discarded.
466 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
468 <dd>"<tt>weak</tt>" linkage is exactly the same as <tt>linkonce</tt> linkage,
469 except that unreferenced <tt>weak</tt> globals may not be discarded. This is
470 used for globals that may be emitted in multiple translation units, but that
471 are not guaranteed to be emitted into every translation unit that uses them.
472 One example of this are common globals in C, such as "<tt>int X;</tt>" at
476 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
478 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
479 pointer to array type. When two global variables with appending linkage are
480 linked together, the two global arrays are appended together. This is the
481 LLVM, typesafe, equivalent of having the system linker append together
482 "sections" with identical names when .o files are linked.
485 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt>: </dt>
486 <dd>The semantics of this linkage follow the ELF model: the symbol is weak
487 until linked, if not linked, the symbol becomes null instead of being an
491 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
493 <dd>If none of the above identifiers are used, the global is externally
494 visible, meaning that it participates in linkage and can be used to resolve
495 external symbol references.
500 The next two types of linkage are targeted for Microsoft Windows platform
501 only. They are designed to support importing (exporting) symbols from (to)
506 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt>: </dt>
508 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
509 or variable via a global pointer to a pointer that is set up by the DLL
510 exporting the symbol. On Microsoft Windows targets, the pointer name is
511 formed by combining <code>_imp__</code> and the function or variable name.
514 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt>: </dt>
516 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
517 pointer to a pointer in a DLL, so that it can be referenced with the
518 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
519 name is formed by combining <code>_imp__</code> and the function or variable
525 <p><a name="linkage_external"></a>For example, since the "<tt>.LC0</tt>"
526 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
527 variable and was linked with this one, one of the two would be renamed,
528 preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
529 external (i.e., lacking any linkage declarations), they are accessible
530 outside of the current module.</p>
531 <p>It is illegal for a function <i>declaration</i>
532 to have any linkage type other than "externally visible", <tt>dllimport</tt>,
533 or <tt>extern_weak</tt>.</p>
534 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt> and <tt>weak</tt>
538 <!-- ======================================================================= -->
539 <div class="doc_subsection">
540 <a name="callingconv">Calling Conventions</a>
543 <div class="doc_text">
545 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
546 and <a href="#i_invoke">invokes</a> can all have an optional calling convention
547 specified for the call. The calling convention of any pair of dynamic
548 caller/callee must match, or the behavior of the program is undefined. The
549 following calling conventions are supported by LLVM, and more may be added in
553 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
555 <dd>This calling convention (the default if no other calling convention is
556 specified) matches the target C calling conventions. This calling convention
557 supports varargs function calls and tolerates some mismatch in the declared
558 prototype and implemented declaration of the function (as does normal C).
561 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
563 <dd>This calling convention attempts to make calls as fast as possible
564 (e.g. by passing things in registers). This calling convention allows the
565 target to use whatever tricks it wants to produce fast code for the target,
566 without having to conform to an externally specified ABI. Implementations of
567 this convention should allow arbitrary tail call optimization to be supported.
568 This calling convention does not support varargs and requires the prototype of
569 all callees to exactly match the prototype of the function definition.
572 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
574 <dd>This calling convention attempts to make code in the caller as efficient
575 as possible under the assumption that the call is not commonly executed. As
576 such, these calls often preserve all registers so that the call does not break
577 any live ranges in the caller side. This calling convention does not support
578 varargs and requires the prototype of all callees to exactly match the
579 prototype of the function definition.
582 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
584 <dd>Any calling convention may be specified by number, allowing
585 target-specific calling conventions to be used. Target specific calling
586 conventions start at 64.
590 <p>More calling conventions can be added/defined on an as-needed basis, to
591 support pascal conventions or any other well-known target-independent
596 <!-- ======================================================================= -->
597 <div class="doc_subsection">
598 <a name="visibility">Visibility Styles</a>
601 <div class="doc_text">
604 All Global Variables and Functions have one of the following visibility styles:
608 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
610 <dd>On ELF, default visibility means that the declaration is visible to other
611 modules and, in shared libraries, means that the declared entity may be
612 overridden. On Darwin, default visibility means that the declaration is
613 visible to other modules. Default visibility corresponds to "external
614 linkage" in the language.
617 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
619 <dd>Two declarations of an object with hidden visibility refer to the same
620 object if they are in the same shared object. Usually, hidden visibility
621 indicates that the symbol will not be placed into the dynamic symbol table,
622 so no other module (executable or shared library) can reference it
626 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
628 <dd>On ELF, protected visibility indicates that the symbol will be placed in
629 the dynamic symbol table, but that references within the defining module will
630 bind to the local symbol. That is, the symbol cannot be overridden by another
637 <!-- ======================================================================= -->
638 <div class="doc_subsection">
639 <a name="globalvars">Global Variables</a>
642 <div class="doc_text">
644 <p>Global variables define regions of memory allocated at compilation time
645 instead of run-time. Global variables may optionally be initialized, may have
646 an explicit section to be placed in, and may have an optional explicit alignment
647 specified. A variable may be defined as "thread_local", which means that it
648 will not be shared by threads (each thread will have a separated copy of the
649 variable). A variable may be defined as a global "constant," which indicates
650 that the contents of the variable will <b>never</b> be modified (enabling better
651 optimization, allowing the global data to be placed in the read-only section of
652 an executable, etc). Note that variables that need runtime initialization
653 cannot be marked "constant" as there is a store to the variable.</p>
656 LLVM explicitly allows <em>declarations</em> of global variables to be marked
657 constant, even if the final definition of the global is not. This capability
658 can be used to enable slightly better optimization of the program, but requires
659 the language definition to guarantee that optimizations based on the
660 'constantness' are valid for the translation units that do not include the
664 <p>As SSA values, global variables define pointer values that are in
665 scope (i.e. they dominate) all basic blocks in the program. Global
666 variables always define a pointer to their "content" type because they
667 describe a region of memory, and all memory objects in LLVM are
668 accessed through pointers.</p>
670 <p>LLVM allows an explicit section to be specified for globals. If the target
671 supports it, it will emit globals to the section specified.</p>
673 <p>An explicit alignment may be specified for a global. If not present, or if
674 the alignment is set to zero, the alignment of the global is set by the target
675 to whatever it feels convenient. If an explicit alignment is specified, the
676 global is forced to have at least that much alignment. All alignments must be
679 <p>For example, the following defines a global with an initializer, section,
682 <div class="doc_code">
684 @G = constant float 1.0, section "foo", align 4
691 <!-- ======================================================================= -->
692 <div class="doc_subsection">
693 <a name="functionstructure">Functions</a>
696 <div class="doc_text">
698 <p>LLVM function definitions consist of the "<tt>define</tt>" keyord,
699 an optional <a href="#linkage">linkage type</a>, an optional
700 <a href="#visibility">visibility style</a>, an optional
701 <a href="#callingconv">calling convention</a>, a return type, an optional
702 <a href="#paramattrs">parameter attribute</a> for the return type, a function
703 name, a (possibly empty) argument list (each with optional
704 <a href="#paramattrs">parameter attributes</a>), an optional section, an
705 optional alignment, an opening curly brace, a list of basic blocks, and a
708 LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
709 optional <a href="#linkage">linkage type</a>, an optional
710 <a href="#visibility">visibility style</a>, an optional
711 <a href="#callingconv">calling convention</a>, a return type, an optional
712 <a href="#paramattrs">parameter attribute</a> for the return type, a function
713 name, a possibly empty list of arguments, and an optional alignment.</p>
715 <p>A function definition contains a list of basic blocks, forming the CFG for
716 the function. Each basic block may optionally start with a label (giving the
717 basic block a symbol table entry), contains a list of instructions, and ends
718 with a <a href="#terminators">terminator</a> instruction (such as a branch or
719 function return).</p>
721 <p>The first basic block in a function is special in two ways: it is immediately
722 executed on entrance to the function, and it is not allowed to have predecessor
723 basic blocks (i.e. there can not be any branches to the entry block of a
724 function). Because the block can have no predecessors, it also cannot have any
725 <a href="#i_phi">PHI nodes</a>.</p>
727 <p>LLVM allows an explicit section to be specified for functions. If the target
728 supports it, it will emit functions to the section specified.</p>
730 <p>An explicit alignment may be specified for a function. If not present, or if
731 the alignment is set to zero, the alignment of the function is set by the target
732 to whatever it feels convenient. If an explicit alignment is specified, the
733 function is forced to have at least that much alignment. All alignments must be
739 <!-- ======================================================================= -->
740 <div class="doc_subsection">
741 <a name="aliasstructure">Aliases</a>
743 <div class="doc_text">
744 <p>Aliases act as "second name" for the aliasee value (which can be either
745 function or global variable or bitcast of global value). Aliases may have an
746 optional <a href="#linkage">linkage type</a>, and an
747 optional <a href="#visibility">visibility style</a>.</p>
751 <div class="doc_code">
753 @<Name> = [Linkage] [Visibility] alias <AliaseeTy> @<Aliasee>
761 <!-- ======================================================================= -->
762 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
763 <div class="doc_text">
764 <p>The return type and each parameter of a function type may have a set of
765 <i>parameter attributes</i> associated with them. Parameter attributes are
766 used to communicate additional information about the result or parameters of
767 a function. Parameter attributes are considered to be part of the function
768 type so two functions types that differ only by the parameter attributes
769 are different function types.</p>
771 <p>Parameter attributes are simple keywords that follow the type specified. If
772 multiple parameter attributes are needed, they are space separated. For
775 <div class="doc_code">
777 %someFunc = i16 (i8 signext %someParam) zeroext
778 %someFunc = i16 (i8 zeroext %someParam) zeroext
782 <p>Note that the two function types above are unique because the parameter has
783 a different attribute (<tt>signext</tt> in the first one, <tt>zeroext</tt> in
784 the second). Also note that the attribute for the function result
785 (<tt>zeroext</tt>) comes immediately after the argument list.</p>
787 <p>Currently, only the following parameter attributes are defined:</p>
789 <dt><tt>zeroext</tt></dt>
790 <dd>This indicates that the parameter should be zero extended just before
791 a call to this function.</dd>
792 <dt><tt>signext</tt></dt>
793 <dd>This indicates that the parameter should be sign extended just before
794 a call to this function.</dd>
795 <dt><tt>inreg</tt></dt>
796 <dd>This indicates that the parameter should be placed in register (if
797 possible) during assembling function call. Support for this attribute is
799 <dt><tt>sret</tt></dt>
800 <dd>This indicates that the parameter specifies the address of a structure
801 that is the return value of the function in the source program.</dd>
802 <dt><tt>noalias</tt></dt>
803 <dd>This indicates that the parameter not alias any other object or any
804 other "noalias" objects during the function call.
805 <dt><tt>noreturn</tt></dt>
806 <dd>This function attribute indicates that the function never returns. This
807 indicates to LLVM that every call to this function should be treated as if
808 an <tt>unreachable</tt> instruction immediately followed the call.</dd>
809 <dt><tt>nounwind</tt></dt>
810 <dd>This function attribute indicates that the function type does not use
811 the unwind instruction and does not allow stack unwinding to propagate
813 <dt><tt>nest</tt></dt>
814 <dd>This indicates that the parameter can be excised using the
815 <a href="#int_trampoline">trampoline intrinsics</a>.</dd>
816 <dt><tt>readonly</tt></dt>
817 <dd>This function attribute indicates that the function has no side-effects
818 except for producing a return value or throwing an exception. The value
819 returned must only depend on the function arguments and/or global variables.
820 It may use values obtained by dereferencing pointers.</dd>
821 <dt><tt>readnone</tt></dt>
822 <dd>A <tt>readnone</tt> function has the same restrictions as a <tt>readonly</tt>
823 function, but in addition it is not allowed to dereference any pointer arguments
829 <!-- ======================================================================= -->
830 <div class="doc_subsection">
831 <a name="moduleasm">Module-Level Inline Assembly</a>
834 <div class="doc_text">
836 Modules may contain "module-level inline asm" blocks, which corresponds to the
837 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
838 LLVM and treated as a single unit, but may be separated in the .ll file if
839 desired. The syntax is very simple:
842 <div class="doc_code">
844 module asm "inline asm code goes here"
845 module asm "more can go here"
849 <p>The strings can contain any character by escaping non-printable characters.
850 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
855 The inline asm code is simply printed to the machine code .s file when
856 assembly code is generated.
860 <!-- ======================================================================= -->
861 <div class="doc_subsection">
862 <a name="datalayout">Data Layout</a>
865 <div class="doc_text">
866 <p>A module may specify a target specific data layout string that specifies how
867 data is to be laid out in memory. The syntax for the data layout is simply:</p>
868 <pre> target datalayout = "<i>layout specification</i>"</pre>
869 <p>The <i>layout specification</i> consists of a list of specifications
870 separated by the minus sign character ('-'). Each specification starts with a
871 letter and may include other information after the letter to define some
872 aspect of the data layout. The specifications accepted are as follows: </p>
875 <dd>Specifies that the target lays out data in big-endian form. That is, the
876 bits with the most significance have the lowest address location.</dd>
878 <dd>Specifies that hte target lays out data in little-endian form. That is,
879 the bits with the least significance have the lowest address location.</dd>
880 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
881 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
882 <i>preferred</i> alignments. All sizes are in bits. Specifying the <i>pref</i>
883 alignment is optional. If omitted, the preceding <tt>:</tt> should be omitted
885 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
886 <dd>This specifies the alignment for an integer type of a given bit
887 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
888 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
889 <dd>This specifies the alignment for a vector type of a given bit
891 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
892 <dd>This specifies the alignment for a floating point type of a given bit
893 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
895 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
896 <dd>This specifies the alignment for an aggregate type of a given bit
899 <p>When constructing the data layout for a given target, LLVM starts with a
900 default set of specifications which are then (possibly) overriden by the
901 specifications in the <tt>datalayout</tt> keyword. The default specifications
902 are given in this list:</p>
904 <li><tt>E</tt> - big endian</li>
905 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
906 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
907 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
908 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
909 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
910 <li><tt>i64:32:64</tt> - i64 has abi alignment of 32-bits but preferred
911 alignment of 64-bits</li>
912 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
913 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
914 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
915 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
916 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
918 <p>When llvm is determining the alignment for a given type, it uses the
921 <li>If the type sought is an exact match for one of the specifications, that
922 specification is used.</li>
923 <li>If no match is found, and the type sought is an integer type, then the
924 smallest integer type that is larger than the bitwidth of the sought type is
925 used. If none of the specifications are larger than the bitwidth then the the
926 largest integer type is used. For example, given the default specifications
927 above, the i7 type will use the alignment of i8 (next largest) while both
928 i65 and i256 will use the alignment of i64 (largest specified).</li>
929 <li>If no match is found, and the type sought is a vector type, then the
930 largest vector type that is smaller than the sought vector type will be used
931 as a fall back. This happens because <128 x double> can be implemented in
932 terms of 64 <2 x double>, for example.</li>
936 <!-- *********************************************************************** -->
937 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
938 <!-- *********************************************************************** -->
940 <div class="doc_text">
942 <p>The LLVM type system is one of the most important features of the
943 intermediate representation. Being typed enables a number of
944 optimizations to be performed on the IR directly, without having to do
945 extra analyses on the side before the transformation. A strong type
946 system makes it easier to read the generated code and enables novel
947 analyses and transformations that are not feasible to perform on normal
948 three address code representations.</p>
952 <!-- ======================================================================= -->
953 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
954 <div class="doc_text">
955 <p>The primitive types are the fundamental building blocks of the LLVM
956 system. The current set of primitive types is as follows:</p>
958 <table class="layout">
963 <tr><th>Type</th><th>Description</th></tr>
964 <tr><td><tt><a name="t_void">void</a></tt></td><td>No value</td></tr>
965 <tr><td><tt>label</tt></td><td>Branch destination</td></tr>
972 <tr><th>Type</th><th>Description</th></tr>
973 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
974 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
982 <!-- _______________________________________________________________________ -->
983 <div class="doc_subsubsection"> <a name="t_classifications">Type
984 Classifications</a> </div>
985 <div class="doc_text">
986 <p>These different primitive types fall into a few useful
989 <table border="1" cellspacing="0" cellpadding="4">
991 <tr><th>Classification</th><th>Types</th></tr>
993 <td><a name="t_integer">integer</a></td>
994 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
997 <td><a name="t_floating">floating point</a></td>
998 <td><tt>float, double</tt></td>
1001 <td><a name="t_firstclass">first class</a></td>
1002 <td><tt>i1, ..., float, double, <br/>
1003 <a href="#t_pointer">pointer</a>,<a href="#t_vector">vector</a></tt>
1009 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
1010 most important. Values of these types are the only ones which can be
1011 produced by instructions, passed as arguments, or used as operands to
1012 instructions. This means that all structures and arrays must be
1013 manipulated either by pointer or by component.</p>
1016 <!-- ======================================================================= -->
1017 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1019 <div class="doc_text">
1021 <p>The real power in LLVM comes from the derived types in the system.
1022 This is what allows a programmer to represent arrays, functions,
1023 pointers, and other useful types. Note that these derived types may be
1024 recursive: For example, it is possible to have a two dimensional array.</p>
1028 <!-- _______________________________________________________________________ -->
1029 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1031 <div class="doc_text">
1034 <p>The integer type is a very simple derived type that simply specifies an
1035 arbitrary bit width for the integer type desired. Any bit width from 1 bit to
1036 2^23-1 (about 8 million) can be specified.</p>
1044 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1048 <table class="layout">
1058 <tt>i1942652</tt><br/>
1061 A boolean integer of 1 bit<br/>
1062 A nibble sized integer of 4 bits.<br/>
1063 A byte sized integer of 8 bits.<br/>
1064 A half word sized integer of 16 bits.<br/>
1065 A word sized integer of 32 bits.<br/>
1066 An integer whose bit width is the answer. <br/>
1067 A double word sized integer of 64 bits.<br/>
1068 A really big integer of over 1 million bits.<br/>
1074 <!-- _______________________________________________________________________ -->
1075 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1077 <div class="doc_text">
1081 <p>The array type is a very simple derived type that arranges elements
1082 sequentially in memory. The array type requires a size (number of
1083 elements) and an underlying data type.</p>
1088 [<# elements> x <elementtype>]
1091 <p>The number of elements is a constant integer value; elementtype may
1092 be any type with a size.</p>
1095 <table class="layout">
1098 <tt>[40 x i32 ]</tt><br/>
1099 <tt>[41 x i32 ]</tt><br/>
1100 <tt>[40 x i8]</tt><br/>
1103 Array of 40 32-bit integer values.<br/>
1104 Array of 41 32-bit integer values.<br/>
1105 Array of 40 8-bit integer values.<br/>
1109 <p>Here are some examples of multidimensional arrays:</p>
1110 <table class="layout">
1113 <tt>[3 x [4 x i32]]</tt><br/>
1114 <tt>[12 x [10 x float]]</tt><br/>
1115 <tt>[2 x [3 x [4 x i16]]]</tt><br/>
1118 3x4 array of 32-bit integer values.<br/>
1119 12x10 array of single precision floating point values.<br/>
1120 2x3x4 array of 16-bit integer values.<br/>
1125 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1126 length array. Normally, accesses past the end of an array are undefined in
1127 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
1128 As a special case, however, zero length arrays are recognized to be variable
1129 length. This allows implementation of 'pascal style arrays' with the LLVM
1130 type "{ i32, [0 x float]}", for example.</p>
1134 <!-- _______________________________________________________________________ -->
1135 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1136 <div class="doc_text">
1138 <p>The function type can be thought of as a function signature. It
1139 consists of a return type and a list of formal parameter types.
1140 Function types are usually used to build virtual function tables
1141 (which are structures of pointers to functions), for indirect function
1142 calls, and when defining a function.</p>
1144 The return type of a function type cannot be an aggregate type.
1147 <pre> <returntype> (<parameter list>)<br></pre>
1148 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1149 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1150 which indicates that the function takes a variable number of arguments.
1151 Variable argument functions can access their arguments with the <a
1152 href="#int_varargs">variable argument handling intrinsic</a> functions.</p>
1154 <table class="layout">
1156 <td class="left"><tt>i32 (i32)</tt></td>
1157 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1159 </tr><tr class="layout">
1160 <td class="left"><tt>float (i16 signext, i32 *) *
1162 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1163 an <tt>i16</tt> that should be sign extended and a
1164 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1167 </tr><tr class="layout">
1168 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1169 <td class="left">A vararg function that takes at least one
1170 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1171 which returns an integer. This is the signature for <tt>printf</tt> in
1178 <!-- _______________________________________________________________________ -->
1179 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1180 <div class="doc_text">
1182 <p>The structure type is used to represent a collection of data members
1183 together in memory. The packing of the field types is defined to match
1184 the ABI of the underlying processor. The elements of a structure may
1185 be any type that has a size.</p>
1186 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1187 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1188 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1191 <pre> { <type list> }<br></pre>
1193 <table class="layout">
1195 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1196 <td class="left">A triple of three <tt>i32</tt> values</td>
1197 </tr><tr class="layout">
1198 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1199 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1200 second element is a <a href="#t_pointer">pointer</a> to a
1201 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1202 an <tt>i32</tt>.</td>
1207 <!-- _______________________________________________________________________ -->
1208 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1210 <div class="doc_text">
1212 <p>The packed structure type is used to represent a collection of data members
1213 together in memory. There is no padding between fields. Further, the alignment
1214 of a packed structure is 1 byte. The elements of a packed structure may
1215 be any type that has a size.</p>
1216 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1217 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1218 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1221 <pre> < { <type list> } > <br></pre>
1223 <table class="layout">
1225 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1226 <td class="left">A triple of three <tt>i32</tt> values</td>
1227 </tr><tr class="layout">
1228 <td class="left"><tt>< { float, i32 (i32) * } ></tt></td>
1229 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1230 second element is a <a href="#t_pointer">pointer</a> to a
1231 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1232 an <tt>i32</tt>.</td>
1237 <!-- _______________________________________________________________________ -->
1238 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1239 <div class="doc_text">
1241 <p>As in many languages, the pointer type represents a pointer or
1242 reference to another object, which must live in memory.</p>
1244 <pre> <type> *<br></pre>
1246 <table class="layout">
1249 <tt>[4x i32]*</tt><br/>
1250 <tt>i32 (i32 *) *</tt><br/>
1253 A <a href="#t_pointer">pointer</a> to <a href="#t_array">array</a> of
1254 four <tt>i32</tt> values<br/>
1255 A <a href="#t_pointer">pointer</a> to a <a
1256 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1263 <!-- _______________________________________________________________________ -->
1264 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1265 <div class="doc_text">
1269 <p>A vector type is a simple derived type that represents a vector
1270 of elements. Vector types are used when multiple primitive data
1271 are operated in parallel using a single instruction (SIMD).
1272 A vector type requires a size (number of
1273 elements) and an underlying primitive data type. Vectors must have a power
1274 of two length (1, 2, 4, 8, 16 ...). Vector types are
1275 considered <a href="#t_firstclass">first class</a>.</p>
1280 < <# elements> x <elementtype> >
1283 <p>The number of elements is a constant integer value; elementtype may
1284 be any integer or floating point type.</p>
1288 <table class="layout">
1291 <tt><4 x i32></tt><br/>
1292 <tt><8 x float></tt><br/>
1293 <tt><2 x i64></tt><br/>
1296 Vector of 4 32-bit integer values.<br/>
1297 Vector of 8 floating-point values.<br/>
1298 Vector of 2 64-bit integer values.<br/>
1304 <!-- _______________________________________________________________________ -->
1305 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1306 <div class="doc_text">
1310 <p>Opaque types are used to represent unknown types in the system. This
1311 corresponds (for example) to the C notion of a forward declared structure type.
1312 In LLVM, opaque types can eventually be resolved to any type (not just a
1313 structure type).</p>
1323 <table class="layout">
1329 An opaque type.<br/>
1336 <!-- *********************************************************************** -->
1337 <div class="doc_section"> <a name="constants">Constants</a> </div>
1338 <!-- *********************************************************************** -->
1340 <div class="doc_text">
1342 <p>LLVM has several different basic types of constants. This section describes
1343 them all and their syntax.</p>
1347 <!-- ======================================================================= -->
1348 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1350 <div class="doc_text">
1353 <dt><b>Boolean constants</b></dt>
1355 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1356 constants of the <tt><a href="#t_primitive">i1</a></tt> type.
1359 <dt><b>Integer constants</b></dt>
1361 <dd>Standard integers (such as '4') are constants of the <a
1362 href="#t_integer">integer</a> type. Negative numbers may be used with
1366 <dt><b>Floating point constants</b></dt>
1368 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1369 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1370 notation (see below). Floating point constants must have a <a
1371 href="#t_floating">floating point</a> type. </dd>
1373 <dt><b>Null pointer constants</b></dt>
1375 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1376 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1380 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1381 of floating point constants. For example, the form '<tt>double
1382 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1383 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1384 (and the only time that they are generated by the disassembler) is when a
1385 floating point constant must be emitted but it cannot be represented as a
1386 decimal floating point number. For example, NaN's, infinities, and other
1387 special values are represented in their IEEE hexadecimal format so that
1388 assembly and disassembly do not cause any bits to change in the constants.</p>
1392 <!-- ======================================================================= -->
1393 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1396 <div class="doc_text">
1397 <p>Aggregate constants arise from aggregation of simple constants
1398 and smaller aggregate constants.</p>
1401 <dt><b>Structure constants</b></dt>
1403 <dd>Structure constants are represented with notation similar to structure
1404 type definitions (a comma separated list of elements, surrounded by braces
1405 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* %G }</tt>",
1406 where "<tt>%G</tt>" is declared as "<tt>@G = external global i32</tt>". Structure constants
1407 must have <a href="#t_struct">structure type</a>, and the number and
1408 types of elements must match those specified by the type.
1411 <dt><b>Array constants</b></dt>
1413 <dd>Array constants are represented with notation similar to array type
1414 definitions (a comma separated list of elements, surrounded by square brackets
1415 (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74 ]</tt>". Array
1416 constants must have <a href="#t_array">array type</a>, and the number and
1417 types of elements must match those specified by the type.
1420 <dt><b>Vector constants</b></dt>
1422 <dd>Vector constants are represented with notation similar to vector type
1423 definitions (a comma separated list of elements, surrounded by
1424 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32 42,
1425 i32 11, i32 74, i32 100 ></tt>". Vector constants must have <a
1426 href="#t_vector">vector type</a>, and the number and types of elements must
1427 match those specified by the type.
1430 <dt><b>Zero initialization</b></dt>
1432 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1433 value to zero of <em>any</em> type, including scalar and aggregate types.
1434 This is often used to avoid having to print large zero initializers (e.g. for
1435 large arrays) and is always exactly equivalent to using explicit zero
1442 <!-- ======================================================================= -->
1443 <div class="doc_subsection">
1444 <a name="globalconstants">Global Variable and Function Addresses</a>
1447 <div class="doc_text">
1449 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1450 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1451 constants. These constants are explicitly referenced when the <a
1452 href="#identifiers">identifier for the global</a> is used and always have <a
1453 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1456 <div class="doc_code">
1460 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1466 <!-- ======================================================================= -->
1467 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1468 <div class="doc_text">
1469 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1470 no specific value. Undefined values may be of any type and be used anywhere
1471 a constant is permitted.</p>
1473 <p>Undefined values indicate to the compiler that the program is well defined
1474 no matter what value is used, giving the compiler more freedom to optimize.
1478 <!-- ======================================================================= -->
1479 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1482 <div class="doc_text">
1484 <p>Constant expressions are used to allow expressions involving other constants
1485 to be used as constants. Constant expressions may be of any <a
1486 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1487 that does not have side effects (e.g. load and call are not supported). The
1488 following is the syntax for constant expressions:</p>
1491 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1492 <dd>Truncate a constant to another type. The bit size of CST must be larger
1493 than the bit size of TYPE. Both types must be integers.</dd>
1495 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1496 <dd>Zero extend a constant to another type. The bit size of CST must be
1497 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1499 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1500 <dd>Sign extend a constant to another type. The bit size of CST must be
1501 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1503 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1504 <dd>Truncate a floating point constant to another floating point type. The
1505 size of CST must be larger than the size of TYPE. Both types must be
1506 floating point.</dd>
1508 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1509 <dd>Floating point extend a constant to another type. The size of CST must be
1510 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1512 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
1513 <dd>Convert a floating point constant to the corresponding unsigned integer
1514 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1515 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1516 of the same number of elements. If the value won't fit in the integer type,
1517 the results are undefined.</dd>
1519 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
1520 <dd>Convert a floating point constant to the corresponding signed integer
1521 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1522 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1523 of the same number of elements. If the value won't fit in the integer type,
1524 the results are undefined.</dd>
1526 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
1527 <dd>Convert an unsigned integer constant to the corresponding floating point
1528 constant. TYPE must be a scalar or vector floating point type. CST must be of
1529 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1530 of the same number of elements. If the value won't fit in the floating point
1531 type, the results are undefined.</dd>
1533 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
1534 <dd>Convert a signed integer constant to the corresponding floating point
1535 constant. TYPE must be a scalar or vector floating point type. CST must be of
1536 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1537 of the same number of elements. If the value won't fit in the floating point
1538 type, the results are undefined.</dd>
1540 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
1541 <dd>Convert a pointer typed constant to the corresponding integer constant
1542 TYPE must be an integer type. CST must be of pointer type. The CST value is
1543 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1545 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
1546 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1547 pointer type. CST must be of integer type. The CST value is zero extended,
1548 truncated, or unchanged to make it fit in a pointer size. This one is
1549 <i>really</i> dangerous!</dd>
1551 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
1552 <dd>Convert a constant, CST, to another TYPE. The size of CST and TYPE must be
1553 identical (same number of bits). The conversion is done as if the CST value
1554 was stored to memory and read back as TYPE. In other words, no bits change
1555 with this operator, just the type. This can be used for conversion of
1556 vector types to any other type, as long as they have the same bit width. For
1557 pointers it is only valid to cast to another pointer type.
1560 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1562 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1563 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1564 instruction, the index list may have zero or more indexes, which are required
1565 to make sense for the type of "CSTPTR".</dd>
1567 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1569 <dd>Perform the <a href="#i_select">select operation</a> on
1572 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
1573 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
1575 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
1576 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
1578 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1580 <dd>Perform the <a href="#i_extractelement">extractelement
1581 operation</a> on constants.
1583 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1585 <dd>Perform the <a href="#i_insertelement">insertelement
1586 operation</a> on constants.</dd>
1589 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
1591 <dd>Perform the <a href="#i_shufflevector">shufflevector
1592 operation</a> on constants.</dd>
1594 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1596 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1597 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1598 binary</a> operations. The constraints on operands are the same as those for
1599 the corresponding instruction (e.g. no bitwise operations on floating point
1600 values are allowed).</dd>
1604 <!-- *********************************************************************** -->
1605 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1606 <!-- *********************************************************************** -->
1608 <!-- ======================================================================= -->
1609 <div class="doc_subsection">
1610 <a name="inlineasm">Inline Assembler Expressions</a>
1613 <div class="doc_text">
1616 LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
1617 Module-Level Inline Assembly</a>) through the use of a special value. This
1618 value represents the inline assembler as a string (containing the instructions
1619 to emit), a list of operand constraints (stored as a string), and a flag that
1620 indicates whether or not the inline asm expression has side effects. An example
1621 inline assembler expression is:
1624 <div class="doc_code">
1626 i32 (i32) asm "bswap $0", "=r,r"
1631 Inline assembler expressions may <b>only</b> be used as the callee operand of
1632 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
1635 <div class="doc_code">
1637 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
1642 Inline asms with side effects not visible in the constraint list must be marked
1643 as having side effects. This is done through the use of the
1644 '<tt>sideeffect</tt>' keyword, like so:
1647 <div class="doc_code">
1649 call void asm sideeffect "eieio", ""()
1653 <p>TODO: The format of the asm and constraints string still need to be
1654 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1655 need to be documented).
1660 <!-- *********************************************************************** -->
1661 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1662 <!-- *********************************************************************** -->
1664 <div class="doc_text">
1666 <p>The LLVM instruction set consists of several different
1667 classifications of instructions: <a href="#terminators">terminator
1668 instructions</a>, <a href="#binaryops">binary instructions</a>,
1669 <a href="#bitwiseops">bitwise binary instructions</a>, <a
1670 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1671 instructions</a>.</p>
1675 <!-- ======================================================================= -->
1676 <div class="doc_subsection"> <a name="terminators">Terminator
1677 Instructions</a> </div>
1679 <div class="doc_text">
1681 <p>As mentioned <a href="#functionstructure">previously</a>, every
1682 basic block in a program ends with a "Terminator" instruction, which
1683 indicates which block should be executed after the current block is
1684 finished. These terminator instructions typically yield a '<tt>void</tt>'
1685 value: they produce control flow, not values (the one exception being
1686 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1687 <p>There are six different terminator instructions: the '<a
1688 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1689 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1690 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1691 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1692 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1696 <!-- _______________________________________________________________________ -->
1697 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1698 Instruction</a> </div>
1699 <div class="doc_text">
1701 <pre> ret <type> <value> <i>; Return a value from a non-void function</i>
1702 ret void <i>; Return from void function</i>
1705 <p>The '<tt>ret</tt>' instruction is used to return control flow (and a
1706 value) from a function back to the caller.</p>
1707 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
1708 returns a value and then causes control flow, and one that just causes
1709 control flow to occur.</p>
1711 <p>The '<tt>ret</tt>' instruction may return any '<a
1712 href="#t_firstclass">first class</a>' type. Notice that a function is
1713 not <a href="#wellformed">well formed</a> if there exists a '<tt>ret</tt>'
1714 instruction inside of the function that returns a value that does not
1715 match the return type of the function.</p>
1717 <p>When the '<tt>ret</tt>' instruction is executed, control flow
1718 returns back to the calling function's context. If the caller is a "<a
1719 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1720 the instruction after the call. If the caller was an "<a
1721 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1722 at the beginning of the "normal" destination block. If the instruction
1723 returns a value, that value shall set the call or invoke instruction's
1726 <pre> ret i32 5 <i>; Return an integer value of 5</i>
1727 ret void <i>; Return from a void function</i>
1730 <!-- _______________________________________________________________________ -->
1731 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1732 <div class="doc_text">
1734 <pre> br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1737 <p>The '<tt>br</tt>' instruction is used to cause control flow to
1738 transfer to a different basic block in the current function. There are
1739 two forms of this instruction, corresponding to a conditional branch
1740 and an unconditional branch.</p>
1742 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1743 single '<tt>i1</tt>' value and two '<tt>label</tt>' values. The
1744 unconditional form of the '<tt>br</tt>' instruction takes a single
1745 '<tt>label</tt>' value as a target.</p>
1747 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
1748 argument is evaluated. If the value is <tt>true</tt>, control flows
1749 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1750 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1752 <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
1753 href="#i_ret">ret</a> i32 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> i32 0<br></pre>
1755 <!-- _______________________________________________________________________ -->
1756 <div class="doc_subsubsection">
1757 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
1760 <div class="doc_text">
1764 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
1769 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
1770 several different places. It is a generalization of the '<tt>br</tt>'
1771 instruction, allowing a branch to occur to one of many possible
1777 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
1778 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
1779 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
1780 table is not allowed to contain duplicate constant entries.</p>
1784 <p>The <tt>switch</tt> instruction specifies a table of values and
1785 destinations. When the '<tt>switch</tt>' instruction is executed, this
1786 table is searched for the given value. If the value is found, control flow is
1787 transfered to the corresponding destination; otherwise, control flow is
1788 transfered to the default destination.</p>
1790 <h5>Implementation:</h5>
1792 <p>Depending on properties of the target machine and the particular
1793 <tt>switch</tt> instruction, this instruction may be code generated in different
1794 ways. For example, it could be generated as a series of chained conditional
1795 branches or with a lookup table.</p>
1800 <i>; Emulate a conditional br instruction</i>
1801 %Val = <a href="#i_zext">zext</a> i1 %value to i32
1802 switch i32 %Val, label %truedest [i32 0, label %falsedest ]
1804 <i>; Emulate an unconditional br instruction</i>
1805 switch i32 0, label %dest [ ]
1807 <i>; Implement a jump table:</i>
1808 switch i32 %val, label %otherwise [ i32 0, label %onzero
1810 i32 2, label %ontwo ]
1814 <!-- _______________________________________________________________________ -->
1815 <div class="doc_subsubsection">
1816 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
1819 <div class="doc_text">
1824 <result> = invoke [<a href="#callingconv">cconv</a>] <ptr to function ty> %<function ptr val>(<function args>)
1825 to label <normal label> unwind label <exception label>
1830 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
1831 function, with the possibility of control flow transfer to either the
1832 '<tt>normal</tt>' label or the
1833 '<tt>exception</tt>' label. If the callee function returns with the
1834 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
1835 "normal" label. If the callee (or any indirect callees) returns with the "<a
1836 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
1837 continued at the dynamically nearest "exception" label.</p>
1841 <p>This instruction requires several arguments:</p>
1845 The optional "cconv" marker indicates which <a href="#callingconv">calling
1846 convention</a> the call should use. If none is specified, the call defaults
1847 to using C calling conventions.
1849 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
1850 function value being invoked. In most cases, this is a direct function
1851 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
1852 an arbitrary pointer to function value.
1855 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
1856 function to be invoked. </li>
1858 <li>'<tt>function args</tt>': argument list whose types match the function
1859 signature argument types. If the function signature indicates the function
1860 accepts a variable number of arguments, the extra arguments can be
1863 <li>'<tt>normal label</tt>': the label reached when the called function
1864 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
1866 <li>'<tt>exception label</tt>': the label reached when a callee returns with
1867 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
1873 <p>This instruction is designed to operate as a standard '<tt><a
1874 href="#i_call">call</a></tt>' instruction in most regards. The primary
1875 difference is that it establishes an association with a label, which is used by
1876 the runtime library to unwind the stack.</p>
1878 <p>This instruction is used in languages with destructors to ensure that proper
1879 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
1880 exception. Additionally, this is important for implementation of
1881 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
1885 %retval = invoke i32 %Test(i32 15) to label %Continue
1886 unwind label %TestCleanup <i>; {i32}:retval set</i>
1887 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Test(i32 15) to label %Continue
1888 unwind label %TestCleanup <i>; {i32}:retval set</i>
1893 <!-- _______________________________________________________________________ -->
1895 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
1896 Instruction</a> </div>
1898 <div class="doc_text">
1907 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
1908 at the first callee in the dynamic call stack which used an <a
1909 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
1910 primarily used to implement exception handling.</p>
1914 <p>The '<tt>unwind</tt>' intrinsic causes execution of the current function to
1915 immediately halt. The dynamic call stack is then searched for the first <a
1916 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
1917 execution continues at the "exceptional" destination block specified by the
1918 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
1919 dynamic call chain, undefined behavior results.</p>
1922 <!-- _______________________________________________________________________ -->
1924 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
1925 Instruction</a> </div>
1927 <div class="doc_text">
1936 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
1937 instruction is used to inform the optimizer that a particular portion of the
1938 code is not reachable. This can be used to indicate that the code after a
1939 no-return function cannot be reached, and other facts.</p>
1943 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
1948 <!-- ======================================================================= -->
1949 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
1950 <div class="doc_text">
1951 <p>Binary operators are used to do most of the computation in a
1952 program. They require two operands, execute an operation on them, and
1953 produce a single value. The operands might represent
1954 multiple data, as is the case with the <a href="#t_vector">vector</a> data type.
1955 The result value of a binary operator is not
1956 necessarily the same type as its operands.</p>
1957 <p>There are several different binary operators:</p>
1959 <!-- _______________________________________________________________________ -->
1960 <div class="doc_subsubsection"> <a name="i_add">'<tt>add</tt>'
1961 Instruction</a> </div>
1962 <div class="doc_text">
1964 <pre> <result> = add <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1967 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
1969 <p>The two arguments to the '<tt>add</tt>' instruction must be either <a
1970 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a> values.
1971 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
1972 Both arguments must have identical types.</p>
1974 <p>The value produced is the integer or floating point sum of the two
1977 <pre> <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
1980 <!-- _______________________________________________________________________ -->
1981 <div class="doc_subsubsection"> <a name="i_sub">'<tt>sub</tt>'
1982 Instruction</a> </div>
1983 <div class="doc_text">
1985 <pre> <result> = sub <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1988 <p>The '<tt>sub</tt>' instruction returns the difference of its two
1990 <p>Note that the '<tt>sub</tt>' instruction is used to represent the '<tt>neg</tt>'
1991 instruction present in most other intermediate representations.</p>
1993 <p>The two arguments to the '<tt>sub</tt>' instruction must be either <a
1994 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1996 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
1997 Both arguments must have identical types.</p>
1999 <p>The value produced is the integer or floating point difference of
2000 the two operands.</p>
2003 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
2004 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
2007 <!-- _______________________________________________________________________ -->
2008 <div class="doc_subsubsection"> <a name="i_mul">'<tt>mul</tt>'
2009 Instruction</a> </div>
2010 <div class="doc_text">
2012 <pre> <result> = mul <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2015 <p>The '<tt>mul</tt>' instruction returns the product of its two
2018 <p>The two arguments to the '<tt>mul</tt>' instruction must be either <a
2019 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
2021 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2022 Both arguments must have identical types.</p>
2024 <p>The value produced is the integer or floating point product of the
2026 <p>Because the operands are the same width, the result of an integer
2027 multiplication is the same whether the operands should be deemed unsigned or
2030 <pre> <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
2033 <!-- _______________________________________________________________________ -->
2034 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
2036 <div class="doc_text">
2038 <pre> <result> = udiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2041 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two
2044 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
2045 <a href="#t_integer">integer</a> values. Both arguments must have identical
2046 types. This instruction can also take <a href="#t_vector">vector</a> versions
2047 of the values in which case the elements must be integers.</p>
2049 <p>The value produced is the unsigned integer quotient of the two operands. This
2050 instruction always performs an unsigned division operation, regardless of
2051 whether the arguments are unsigned or not.</p>
2053 <pre> <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2056 <!-- _______________________________________________________________________ -->
2057 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
2059 <div class="doc_text">
2061 <pre> <result> = sdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2064 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
2067 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
2068 <a href="#t_integer">integer</a> values. Both arguments must have identical
2069 types. This instruction can also take <a href="#t_vector">vector</a> versions
2070 of the values in which case the elements must be integers.</p>
2072 <p>The value produced is the signed integer quotient of the two operands. This
2073 instruction always performs a signed division operation, regardless of whether
2074 the arguments are signed or not.</p>
2076 <pre> <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2079 <!-- _______________________________________________________________________ -->
2080 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
2081 Instruction</a> </div>
2082 <div class="doc_text">
2084 <pre> <result> = fdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2087 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
2090 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
2091 <a href="#t_floating">floating point</a> values. Both arguments must have
2092 identical types. This instruction can also take <a href="#t_vector">vector</a>
2093 versions of floating point values.</p>
2095 <p>The value produced is the floating point quotient of the two operands.</p>
2097 <pre> <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
2100 <!-- _______________________________________________________________________ -->
2101 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
2103 <div class="doc_text">
2105 <pre> <result> = urem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2108 <p>The '<tt>urem</tt>' instruction returns the remainder from the
2109 unsigned division of its two arguments.</p>
2111 <p>The two arguments to the '<tt>urem</tt>' instruction must be
2112 <a href="#t_integer">integer</a> values. Both arguments must have identical
2113 types. This instruction can also take <a href="#t_vector">vector</a> versions
2114 of the values in which case the elements must be integers.</p>
2116 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
2117 This instruction always performs an unsigned division to get the remainder,
2118 regardless of whether the arguments are unsigned or not.</p>
2120 <pre> <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2124 <!-- _______________________________________________________________________ -->
2125 <div class="doc_subsubsection"> <a name="i_srem">'<tt>srem</tt>'
2126 Instruction</a> </div>
2127 <div class="doc_text">
2129 <pre> <result> = srem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2132 <p>The '<tt>srem</tt>' instruction returns the remainder from the
2133 signed division of its two operands. This instruction can also take
2134 <a href="#t_vector">vector</a> versions of the values in which case
2135 the elements must be integers.</p>
2138 <p>The two arguments to the '<tt>srem</tt>' instruction must be
2139 <a href="#t_integer">integer</a> values. Both arguments must have identical
2142 <p>This instruction returns the <i>remainder</i> of a division (where the result
2143 has the same sign as the dividend, <tt>var1</tt>), not the <i>modulo</i>
2144 operator (where the result has the same sign as the divisor, <tt>var2</tt>) of
2145 a value. For more information about the difference, see <a
2146 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
2147 Math Forum</a>. For a table of how this is implemented in various languages,
2148 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
2149 Wikipedia: modulo operation</a>.</p>
2151 <pre> <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2155 <!-- _______________________________________________________________________ -->
2156 <div class="doc_subsubsection"> <a name="i_frem">'<tt>frem</tt>'
2157 Instruction</a> </div>
2158 <div class="doc_text">
2160 <pre> <result> = frem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2163 <p>The '<tt>frem</tt>' instruction returns the remainder from the
2164 division of its two operands.</p>
2166 <p>The two arguments to the '<tt>frem</tt>' instruction must be
2167 <a href="#t_floating">floating point</a> values. Both arguments must have
2168 identical types. This instruction can also take <a href="#t_vector">vector</a>
2169 versions of floating point values.</p>
2171 <p>This instruction returns the <i>remainder</i> of a division.</p>
2173 <pre> <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
2177 <!-- ======================================================================= -->
2178 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
2179 Operations</a> </div>
2180 <div class="doc_text">
2181 <p>Bitwise binary operators are used to do various forms of
2182 bit-twiddling in a program. They are generally very efficient
2183 instructions and can commonly be strength reduced from other
2184 instructions. They require two operands, execute an operation on them,
2185 and produce a single value. The resulting value of the bitwise binary
2186 operators is always the same type as its first operand.</p>
2189 <!-- _______________________________________________________________________ -->
2190 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
2191 Instruction</a> </div>
2192 <div class="doc_text">
2194 <pre> <result> = shl <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2199 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2200 the left a specified number of bits.</p>
2204 <p>Both arguments to the '<tt>shl</tt>' instruction must be the same <a
2205 href="#t_integer">integer</a> type.</p>
2209 <p>The value produced is <tt>var1</tt> * 2<sup><tt>var2</tt></sup>. If
2210 <tt>var2</tt> is (statically or dynamically) equal to or larger than the number
2211 of bits in <tt>var1</tt>, the result is undefined.</p>
2213 <h5>Example:</h5><pre>
2214 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
2215 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
2216 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
2217 <result> = shl i32 1, 32 <i>; undefined</i>
2220 <!-- _______________________________________________________________________ -->
2221 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
2222 Instruction</a> </div>
2223 <div class="doc_text">
2225 <pre> <result> = lshr <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2229 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
2230 operand shifted to the right a specified number of bits with zero fill.</p>
2233 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
2234 <a href="#t_integer">integer</a> type.</p>
2238 <p>This instruction always performs a logical shift right operation. The most
2239 significant bits of the result will be filled with zero bits after the
2240 shift. If <tt>var2</tt> is (statically or dynamically) equal to or larger than
2241 the number of bits in <tt>var1</tt>, the result is undefined.</p>
2245 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
2246 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
2247 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
2248 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
2249 <result> = lshr i32 1, 32 <i>; undefined</i>
2253 <!-- _______________________________________________________________________ -->
2254 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
2255 Instruction</a> </div>
2256 <div class="doc_text">
2259 <pre> <result> = ashr <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2263 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
2264 operand shifted to the right a specified number of bits with sign extension.</p>
2267 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
2268 <a href="#t_integer">integer</a> type.</p>
2271 <p>This instruction always performs an arithmetic shift right operation,
2272 The most significant bits of the result will be filled with the sign bit
2273 of <tt>var1</tt>. If <tt>var2</tt> is (statically or dynamically) equal to or
2274 larger than the number of bits in <tt>var1</tt>, the result is undefined.
2279 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
2280 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
2281 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
2282 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
2283 <result> = ashr i32 1, 32 <i>; undefined</i>
2287 <!-- _______________________________________________________________________ -->
2288 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
2289 Instruction</a> </div>
2290 <div class="doc_text">
2292 <pre> <result> = and <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2295 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
2296 its two operands.</p>
2298 <p>The two arguments to the '<tt>and</tt>' instruction must be <a
2299 href="#t_integer">integer</a> values. Both arguments must have
2300 identical types.</p>
2302 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
2304 <div style="align: center">
2305 <table border="1" cellspacing="0" cellpadding="4">
2336 <pre> <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
2337 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
2338 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
2341 <!-- _______________________________________________________________________ -->
2342 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
2343 <div class="doc_text">
2345 <pre> <result> = or <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2348 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
2349 or of its two operands.</p>
2351 <p>The two arguments to the '<tt>or</tt>' instruction must be <a
2352 href="#t_integer">integer</a> values. Both arguments must have
2353 identical types.</p>
2355 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
2357 <div style="align: center">
2358 <table border="1" cellspacing="0" cellpadding="4">
2389 <pre> <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
2390 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
2391 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
2394 <!-- _______________________________________________________________________ -->
2395 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
2396 Instruction</a> </div>
2397 <div class="doc_text">
2399 <pre> <result> = xor <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2402 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
2403 or of its two operands. The <tt>xor</tt> is used to implement the
2404 "one's complement" operation, which is the "~" operator in C.</p>
2406 <p>The two arguments to the '<tt>xor</tt>' instruction must be <a
2407 href="#t_integer">integer</a> values. Both arguments must have
2408 identical types.</p>
2410 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
2412 <div style="align: center">
2413 <table border="1" cellspacing="0" cellpadding="4">
2445 <pre> <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
2446 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
2447 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
2448 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
2452 <!-- ======================================================================= -->
2453 <div class="doc_subsection">
2454 <a name="vectorops">Vector Operations</a>
2457 <div class="doc_text">
2459 <p>LLVM supports several instructions to represent vector operations in a
2460 target-independent manner. These instructions cover the element-access and
2461 vector-specific operations needed to process vectors effectively. While LLVM
2462 does directly support these vector operations, many sophisticated algorithms
2463 will want to use target-specific intrinsics to take full advantage of a specific
2468 <!-- _______________________________________________________________________ -->
2469 <div class="doc_subsubsection">
2470 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2473 <div class="doc_text">
2478 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
2484 The '<tt>extractelement</tt>' instruction extracts a single scalar
2485 element from a vector at a specified index.
2492 The first operand of an '<tt>extractelement</tt>' instruction is a
2493 value of <a href="#t_vector">vector</a> type. The second operand is
2494 an index indicating the position from which to extract the element.
2495 The index may be a variable.</p>
2500 The result is a scalar of the same type as the element type of
2501 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2502 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2503 results are undefined.
2509 %result = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
2514 <!-- _______________________________________________________________________ -->
2515 <div class="doc_subsubsection">
2516 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
2519 <div class="doc_text">
2524 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
2530 The '<tt>insertelement</tt>' instruction inserts a scalar
2531 element into a vector at a specified index.
2538 The first operand of an '<tt>insertelement</tt>' instruction is a
2539 value of <a href="#t_vector">vector</a> type. The second operand is a
2540 scalar value whose type must equal the element type of the first
2541 operand. The third operand is an index indicating the position at
2542 which to insert the value. The index may be a variable.</p>
2547 The result is a vector of the same type as <tt>val</tt>. Its
2548 element values are those of <tt>val</tt> except at position
2549 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
2550 exceeds the length of <tt>val</tt>, the results are undefined.
2556 %result = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
2560 <!-- _______________________________________________________________________ -->
2561 <div class="doc_subsubsection">
2562 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
2565 <div class="doc_text">
2570 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <n x i32> <mask> <i>; yields <n x <ty>></i>
2576 The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
2577 from two input vectors, returning a vector of the same type.
2583 The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
2584 with types that match each other and types that match the result of the
2585 instruction. The third argument is a shuffle mask, which has the same number
2586 of elements as the other vector type, but whose element type is always 'i32'.
2590 The shuffle mask operand is required to be a constant vector with either
2591 constant integer or undef values.
2597 The elements of the two input vectors are numbered from left to right across
2598 both of the vectors. The shuffle mask operand specifies, for each element of
2599 the result vector, which element of the two input registers the result element
2600 gets. The element selector may be undef (meaning "don't care") and the second
2601 operand may be undef if performing a shuffle from only one vector.
2607 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
2608 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
2609 %result = shufflevector <4 x i32> %v1, <4 x i32> undef,
2610 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
2615 <!-- ======================================================================= -->
2616 <div class="doc_subsection">
2617 <a name="memoryops">Memory Access and Addressing Operations</a>
2620 <div class="doc_text">
2622 <p>A key design point of an SSA-based representation is how it
2623 represents memory. In LLVM, no memory locations are in SSA form, which
2624 makes things very simple. This section describes how to read, write,
2625 allocate, and free memory in LLVM.</p>
2629 <!-- _______________________________________________________________________ -->
2630 <div class="doc_subsubsection">
2631 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
2634 <div class="doc_text">
2639 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2644 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
2645 heap and returns a pointer to it.</p>
2649 <p>The '<tt>malloc</tt>' instruction allocates
2650 <tt>sizeof(<type>)*NumElements</tt>
2651 bytes of memory from the operating system and returns a pointer of the
2652 appropriate type to the program. If "NumElements" is specified, it is the
2653 number of elements allocated. If an alignment is specified, the value result
2654 of the allocation is guaranteed to be aligned to at least that boundary. If
2655 not specified, or if zero, the target can choose to align the allocation on any
2656 convenient boundary.</p>
2658 <p>'<tt>type</tt>' must be a sized type.</p>
2662 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
2663 a pointer is returned.</p>
2668 %array = malloc [4 x i8 ] <i>; yields {[%4 x i8]*}:array</i>
2670 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
2671 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
2672 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
2673 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
2674 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
2678 <!-- _______________________________________________________________________ -->
2679 <div class="doc_subsubsection">
2680 <a name="i_free">'<tt>free</tt>' Instruction</a>
2683 <div class="doc_text">
2688 free <type> <value> <i>; yields {void}</i>
2693 <p>The '<tt>free</tt>' instruction returns memory back to the unused
2694 memory heap to be reallocated in the future.</p>
2698 <p>'<tt>value</tt>' shall be a pointer value that points to a value
2699 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
2704 <p>Access to the memory pointed to by the pointer is no longer defined
2705 after this instruction executes.</p>
2710 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
2711 free [4 x i8]* %array
2715 <!-- _______________________________________________________________________ -->
2716 <div class="doc_subsubsection">
2717 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
2720 <div class="doc_text">
2725 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2730 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
2731 currently executing function, to be automatically released when this function
2732 returns to its caller.</p>
2736 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
2737 bytes of memory on the runtime stack, returning a pointer of the
2738 appropriate type to the program. If "NumElements" is specified, it is the
2739 number of elements allocated. If an alignment is specified, the value result
2740 of the allocation is guaranteed to be aligned to at least that boundary. If
2741 not specified, or if zero, the target can choose to align the allocation on any
2742 convenient boundary.</p>
2744 <p>'<tt>type</tt>' may be any sized type.</p>
2748 <p>Memory is allocated; a pointer is returned. '<tt>alloca</tt>'d
2749 memory is automatically released when the function returns. The '<tt>alloca</tt>'
2750 instruction is commonly used to represent automatic variables that must
2751 have an address available. When the function returns (either with the <tt><a
2752 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
2753 instructions), the memory is reclaimed.</p>
2758 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
2759 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
2760 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
2761 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
2765 <!-- _______________________________________________________________________ -->
2766 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
2767 Instruction</a> </div>
2768 <div class="doc_text">
2770 <pre> <result> = load <ty>* <pointer>[, align <alignment>]<br> <result> = volatile load <ty>* <pointer>[, align <alignment>]<br></pre>
2772 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
2774 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
2775 address from which to load. The pointer must point to a <a
2776 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
2777 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
2778 the number or order of execution of this <tt>load</tt> with other
2779 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
2782 <p>The location of memory pointed to is loaded.</p>
2784 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
2786 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
2787 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
2790 <!-- _______________________________________________________________________ -->
2791 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
2792 Instruction</a> </div>
2793 <div class="doc_text">
2795 <pre> store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
2796 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
2799 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
2801 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
2802 to store and an address at which to store it. The type of the '<tt><pointer></tt>'
2803 operand must be a pointer to the type of the '<tt><value></tt>'
2804 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
2805 optimizer is not allowed to modify the number or order of execution of
2806 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
2807 href="#i_store">store</a></tt> instructions.</p>
2809 <p>The contents of memory are updated to contain '<tt><value></tt>'
2810 at the location specified by the '<tt><pointer></tt>' operand.</p>
2812 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
2813 store i32 3, i32* %ptr <i>; yields {void}</i>
2814 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
2818 <!-- _______________________________________________________________________ -->
2819 <div class="doc_subsubsection">
2820 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
2823 <div class="doc_text">
2826 <result> = getelementptr <ty>* <ptrval>{, <ty> <idx>}*
2832 The '<tt>getelementptr</tt>' instruction is used to get the address of a
2833 subelement of an aggregate data structure.</p>
2837 <p>This instruction takes a list of integer operands that indicate what
2838 elements of the aggregate object to index to. The actual types of the arguments
2839 provided depend on the type of the first pointer argument. The
2840 '<tt>getelementptr</tt>' instruction is used to index down through the type
2841 levels of a structure or to a specific index in an array. When indexing into a
2842 structure, only <tt>i32</tt> integer constants are allowed. When indexing
2843 into an array or pointer, only integers of 32 or 64 bits are allowed, and will
2844 be sign extended to 64-bit values.</p>
2846 <p>For example, let's consider a C code fragment and how it gets
2847 compiled to LLVM:</p>
2849 <div class="doc_code">
2862 int *foo(struct ST *s) {
2863 return &s[1].Z.B[5][13];
2868 <p>The LLVM code generated by the GCC frontend is:</p>
2870 <div class="doc_code">
2872 %RT = type { i8 , [10 x [20 x i32]], i8 }
2873 %ST = type { i32, double, %RT }
2875 define i32* %foo(%ST* %s) {
2877 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
2885 <p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
2886 on the pointer type that is being indexed into. <a href="#t_pointer">Pointer</a>
2887 and <a href="#t_array">array</a> types can use a 32-bit or 64-bit
2888 <a href="#t_integer">integer</a> type but the value will always be sign extended
2889 to 64-bits. <a href="#t_struct">Structure</a> types require <tt>i32</tt>
2890 <b>constants</b>.</p>
2892 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
2893 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
2894 }</tt>' type, a structure. The second index indexes into the third element of
2895 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
2896 i8 }</tt>' type, another structure. The third index indexes into the second
2897 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
2898 array. The two dimensions of the array are subscripted into, yielding an
2899 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
2900 to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
2902 <p>Note that it is perfectly legal to index partially through a
2903 structure, returning a pointer to an inner element. Because of this,
2904 the LLVM code for the given testcase is equivalent to:</p>
2907 define i32* %foo(%ST* %s) {
2908 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
2909 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
2910 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
2911 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
2912 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
2917 <p>Note that it is undefined to access an array out of bounds: array and
2918 pointer indexes must always be within the defined bounds of the array type.
2919 The one exception for this rules is zero length arrays. These arrays are
2920 defined to be accessible as variable length arrays, which requires access
2921 beyond the zero'th element.</p>
2923 <p>The getelementptr instruction is often confusing. For some more insight
2924 into how it works, see <a href="GetElementPtr.html">the getelementptr
2930 <i>; yields [12 x i8]*:aptr</i>
2931 %aptr = getelementptr {i32, [12 x i8]}* %sptr, i64 0, i32 1
2935 <!-- ======================================================================= -->
2936 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
2938 <div class="doc_text">
2939 <p>The instructions in this category are the conversion instructions (casting)
2940 which all take a single operand and a type. They perform various bit conversions
2944 <!-- _______________________________________________________________________ -->
2945 <div class="doc_subsubsection">
2946 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
2948 <div class="doc_text">
2952 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
2957 The '<tt>trunc</tt>' instruction truncates its operand to the type <tt>ty2</tt>.
2962 The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
2963 be an <a href="#t_integer">integer</a> type, and a type that specifies the size
2964 and type of the result, which must be an <a href="#t_integer">integer</a>
2965 type. The bit size of <tt>value</tt> must be larger than the bit size of
2966 <tt>ty2</tt>. Equal sized types are not allowed.</p>
2970 The '<tt>trunc</tt>' instruction truncates the high order bits in <tt>value</tt>
2971 and converts the remaining bits to <tt>ty2</tt>. Since the source size must be
2972 larger than the destination size, <tt>trunc</tt> cannot be a <i>no-op cast</i>.
2973 It will always truncate bits.</p>
2977 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
2978 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
2979 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
2983 <!-- _______________________________________________________________________ -->
2984 <div class="doc_subsubsection">
2985 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
2987 <div class="doc_text">
2991 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
2995 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
3000 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
3001 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3002 also be of <a href="#t_integer">integer</a> type. The bit size of the
3003 <tt>value</tt> must be smaller than the bit size of the destination type,
3007 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
3008 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
3010 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
3014 %X = zext i32 257 to i64 <i>; yields i64:257</i>
3015 %Y = zext i1 true to i32 <i>; yields i32:1</i>
3019 <!-- _______________________________________________________________________ -->
3020 <div class="doc_subsubsection">
3021 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
3023 <div class="doc_text">
3027 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
3031 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
3035 The '<tt>sext</tt>' instruction takes a value to cast, which must be of
3036 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3037 also be of <a href="#t_integer">integer</a> type. The bit size of the
3038 <tt>value</tt> must be smaller than the bit size of the destination type,
3043 The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
3044 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size of
3045 the type <tt>ty2</tt>.</p>
3047 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
3051 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
3052 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
3056 <!-- _______________________________________________________________________ -->
3057 <div class="doc_subsubsection">
3058 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
3061 <div class="doc_text">
3066 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
3070 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
3075 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
3076 point</a> value to cast and a <a href="#t_floating">floating point</a> type to
3077 cast it to. The size of <tt>value</tt> must be larger than the size of
3078 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
3079 <i>no-op cast</i>.</p>
3082 <p> The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
3083 <a href="#t_floating">floating point</a> type to a smaller
3084 <a href="#t_floating">floating point</a> type. If the value cannot fit within
3085 the destination type, <tt>ty2</tt>, then the results are undefined.</p>
3089 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
3090 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
3094 <!-- _______________________________________________________________________ -->
3095 <div class="doc_subsubsection">
3096 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
3098 <div class="doc_text">
3102 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
3106 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
3107 floating point value.</p>
3110 <p>The '<tt>fpext</tt>' instruction takes a
3111 <a href="#t_floating">floating point</a> <tt>value</tt> to cast,
3112 and a <a href="#t_floating">floating point</a> type to cast it to. The source
3113 type must be smaller than the destination type.</p>
3116 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
3117 <a href="#t_floating">floating point</a> type to a larger
3118 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
3119 used to make a <i>no-op cast</i> because it always changes bits. Use
3120 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
3124 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
3125 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
3129 <!-- _______________________________________________________________________ -->
3130 <div class="doc_subsubsection">
3131 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
3133 <div class="doc_text">
3137 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
3141 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
3142 unsigned integer equivalent of type <tt>ty2</tt>.
3146 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
3147 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3148 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3149 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3150 vector integer type with the same number of elements as <tt>ty</tt></p>
3153 <p> The '<tt>fptoui</tt>' instruction converts its
3154 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3155 towards zero) unsigned integer value. If the value cannot fit in <tt>ty2</tt>,
3156 the results are undefined.</p>
3160 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
3161 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
3162 %X = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
3166 <!-- _______________________________________________________________________ -->
3167 <div class="doc_subsubsection">
3168 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
3170 <div class="doc_text">
3174 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
3178 <p>The '<tt>fptosi</tt>' instruction converts
3179 <a href="#t_floating">floating point</a> <tt>value</tt> to type <tt>ty2</tt>.
3183 <p> The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
3184 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3185 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3186 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3187 vector integer type with the same number of elements as <tt>ty</tt></p>
3190 <p>The '<tt>fptosi</tt>' instruction converts its
3191 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3192 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
3193 the results are undefined.</p>
3197 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
3198 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
3199 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
3203 <!-- _______________________________________________________________________ -->
3204 <div class="doc_subsubsection">
3205 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
3207 <div class="doc_text">
3211 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3215 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
3216 integer and converts that value to the <tt>ty2</tt> type.</p>
3219 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
3220 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3221 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3222 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3223 floating point type with the same number of elements as <tt>ty</tt></p>
3226 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
3227 integer quantity and converts it to the corresponding floating point value. If
3228 the value cannot fit in the floating point value, the results are undefined.</p>
3232 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
3233 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
3237 <!-- _______________________________________________________________________ -->
3238 <div class="doc_subsubsection">
3239 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
3241 <div class="doc_text">
3245 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3249 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed
3250 integer and converts that value to the <tt>ty2</tt> type.</p>
3253 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
3254 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3255 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3256 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3257 floating point type with the same number of elements as <tt>ty</tt></p>
3260 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed
3261 integer quantity and converts it to the corresponding floating point value. If
3262 the value cannot fit in the floating point value, the results are undefined.</p>
3266 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
3267 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
3271 <!-- _______________________________________________________________________ -->
3272 <div class="doc_subsubsection">
3273 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
3275 <div class="doc_text">
3279 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
3283 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
3284 the integer type <tt>ty2</tt>.</p>
3287 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
3288 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
3289 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.
3292 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
3293 <tt>ty2</tt> by interpreting the pointer value as an integer and either
3294 truncating or zero extending that value to the size of the integer type. If
3295 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
3296 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
3297 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
3302 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
3303 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
3307 <!-- _______________________________________________________________________ -->
3308 <div class="doc_subsubsection">
3309 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
3311 <div class="doc_text">
3315 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
3319 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to
3320 a pointer type, <tt>ty2</tt>.</p>
3323 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
3324 value to cast, and a type to cast it to, which must be a
3325 <a href="#t_pointer">pointer</a> type.
3328 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
3329 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
3330 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
3331 size of a pointer then a truncation is done. If <tt>value</tt> is smaller than
3332 the size of a pointer then a zero extension is done. If they are the same size,
3333 nothing is done (<i>no-op cast</i>).</p>
3337 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
3338 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
3339 %Y = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
3343 <!-- _______________________________________________________________________ -->
3344 <div class="doc_subsubsection">
3345 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
3347 <div class="doc_text">
3351 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
3355 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3356 <tt>ty2</tt> without changing any bits.</p>
3359 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be
3360 a first class value, and a type to cast it to, which must also be a <a
3361 href="#t_firstclass">first class</a> type. The bit sizes of <tt>value</tt>
3362 and the destination type, <tt>ty2</tt>, must be identical. If the source
3363 type is a pointer, the destination type must also be a pointer.</p>
3366 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3367 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
3368 this conversion. The conversion is done as if the <tt>value</tt> had been
3369 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only be
3370 converted to other pointer types with this instruction. To convert pointers to
3371 other types, use the <a href="#i_inttoptr">inttoptr</a> or
3372 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
3376 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
3377 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
3378 %Z = bitcast <2xint> %V to i64; <i>; yields i64: %V</i>
3382 <!-- ======================================================================= -->
3383 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
3384 <div class="doc_text">
3385 <p>The instructions in this category are the "miscellaneous"
3386 instructions, which defy better classification.</p>
3389 <!-- _______________________________________________________________________ -->
3390 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
3392 <div class="doc_text">
3394 <pre> <result> = icmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3397 <p>The '<tt>icmp</tt>' instruction returns a boolean value based on comparison
3398 of its two integer operands.</p>
3400 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
3401 the condition code indicating the kind of comparison to perform. It is not
3402 a value, just a keyword. The possible condition code are:
3404 <li><tt>eq</tt>: equal</li>
3405 <li><tt>ne</tt>: not equal </li>
3406 <li><tt>ugt</tt>: unsigned greater than</li>
3407 <li><tt>uge</tt>: unsigned greater or equal</li>
3408 <li><tt>ult</tt>: unsigned less than</li>
3409 <li><tt>ule</tt>: unsigned less or equal</li>
3410 <li><tt>sgt</tt>: signed greater than</li>
3411 <li><tt>sge</tt>: signed greater or equal</li>
3412 <li><tt>slt</tt>: signed less than</li>
3413 <li><tt>sle</tt>: signed less or equal</li>
3415 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
3416 <a href="#t_pointer">pointer</a> typed. They must also be identical types.</p>
3418 <p>The '<tt>icmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3419 the condition code given as <tt>cond</tt>. The comparison performed always
3420 yields a <a href="#t_primitive">i1</a> result, as follows:
3422 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
3423 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3425 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
3426 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3427 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
3428 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3429 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
3430 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3431 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
3432 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3433 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
3434 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3435 <li><tt>sgt</tt>: interprets the operands as signed values and yields
3436 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3437 <li><tt>sge</tt>: interprets the operands as signed values and yields
3438 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3439 <li><tt>slt</tt>: interprets the operands as signed values and yields
3440 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3441 <li><tt>sle</tt>: interprets the operands as signed values and yields
3442 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3444 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
3445 values are compared as if they were integers.</p>
3448 <pre> <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
3449 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
3450 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
3451 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
3452 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
3453 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
3457 <!-- _______________________________________________________________________ -->
3458 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
3460 <div class="doc_text">
3462 <pre> <result> = fcmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3465 <p>The '<tt>fcmp</tt>' instruction returns a boolean value based on comparison
3466 of its floating point operands.</p>
3468 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
3469 the condition code indicating the kind of comparison to perform. It is not
3470 a value, just a keyword. The possible condition code are:
3472 <li><tt>false</tt>: no comparison, always returns false</li>
3473 <li><tt>oeq</tt>: ordered and equal</li>
3474 <li><tt>ogt</tt>: ordered and greater than </li>
3475 <li><tt>oge</tt>: ordered and greater than or equal</li>
3476 <li><tt>olt</tt>: ordered and less than </li>
3477 <li><tt>ole</tt>: ordered and less than or equal</li>
3478 <li><tt>one</tt>: ordered and not equal</li>
3479 <li><tt>ord</tt>: ordered (no nans)</li>
3480 <li><tt>ueq</tt>: unordered or equal</li>
3481 <li><tt>ugt</tt>: unordered or greater than </li>
3482 <li><tt>uge</tt>: unordered or greater than or equal</li>
3483 <li><tt>ult</tt>: unordered or less than </li>
3484 <li><tt>ule</tt>: unordered or less than or equal</li>
3485 <li><tt>une</tt>: unordered or not equal</li>
3486 <li><tt>uno</tt>: unordered (either nans)</li>
3487 <li><tt>true</tt>: no comparison, always returns true</li>
3489 <p><i>Ordered</i> means that neither operand is a QNAN while
3490 <i>unordered</i> means that either operand may be a QNAN.</p>
3491 <p>The <tt>val1</tt> and <tt>val2</tt> arguments must be
3492 <a href="#t_floating">floating point</a> typed. They must have identical
3495 <p>The '<tt>fcmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3496 the condition code given as <tt>cond</tt>. The comparison performed always
3497 yields a <a href="#t_primitive">i1</a> result, as follows:
3499 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
3500 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3501 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3502 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3503 <tt>var1</tt> is greather than <tt>var2</tt>.</li>
3504 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3505 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3506 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3507 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3508 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3509 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3510 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3511 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3512 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
3513 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
3514 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3515 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
3516 <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3517 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
3518 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3519 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
3520 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3521 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
3522 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3523 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
3524 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3525 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
3526 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
3530 <pre> <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
3531 <result> = icmp one float 4.0, 5.0 <i>; yields: result=true</i>
3532 <result> = icmp olt float 4.0, 5.0 <i>; yields: result=true</i>
3533 <result> = icmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
3537 <!-- _______________________________________________________________________ -->
3538 <div class="doc_subsubsection"> <a name="i_phi">'<tt>phi</tt>'
3539 Instruction</a> </div>
3540 <div class="doc_text">
3542 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
3544 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
3545 the SSA graph representing the function.</p>
3547 <p>The type of the incoming values is specified with the first type
3548 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
3549 as arguments, with one pair for each predecessor basic block of the
3550 current block. Only values of <a href="#t_firstclass">first class</a>
3551 type may be used as the value arguments to the PHI node. Only labels
3552 may be used as the label arguments.</p>
3553 <p>There must be no non-phi instructions between the start of a basic
3554 block and the PHI instructions: i.e. PHI instructions must be first in
3557 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
3558 specified by the pair corresponding to the predecessor basic block that executed
3559 just prior to the current block.</p>
3561 <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>
3564 <!-- _______________________________________________________________________ -->
3565 <div class="doc_subsubsection">
3566 <a name="i_select">'<tt>select</tt>' Instruction</a>
3569 <div class="doc_text">
3574 <result> = select i1 <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
3580 The '<tt>select</tt>' instruction is used to choose one value based on a
3581 condition, without branching.
3588 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.
3594 If the boolean condition evaluates to true, the instruction returns the first
3595 value argument; otherwise, it returns the second value argument.
3601 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
3606 <!-- _______________________________________________________________________ -->
3607 <div class="doc_subsubsection">
3608 <a name="i_call">'<tt>call</tt>' Instruction</a>
3611 <div class="doc_text">
3615 <result> = [tail] call [<a href="#callingconv">cconv</a>] <ty> [<fnty>*] <fnptrval>(<param list>)
3620 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
3624 <p>This instruction requires several arguments:</p>
3628 <p>The optional "tail" marker indicates whether the callee function accesses
3629 any allocas or varargs in the caller. If the "tail" marker is present, the
3630 function call is eligible for tail call optimization. Note that calls may
3631 be marked "tail" even if they do not occur before a <a
3632 href="#i_ret"><tt>ret</tt></a> instruction.
3635 <p>The optional "cconv" marker indicates which <a href="#callingconv">calling
3636 convention</a> the call should use. If none is specified, the call defaults
3637 to using C calling conventions.
3640 <p>'<tt>ty</tt>': the type of the call instruction itself which is also
3641 the type of the return value. Functions that return no value are marked
3642 <tt><a href="#t_void">void</a></tt>.</p>
3645 <p>'<tt>fnty</tt>': shall be the signature of the pointer to function
3646 value being invoked. The argument types must match the types implied by
3647 this signature. This type can be omitted if the function is not varargs
3648 and if the function type does not return a pointer to a function.</p>
3651 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
3652 be invoked. In most cases, this is a direct function invocation, but
3653 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
3654 to function value.</p>
3657 <p>'<tt>function args</tt>': argument list whose types match the
3658 function signature argument types. All arguments must be of
3659 <a href="#t_firstclass">first class</a> type. If the function signature
3660 indicates the function accepts a variable number of arguments, the extra
3661 arguments can be specified.</p>
3667 <p>The '<tt>call</tt>' instruction is used to cause control flow to
3668 transfer to a specified function, with its incoming arguments bound to
3669 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
3670 instruction in the called function, control flow continues with the
3671 instruction after the function call, and the return value of the
3672 function is bound to the result argument. This is a simpler case of
3673 the <a href="#i_invoke">invoke</a> instruction.</p>
3678 %retval = call i32 @test(i32 %argc)
3679 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42);
3680 %X = tail call i32 @foo()
3681 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo()
3682 %Z = call void %foo(i8 97 signext)
3687 <!-- _______________________________________________________________________ -->
3688 <div class="doc_subsubsection">
3689 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
3692 <div class="doc_text">
3697 <resultval> = va_arg <va_list*> <arglist>, <argty>
3702 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
3703 the "variable argument" area of a function call. It is used to implement the
3704 <tt>va_arg</tt> macro in C.</p>
3708 <p>This instruction takes a <tt>va_list*</tt> value and the type of
3709 the argument. It returns a value of the specified argument type and
3710 increments the <tt>va_list</tt> to point to the next argument. The
3711 actual type of <tt>va_list</tt> is target specific.</p>
3715 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
3716 type from the specified <tt>va_list</tt> and causes the
3717 <tt>va_list</tt> to point to the next argument. For more information,
3718 see the variable argument handling <a href="#int_varargs">Intrinsic
3721 <p>It is legal for this instruction to be called in a function which does not
3722 take a variable number of arguments, for example, the <tt>vfprintf</tt>
3725 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
3726 href="#intrinsics">intrinsic function</a> because it takes a type as an
3731 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
3735 <!-- *********************************************************************** -->
3736 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
3737 <!-- *********************************************************************** -->
3739 <div class="doc_text">
3741 <p>LLVM supports the notion of an "intrinsic function". These functions have
3742 well known names and semantics and are required to follow certain restrictions.
3743 Overall, these intrinsics represent an extension mechanism for the LLVM
3744 language that does not require changing all of the transformations in LLVM when
3745 adding to the language (or the bitcode reader/writer, the parser, etc...).</p>
3747 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
3748 prefix is reserved in LLVM for intrinsic names; thus, function names may not
3749 begin with this prefix. Intrinsic functions must always be external functions:
3750 you cannot define the body of intrinsic functions. Intrinsic functions may
3751 only be used in call or invoke instructions: it is illegal to take the address
3752 of an intrinsic function. Additionally, because intrinsic functions are part
3753 of the LLVM language, it is required if any are added that they be documented
3756 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
3757 a family of functions that perform the same operation but on different data
3758 types. Because LLVM can represent over 8 million different integer types,
3759 overloading is used commonly to allow an intrinsic function to operate on any
3760 integer type. One or more of the argument types or the result type can be
3761 overloaded to accept any integer type. Argument types may also be defined as
3762 exactly matching a previous argument's type or the result type. This allows an
3763 intrinsic function which accepts multiple arguments, but needs all of them to
3764 be of the same type, to only be overloaded with respect to a single argument or
3767 <p>Overloaded intrinsics will have the names of its overloaded argument types
3768 encoded into its function name, each preceded by a period. Only those types
3769 which are overloaded result in a name suffix. Arguments whose type is matched
3770 against another type do not. For example, the <tt>llvm.ctpop</tt> function can
3771 take an integer of any width and returns an integer of exactly the same integer
3772 width. This leads to a family of functions such as
3773 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29 %val)</tt>.
3774 Only one type, the return type, is overloaded, and only one type suffix is
3775 required. Because the argument's type is matched against the return type, it
3776 does not require its own name suffix.</p>
3778 <p>To learn how to add an intrinsic function, please see the
3779 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.
3784 <!-- ======================================================================= -->
3785 <div class="doc_subsection">
3786 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
3789 <div class="doc_text">
3791 <p>Variable argument support is defined in LLVM with the <a
3792 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
3793 intrinsic functions. These functions are related to the similarly
3794 named macros defined in the <tt><stdarg.h></tt> header file.</p>
3796 <p>All of these functions operate on arguments that use a
3797 target-specific value type "<tt>va_list</tt>". The LLVM assembly
3798 language reference manual does not define what this type is, so all
3799 transformations should be prepared to handle these functions regardless of
3802 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
3803 instruction and the variable argument handling intrinsic functions are
3806 <div class="doc_code">
3808 define i32 @test(i32 %X, ...) {
3809 ; Initialize variable argument processing
3811 %ap2 = bitcast i8** %ap to i8*
3812 call void @llvm.va_start(i8* %ap2)
3814 ; Read a single integer argument
3815 %tmp = va_arg i8** %ap, i32
3817 ; Demonstrate usage of llvm.va_copy and llvm.va_end
3819 %aq2 = bitcast i8** %aq to i8*
3820 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
3821 call void @llvm.va_end(i8* %aq2)
3823 ; Stop processing of arguments.
3824 call void @llvm.va_end(i8* %ap2)
3828 declare void @llvm.va_start(i8*)
3829 declare void @llvm.va_copy(i8*, i8*)
3830 declare void @llvm.va_end(i8*)
3836 <!-- _______________________________________________________________________ -->
3837 <div class="doc_subsubsection">
3838 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
3842 <div class="doc_text">
3844 <pre> declare void %llvm.va_start(i8* <arglist>)<br></pre>
3846 <P>The '<tt>llvm.va_start</tt>' intrinsic initializes
3847 <tt>*<arglist></tt> for subsequent use by <tt><a
3848 href="#i_va_arg">va_arg</a></tt>.</p>
3852 <P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
3856 <P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
3857 macro available in C. In a target-dependent way, it initializes the
3858 <tt>va_list</tt> element to which the argument points, so that the next call to
3859 <tt>va_arg</tt> will produce the first variable argument passed to the function.
3860 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
3861 last argument of the function as the compiler can figure that out.</p>
3865 <!-- _______________________________________________________________________ -->
3866 <div class="doc_subsubsection">
3867 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
3870 <div class="doc_text">
3872 <pre> declare void @llvm.va_end(i8* <arglist>)<br></pre>
3875 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
3876 which has been initialized previously with <tt><a href="#int_va_start">llvm.va_start</a></tt>
3877 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
3881 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
3885 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
3886 macro available in C. In a target-dependent way, it destroys the
3887 <tt>va_list</tt> element to which the argument points. Calls to <a
3888 href="#int_va_start"><tt>llvm.va_start</tt></a> and <a href="#int_va_copy">
3889 <tt>llvm.va_copy</tt></a> must be matched exactly with calls to
3890 <tt>llvm.va_end</tt>.</p>
3894 <!-- _______________________________________________________________________ -->
3895 <div class="doc_subsubsection">
3896 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
3899 <div class="doc_text">
3904 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
3909 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
3910 from the source argument list to the destination argument list.</p>
3914 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
3915 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
3920 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
3921 macro available in C. In a target-dependent way, it copies the source
3922 <tt>va_list</tt> element into the destination <tt>va_list</tt> element. This
3923 intrinsic is necessary because the <tt><a href="#int_va_start">
3924 llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
3925 example, memory allocation.</p>
3929 <!-- ======================================================================= -->
3930 <div class="doc_subsection">
3931 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
3934 <div class="doc_text">
3937 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
3938 Collection</a> requires the implementation and generation of these intrinsics.
3939 These intrinsics allow identification of <a href="#int_gcroot">GC roots on the
3940 stack</a>, as well as garbage collector implementations that require <a
3941 href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a> barriers.
3942 Front-ends for type-safe garbage collected languages should generate these
3943 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
3944 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
3948 <!-- _______________________________________________________________________ -->
3949 <div class="doc_subsubsection">
3950 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
3953 <div class="doc_text">
3958 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
3963 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
3964 the code generator, and allows some metadata to be associated with it.</p>
3968 <p>The first argument specifies the address of a stack object that contains the
3969 root pointer. The second pointer (which must be either a constant or a global
3970 value address) contains the meta-data to be associated with the root.</p>
3974 <p>At runtime, a call to this intrinsics stores a null pointer into the "ptrloc"
3975 location. At compile-time, the code generator generates information to allow
3976 the runtime to find the pointer at GC safe points.
3982 <!-- _______________________________________________________________________ -->
3983 <div class="doc_subsubsection">
3984 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
3987 <div class="doc_text">
3992 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
3997 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
3998 locations, allowing garbage collector implementations that require read
4003 <p>The second argument is the address to read from, which should be an address
4004 allocated from the garbage collector. The first object is a pointer to the
4005 start of the referenced object, if needed by the language runtime (otherwise
4010 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
4011 instruction, but may be replaced with substantially more complex code by the
4012 garbage collector runtime, as needed.</p>
4017 <!-- _______________________________________________________________________ -->
4018 <div class="doc_subsubsection">
4019 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
4022 <div class="doc_text">
4027 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
4032 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
4033 locations, allowing garbage collector implementations that require write
4034 barriers (such as generational or reference counting collectors).</p>
4038 <p>The first argument is the reference to store, the second is the start of the
4039 object to store it to, and the third is the address of the field of Obj to
4040 store to. If the runtime does not require a pointer to the object, Obj may be
4045 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
4046 instruction, but may be replaced with substantially more complex code by the
4047 garbage collector runtime, as needed.</p>
4053 <!-- ======================================================================= -->
4054 <div class="doc_subsection">
4055 <a name="int_codegen">Code Generator Intrinsics</a>
4058 <div class="doc_text">
4060 These intrinsics are provided by LLVM to expose special features that may only
4061 be implemented with code generator support.
4066 <!-- _______________________________________________________________________ -->
4067 <div class="doc_subsubsection">
4068 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
4071 <div class="doc_text">
4075 declare i8 *@llvm.returnaddress(i32 <level>)
4081 The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
4082 target-specific value indicating the return address of the current function
4083 or one of its callers.
4089 The argument to this intrinsic indicates which function to return the address
4090 for. Zero indicates the calling function, one indicates its caller, etc. The
4091 argument is <b>required</b> to be a constant integer value.
4097 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
4098 the return address of the specified call frame, or zero if it cannot be
4099 identified. The value returned by this intrinsic is likely to be incorrect or 0
4100 for arguments other than zero, so it should only be used for debugging purposes.
4104 Note that calling this intrinsic does not prevent function inlining or other
4105 aggressive transformations, so the value returned may not be that of the obvious
4106 source-language caller.
4111 <!-- _______________________________________________________________________ -->
4112 <div class="doc_subsubsection">
4113 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
4116 <div class="doc_text">
4120 declare i8 *@llvm.frameaddress(i32 <level>)
4126 The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
4127 target-specific frame pointer value for the specified stack frame.
4133 The argument to this intrinsic indicates which function to return the frame
4134 pointer for. Zero indicates the calling function, one indicates its caller,
4135 etc. The argument is <b>required</b> to be a constant integer value.
4141 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
4142 the frame address of the specified call frame, or zero if it cannot be
4143 identified. The value returned by this intrinsic is likely to be incorrect or 0
4144 for arguments other than zero, so it should only be used for debugging purposes.
4148 Note that calling this intrinsic does not prevent function inlining or other
4149 aggressive transformations, so the value returned may not be that of the obvious
4150 source-language caller.
4154 <!-- _______________________________________________________________________ -->
4155 <div class="doc_subsubsection">
4156 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
4159 <div class="doc_text">
4163 declare i8 *@llvm.stacksave()
4169 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
4170 the function stack, for use with <a href="#int_stackrestore">
4171 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
4172 features like scoped automatic variable sized arrays in C99.
4178 This intrinsic returns a opaque pointer value that can be passed to <a
4179 href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
4180 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
4181 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
4182 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
4183 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
4184 that were allocated after the <tt>llvm.stacksave</tt> was executed.
4189 <!-- _______________________________________________________________________ -->
4190 <div class="doc_subsubsection">
4191 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
4194 <div class="doc_text">
4198 declare void @llvm.stackrestore(i8 * %ptr)
4204 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
4205 the function stack to the state it was in when the corresponding <a
4206 href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
4207 useful for implementing language features like scoped automatic variable sized
4214 See the description for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.
4220 <!-- _______________________________________________________________________ -->
4221 <div class="doc_subsubsection">
4222 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
4225 <div class="doc_text">
4229 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>)
4236 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
4237 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
4239 effect on the behavior of the program but can change its performance
4246 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
4247 determining if the fetch should be for a read (0) or write (1), and
4248 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
4249 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
4250 <tt>locality</tt> arguments must be constant integers.
4256 This intrinsic does not modify the behavior of the program. In particular,
4257 prefetches cannot trap and do not produce a value. On targets that support this
4258 intrinsic, the prefetch can provide hints to the processor cache for better
4264 <!-- _______________________________________________________________________ -->
4265 <div class="doc_subsubsection">
4266 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
4269 <div class="doc_text">
4273 declare void @llvm.pcmarker(i32 <id>)
4280 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
4282 code to simulators and other tools. The method is target specific, but it is
4283 expected that the marker will use exported symbols to transmit the PC of the marker.
4284 The marker makes no guarantees that it will remain with any specific instruction
4285 after optimizations. It is possible that the presence of a marker will inhibit
4286 optimizations. The intended use is to be inserted after optimizations to allow
4287 correlations of simulation runs.
4293 <tt>id</tt> is a numerical id identifying the marker.
4299 This intrinsic does not modify the behavior of the program. Backends that do not
4300 support this intrinisic may ignore it.
4305 <!-- _______________________________________________________________________ -->
4306 <div class="doc_subsubsection">
4307 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
4310 <div class="doc_text">
4314 declare i64 @llvm.readcyclecounter( )
4321 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
4322 counter register (or similar low latency, high accuracy clocks) on those targets
4323 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
4324 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
4325 should only be used for small timings.
4331 When directly supported, reading the cycle counter should not modify any memory.
4332 Implementations are allowed to either return a application specific value or a
4333 system wide value. On backends without support, this is lowered to a constant 0.
4338 <!-- ======================================================================= -->
4339 <div class="doc_subsection">
4340 <a name="int_libc">Standard C Library Intrinsics</a>
4343 <div class="doc_text">
4345 LLVM provides intrinsics for a few important standard C library functions.
4346 These intrinsics allow source-language front-ends to pass information about the
4347 alignment of the pointer arguments to the code generator, providing opportunity
4348 for more efficient code generation.
4353 <!-- _______________________________________________________________________ -->
4354 <div class="doc_subsubsection">
4355 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
4358 <div class="doc_text">
4362 declare void @llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
4363 i32 <len>, i32 <align>)
4364 declare void @llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
4365 i64 <len>, i32 <align>)
4371 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4372 location to the destination location.
4376 Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
4377 intrinsics do not return a value, and takes an extra alignment argument.
4383 The first argument is a pointer to the destination, the second is a pointer to
4384 the source. The third argument is an integer argument
4385 specifying the number of bytes to copy, and the fourth argument is the alignment
4386 of the source and destination locations.
4390 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4391 the caller guarantees that both the source and destination pointers are aligned
4398 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4399 location to the destination location, which are not allowed to overlap. It
4400 copies "len" bytes of memory over. If the argument is known to be aligned to
4401 some boundary, this can be specified as the fourth argument, otherwise it should
4407 <!-- _______________________________________________________________________ -->
4408 <div class="doc_subsubsection">
4409 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
4412 <div class="doc_text">
4416 declare void @llvm.memmove.i32(i8 * <dest>, i8 * <src>,
4417 i32 <len>, i32 <align>)
4418 declare void @llvm.memmove.i64(i8 * <dest>, i8 * <src>,
4419 i64 <len>, i32 <align>)
4425 The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
4426 location to the destination location. It is similar to the
4427 '<tt>llvm.memcmp</tt>' intrinsic but allows the two memory locations to overlap.
4431 Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
4432 intrinsics do not return a value, and takes an extra alignment argument.
4438 The first argument is a pointer to the destination, the second is a pointer to
4439 the source. The third argument is an integer argument
4440 specifying the number of bytes to copy, and the fourth argument is the alignment
4441 of the source and destination locations.
4445 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4446 the caller guarantees that the source and destination pointers are aligned to
4453 The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
4454 location to the destination location, which may overlap. It
4455 copies "len" bytes of memory over. If the argument is known to be aligned to
4456 some boundary, this can be specified as the fourth argument, otherwise it should
4462 <!-- _______________________________________________________________________ -->
4463 <div class="doc_subsubsection">
4464 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
4467 <div class="doc_text">
4471 declare void @llvm.memset.i32(i8 * <dest>, i8 <val>,
4472 i32 <len>, i32 <align>)
4473 declare void @llvm.memset.i64(i8 * <dest>, i8 <val>,
4474 i64 <len>, i32 <align>)
4480 The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
4485 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
4486 does not return a value, and takes an extra alignment argument.
4492 The first argument is a pointer to the destination to fill, the second is the
4493 byte value to fill it with, the third argument is an integer
4494 argument specifying the number of bytes to fill, and the fourth argument is the
4495 known alignment of destination location.
4499 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4500 the caller guarantees that the destination pointer is aligned to that boundary.
4506 The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
4508 destination location. If the argument is known to be aligned to some boundary,
4509 this can be specified as the fourth argument, otherwise it should be set to 0 or
4515 <!-- _______________________________________________________________________ -->
4516 <div class="doc_subsubsection">
4517 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
4520 <div class="doc_text">
4523 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
4524 floating point or vector of floating point type. Not all targets support all
4527 declare float @llvm.sqrt.f32(float %Val)
4528 declare double @llvm.sqrt.f64(double %Val)
4529 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
4530 declare fp128 @llvm.sqrt.f128(fp128 %Val)
4531 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
4537 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
4538 returning the same value as the libm '<tt>sqrt</tt>' functions would. Unlike
4539 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
4540 negative numbers (which allows for better optimization).
4546 The argument and return value are floating point numbers of the same type.
4552 This function returns the sqrt of the specified operand if it is a nonnegative
4553 floating point number.
4557 <!-- _______________________________________________________________________ -->
4558 <div class="doc_subsubsection">
4559 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
4562 <div class="doc_text">
4565 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
4566 floating point or vector of floating point type. Not all targets support all
4569 declare float @llvm.powi.f32(float %Val, i32 %power)
4570 declare double @llvm.powi.f64(double %Val, i32 %power)
4571 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
4572 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
4573 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
4579 The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
4580 specified (positive or negative) power. The order of evaluation of
4581 multiplications is not defined. When a vector of floating point type is
4582 used, the second argument remains a scalar integer value.
4588 The second argument is an integer power, and the first is a value to raise to
4595 This function returns the first value raised to the second power with an
4596 unspecified sequence of rounding operations.</p>
4599 <!-- _______________________________________________________________________ -->
4600 <div class="doc_subsubsection">
4601 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
4604 <div class="doc_text">
4607 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
4608 floating point or vector of floating point type. Not all targets support all
4611 declare float @llvm.sin.f32(float %Val)
4612 declare double @llvm.sin.f64(double %Val)
4613 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
4614 declare fp128 @llvm.sin.f128(fp128 %Val)
4615 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
4621 The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.
4627 The argument and return value are floating point numbers of the same type.
4633 This function returns the sine of the specified operand, returning the
4634 same values as the libm <tt>sin</tt> functions would, and handles error
4635 conditions in the same way.</p>
4638 <!-- _______________________________________________________________________ -->
4639 <div class="doc_subsubsection">
4640 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
4643 <div class="doc_text">
4646 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
4647 floating point or vector of floating point type. Not all targets support all
4650 declare float @llvm.cos.f32(float %Val)
4651 declare double @llvm.cos.f64(double %Val)
4652 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
4653 declare fp128 @llvm.cos.f128(fp128 %Val)
4654 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
4660 The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.
4666 The argument and return value are floating point numbers of the same type.
4672 This function returns the cosine of the specified operand, returning the
4673 same values as the libm <tt>cos</tt> functions would, and handles error
4674 conditions in the same way.</p>
4677 <!-- _______________________________________________________________________ -->
4678 <div class="doc_subsubsection">
4679 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
4682 <div class="doc_text">
4685 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
4686 floating point or vector of floating point type. Not all targets support all
4689 declare float @llvm.pow.f32(float %Val, float %Power)
4690 declare double @llvm.pow.f64(double %Val, double %Power)
4691 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
4692 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
4693 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
4699 The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
4700 specified (positive or negative) power.
4706 The second argument is a floating point power, and the first is a value to
4707 raise to that power.
4713 This function returns the first value raised to the second power,
4715 same values as the libm <tt>pow</tt> functions would, and handles error
4716 conditions in the same way.</p>
4720 <!-- ======================================================================= -->
4721 <div class="doc_subsection">
4722 <a name="int_manip">Bit Manipulation Intrinsics</a>
4725 <div class="doc_text">
4727 LLVM provides intrinsics for a few important bit manipulation operations.
4728 These allow efficient code generation for some algorithms.
4733 <!-- _______________________________________________________________________ -->
4734 <div class="doc_subsubsection">
4735 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
4738 <div class="doc_text">
4741 <p>This is an overloaded intrinsic function. You can use bswap on any integer
4742 type that is an even number of bytes (i.e. BitWidth % 16 == 0).
4744 declare i16 @llvm.bswap.i16(i16 <id>)
4745 declare i32 @llvm.bswap.i32(i32 <id>)
4746 declare i64 @llvm.bswap.i64(i64 <id>)
4752 The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
4753 values with an even number of bytes (positive multiple of 16 bits). These are
4754 useful for performing operations on data that is not in the target's native
4761 The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
4762 and low byte of the input i16 swapped. Similarly, the <tt>llvm.bswap.i32</tt>
4763 intrinsic returns an i32 value that has the four bytes of the input i32
4764 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the returned
4765 i32 will have its bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i48</tt>,
4766 <tt>llvm.bswap.i64</tt> and other intrinsics extend this concept to
4767 additional even-byte lengths (6 bytes, 8 bytes and more, respectively).
4772 <!-- _______________________________________________________________________ -->
4773 <div class="doc_subsubsection">
4774 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
4777 <div class="doc_text">
4780 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
4781 width. Not all targets support all bit widths however.
4783 declare i8 @llvm.ctpop.i8 (i8 <src>)
4784 declare i16 @llvm.ctpop.i16(i16 <src>)
4785 declare i32 @llvm.ctpop.i32(i32 <src>)
4786 declare i64 @llvm.ctpop.i64(i64 <src>)
4787 declare i256 @llvm.ctpop.i256(i256 <src>)
4793 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
4800 The only argument is the value to be counted. The argument may be of any
4801 integer type. The return type must match the argument type.
4807 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
4811 <!-- _______________________________________________________________________ -->
4812 <div class="doc_subsubsection">
4813 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
4816 <div class="doc_text">
4819 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
4820 integer bit width. Not all targets support all bit widths however.
4822 declare i8 @llvm.ctlz.i8 (i8 <src>)
4823 declare i16 @llvm.ctlz.i16(i16 <src>)
4824 declare i32 @llvm.ctlz.i32(i32 <src>)
4825 declare i64 @llvm.ctlz.i64(i64 <src>)
4826 declare i256 @llvm.ctlz.i256(i256 <src>)
4832 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
4833 leading zeros in a variable.
4839 The only argument is the value to be counted. The argument may be of any
4840 integer type. The return type must match the argument type.
4846 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
4847 in a variable. If the src == 0 then the result is the size in bits of the type
4848 of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.
4854 <!-- _______________________________________________________________________ -->
4855 <div class="doc_subsubsection">
4856 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
4859 <div class="doc_text">
4862 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
4863 integer bit width. Not all targets support all bit widths however.
4865 declare i8 @llvm.cttz.i8 (i8 <src>)
4866 declare i16 @llvm.cttz.i16(i16 <src>)
4867 declare i32 @llvm.cttz.i32(i32 <src>)
4868 declare i64 @llvm.cttz.i64(i64 <src>)
4869 declare i256 @llvm.cttz.i256(i256 <src>)
4875 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
4882 The only argument is the value to be counted. The argument may be of any
4883 integer type. The return type must match the argument type.
4889 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
4890 in a variable. If the src == 0 then the result is the size in bits of the type
4891 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
4895 <!-- _______________________________________________________________________ -->
4896 <div class="doc_subsubsection">
4897 <a name="int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic</a>
4900 <div class="doc_text">
4903 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.select</tt>
4904 on any integer bit width.
4906 declare i17 @llvm.part.select.i17 (i17 %val, i32 %loBit, i32 %hiBit)
4907 declare i29 @llvm.part.select.i29 (i29 %val, i32 %loBit, i32 %hiBit)
4911 <p>The '<tt>llvm.part.select</tt>' family of intrinsic functions selects a
4912 range of bits from an integer value and returns them in the same bit width as
4913 the original value.</p>
4916 <p>The first argument, <tt>%val</tt> and the result may be integer types of
4917 any bit width but they must have the same bit width. The second and third
4918 arguments must be <tt>i32</tt> type since they specify only a bit index.</p>
4921 <p>The operation of the '<tt>llvm.part.select</tt>' intrinsic has two modes
4922 of operation: forwards and reverse. If <tt>%loBit</tt> is greater than
4923 <tt>%hiBits</tt> then the intrinsic operates in reverse mode. Otherwise it
4924 operates in forward mode.</p>
4925 <p>In forward mode, this intrinsic is the equivalent of shifting <tt>%val</tt>
4926 right by <tt>%loBit</tt> bits and then ANDing it with a mask with
4927 only the <tt>%hiBit - %loBit</tt> bits set, as follows:</p>
4929 <li>The <tt>%val</tt> is shifted right (LSHR) by the number of bits specified
4930 by <tt>%loBits</tt>. This normalizes the value to the low order bits.</li>
4931 <li>The <tt>%loBits</tt> value is subtracted from the <tt>%hiBits</tt> value
4932 to determine the number of bits to retain.</li>
4933 <li>A mask of the retained bits is created by shifting a -1 value.</li>
4934 <li>The mask is ANDed with <tt>%val</tt> to produce the result.
4936 <p>In reverse mode, a similar computation is made except that the bits are
4937 returned in the reverse order. So, for example, if <tt>X</tt> has the value
4938 <tt>i16 0x0ACF (101011001111)</tt> and we apply
4939 <tt>part.select(i16 X, 8, 3)</tt> to it, we get back the value
4940 <tt>i16 0x0026 (000000100110)</tt>.</p>
4943 <div class="doc_subsubsection">
4944 <a name="int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic</a>
4947 <div class="doc_text">
4950 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.set</tt>
4951 on any integer bit width.
4953 declare i17 @llvm.part.set.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
4954 declare i29 @llvm.part.set.i29.i9 (i29 %val, i9 %repl, i32 %lo, i32 %hi)
4958 <p>The '<tt>llvm.part.set</tt>' family of intrinsic functions replaces a range
4959 of bits in an integer value with another integer value. It returns the integer
4960 with the replaced bits.</p>
4963 <p>The first argument, <tt>%val</tt> and the result may be integer types of
4964 any bit width but they must have the same bit width. <tt>%val</tt> is the value
4965 whose bits will be replaced. The second argument, <tt>%repl</tt> may be an
4966 integer of any bit width. The third and fourth arguments must be <tt>i32</tt>
4967 type since they specify only a bit index.</p>
4970 <p>The operation of the '<tt>llvm.part.set</tt>' intrinsic has two modes
4971 of operation: forwards and reverse. If <tt>%lo</tt> is greater than
4972 <tt>%hi</tt> then the intrinsic operates in reverse mode. Otherwise it
4973 operates in forward mode.</p>
4974 <p>For both modes, the <tt>%repl</tt> value is prepared for use by either
4975 truncating it down to the size of the replacement area or zero extending it
4976 up to that size.</p>
4977 <p>In forward mode, the bits between <tt>%lo</tt> and <tt>%hi</tt> (inclusive)
4978 are replaced with corresponding bits from <tt>%repl</tt>. That is the 0th bit
4979 in <tt>%repl</tt> replaces the <tt>%lo</tt>th bit in <tt>%val</tt> and etc. up
4980 to the <tt>%hi</tt>th bit.
4981 <p>In reverse mode, a similar computation is made except that the bits are
4982 reversed. That is, the <tt>0</tt>th bit in <tt>%repl</tt> replaces the
4983 <tt>%hi</tt> bit in <tt>%val</tt> and etc. down to the <tt>%lo</tt>th bit.
4986 llvm.part.set(0xFFFF, 0, 4, 7) -> 0xFF0F
4987 llvm.part.set(0xFFFF, 0, 7, 4) -> 0xFF0F
4988 llvm.part.set(0xFFFF, 1, 7, 4) -> 0xFF8F
4989 llvm.part.set(0xFFFF, F, 8, 3) -> 0xFFE7
4990 llvm.part.set(0xFFFF, 0, 3, 8) -> 0xFE07
4994 <!-- ======================================================================= -->
4995 <div class="doc_subsection">
4996 <a name="int_debugger">Debugger Intrinsics</a>
4999 <div class="doc_text">
5001 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
5002 are described in the <a
5003 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
5004 Debugging</a> document.
5009 <!-- ======================================================================= -->
5010 <div class="doc_subsection">
5011 <a name="int_eh">Exception Handling Intrinsics</a>
5014 <div class="doc_text">
5015 <p> The LLVM exception handling intrinsics (which all start with
5016 <tt>llvm.eh.</tt> prefix), are described in the <a
5017 href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
5018 Handling</a> document. </p>
5021 <!-- ======================================================================= -->
5022 <div class="doc_subsection">
5023 <a name="int_trampoline">Trampoline Intrinsic</a>
5026 <div class="doc_text">
5028 This intrinsic makes it possible to excise one parameter, marked with
5029 the <tt>nest</tt> attribute, from a function. The result is a callable
5030 function pointer lacking the nest parameter - the caller does not need
5031 to provide a value for it. Instead, the value to use is stored in
5032 advance in a "trampoline", a block of memory usually allocated
5033 on the stack, which also contains code to splice the nest value into the
5034 argument list. This is used to implement the GCC nested function address
5038 For example, if the function is
5039 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
5040 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as follows:</p>
5042 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
5043 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
5044 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
5045 %fp = bitcast i8* %p to i32 (i32, i32)*
5047 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
5048 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
5051 <!-- _______________________________________________________________________ -->
5052 <div class="doc_subsubsection">
5053 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
5055 <div class="doc_text">
5058 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
5062 This fills the memory pointed to by <tt>tramp</tt> with code
5063 and returns a function pointer suitable for executing it.
5067 The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
5068 pointers. The <tt>tramp</tt> argument must point to a sufficiently large
5069 and sufficiently aligned block of memory; this memory is written to by the
5070 intrinsic. Note that the size and the alignment are target-specific - LLVM
5071 currently provides no portable way of determining them, so a front-end that
5072 generates this intrinsic needs to have some target-specific knowledge.
5073 The <tt>func</tt> argument must hold a function bitcast to an <tt>i8*</tt>.
5077 The block of memory pointed to by <tt>tramp</tt> is filled with target
5078 dependent code, turning it into a function. A pointer to this function is
5079 returned, but needs to be bitcast to an
5080 <a href="#int_trampoline">appropriate function pointer type</a>
5081 before being called. The new function's signature is the same as that of
5082 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
5083 removed. At most one such <tt>nest</tt> argument is allowed, and it must be
5084 of pointer type. Calling the new function is equivalent to calling
5085 <tt>func</tt> with the same argument list, but with <tt>nval</tt> used for the
5086 missing <tt>nest</tt> argument. If, after calling
5087 <tt>llvm.init.trampoline</tt>, the memory pointed to by <tt>tramp</tt> is
5088 modified, then the effect of any later call to the returned function pointer is
5093 <!-- ======================================================================= -->
5094 <div class="doc_subsection">
5095 <a name="int_general">General Intrinsics</a>
5098 <div class="doc_text">
5099 <p> This class of intrinsics is designed to be generic and has
5100 no specific purpose. </p>
5103 <!-- _______________________________________________________________________ -->
5104 <div class="doc_subsubsection">
5105 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
5108 <div class="doc_text">
5112 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
5118 The '<tt>llvm.var.annotation</tt>' intrinsic
5124 The first argument is a pointer to a value, the second is a pointer to a
5125 global string, the third is a pointer to a global string which is the source
5126 file name, and the last argument is the line number.
5132 This intrinsic allows annotation of local variables with arbitrary strings.
5133 This can be useful for special purpose optimizations that want to look for these
5134 annotations. These have no other defined use, they are ignored by code
5135 generation and optimization.
5138 <!-- _______________________________________________________________________ -->
5139 <div class="doc_subsubsection">
5140 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
5143 <div class="doc_text">
5146 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
5147 any integer bit width.
5150 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
5151 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
5152 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
5153 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
5154 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
5160 The '<tt>llvm.annotation</tt>' intrinsic.
5166 The first argument is an integer value (result of some expression),
5167 the second is a pointer to a global string, the third is a pointer to a global
5168 string which is the source file name, and the last argument is the line number.
5169 It returns the value of the first argument.
5175 This intrinsic allows annotations to be put on arbitrary expressions
5176 with arbitrary strings. This can be useful for special purpose optimizations
5177 that want to look for these annotations. These have no other defined use, they
5178 are ignored by code generation and optimization.
5181 <!-- *********************************************************************** -->
5184 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
5185 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
5186 <a href="http://validator.w3.org/check/referer"><img
5187 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a>
5189 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
5190 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
5191 Last modified: $Date$