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>
820 <!-- ======================================================================= -->
821 <div class="doc_subsection">
822 <a name="moduleasm">Module-Level Inline Assembly</a>
825 <div class="doc_text">
827 Modules may contain "module-level inline asm" blocks, which corresponds to the
828 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
829 LLVM and treated as a single unit, but may be separated in the .ll file if
830 desired. The syntax is very simple:
833 <div class="doc_code">
835 module asm "inline asm code goes here"
836 module asm "more can go here"
840 <p>The strings can contain any character by escaping non-printable characters.
841 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
846 The inline asm code is simply printed to the machine code .s file when
847 assembly code is generated.
851 <!-- ======================================================================= -->
852 <div class="doc_subsection">
853 <a name="datalayout">Data Layout</a>
856 <div class="doc_text">
857 <p>A module may specify a target specific data layout string that specifies how
858 data is to be laid out in memory. The syntax for the data layout is simply:</p>
859 <pre> target datalayout = "<i>layout specification</i>"</pre>
860 <p>The <i>layout specification</i> consists of a list of specifications
861 separated by the minus sign character ('-'). Each specification starts with a
862 letter and may include other information after the letter to define some
863 aspect of the data layout. The specifications accepted are as follows: </p>
866 <dd>Specifies that the target lays out data in big-endian form. That is, the
867 bits with the most significance have the lowest address location.</dd>
869 <dd>Specifies that hte target lays out data in little-endian form. That is,
870 the bits with the least significance have the lowest address location.</dd>
871 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
872 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
873 <i>preferred</i> alignments. All sizes are in bits. Specifying the <i>pref</i>
874 alignment is optional. If omitted, the preceding <tt>:</tt> should be omitted
876 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
877 <dd>This specifies the alignment for an integer type of a given bit
878 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
879 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
880 <dd>This specifies the alignment for a vector type of a given bit
882 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
883 <dd>This specifies the alignment for a floating point type of a given bit
884 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
886 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
887 <dd>This specifies the alignment for an aggregate type of a given bit
890 <p>When constructing the data layout for a given target, LLVM starts with a
891 default set of specifications which are then (possibly) overriden by the
892 specifications in the <tt>datalayout</tt> keyword. The default specifications
893 are given in this list:</p>
895 <li><tt>E</tt> - big endian</li>
896 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
897 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
898 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
899 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
900 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
901 <li><tt>i64:32:64</tt> - i64 has abi alignment of 32-bits but preferred
902 alignment of 64-bits</li>
903 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
904 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
905 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
906 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
907 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
909 <p>When llvm is determining the alignment for a given type, it uses the
912 <li>If the type sought is an exact match for one of the specifications, that
913 specification is used.</li>
914 <li>If no match is found, and the type sought is an integer type, then the
915 smallest integer type that is larger than the bitwidth of the sought type is
916 used. If none of the specifications are larger than the bitwidth then the the
917 largest integer type is used. For example, given the default specifications
918 above, the i7 type will use the alignment of i8 (next largest) while both
919 i65 and i256 will use the alignment of i64 (largest specified).</li>
920 <li>If no match is found, and the type sought is a vector type, then the
921 largest vector type that is smaller than the sought vector type will be used
922 as a fall back. This happens because <128 x double> can be implemented in
923 terms of 64 <2 x double>, for example.</li>
927 <!-- *********************************************************************** -->
928 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
929 <!-- *********************************************************************** -->
931 <div class="doc_text">
933 <p>The LLVM type system is one of the most important features of the
934 intermediate representation. Being typed enables a number of
935 optimizations to be performed on the IR directly, without having to do
936 extra analyses on the side before the transformation. A strong type
937 system makes it easier to read the generated code and enables novel
938 analyses and transformations that are not feasible to perform on normal
939 three address code representations.</p>
943 <!-- ======================================================================= -->
944 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
945 <div class="doc_text">
946 <p>The primitive types are the fundamental building blocks of the LLVM
947 system. The current set of primitive types is as follows:</p>
949 <table class="layout">
954 <tr><th>Type</th><th>Description</th></tr>
955 <tr><td><tt><a name="t_void">void</a></tt></td><td>No value</td></tr>
956 <tr><td><tt>label</tt></td><td>Branch destination</td></tr>
963 <tr><th>Type</th><th>Description</th></tr>
964 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
965 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
973 <!-- _______________________________________________________________________ -->
974 <div class="doc_subsubsection"> <a name="t_classifications">Type
975 Classifications</a> </div>
976 <div class="doc_text">
977 <p>These different primitive types fall into a few useful
980 <table border="1" cellspacing="0" cellpadding="4">
982 <tr><th>Classification</th><th>Types</th></tr>
984 <td><a name="t_integer">integer</a></td>
985 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
988 <td><a name="t_floating">floating point</a></td>
989 <td><tt>float, double</tt></td>
992 <td><a name="t_firstclass">first class</a></td>
993 <td><tt>i1, ..., float, double, <br/>
994 <a href="#t_pointer">pointer</a>,<a href="#t_vector">vector</a></tt>
1000 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
1001 most important. Values of these types are the only ones which can be
1002 produced by instructions, passed as arguments, or used as operands to
1003 instructions. This means that all structures and arrays must be
1004 manipulated either by pointer or by component.</p>
1007 <!-- ======================================================================= -->
1008 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1010 <div class="doc_text">
1012 <p>The real power in LLVM comes from the derived types in the system.
1013 This is what allows a programmer to represent arrays, functions,
1014 pointers, and other useful types. Note that these derived types may be
1015 recursive: For example, it is possible to have a two dimensional array.</p>
1019 <!-- _______________________________________________________________________ -->
1020 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1022 <div class="doc_text">
1025 <p>The integer type is a very simple derived type that simply specifies an
1026 arbitrary bit width for the integer type desired. Any bit width from 1 bit to
1027 2^23-1 (about 8 million) can be specified.</p>
1035 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1039 <table class="layout">
1049 <tt>i1942652</tt><br/>
1052 A boolean integer of 1 bit<br/>
1053 A nibble sized integer of 4 bits.<br/>
1054 A byte sized integer of 8 bits.<br/>
1055 A half word sized integer of 16 bits.<br/>
1056 A word sized integer of 32 bits.<br/>
1057 An integer whose bit width is the answer. <br/>
1058 A double word sized integer of 64 bits.<br/>
1059 A really big integer of over 1 million bits.<br/>
1065 <!-- _______________________________________________________________________ -->
1066 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1068 <div class="doc_text">
1072 <p>The array type is a very simple derived type that arranges elements
1073 sequentially in memory. The array type requires a size (number of
1074 elements) and an underlying data type.</p>
1079 [<# elements> x <elementtype>]
1082 <p>The number of elements is a constant integer value; elementtype may
1083 be any type with a size.</p>
1086 <table class="layout">
1089 <tt>[40 x i32 ]</tt><br/>
1090 <tt>[41 x i32 ]</tt><br/>
1091 <tt>[40 x i8]</tt><br/>
1094 Array of 40 32-bit integer values.<br/>
1095 Array of 41 32-bit integer values.<br/>
1096 Array of 40 8-bit integer values.<br/>
1100 <p>Here are some examples of multidimensional arrays:</p>
1101 <table class="layout">
1104 <tt>[3 x [4 x i32]]</tt><br/>
1105 <tt>[12 x [10 x float]]</tt><br/>
1106 <tt>[2 x [3 x [4 x i16]]]</tt><br/>
1109 3x4 array of 32-bit integer values.<br/>
1110 12x10 array of single precision floating point values.<br/>
1111 2x3x4 array of 16-bit integer values.<br/>
1116 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1117 length array. Normally, accesses past the end of an array are undefined in
1118 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
1119 As a special case, however, zero length arrays are recognized to be variable
1120 length. This allows implementation of 'pascal style arrays' with the LLVM
1121 type "{ i32, [0 x float]}", for example.</p>
1125 <!-- _______________________________________________________________________ -->
1126 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1127 <div class="doc_text">
1129 <p>The function type can be thought of as a function signature. It
1130 consists of a return type and a list of formal parameter types.
1131 Function types are usually used to build virtual function tables
1132 (which are structures of pointers to functions), for indirect function
1133 calls, and when defining a function.</p>
1135 The return type of a function type cannot be an aggregate type.
1138 <pre> <returntype> (<parameter list>)<br></pre>
1139 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1140 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1141 which indicates that the function takes a variable number of arguments.
1142 Variable argument functions can access their arguments with the <a
1143 href="#int_varargs">variable argument handling intrinsic</a> functions.</p>
1145 <table class="layout">
1147 <td class="left"><tt>i32 (i32)</tt></td>
1148 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1150 </tr><tr class="layout">
1151 <td class="left"><tt>float (i16 signext, i32 *) *
1153 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1154 an <tt>i16</tt> that should be sign extended and a
1155 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1158 </tr><tr class="layout">
1159 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1160 <td class="left">A vararg function that takes at least one
1161 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1162 which returns an integer. This is the signature for <tt>printf</tt> in
1169 <!-- _______________________________________________________________________ -->
1170 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1171 <div class="doc_text">
1173 <p>The structure type is used to represent a collection of data members
1174 together in memory. The packing of the field types is defined to match
1175 the ABI of the underlying processor. The elements of a structure may
1176 be any type that has a size.</p>
1177 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1178 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1179 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1182 <pre> { <type list> }<br></pre>
1184 <table class="layout">
1186 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1187 <td class="left">A triple of three <tt>i32</tt> values</td>
1188 </tr><tr class="layout">
1189 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1190 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1191 second element is a <a href="#t_pointer">pointer</a> to a
1192 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1193 an <tt>i32</tt>.</td>
1198 <!-- _______________________________________________________________________ -->
1199 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1201 <div class="doc_text">
1203 <p>The packed structure type is used to represent a collection of data members
1204 together in memory. There is no padding between fields. Further, the alignment
1205 of a packed structure is 1 byte. The elements of a packed structure may
1206 be any type that has a size.</p>
1207 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1208 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1209 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1212 <pre> < { <type list> } > <br></pre>
1214 <table class="layout">
1216 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1217 <td class="left">A triple of three <tt>i32</tt> values</td>
1218 </tr><tr class="layout">
1219 <td class="left"><tt>< { float, i32 (i32) * } ></tt></td>
1220 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1221 second element is a <a href="#t_pointer">pointer</a> to a
1222 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1223 an <tt>i32</tt>.</td>
1228 <!-- _______________________________________________________________________ -->
1229 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1230 <div class="doc_text">
1232 <p>As in many languages, the pointer type represents a pointer or
1233 reference to another object, which must live in memory.</p>
1235 <pre> <type> *<br></pre>
1237 <table class="layout">
1240 <tt>[4x i32]*</tt><br/>
1241 <tt>i32 (i32 *) *</tt><br/>
1244 A <a href="#t_pointer">pointer</a> to <a href="#t_array">array</a> of
1245 four <tt>i32</tt> values<br/>
1246 A <a href="#t_pointer">pointer</a> to a <a
1247 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1254 <!-- _______________________________________________________________________ -->
1255 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1256 <div class="doc_text">
1260 <p>A vector type is a simple derived type that represents a vector
1261 of elements. Vector types are used when multiple primitive data
1262 are operated in parallel using a single instruction (SIMD).
1263 A vector type requires a size (number of
1264 elements) and an underlying primitive data type. Vectors must have a power
1265 of two length (1, 2, 4, 8, 16 ...). Vector types are
1266 considered <a href="#t_firstclass">first class</a>.</p>
1271 < <# elements> x <elementtype> >
1274 <p>The number of elements is a constant integer value; elementtype may
1275 be any integer or floating point type.</p>
1279 <table class="layout">
1282 <tt><4 x i32></tt><br/>
1283 <tt><8 x float></tt><br/>
1284 <tt><2 x i64></tt><br/>
1287 Vector of 4 32-bit integer values.<br/>
1288 Vector of 8 floating-point values.<br/>
1289 Vector of 2 64-bit integer values.<br/>
1295 <!-- _______________________________________________________________________ -->
1296 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1297 <div class="doc_text">
1301 <p>Opaque types are used to represent unknown types in the system. This
1302 corresponds (for example) to the C notion of a forward declared structure type.
1303 In LLVM, opaque types can eventually be resolved to any type (not just a
1304 structure type).</p>
1314 <table class="layout">
1320 An opaque type.<br/>
1327 <!-- *********************************************************************** -->
1328 <div class="doc_section"> <a name="constants">Constants</a> </div>
1329 <!-- *********************************************************************** -->
1331 <div class="doc_text">
1333 <p>LLVM has several different basic types of constants. This section describes
1334 them all and their syntax.</p>
1338 <!-- ======================================================================= -->
1339 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1341 <div class="doc_text">
1344 <dt><b>Boolean constants</b></dt>
1346 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1347 constants of the <tt><a href="#t_primitive">i1</a></tt> type.
1350 <dt><b>Integer constants</b></dt>
1352 <dd>Standard integers (such as '4') are constants of the <a
1353 href="#t_integer">integer</a> type. Negative numbers may be used with
1357 <dt><b>Floating point constants</b></dt>
1359 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1360 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1361 notation (see below). Floating point constants must have a <a
1362 href="#t_floating">floating point</a> type. </dd>
1364 <dt><b>Null pointer constants</b></dt>
1366 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1367 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1371 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1372 of floating point constants. For example, the form '<tt>double
1373 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1374 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1375 (and the only time that they are generated by the disassembler) is when a
1376 floating point constant must be emitted but it cannot be represented as a
1377 decimal floating point number. For example, NaN's, infinities, and other
1378 special values are represented in their IEEE hexadecimal format so that
1379 assembly and disassembly do not cause any bits to change in the constants.</p>
1383 <!-- ======================================================================= -->
1384 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1387 <div class="doc_text">
1388 <p>Aggregate constants arise from aggregation of simple constants
1389 and smaller aggregate constants.</p>
1392 <dt><b>Structure constants</b></dt>
1394 <dd>Structure constants are represented with notation similar to structure
1395 type definitions (a comma separated list of elements, surrounded by braces
1396 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* %G }</tt>",
1397 where "<tt>%G</tt>" is declared as "<tt>@G = external global i32</tt>". Structure constants
1398 must have <a href="#t_struct">structure type</a>, and the number and
1399 types of elements must match those specified by the type.
1402 <dt><b>Array constants</b></dt>
1404 <dd>Array constants are represented with notation similar to array type
1405 definitions (a comma separated list of elements, surrounded by square brackets
1406 (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74 ]</tt>". Array
1407 constants must have <a href="#t_array">array type</a>, and the number and
1408 types of elements must match those specified by the type.
1411 <dt><b>Vector constants</b></dt>
1413 <dd>Vector constants are represented with notation similar to vector type
1414 definitions (a comma separated list of elements, surrounded by
1415 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32 42,
1416 i32 11, i32 74, i32 100 ></tt>". Vector constants must have <a
1417 href="#t_vector">vector type</a>, and the number and types of elements must
1418 match those specified by the type.
1421 <dt><b>Zero initialization</b></dt>
1423 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1424 value to zero of <em>any</em> type, including scalar and aggregate types.
1425 This is often used to avoid having to print large zero initializers (e.g. for
1426 large arrays) and is always exactly equivalent to using explicit zero
1433 <!-- ======================================================================= -->
1434 <div class="doc_subsection">
1435 <a name="globalconstants">Global Variable and Function Addresses</a>
1438 <div class="doc_text">
1440 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1441 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1442 constants. These constants are explicitly referenced when the <a
1443 href="#identifiers">identifier for the global</a> is used and always have <a
1444 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1447 <div class="doc_code">
1451 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1457 <!-- ======================================================================= -->
1458 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1459 <div class="doc_text">
1460 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1461 no specific value. Undefined values may be of any type and be used anywhere
1462 a constant is permitted.</p>
1464 <p>Undefined values indicate to the compiler that the program is well defined
1465 no matter what value is used, giving the compiler more freedom to optimize.
1469 <!-- ======================================================================= -->
1470 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1473 <div class="doc_text">
1475 <p>Constant expressions are used to allow expressions involving other constants
1476 to be used as constants. Constant expressions may be of any <a
1477 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1478 that does not have side effects (e.g. load and call are not supported). The
1479 following is the syntax for constant expressions:</p>
1482 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1483 <dd>Truncate a constant to another type. The bit size of CST must be larger
1484 than the bit size of TYPE. Both types must be integers.</dd>
1486 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1487 <dd>Zero extend a constant to another type. The bit size of CST must be
1488 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1490 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1491 <dd>Sign extend a constant to another type. The bit size of CST must be
1492 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1494 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1495 <dd>Truncate a floating point constant to another floating point type. The
1496 size of CST must be larger than the size of TYPE. Both types must be
1497 floating point.</dd>
1499 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1500 <dd>Floating point extend a constant to another type. The size of CST must be
1501 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1503 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
1504 <dd>Convert a floating point constant to the corresponding unsigned integer
1505 constant. TYPE must be an integer type. CST must be floating point. If the
1506 value won't fit in the integer type, the results are undefined.</dd>
1508 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
1509 <dd>Convert a floating point constant to the corresponding signed integer
1510 constant. TYPE must be an integer type. CST must be floating point. If the
1511 value won't fit in the integer type, the results are undefined.</dd>
1513 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
1514 <dd>Convert an unsigned integer constant to the corresponding floating point
1515 constant. TYPE must be floating point. CST must be of integer type. If the
1516 value won't fit in the floating point type, the results are undefined.</dd>
1518 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
1519 <dd>Convert a signed integer constant to the corresponding floating point
1520 constant. TYPE must be floating point. CST must be of integer type. If the
1521 value won't fit in the floating point type, the results are undefined.</dd>
1523 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
1524 <dd>Convert a pointer typed constant to the corresponding integer constant
1525 TYPE must be an integer type. CST must be of pointer type. The CST value is
1526 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1528 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
1529 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1530 pointer type. CST must be of integer type. The CST value is zero extended,
1531 truncated, or unchanged to make it fit in a pointer size. This one is
1532 <i>really</i> dangerous!</dd>
1534 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
1535 <dd>Convert a constant, CST, to another TYPE. The size of CST and TYPE must be
1536 identical (same number of bits). The conversion is done as if the CST value
1537 was stored to memory and read back as TYPE. In other words, no bits change
1538 with this operator, just the type. This can be used for conversion of
1539 vector types to any other type, as long as they have the same bit width. For
1540 pointers it is only valid to cast to another pointer type.
1543 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1545 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1546 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1547 instruction, the index list may have zero or more indexes, which are required
1548 to make sense for the type of "CSTPTR".</dd>
1550 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1552 <dd>Perform the <a href="#i_select">select operation</a> on
1555 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
1556 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
1558 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
1559 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
1561 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1563 <dd>Perform the <a href="#i_extractelement">extractelement
1564 operation</a> on constants.
1566 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1568 <dd>Perform the <a href="#i_insertelement">insertelement
1569 operation</a> on constants.</dd>
1572 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
1574 <dd>Perform the <a href="#i_shufflevector">shufflevector
1575 operation</a> on constants.</dd>
1577 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1579 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1580 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1581 binary</a> operations. The constraints on operands are the same as those for
1582 the corresponding instruction (e.g. no bitwise operations on floating point
1583 values are allowed).</dd>
1587 <!-- *********************************************************************** -->
1588 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1589 <!-- *********************************************************************** -->
1591 <!-- ======================================================================= -->
1592 <div class="doc_subsection">
1593 <a name="inlineasm">Inline Assembler Expressions</a>
1596 <div class="doc_text">
1599 LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
1600 Module-Level Inline Assembly</a>) through the use of a special value. This
1601 value represents the inline assembler as a string (containing the instructions
1602 to emit), a list of operand constraints (stored as a string), and a flag that
1603 indicates whether or not the inline asm expression has side effects. An example
1604 inline assembler expression is:
1607 <div class="doc_code">
1609 i32 (i32) asm "bswap $0", "=r,r"
1614 Inline assembler expressions may <b>only</b> be used as the callee operand of
1615 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
1618 <div class="doc_code">
1620 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
1625 Inline asms with side effects not visible in the constraint list must be marked
1626 as having side effects. This is done through the use of the
1627 '<tt>sideeffect</tt>' keyword, like so:
1630 <div class="doc_code">
1632 call void asm sideeffect "eieio", ""()
1636 <p>TODO: The format of the asm and constraints string still need to be
1637 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1638 need to be documented).
1643 <!-- *********************************************************************** -->
1644 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1645 <!-- *********************************************************************** -->
1647 <div class="doc_text">
1649 <p>The LLVM instruction set consists of several different
1650 classifications of instructions: <a href="#terminators">terminator
1651 instructions</a>, <a href="#binaryops">binary instructions</a>,
1652 <a href="#bitwiseops">bitwise binary instructions</a>, <a
1653 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1654 instructions</a>.</p>
1658 <!-- ======================================================================= -->
1659 <div class="doc_subsection"> <a name="terminators">Terminator
1660 Instructions</a> </div>
1662 <div class="doc_text">
1664 <p>As mentioned <a href="#functionstructure">previously</a>, every
1665 basic block in a program ends with a "Terminator" instruction, which
1666 indicates which block should be executed after the current block is
1667 finished. These terminator instructions typically yield a '<tt>void</tt>'
1668 value: they produce control flow, not values (the one exception being
1669 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1670 <p>There are six different terminator instructions: the '<a
1671 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1672 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1673 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1674 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1675 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1679 <!-- _______________________________________________________________________ -->
1680 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1681 Instruction</a> </div>
1682 <div class="doc_text">
1684 <pre> ret <type> <value> <i>; Return a value from a non-void function</i>
1685 ret void <i>; Return from void function</i>
1688 <p>The '<tt>ret</tt>' instruction is used to return control flow (and a
1689 value) from a function back to the caller.</p>
1690 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
1691 returns a value and then causes control flow, and one that just causes
1692 control flow to occur.</p>
1694 <p>The '<tt>ret</tt>' instruction may return any '<a
1695 href="#t_firstclass">first class</a>' type. Notice that a function is
1696 not <a href="#wellformed">well formed</a> if there exists a '<tt>ret</tt>'
1697 instruction inside of the function that returns a value that does not
1698 match the return type of the function.</p>
1700 <p>When the '<tt>ret</tt>' instruction is executed, control flow
1701 returns back to the calling function's context. If the caller is a "<a
1702 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1703 the instruction after the call. If the caller was an "<a
1704 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1705 at the beginning of the "normal" destination block. If the instruction
1706 returns a value, that value shall set the call or invoke instruction's
1709 <pre> ret i32 5 <i>; Return an integer value of 5</i>
1710 ret void <i>; Return from a void function</i>
1713 <!-- _______________________________________________________________________ -->
1714 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1715 <div class="doc_text">
1717 <pre> br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1720 <p>The '<tt>br</tt>' instruction is used to cause control flow to
1721 transfer to a different basic block in the current function. There are
1722 two forms of this instruction, corresponding to a conditional branch
1723 and an unconditional branch.</p>
1725 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1726 single '<tt>i1</tt>' value and two '<tt>label</tt>' values. The
1727 unconditional form of the '<tt>br</tt>' instruction takes a single
1728 '<tt>label</tt>' value as a target.</p>
1730 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
1731 argument is evaluated. If the value is <tt>true</tt>, control flows
1732 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1733 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1735 <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
1736 href="#i_ret">ret</a> i32 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> i32 0<br></pre>
1738 <!-- _______________________________________________________________________ -->
1739 <div class="doc_subsubsection">
1740 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
1743 <div class="doc_text">
1747 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
1752 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
1753 several different places. It is a generalization of the '<tt>br</tt>'
1754 instruction, allowing a branch to occur to one of many possible
1760 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
1761 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
1762 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
1763 table is not allowed to contain duplicate constant entries.</p>
1767 <p>The <tt>switch</tt> instruction specifies a table of values and
1768 destinations. When the '<tt>switch</tt>' instruction is executed, this
1769 table is searched for the given value. If the value is found, control flow is
1770 transfered to the corresponding destination; otherwise, control flow is
1771 transfered to the default destination.</p>
1773 <h5>Implementation:</h5>
1775 <p>Depending on properties of the target machine and the particular
1776 <tt>switch</tt> instruction, this instruction may be code generated in different
1777 ways. For example, it could be generated as a series of chained conditional
1778 branches or with a lookup table.</p>
1783 <i>; Emulate a conditional br instruction</i>
1784 %Val = <a href="#i_zext">zext</a> i1 %value to i32
1785 switch i32 %Val, label %truedest [i32 0, label %falsedest ]
1787 <i>; Emulate an unconditional br instruction</i>
1788 switch i32 0, label %dest [ ]
1790 <i>; Implement a jump table:</i>
1791 switch i32 %val, label %otherwise [ i32 0, label %onzero
1793 i32 2, label %ontwo ]
1797 <!-- _______________________________________________________________________ -->
1798 <div class="doc_subsubsection">
1799 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
1802 <div class="doc_text">
1807 <result> = invoke [<a href="#callingconv">cconv</a>] <ptr to function ty> %<function ptr val>(<function args>)
1808 to label <normal label> unwind label <exception label>
1813 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
1814 function, with the possibility of control flow transfer to either the
1815 '<tt>normal</tt>' label or the
1816 '<tt>exception</tt>' label. If the callee function returns with the
1817 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
1818 "normal" label. If the callee (or any indirect callees) returns with the "<a
1819 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
1820 continued at the dynamically nearest "exception" label.</p>
1824 <p>This instruction requires several arguments:</p>
1828 The optional "cconv" marker indicates which <a href="#callingconv">calling
1829 convention</a> the call should use. If none is specified, the call defaults
1830 to using C calling conventions.
1832 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
1833 function value being invoked. In most cases, this is a direct function
1834 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
1835 an arbitrary pointer to function value.
1838 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
1839 function to be invoked. </li>
1841 <li>'<tt>function args</tt>': argument list whose types match the function
1842 signature argument types. If the function signature indicates the function
1843 accepts a variable number of arguments, the extra arguments can be
1846 <li>'<tt>normal label</tt>': the label reached when the called function
1847 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
1849 <li>'<tt>exception label</tt>': the label reached when a callee returns with
1850 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
1856 <p>This instruction is designed to operate as a standard '<tt><a
1857 href="#i_call">call</a></tt>' instruction in most regards. The primary
1858 difference is that it establishes an association with a label, which is used by
1859 the runtime library to unwind the stack.</p>
1861 <p>This instruction is used in languages with destructors to ensure that proper
1862 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
1863 exception. Additionally, this is important for implementation of
1864 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
1868 %retval = invoke i32 %Test(i32 15) to label %Continue
1869 unwind label %TestCleanup <i>; {i32}:retval set</i>
1870 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Test(i32 15) to label %Continue
1871 unwind label %TestCleanup <i>; {i32}:retval set</i>
1876 <!-- _______________________________________________________________________ -->
1878 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
1879 Instruction</a> </div>
1881 <div class="doc_text">
1890 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
1891 at the first callee in the dynamic call stack which used an <a
1892 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
1893 primarily used to implement exception handling.</p>
1897 <p>The '<tt>unwind</tt>' intrinsic causes execution of the current function to
1898 immediately halt. The dynamic call stack is then searched for the first <a
1899 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
1900 execution continues at the "exceptional" destination block specified by the
1901 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
1902 dynamic call chain, undefined behavior results.</p>
1905 <!-- _______________________________________________________________________ -->
1907 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
1908 Instruction</a> </div>
1910 <div class="doc_text">
1919 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
1920 instruction is used to inform the optimizer that a particular portion of the
1921 code is not reachable. This can be used to indicate that the code after a
1922 no-return function cannot be reached, and other facts.</p>
1926 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
1931 <!-- ======================================================================= -->
1932 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
1933 <div class="doc_text">
1934 <p>Binary operators are used to do most of the computation in a
1935 program. They require two operands, execute an operation on them, and
1936 produce a single value. The operands might represent
1937 multiple data, as is the case with the <a href="#t_vector">vector</a> data type.
1938 The result value of a binary operator is not
1939 necessarily the same type as its operands.</p>
1940 <p>There are several different binary operators:</p>
1942 <!-- _______________________________________________________________________ -->
1943 <div class="doc_subsubsection"> <a name="i_add">'<tt>add</tt>'
1944 Instruction</a> </div>
1945 <div class="doc_text">
1947 <pre> <result> = add <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1950 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
1952 <p>The two arguments to the '<tt>add</tt>' instruction must be either <a
1953 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a> values.
1954 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
1955 Both arguments must have identical types.</p>
1957 <p>The value produced is the integer or floating point sum of the two
1960 <pre> <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
1963 <!-- _______________________________________________________________________ -->
1964 <div class="doc_subsubsection"> <a name="i_sub">'<tt>sub</tt>'
1965 Instruction</a> </div>
1966 <div class="doc_text">
1968 <pre> <result> = sub <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1971 <p>The '<tt>sub</tt>' instruction returns the difference of its two
1973 <p>Note that the '<tt>sub</tt>' instruction is used to represent the '<tt>neg</tt>'
1974 instruction present in most other intermediate representations.</p>
1976 <p>The two arguments to the '<tt>sub</tt>' instruction must be either <a
1977 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1979 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
1980 Both arguments must have identical types.</p>
1982 <p>The value produced is the integer or floating point difference of
1983 the two operands.</p>
1986 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
1987 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
1990 <!-- _______________________________________________________________________ -->
1991 <div class="doc_subsubsection"> <a name="i_mul">'<tt>mul</tt>'
1992 Instruction</a> </div>
1993 <div class="doc_text">
1995 <pre> <result> = mul <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1998 <p>The '<tt>mul</tt>' instruction returns the product of its two
2001 <p>The two arguments to the '<tt>mul</tt>' instruction must be either <a
2002 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
2004 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2005 Both arguments must have identical types.</p>
2007 <p>The value produced is the integer or floating point product of the
2009 <p>Because the operands are the same width, the result of an integer
2010 multiplication is the same whether the operands should be deemed unsigned or
2013 <pre> <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
2016 <!-- _______________________________________________________________________ -->
2017 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
2019 <div class="doc_text">
2021 <pre> <result> = udiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2024 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two
2027 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
2028 <a href="#t_integer">integer</a> values. Both arguments must have identical
2029 types. This instruction can also take <a href="#t_vector">vector</a> versions
2030 of the values in which case the elements must be integers.</p>
2032 <p>The value produced is the unsigned integer quotient of the two operands. This
2033 instruction always performs an unsigned division operation, regardless of
2034 whether the arguments are unsigned or not.</p>
2036 <pre> <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2039 <!-- _______________________________________________________________________ -->
2040 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
2042 <div class="doc_text">
2044 <pre> <result> = sdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2047 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
2050 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
2051 <a href="#t_integer">integer</a> values. Both arguments must have identical
2052 types. This instruction can also take <a href="#t_vector">vector</a> versions
2053 of the values in which case the elements must be integers.</p>
2055 <p>The value produced is the signed integer quotient of the two operands. This
2056 instruction always performs a signed division operation, regardless of whether
2057 the arguments are signed or not.</p>
2059 <pre> <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2062 <!-- _______________________________________________________________________ -->
2063 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
2064 Instruction</a> </div>
2065 <div class="doc_text">
2067 <pre> <result> = fdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2070 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
2073 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
2074 <a href="#t_floating">floating point</a> values. Both arguments must have
2075 identical types. This instruction can also take <a href="#t_vector">vector</a>
2076 versions of floating point values.</p>
2078 <p>The value produced is the floating point quotient of the two operands.</p>
2080 <pre> <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
2083 <!-- _______________________________________________________________________ -->
2084 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
2086 <div class="doc_text">
2088 <pre> <result> = urem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2091 <p>The '<tt>urem</tt>' instruction returns the remainder from the
2092 unsigned division of its two arguments.</p>
2094 <p>The two arguments to the '<tt>urem</tt>' instruction must be
2095 <a href="#t_integer">integer</a> values. Both arguments must have identical
2098 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
2099 This instruction always performs an unsigned division to get the remainder,
2100 regardless of whether the arguments are unsigned or not.</p>
2102 <pre> <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2106 <!-- _______________________________________________________________________ -->
2107 <div class="doc_subsubsection"> <a name="i_srem">'<tt>srem</tt>'
2108 Instruction</a> </div>
2109 <div class="doc_text">
2111 <pre> <result> = srem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2114 <p>The '<tt>srem</tt>' instruction returns the remainder from the
2115 signed division of its two operands.</p>
2117 <p>The two arguments to the '<tt>srem</tt>' instruction must be
2118 <a href="#t_integer">integer</a> values. Both arguments must have identical
2121 <p>This instruction returns the <i>remainder</i> of a division (where the result
2122 has the same sign as the dividend, <tt>var1</tt>), not the <i>modulo</i>
2123 operator (where the result has the same sign as the divisor, <tt>var2</tt>) of
2124 a value. For more information about the difference, see <a
2125 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
2126 Math Forum</a>. For a table of how this is implemented in various languages,
2127 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
2128 Wikipedia: modulo operation</a>.</p>
2130 <pre> <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2134 <!-- _______________________________________________________________________ -->
2135 <div class="doc_subsubsection"> <a name="i_frem">'<tt>frem</tt>'
2136 Instruction</a> </div>
2137 <div class="doc_text">
2139 <pre> <result> = frem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2142 <p>The '<tt>frem</tt>' instruction returns the remainder from the
2143 division of its two operands.</p>
2145 <p>The two arguments to the '<tt>frem</tt>' instruction must be
2146 <a href="#t_floating">floating point</a> values. Both arguments must have
2147 identical types.</p>
2149 <p>This instruction returns the <i>remainder</i> of a division.</p>
2151 <pre> <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
2155 <!-- ======================================================================= -->
2156 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
2157 Operations</a> </div>
2158 <div class="doc_text">
2159 <p>Bitwise binary operators are used to do various forms of
2160 bit-twiddling in a program. They are generally very efficient
2161 instructions and can commonly be strength reduced from other
2162 instructions. They require two operands, execute an operation on them,
2163 and produce a single value. The resulting value of the bitwise binary
2164 operators is always the same type as its first operand.</p>
2167 <!-- _______________________________________________________________________ -->
2168 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
2169 Instruction</a> </div>
2170 <div class="doc_text">
2172 <pre> <result> = shl <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2177 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2178 the left a specified number of bits.</p>
2182 <p>Both arguments to the '<tt>shl</tt>' instruction must be the same <a
2183 href="#t_integer">integer</a> type.</p>
2187 <p>The value produced is <tt>var1</tt> * 2<sup><tt>var2</tt></sup>. If
2188 <tt>var2</tt> is (statically or dynamically) equal to or larger than the number
2189 of bits in <tt>var1</tt>, the result is undefined.</p>
2191 <h5>Example:</h5><pre>
2192 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
2193 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
2194 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
2195 <result> = shl i32 1, 32 <i>; undefined</i>
2198 <!-- _______________________________________________________________________ -->
2199 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
2200 Instruction</a> </div>
2201 <div class="doc_text">
2203 <pre> <result> = lshr <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2207 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
2208 operand shifted to the right a specified number of bits with zero fill.</p>
2211 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
2212 <a href="#t_integer">integer</a> type.</p>
2216 <p>This instruction always performs a logical shift right operation. The most
2217 significant bits of the result will be filled with zero bits after the
2218 shift. If <tt>var2</tt> is (statically or dynamically) equal to or larger than
2219 the number of bits in <tt>var1</tt>, the result is undefined.</p>
2223 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
2224 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
2225 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
2226 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
2227 <result> = lshr i32 1, 32 <i>; undefined</i>
2231 <!-- _______________________________________________________________________ -->
2232 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
2233 Instruction</a> </div>
2234 <div class="doc_text">
2237 <pre> <result> = ashr <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2241 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
2242 operand shifted to the right a specified number of bits with sign extension.</p>
2245 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
2246 <a href="#t_integer">integer</a> type.</p>
2249 <p>This instruction always performs an arithmetic shift right operation,
2250 The most significant bits of the result will be filled with the sign bit
2251 of <tt>var1</tt>. If <tt>var2</tt> is (statically or dynamically) equal to or
2252 larger than the number of bits in <tt>var1</tt>, the result is undefined.
2257 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
2258 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
2259 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
2260 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
2261 <result> = ashr i32 1, 32 <i>; undefined</i>
2265 <!-- _______________________________________________________________________ -->
2266 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
2267 Instruction</a> </div>
2268 <div class="doc_text">
2270 <pre> <result> = and <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2273 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
2274 its two operands.</p>
2276 <p>The two arguments to the '<tt>and</tt>' instruction must be <a
2277 href="#t_integer">integer</a> values. Both arguments must have
2278 identical types.</p>
2280 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
2282 <div style="align: center">
2283 <table border="1" cellspacing="0" cellpadding="4">
2314 <pre> <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
2315 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
2316 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
2319 <!-- _______________________________________________________________________ -->
2320 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
2321 <div class="doc_text">
2323 <pre> <result> = or <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2326 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
2327 or of its two operands.</p>
2329 <p>The two arguments to the '<tt>or</tt>' instruction must be <a
2330 href="#t_integer">integer</a> values. Both arguments must have
2331 identical types.</p>
2333 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
2335 <div style="align: center">
2336 <table border="1" cellspacing="0" cellpadding="4">
2367 <pre> <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
2368 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
2369 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
2372 <!-- _______________________________________________________________________ -->
2373 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
2374 Instruction</a> </div>
2375 <div class="doc_text">
2377 <pre> <result> = xor <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2380 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
2381 or of its two operands. The <tt>xor</tt> is used to implement the
2382 "one's complement" operation, which is the "~" operator in C.</p>
2384 <p>The two arguments to the '<tt>xor</tt>' instruction must be <a
2385 href="#t_integer">integer</a> values. Both arguments must have
2386 identical types.</p>
2388 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
2390 <div style="align: center">
2391 <table border="1" cellspacing="0" cellpadding="4">
2423 <pre> <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
2424 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
2425 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
2426 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
2430 <!-- ======================================================================= -->
2431 <div class="doc_subsection">
2432 <a name="vectorops">Vector Operations</a>
2435 <div class="doc_text">
2437 <p>LLVM supports several instructions to represent vector operations in a
2438 target-independent manner. These instructions cover the element-access and
2439 vector-specific operations needed to process vectors effectively. While LLVM
2440 does directly support these vector operations, many sophisticated algorithms
2441 will want to use target-specific intrinsics to take full advantage of a specific
2446 <!-- _______________________________________________________________________ -->
2447 <div class="doc_subsubsection">
2448 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2451 <div class="doc_text">
2456 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
2462 The '<tt>extractelement</tt>' instruction extracts a single scalar
2463 element from a vector at a specified index.
2470 The first operand of an '<tt>extractelement</tt>' instruction is a
2471 value of <a href="#t_vector">vector</a> type. The second operand is
2472 an index indicating the position from which to extract the element.
2473 The index may be a variable.</p>
2478 The result is a scalar of the same type as the element type of
2479 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2480 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2481 results are undefined.
2487 %result = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
2492 <!-- _______________________________________________________________________ -->
2493 <div class="doc_subsubsection">
2494 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
2497 <div class="doc_text">
2502 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
2508 The '<tt>insertelement</tt>' instruction inserts a scalar
2509 element into a vector at a specified index.
2516 The first operand of an '<tt>insertelement</tt>' instruction is a
2517 value of <a href="#t_vector">vector</a> type. The second operand is a
2518 scalar value whose type must equal the element type of the first
2519 operand. The third operand is an index indicating the position at
2520 which to insert the value. The index may be a variable.</p>
2525 The result is a vector of the same type as <tt>val</tt>. Its
2526 element values are those of <tt>val</tt> except at position
2527 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
2528 exceeds the length of <tt>val</tt>, the results are undefined.
2534 %result = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
2538 <!-- _______________________________________________________________________ -->
2539 <div class="doc_subsubsection">
2540 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
2543 <div class="doc_text">
2548 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <n x i32> <mask> <i>; yields <n x <ty>></i>
2554 The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
2555 from two input vectors, returning a vector of the same type.
2561 The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
2562 with types that match each other and types that match the result of the
2563 instruction. The third argument is a shuffle mask, which has the same number
2564 of elements as the other vector type, but whose element type is always 'i32'.
2568 The shuffle mask operand is required to be a constant vector with either
2569 constant integer or undef values.
2575 The elements of the two input vectors are numbered from left to right across
2576 both of the vectors. The shuffle mask operand specifies, for each element of
2577 the result vector, which element of the two input registers the result element
2578 gets. The element selector may be undef (meaning "don't care") and the second
2579 operand may be undef if performing a shuffle from only one vector.
2585 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
2586 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
2587 %result = shufflevector <4 x i32> %v1, <4 x i32> undef,
2588 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
2593 <!-- ======================================================================= -->
2594 <div class="doc_subsection">
2595 <a name="memoryops">Memory Access and Addressing Operations</a>
2598 <div class="doc_text">
2600 <p>A key design point of an SSA-based representation is how it
2601 represents memory. In LLVM, no memory locations are in SSA form, which
2602 makes things very simple. This section describes how to read, write,
2603 allocate, and free memory in LLVM.</p>
2607 <!-- _______________________________________________________________________ -->
2608 <div class="doc_subsubsection">
2609 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
2612 <div class="doc_text">
2617 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2622 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
2623 heap and returns a pointer to it.</p>
2627 <p>The '<tt>malloc</tt>' instruction allocates
2628 <tt>sizeof(<type>)*NumElements</tt>
2629 bytes of memory from the operating system and returns a pointer of the
2630 appropriate type to the program. If "NumElements" is specified, it is the
2631 number of elements allocated. If an alignment is specified, the value result
2632 of the allocation is guaranteed to be aligned to at least that boundary. If
2633 not specified, or if zero, the target can choose to align the allocation on any
2634 convenient boundary.</p>
2636 <p>'<tt>type</tt>' must be a sized type.</p>
2640 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
2641 a pointer is returned.</p>
2646 %array = malloc [4 x i8 ] <i>; yields {[%4 x i8]*}:array</i>
2648 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
2649 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
2650 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
2651 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
2652 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
2656 <!-- _______________________________________________________________________ -->
2657 <div class="doc_subsubsection">
2658 <a name="i_free">'<tt>free</tt>' Instruction</a>
2661 <div class="doc_text">
2666 free <type> <value> <i>; yields {void}</i>
2671 <p>The '<tt>free</tt>' instruction returns memory back to the unused
2672 memory heap to be reallocated in the future.</p>
2676 <p>'<tt>value</tt>' shall be a pointer value that points to a value
2677 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
2682 <p>Access to the memory pointed to by the pointer is no longer defined
2683 after this instruction executes.</p>
2688 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
2689 free [4 x i8]* %array
2693 <!-- _______________________________________________________________________ -->
2694 <div class="doc_subsubsection">
2695 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
2698 <div class="doc_text">
2703 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2708 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
2709 currently executing function, to be automatically released when this function
2710 returns to its caller.</p>
2714 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
2715 bytes of memory on the runtime stack, returning a pointer of the
2716 appropriate type to the program. If "NumElements" is specified, it is the
2717 number of elements allocated. If an alignment is specified, the value result
2718 of the allocation is guaranteed to be aligned to at least that boundary. If
2719 not specified, or if zero, the target can choose to align the allocation on any
2720 convenient boundary.</p>
2722 <p>'<tt>type</tt>' may be any sized type.</p>
2726 <p>Memory is allocated; a pointer is returned. '<tt>alloca</tt>'d
2727 memory is automatically released when the function returns. The '<tt>alloca</tt>'
2728 instruction is commonly used to represent automatic variables that must
2729 have an address available. When the function returns (either with the <tt><a
2730 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
2731 instructions), the memory is reclaimed.</p>
2736 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
2737 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
2738 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
2739 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
2743 <!-- _______________________________________________________________________ -->
2744 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
2745 Instruction</a> </div>
2746 <div class="doc_text">
2748 <pre> <result> = load <ty>* <pointer>[, align <alignment>]<br> <result> = volatile load <ty>* <pointer>[, align <alignment>]<br></pre>
2750 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
2752 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
2753 address from which to load. The pointer must point to a <a
2754 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
2755 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
2756 the number or order of execution of this <tt>load</tt> with other
2757 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
2760 <p>The location of memory pointed to is loaded.</p>
2762 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
2764 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
2765 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
2768 <!-- _______________________________________________________________________ -->
2769 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
2770 Instruction</a> </div>
2771 <div class="doc_text">
2773 <pre> store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
2774 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
2777 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
2779 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
2780 to store and an address at which to store it. The type of the '<tt><pointer></tt>'
2781 operand must be a pointer to the type of the '<tt><value></tt>'
2782 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
2783 optimizer is not allowed to modify the number or order of execution of
2784 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
2785 href="#i_store">store</a></tt> instructions.</p>
2787 <p>The contents of memory are updated to contain '<tt><value></tt>'
2788 at the location specified by the '<tt><pointer></tt>' operand.</p>
2790 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
2791 store i32 3, i32* %ptr <i>; yields {void}</i>
2792 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
2796 <!-- _______________________________________________________________________ -->
2797 <div class="doc_subsubsection">
2798 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
2801 <div class="doc_text">
2804 <result> = getelementptr <ty>* <ptrval>{, <ty> <idx>}*
2810 The '<tt>getelementptr</tt>' instruction is used to get the address of a
2811 subelement of an aggregate data structure.</p>
2815 <p>This instruction takes a list of integer operands that indicate what
2816 elements of the aggregate object to index to. The actual types of the arguments
2817 provided depend on the type of the first pointer argument. The
2818 '<tt>getelementptr</tt>' instruction is used to index down through the type
2819 levels of a structure or to a specific index in an array. When indexing into a
2820 structure, only <tt>i32</tt> integer constants are allowed. When indexing
2821 into an array or pointer, only integers of 32 or 64 bits are allowed, and will
2822 be sign extended to 64-bit values.</p>
2824 <p>For example, let's consider a C code fragment and how it gets
2825 compiled to LLVM:</p>
2827 <div class="doc_code">
2840 int *foo(struct ST *s) {
2841 return &s[1].Z.B[5][13];
2846 <p>The LLVM code generated by the GCC frontend is:</p>
2848 <div class="doc_code">
2850 %RT = type { i8 , [10 x [20 x i32]], i8 }
2851 %ST = type { i32, double, %RT }
2853 define i32* %foo(%ST* %s) {
2855 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
2863 <p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
2864 on the pointer type that is being indexed into. <a href="#t_pointer">Pointer</a>
2865 and <a href="#t_array">array</a> types can use a 32-bit or 64-bit
2866 <a href="#t_integer">integer</a> type but the value will always be sign extended
2867 to 64-bits. <a href="#t_struct">Structure</a> types require <tt>i32</tt>
2868 <b>constants</b>.</p>
2870 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
2871 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
2872 }</tt>' type, a structure. The second index indexes into the third element of
2873 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
2874 i8 }</tt>' type, another structure. The third index indexes into the second
2875 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
2876 array. The two dimensions of the array are subscripted into, yielding an
2877 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
2878 to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
2880 <p>Note that it is perfectly legal to index partially through a
2881 structure, returning a pointer to an inner element. Because of this,
2882 the LLVM code for the given testcase is equivalent to:</p>
2885 define i32* %foo(%ST* %s) {
2886 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
2887 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
2888 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
2889 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
2890 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
2895 <p>Note that it is undefined to access an array out of bounds: array and
2896 pointer indexes must always be within the defined bounds of the array type.
2897 The one exception for this rules is zero length arrays. These arrays are
2898 defined to be accessible as variable length arrays, which requires access
2899 beyond the zero'th element.</p>
2901 <p>The getelementptr instruction is often confusing. For some more insight
2902 into how it works, see <a href="GetElementPtr.html">the getelementptr
2908 <i>; yields [12 x i8]*:aptr</i>
2909 %aptr = getelementptr {i32, [12 x i8]}* %sptr, i64 0, i32 1
2913 <!-- ======================================================================= -->
2914 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
2916 <div class="doc_text">
2917 <p>The instructions in this category are the conversion instructions (casting)
2918 which all take a single operand and a type. They perform various bit conversions
2922 <!-- _______________________________________________________________________ -->
2923 <div class="doc_subsubsection">
2924 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
2926 <div class="doc_text">
2930 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
2935 The '<tt>trunc</tt>' instruction truncates its operand to the type <tt>ty2</tt>.
2940 The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
2941 be an <a href="#t_integer">integer</a> type, and a type that specifies the size
2942 and type of the result, which must be an <a href="#t_integer">integer</a>
2943 type. The bit size of <tt>value</tt> must be larger than the bit size of
2944 <tt>ty2</tt>. Equal sized types are not allowed.</p>
2948 The '<tt>trunc</tt>' instruction truncates the high order bits in <tt>value</tt>
2949 and converts the remaining bits to <tt>ty2</tt>. Since the source size must be
2950 larger than the destination size, <tt>trunc</tt> cannot be a <i>no-op cast</i>.
2951 It will always truncate bits.</p>
2955 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
2956 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
2957 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
2961 <!-- _______________________________________________________________________ -->
2962 <div class="doc_subsubsection">
2963 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
2965 <div class="doc_text">
2969 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
2973 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
2978 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
2979 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
2980 also be of <a href="#t_integer">integer</a> type. The bit size of the
2981 <tt>value</tt> must be smaller than the bit size of the destination type,
2985 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
2986 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
2988 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
2992 %X = zext i32 257 to i64 <i>; yields i64:257</i>
2993 %Y = zext i1 true to i32 <i>; yields i32:1</i>
2997 <!-- _______________________________________________________________________ -->
2998 <div class="doc_subsubsection">
2999 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
3001 <div class="doc_text">
3005 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
3009 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
3013 The '<tt>sext</tt>' instruction takes a value to cast, which must be of
3014 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3015 also be of <a href="#t_integer">integer</a> type. The bit size of the
3016 <tt>value</tt> must be smaller than the bit size of the destination type,
3021 The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
3022 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size of
3023 the type <tt>ty2</tt>.</p>
3025 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
3029 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
3030 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
3034 <!-- _______________________________________________________________________ -->
3035 <div class="doc_subsubsection">
3036 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
3039 <div class="doc_text">
3044 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
3048 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
3053 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
3054 point</a> value to cast and a <a href="#t_floating">floating point</a> type to
3055 cast it to. The size of <tt>value</tt> must be larger than the size of
3056 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
3057 <i>no-op cast</i>.</p>
3060 <p> The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
3061 <a href="#t_floating">floating point</a> type to a smaller
3062 <a href="#t_floating">floating point</a> type. If the value cannot fit within
3063 the destination type, <tt>ty2</tt>, then the results are undefined.</p>
3067 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
3068 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
3072 <!-- _______________________________________________________________________ -->
3073 <div class="doc_subsubsection">
3074 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
3076 <div class="doc_text">
3080 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
3084 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
3085 floating point value.</p>
3088 <p>The '<tt>fpext</tt>' instruction takes a
3089 <a href="#t_floating">floating point</a> <tt>value</tt> to cast,
3090 and a <a href="#t_floating">floating point</a> type to cast it to. The source
3091 type must be smaller than the destination type.</p>
3094 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
3095 <a href="#t_floating">floating point</a> type to a larger
3096 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
3097 used to make a <i>no-op cast</i> because it always changes bits. Use
3098 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
3102 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
3103 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
3107 <!-- _______________________________________________________________________ -->
3108 <div class="doc_subsubsection">
3109 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
3111 <div class="doc_text">
3115 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
3119 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
3120 unsigned integer equivalent of type <tt>ty2</tt>.
3124 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
3125 <a href="#t_floating">floating point</a> value, and a type to cast it to, which
3126 must be an <a href="#t_integer">integer</a> type.</p>
3129 <p> The '<tt>fptoui</tt>' instruction converts its
3130 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3131 towards zero) unsigned integer value. If the value cannot fit in <tt>ty2</tt>,
3132 the results are undefined.</p>
3136 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
3137 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
3138 %X = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
3142 <!-- _______________________________________________________________________ -->
3143 <div class="doc_subsubsection">
3144 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
3146 <div class="doc_text">
3150 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
3154 <p>The '<tt>fptosi</tt>' instruction converts
3155 <a href="#t_floating">floating point</a> <tt>value</tt> to type <tt>ty2</tt>.
3160 <p> The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
3161 <a href="#t_floating">floating point</a> value, and a type to cast it to, which
3162 must also be an <a href="#t_integer">integer</a> type.</p>
3165 <p>The '<tt>fptosi</tt>' instruction converts its
3166 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3167 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
3168 the results are undefined.</p>
3172 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
3173 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
3174 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
3178 <!-- _______________________________________________________________________ -->
3179 <div class="doc_subsubsection">
3180 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
3182 <div class="doc_text">
3186 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3190 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
3191 integer and converts that value to the <tt>ty2</tt> type.</p>
3195 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be an
3196 <a href="#t_integer">integer</a> value, and a type to cast it to, which must
3197 be a <a href="#t_floating">floating point</a> type.</p>
3200 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
3201 integer quantity and converts it to the corresponding floating point value. If
3202 the value cannot fit in the floating point value, the results are undefined.</p>
3207 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
3208 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
3212 <!-- _______________________________________________________________________ -->
3213 <div class="doc_subsubsection">
3214 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
3216 <div class="doc_text">
3220 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3224 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed
3225 integer and converts that value to the <tt>ty2</tt> type.</p>
3228 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be an
3229 <a href="#t_integer">integer</a> value, and a type to cast it to, which must be
3230 a <a href="#t_floating">floating point</a> type.</p>
3233 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed
3234 integer quantity and converts it to the corresponding floating point value. If
3235 the value cannot fit in the floating point value, the results are undefined.</p>
3239 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
3240 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
3244 <!-- _______________________________________________________________________ -->
3245 <div class="doc_subsubsection">
3246 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
3248 <div class="doc_text">
3252 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
3256 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
3257 the integer type <tt>ty2</tt>.</p>
3260 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
3261 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
3262 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.
3265 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
3266 <tt>ty2</tt> by interpreting the pointer value as an integer and either
3267 truncating or zero extending that value to the size of the integer type. If
3268 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
3269 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
3270 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
3275 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
3276 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
3280 <!-- _______________________________________________________________________ -->
3281 <div class="doc_subsubsection">
3282 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
3284 <div class="doc_text">
3288 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
3292 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to
3293 a pointer type, <tt>ty2</tt>.</p>
3296 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
3297 value to cast, and a type to cast it to, which must be a
3298 <a href="#t_pointer">pointer</a> type.
3301 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
3302 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
3303 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
3304 size of a pointer then a truncation is done. If <tt>value</tt> is smaller than
3305 the size of a pointer then a zero extension is done. If they are the same size,
3306 nothing is done (<i>no-op cast</i>).</p>
3310 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
3311 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
3312 %Y = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
3316 <!-- _______________________________________________________________________ -->
3317 <div class="doc_subsubsection">
3318 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
3320 <div class="doc_text">
3324 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
3328 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3329 <tt>ty2</tt> without changing any bits.</p>
3332 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be
3333 a first class value, and a type to cast it to, which must also be a <a
3334 href="#t_firstclass">first class</a> type. The bit sizes of <tt>value</tt>
3335 and the destination type, <tt>ty2</tt>, must be identical. If the source
3336 type is a pointer, the destination type must also be a pointer.</p>
3339 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3340 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
3341 this conversion. The conversion is done as if the <tt>value</tt> had been
3342 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only be
3343 converted to other pointer types with this instruction. To convert pointers to
3344 other types, use the <a href="#i_inttoptr">inttoptr</a> or
3345 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
3349 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
3350 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
3351 %Z = bitcast <2xint> %V to i64; <i>; yields i64: %V</i>
3355 <!-- ======================================================================= -->
3356 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
3357 <div class="doc_text">
3358 <p>The instructions in this category are the "miscellaneous"
3359 instructions, which defy better classification.</p>
3362 <!-- _______________________________________________________________________ -->
3363 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
3365 <div class="doc_text">
3367 <pre> <result> = icmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3370 <p>The '<tt>icmp</tt>' instruction returns a boolean value based on comparison
3371 of its two integer operands.</p>
3373 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
3374 the condition code indicating the kind of comparison to perform. It is not
3375 a value, just a keyword. The possible condition code are:
3377 <li><tt>eq</tt>: equal</li>
3378 <li><tt>ne</tt>: not equal </li>
3379 <li><tt>ugt</tt>: unsigned greater than</li>
3380 <li><tt>uge</tt>: unsigned greater or equal</li>
3381 <li><tt>ult</tt>: unsigned less than</li>
3382 <li><tt>ule</tt>: unsigned less or equal</li>
3383 <li><tt>sgt</tt>: signed greater than</li>
3384 <li><tt>sge</tt>: signed greater or equal</li>
3385 <li><tt>slt</tt>: signed less than</li>
3386 <li><tt>sle</tt>: signed less or equal</li>
3388 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
3389 <a href="#t_pointer">pointer</a> typed. They must also be identical types.</p>
3391 <p>The '<tt>icmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3392 the condition code given as <tt>cond</tt>. The comparison performed always
3393 yields a <a href="#t_primitive">i1</a> result, as follows:
3395 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
3396 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3398 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
3399 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3400 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
3401 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3402 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
3403 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3404 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
3405 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3406 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
3407 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3408 <li><tt>sgt</tt>: interprets the operands as signed values and yields
3409 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3410 <li><tt>sge</tt>: interprets the operands as signed values and yields
3411 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3412 <li><tt>slt</tt>: interprets the operands as signed values and yields
3413 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3414 <li><tt>sle</tt>: interprets the operands as signed values and yields
3415 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3417 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
3418 values are compared as if they were integers.</p>
3421 <pre> <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
3422 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
3423 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
3424 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
3425 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
3426 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
3430 <!-- _______________________________________________________________________ -->
3431 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
3433 <div class="doc_text">
3435 <pre> <result> = fcmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3438 <p>The '<tt>fcmp</tt>' instruction returns a boolean value based on comparison
3439 of its floating point operands.</p>
3441 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
3442 the condition code indicating the kind of comparison to perform. It is not
3443 a value, just a keyword. The possible condition code are:
3445 <li><tt>false</tt>: no comparison, always returns false</li>
3446 <li><tt>oeq</tt>: ordered and equal</li>
3447 <li><tt>ogt</tt>: ordered and greater than </li>
3448 <li><tt>oge</tt>: ordered and greater than or equal</li>
3449 <li><tt>olt</tt>: ordered and less than </li>
3450 <li><tt>ole</tt>: ordered and less than or equal</li>
3451 <li><tt>one</tt>: ordered and not equal</li>
3452 <li><tt>ord</tt>: ordered (no nans)</li>
3453 <li><tt>ueq</tt>: unordered or equal</li>
3454 <li><tt>ugt</tt>: unordered or greater than </li>
3455 <li><tt>uge</tt>: unordered or greater than or equal</li>
3456 <li><tt>ult</tt>: unordered or less than </li>
3457 <li><tt>ule</tt>: unordered or less than or equal</li>
3458 <li><tt>une</tt>: unordered or not equal</li>
3459 <li><tt>uno</tt>: unordered (either nans)</li>
3460 <li><tt>true</tt>: no comparison, always returns true</li>
3462 <p><i>Ordered</i> means that neither operand is a QNAN while
3463 <i>unordered</i> means that either operand may be a QNAN.</p>
3464 <p>The <tt>val1</tt> and <tt>val2</tt> arguments must be
3465 <a href="#t_floating">floating point</a> typed. They must have identical
3468 <p>The '<tt>fcmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3469 the condition code given as <tt>cond</tt>. The comparison performed always
3470 yields a <a href="#t_primitive">i1</a> result, as follows:
3472 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
3473 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3474 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3475 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3476 <tt>var1</tt> is greather than <tt>var2</tt>.</li>
3477 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3478 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3479 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3480 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3481 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3482 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3483 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3484 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3485 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
3486 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
3487 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3488 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
3489 <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3490 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
3491 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3492 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
3493 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3494 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
3495 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3496 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
3497 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3498 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
3499 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
3503 <pre> <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
3504 <result> = icmp one float 4.0, 5.0 <i>; yields: result=true</i>
3505 <result> = icmp olt float 4.0, 5.0 <i>; yields: result=true</i>
3506 <result> = icmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
3510 <!-- _______________________________________________________________________ -->
3511 <div class="doc_subsubsection"> <a name="i_phi">'<tt>phi</tt>'
3512 Instruction</a> </div>
3513 <div class="doc_text">
3515 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
3517 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
3518 the SSA graph representing the function.</p>
3520 <p>The type of the incoming values is specified with the first type
3521 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
3522 as arguments, with one pair for each predecessor basic block of the
3523 current block. Only values of <a href="#t_firstclass">first class</a>
3524 type may be used as the value arguments to the PHI node. Only labels
3525 may be used as the label arguments.</p>
3526 <p>There must be no non-phi instructions between the start of a basic
3527 block and the PHI instructions: i.e. PHI instructions must be first in
3530 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
3531 specified by the pair corresponding to the predecessor basic block that executed
3532 just prior to the current block.</p>
3534 <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>
3537 <!-- _______________________________________________________________________ -->
3538 <div class="doc_subsubsection">
3539 <a name="i_select">'<tt>select</tt>' Instruction</a>
3542 <div class="doc_text">
3547 <result> = select i1 <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
3553 The '<tt>select</tt>' instruction is used to choose one value based on a
3554 condition, without branching.
3561 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.
3567 If the boolean condition evaluates to true, the instruction returns the first
3568 value argument; otherwise, it returns the second value argument.
3574 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
3579 <!-- _______________________________________________________________________ -->
3580 <div class="doc_subsubsection">
3581 <a name="i_call">'<tt>call</tt>' Instruction</a>
3584 <div class="doc_text">
3588 <result> = [tail] call [<a href="#callingconv">cconv</a>] <ty> [<fnty>*] <fnptrval>(<param list>)
3593 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
3597 <p>This instruction requires several arguments:</p>
3601 <p>The optional "tail" marker indicates whether the callee function accesses
3602 any allocas or varargs in the caller. If the "tail" marker is present, the
3603 function call is eligible for tail call optimization. Note that calls may
3604 be marked "tail" even if they do not occur before a <a
3605 href="#i_ret"><tt>ret</tt></a> instruction.
3608 <p>The optional "cconv" marker indicates which <a href="#callingconv">calling
3609 convention</a> the call should use. If none is specified, the call defaults
3610 to using C calling conventions.
3613 <p>'<tt>ty</tt>': the type of the call instruction itself which is also
3614 the type of the return value. Functions that return no value are marked
3615 <tt><a href="#t_void">void</a></tt>.</p>
3618 <p>'<tt>fnty</tt>': shall be the signature of the pointer to function
3619 value being invoked. The argument types must match the types implied by
3620 this signature. This type can be omitted if the function is not varargs
3621 and if the function type does not return a pointer to a function.</p>
3624 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
3625 be invoked. In most cases, this is a direct function invocation, but
3626 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
3627 to function value.</p>
3630 <p>'<tt>function args</tt>': argument list whose types match the
3631 function signature argument types. All arguments must be of
3632 <a href="#t_firstclass">first class</a> type. If the function signature
3633 indicates the function accepts a variable number of arguments, the extra
3634 arguments can be specified.</p>
3640 <p>The '<tt>call</tt>' instruction is used to cause control flow to
3641 transfer to a specified function, with its incoming arguments bound to
3642 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
3643 instruction in the called function, control flow continues with the
3644 instruction after the function call, and the return value of the
3645 function is bound to the result argument. This is a simpler case of
3646 the <a href="#i_invoke">invoke</a> instruction.</p>
3651 %retval = call i32 @test(i32 %argc)
3652 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42);
3653 %X = tail call i32 @foo()
3654 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo()
3655 %Z = call void %foo(i8 97 signext)
3660 <!-- _______________________________________________________________________ -->
3661 <div class="doc_subsubsection">
3662 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
3665 <div class="doc_text">
3670 <resultval> = va_arg <va_list*> <arglist>, <argty>
3675 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
3676 the "variable argument" area of a function call. It is used to implement the
3677 <tt>va_arg</tt> macro in C.</p>
3681 <p>This instruction takes a <tt>va_list*</tt> value and the type of
3682 the argument. It returns a value of the specified argument type and
3683 increments the <tt>va_list</tt> to point to the next argument. The
3684 actual type of <tt>va_list</tt> is target specific.</p>
3688 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
3689 type from the specified <tt>va_list</tt> and causes the
3690 <tt>va_list</tt> to point to the next argument. For more information,
3691 see the variable argument handling <a href="#int_varargs">Intrinsic
3694 <p>It is legal for this instruction to be called in a function which does not
3695 take a variable number of arguments, for example, the <tt>vfprintf</tt>
3698 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
3699 href="#intrinsics">intrinsic function</a> because it takes a type as an
3704 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
3708 <!-- *********************************************************************** -->
3709 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
3710 <!-- *********************************************************************** -->
3712 <div class="doc_text">
3714 <p>LLVM supports the notion of an "intrinsic function". These functions have
3715 well known names and semantics and are required to follow certain restrictions.
3716 Overall, these intrinsics represent an extension mechanism for the LLVM
3717 language that does not require changing all of the transformations in LLVM when
3718 adding to the language (or the bitcode reader/writer, the parser, etc...).</p>
3720 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
3721 prefix is reserved in LLVM for intrinsic names; thus, function names may not
3722 begin with this prefix. Intrinsic functions must always be external functions:
3723 you cannot define the body of intrinsic functions. Intrinsic functions may
3724 only be used in call or invoke instructions: it is illegal to take the address
3725 of an intrinsic function. Additionally, because intrinsic functions are part
3726 of the LLVM language, it is required if any are added that they be documented
3729 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
3730 a family of functions that perform the same operation but on different data
3731 types. Because LLVM can represent over 8 million different integer types,
3732 overloading is used commonly to allow an intrinsic function to operate on any
3733 integer type. One or more of the argument types or the result type can be
3734 overloaded to accept any integer type. Argument types may also be defined as
3735 exactly matching a previous argument's type or the result type. This allows an
3736 intrinsic function which accepts multiple arguments, but needs all of them to
3737 be of the same type, to only be overloaded with respect to a single argument or
3740 <p>Overloaded intrinsics will have the names of its overloaded argument types
3741 encoded into its function name, each preceded by a period. Only those types
3742 which are overloaded result in a name suffix. Arguments whose type is matched
3743 against another type do not. For example, the <tt>llvm.ctpop</tt> function can
3744 take an integer of any width and returns an integer of exactly the same integer
3745 width. This leads to a family of functions such as
3746 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29 %val)</tt>.
3747 Only one type, the return type, is overloaded, and only one type suffix is
3748 required. Because the argument's type is matched against the return type, it
3749 does not require its own name suffix.</p>
3751 <p>To learn how to add an intrinsic function, please see the
3752 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.
3757 <!-- ======================================================================= -->
3758 <div class="doc_subsection">
3759 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
3762 <div class="doc_text">
3764 <p>Variable argument support is defined in LLVM with the <a
3765 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
3766 intrinsic functions. These functions are related to the similarly
3767 named macros defined in the <tt><stdarg.h></tt> header file.</p>
3769 <p>All of these functions operate on arguments that use a
3770 target-specific value type "<tt>va_list</tt>". The LLVM assembly
3771 language reference manual does not define what this type is, so all
3772 transformations should be prepared to handle these functions regardless of
3775 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
3776 instruction and the variable argument handling intrinsic functions are
3779 <div class="doc_code">
3781 define i32 @test(i32 %X, ...) {
3782 ; Initialize variable argument processing
3784 %ap2 = bitcast i8** %ap to i8*
3785 call void @llvm.va_start(i8* %ap2)
3787 ; Read a single integer argument
3788 %tmp = va_arg i8** %ap, i32
3790 ; Demonstrate usage of llvm.va_copy and llvm.va_end
3792 %aq2 = bitcast i8** %aq to i8*
3793 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
3794 call void @llvm.va_end(i8* %aq2)
3796 ; Stop processing of arguments.
3797 call void @llvm.va_end(i8* %ap2)
3801 declare void @llvm.va_start(i8*)
3802 declare void @llvm.va_copy(i8*, i8*)
3803 declare void @llvm.va_end(i8*)
3809 <!-- _______________________________________________________________________ -->
3810 <div class="doc_subsubsection">
3811 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
3815 <div class="doc_text">
3817 <pre> declare void %llvm.va_start(i8* <arglist>)<br></pre>
3819 <P>The '<tt>llvm.va_start</tt>' intrinsic initializes
3820 <tt>*<arglist></tt> for subsequent use by <tt><a
3821 href="#i_va_arg">va_arg</a></tt>.</p>
3825 <P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
3829 <P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
3830 macro available in C. In a target-dependent way, it initializes the
3831 <tt>va_list</tt> element to which the argument points, so that the next call to
3832 <tt>va_arg</tt> will produce the first variable argument passed to the function.
3833 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
3834 last argument of the function as the compiler can figure that out.</p>
3838 <!-- _______________________________________________________________________ -->
3839 <div class="doc_subsubsection">
3840 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
3843 <div class="doc_text">
3845 <pre> declare void @llvm.va_end(i8* <arglist>)<br></pre>
3848 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
3849 which has been initialized previously with <tt><a href="#int_va_start">llvm.va_start</a></tt>
3850 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
3854 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
3858 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
3859 macro available in C. In a target-dependent way, it destroys the
3860 <tt>va_list</tt> element to which the argument points. Calls to <a
3861 href="#int_va_start"><tt>llvm.va_start</tt></a> and <a href="#int_va_copy">
3862 <tt>llvm.va_copy</tt></a> must be matched exactly with calls to
3863 <tt>llvm.va_end</tt>.</p>
3867 <!-- _______________________________________________________________________ -->
3868 <div class="doc_subsubsection">
3869 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
3872 <div class="doc_text">
3877 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
3882 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
3883 from the source argument list to the destination argument list.</p>
3887 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
3888 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
3893 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
3894 macro available in C. In a target-dependent way, it copies the source
3895 <tt>va_list</tt> element into the destination <tt>va_list</tt> element. This
3896 intrinsic is necessary because the <tt><a href="#int_va_start">
3897 llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
3898 example, memory allocation.</p>
3902 <!-- ======================================================================= -->
3903 <div class="doc_subsection">
3904 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
3907 <div class="doc_text">
3910 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
3911 Collection</a> requires the implementation and generation of these intrinsics.
3912 These intrinsics allow identification of <a href="#int_gcroot">GC roots on the
3913 stack</a>, as well as garbage collector implementations that require <a
3914 href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a> barriers.
3915 Front-ends for type-safe garbage collected languages should generate these
3916 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
3917 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
3921 <!-- _______________________________________________________________________ -->
3922 <div class="doc_subsubsection">
3923 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
3926 <div class="doc_text">
3931 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
3936 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
3937 the code generator, and allows some metadata to be associated with it.</p>
3941 <p>The first argument specifies the address of a stack object that contains the
3942 root pointer. The second pointer (which must be either a constant or a global
3943 value address) contains the meta-data to be associated with the root.</p>
3947 <p>At runtime, a call to this intrinsics stores a null pointer into the "ptrloc"
3948 location. At compile-time, the code generator generates information to allow
3949 the runtime to find the pointer at GC safe points.
3955 <!-- _______________________________________________________________________ -->
3956 <div class="doc_subsubsection">
3957 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
3960 <div class="doc_text">
3965 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
3970 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
3971 locations, allowing garbage collector implementations that require read
3976 <p>The second argument is the address to read from, which should be an address
3977 allocated from the garbage collector. The first object is a pointer to the
3978 start of the referenced object, if needed by the language runtime (otherwise
3983 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
3984 instruction, but may be replaced with substantially more complex code by the
3985 garbage collector runtime, as needed.</p>
3990 <!-- _______________________________________________________________________ -->
3991 <div class="doc_subsubsection">
3992 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
3995 <div class="doc_text">
4000 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
4005 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
4006 locations, allowing garbage collector implementations that require write
4007 barriers (such as generational or reference counting collectors).</p>
4011 <p>The first argument is the reference to store, the second is the start of the
4012 object to store it to, and the third is the address of the field of Obj to
4013 store to. If the runtime does not require a pointer to the object, Obj may be
4018 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
4019 instruction, but may be replaced with substantially more complex code by the
4020 garbage collector runtime, as needed.</p>
4026 <!-- ======================================================================= -->
4027 <div class="doc_subsection">
4028 <a name="int_codegen">Code Generator Intrinsics</a>
4031 <div class="doc_text">
4033 These intrinsics are provided by LLVM to expose special features that may only
4034 be implemented with code generator support.
4039 <!-- _______________________________________________________________________ -->
4040 <div class="doc_subsubsection">
4041 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
4044 <div class="doc_text">
4048 declare i8 *@llvm.returnaddress(i32 <level>)
4054 The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
4055 target-specific value indicating the return address of the current function
4056 or one of its callers.
4062 The argument to this intrinsic indicates which function to return the address
4063 for. Zero indicates the calling function, one indicates its caller, etc. The
4064 argument is <b>required</b> to be a constant integer value.
4070 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
4071 the return address of the specified call frame, or zero if it cannot be
4072 identified. The value returned by this intrinsic is likely to be incorrect or 0
4073 for arguments other than zero, so it should only be used for debugging purposes.
4077 Note that calling this intrinsic does not prevent function inlining or other
4078 aggressive transformations, so the value returned may not be that of the obvious
4079 source-language caller.
4084 <!-- _______________________________________________________________________ -->
4085 <div class="doc_subsubsection">
4086 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
4089 <div class="doc_text">
4093 declare i8 *@llvm.frameaddress(i32 <level>)
4099 The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
4100 target-specific frame pointer value for the specified stack frame.
4106 The argument to this intrinsic indicates which function to return the frame
4107 pointer for. Zero indicates the calling function, one indicates its caller,
4108 etc. The argument is <b>required</b> to be a constant integer value.
4114 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
4115 the frame address of the specified call frame, or zero if it cannot be
4116 identified. The value returned by this intrinsic is likely to be incorrect or 0
4117 for arguments other than zero, so it should only be used for debugging purposes.
4121 Note that calling this intrinsic does not prevent function inlining or other
4122 aggressive transformations, so the value returned may not be that of the obvious
4123 source-language caller.
4127 <!-- _______________________________________________________________________ -->
4128 <div class="doc_subsubsection">
4129 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
4132 <div class="doc_text">
4136 declare i8 *@llvm.stacksave()
4142 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
4143 the function stack, for use with <a href="#int_stackrestore">
4144 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
4145 features like scoped automatic variable sized arrays in C99.
4151 This intrinsic returns a opaque pointer value that can be passed to <a
4152 href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
4153 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
4154 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
4155 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
4156 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
4157 that were allocated after the <tt>llvm.stacksave</tt> was executed.
4162 <!-- _______________________________________________________________________ -->
4163 <div class="doc_subsubsection">
4164 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
4167 <div class="doc_text">
4171 declare void @llvm.stackrestore(i8 * %ptr)
4177 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
4178 the function stack to the state it was in when the corresponding <a
4179 href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
4180 useful for implementing language features like scoped automatic variable sized
4187 See the description for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.
4193 <!-- _______________________________________________________________________ -->
4194 <div class="doc_subsubsection">
4195 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
4198 <div class="doc_text">
4202 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>)
4209 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
4210 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
4212 effect on the behavior of the program but can change its performance
4219 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
4220 determining if the fetch should be for a read (0) or write (1), and
4221 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
4222 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
4223 <tt>locality</tt> arguments must be constant integers.
4229 This intrinsic does not modify the behavior of the program. In particular,
4230 prefetches cannot trap and do not produce a value. On targets that support this
4231 intrinsic, the prefetch can provide hints to the processor cache for better
4237 <!-- _______________________________________________________________________ -->
4238 <div class="doc_subsubsection">
4239 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
4242 <div class="doc_text">
4246 declare void @llvm.pcmarker(i32 <id>)
4253 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
4255 code to simulators and other tools. The method is target specific, but it is
4256 expected that the marker will use exported symbols to transmit the PC of the marker.
4257 The marker makes no guarantees that it will remain with any specific instruction
4258 after optimizations. It is possible that the presence of a marker will inhibit
4259 optimizations. The intended use is to be inserted after optimizations to allow
4260 correlations of simulation runs.
4266 <tt>id</tt> is a numerical id identifying the marker.
4272 This intrinsic does not modify the behavior of the program. Backends that do not
4273 support this intrinisic may ignore it.
4278 <!-- _______________________________________________________________________ -->
4279 <div class="doc_subsubsection">
4280 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
4283 <div class="doc_text">
4287 declare i64 @llvm.readcyclecounter( )
4294 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
4295 counter register (or similar low latency, high accuracy clocks) on those targets
4296 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
4297 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
4298 should only be used for small timings.
4304 When directly supported, reading the cycle counter should not modify any memory.
4305 Implementations are allowed to either return a application specific value or a
4306 system wide value. On backends without support, this is lowered to a constant 0.
4311 <!-- ======================================================================= -->
4312 <div class="doc_subsection">
4313 <a name="int_libc">Standard C Library Intrinsics</a>
4316 <div class="doc_text">
4318 LLVM provides intrinsics for a few important standard C library functions.
4319 These intrinsics allow source-language front-ends to pass information about the
4320 alignment of the pointer arguments to the code generator, providing opportunity
4321 for more efficient code generation.
4326 <!-- _______________________________________________________________________ -->
4327 <div class="doc_subsubsection">
4328 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
4331 <div class="doc_text">
4335 declare void @llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
4336 i32 <len>, i32 <align>)
4337 declare void @llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
4338 i64 <len>, i32 <align>)
4344 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4345 location to the destination location.
4349 Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
4350 intrinsics do not return a value, and takes an extra alignment argument.
4356 The first argument is a pointer to the destination, the second is a pointer to
4357 the source. The third argument is an integer argument
4358 specifying the number of bytes to copy, and the fourth argument is the alignment
4359 of the source and destination locations.
4363 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4364 the caller guarantees that both the source and destination pointers are aligned
4371 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4372 location to the destination location, which are not allowed to overlap. It
4373 copies "len" bytes of memory over. If the argument is known to be aligned to
4374 some boundary, this can be specified as the fourth argument, otherwise it should
4380 <!-- _______________________________________________________________________ -->
4381 <div class="doc_subsubsection">
4382 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
4385 <div class="doc_text">
4389 declare void @llvm.memmove.i32(i8 * <dest>, i8 * <src>,
4390 i32 <len>, i32 <align>)
4391 declare void @llvm.memmove.i64(i8 * <dest>, i8 * <src>,
4392 i64 <len>, i32 <align>)
4398 The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
4399 location to the destination location. It is similar to the
4400 '<tt>llvm.memcmp</tt>' intrinsic but allows the two memory locations to overlap.
4404 Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
4405 intrinsics do not return a value, and takes an extra alignment argument.
4411 The first argument is a pointer to the destination, the second is a pointer to
4412 the source. The third argument is an integer argument
4413 specifying the number of bytes to copy, and the fourth argument is the alignment
4414 of the source and destination locations.
4418 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4419 the caller guarantees that the source and destination pointers are aligned to
4426 The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
4427 location to the destination location, which may overlap. It
4428 copies "len" bytes of memory over. If the argument is known to be aligned to
4429 some boundary, this can be specified as the fourth argument, otherwise it should
4435 <!-- _______________________________________________________________________ -->
4436 <div class="doc_subsubsection">
4437 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
4440 <div class="doc_text">
4444 declare void @llvm.memset.i32(i8 * <dest>, i8 <val>,
4445 i32 <len>, i32 <align>)
4446 declare void @llvm.memset.i64(i8 * <dest>, i8 <val>,
4447 i64 <len>, i32 <align>)
4453 The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
4458 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
4459 does not return a value, and takes an extra alignment argument.
4465 The first argument is a pointer to the destination to fill, the second is the
4466 byte value to fill it with, the third argument is an integer
4467 argument specifying the number of bytes to fill, and the fourth argument is the
4468 known alignment of destination location.
4472 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4473 the caller guarantees that the destination pointer is aligned to that boundary.
4479 The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
4481 destination location. If the argument is known to be aligned to some boundary,
4482 this can be specified as the fourth argument, otherwise it should be set to 0 or
4488 <!-- _______________________________________________________________________ -->
4489 <div class="doc_subsubsection">
4490 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
4493 <div class="doc_text">
4496 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
4497 floating point or vector of floating point type. Not all targets support all
4500 declare float @llvm.sqrt.f32(float %Val)
4501 declare double @llvm.sqrt.f64(double %Val)
4502 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
4503 declare fp128 @llvm.sqrt.f128(fp128 %Val)
4504 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
4510 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
4511 returning the same value as the libm '<tt>sqrt</tt>' functions would. Unlike
4512 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
4513 negative numbers (which allows for better optimization).
4519 The argument and return value are floating point numbers of the same type.
4525 This function returns the sqrt of the specified operand if it is a nonnegative
4526 floating point number.
4530 <!-- _______________________________________________________________________ -->
4531 <div class="doc_subsubsection">
4532 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
4535 <div class="doc_text">
4538 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
4539 floating point or vector of floating point type. Not all targets support all
4542 declare float @llvm.powi.f32(float %Val, i32 %power)
4543 declare double @llvm.powi.f64(double %Val, i32 %power)
4544 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
4545 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
4546 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
4552 The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
4553 specified (positive or negative) power. The order of evaluation of
4554 multiplications is not defined. When a vector of floating point type is
4555 used, the second argument remains a scalar integer value.
4561 The second argument is an integer power, and the first is a value to raise to
4568 This function returns the first value raised to the second power with an
4569 unspecified sequence of rounding operations.</p>
4572 <!-- _______________________________________________________________________ -->
4573 <div class="doc_subsubsection">
4574 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
4577 <div class="doc_text">
4580 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
4581 floating point or vector of floating point type. Not all targets support all
4584 declare float @llvm.sin.f32(float %Val)
4585 declare double @llvm.sin.f64(double %Val)
4586 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
4587 declare fp128 @llvm.sin.f128(fp128 %Val)
4588 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
4594 The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.
4600 The argument and return value are floating point numbers of the same type.
4606 This function returns the sine of the specified operand, returning the
4607 same values as the libm <tt>sin</tt> functions would, and handles error
4608 conditions in the same way.</p>
4611 <!-- _______________________________________________________________________ -->
4612 <div class="doc_subsubsection">
4613 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
4616 <div class="doc_text">
4619 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
4620 floating point or vector of floating point type. Not all targets support all
4623 declare float @llvm.cos.f32(float %Val)
4624 declare double @llvm.cos.f64(double %Val)
4625 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
4626 declare fp128 @llvm.cos.f128(fp128 %Val)
4627 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
4633 The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.
4639 The argument and return value are floating point numbers of the same type.
4645 This function returns the cosine of the specified operand, returning the
4646 same values as the libm <tt>cos</tt> functions would, and handles error
4647 conditions in the same way.</p>
4650 <!-- _______________________________________________________________________ -->
4651 <div class="doc_subsubsection">
4652 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
4655 <div class="doc_text">
4658 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
4659 floating point or vector of floating point type. Not all targets support all
4662 declare float @llvm.pow.f32(float %Val, float %Power)
4663 declare double @llvm.pow.f64(double %Val, double %Power)
4664 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
4665 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
4666 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
4672 The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
4673 specified (positive or negative) power.
4679 The second argument is a floating point power, and the first is a value to
4680 raise to that power.
4686 This function returns the first value raised to the second power,
4688 same values as the libm <tt>pow</tt> functions would, and handles error
4689 conditions in the same way.</p>
4693 <!-- ======================================================================= -->
4694 <div class="doc_subsection">
4695 <a name="int_manip">Bit Manipulation Intrinsics</a>
4698 <div class="doc_text">
4700 LLVM provides intrinsics for a few important bit manipulation operations.
4701 These allow efficient code generation for some algorithms.
4706 <!-- _______________________________________________________________________ -->
4707 <div class="doc_subsubsection">
4708 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
4711 <div class="doc_text">
4714 <p>This is an overloaded intrinsic function. You can use bswap on any integer
4715 type that is an even number of bytes (i.e. BitWidth % 16 == 0).
4717 declare i16 @llvm.bswap.i16(i16 <id>)
4718 declare i32 @llvm.bswap.i32(i32 <id>)
4719 declare i64 @llvm.bswap.i64(i64 <id>)
4725 The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
4726 values with an even number of bytes (positive multiple of 16 bits). These are
4727 useful for performing operations on data that is not in the target's native
4734 The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
4735 and low byte of the input i16 swapped. Similarly, the <tt>llvm.bswap.i32</tt>
4736 intrinsic returns an i32 value that has the four bytes of the input i32
4737 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the returned
4738 i32 will have its bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i48</tt>,
4739 <tt>llvm.bswap.i64</tt> and other intrinsics extend this concept to
4740 additional even-byte lengths (6 bytes, 8 bytes and more, respectively).
4745 <!-- _______________________________________________________________________ -->
4746 <div class="doc_subsubsection">
4747 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
4750 <div class="doc_text">
4753 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
4754 width. Not all targets support all bit widths however.
4756 declare i8 @llvm.ctpop.i8 (i8 <src>)
4757 declare i16 @llvm.ctpop.i16(i16 <src>)
4758 declare i32 @llvm.ctpop.i32(i32 <src>)
4759 declare i64 @llvm.ctpop.i64(i64 <src>)
4760 declare i256 @llvm.ctpop.i256(i256 <src>)
4766 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
4773 The only argument is the value to be counted. The argument may be of any
4774 integer type. The return type must match the argument type.
4780 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
4784 <!-- _______________________________________________________________________ -->
4785 <div class="doc_subsubsection">
4786 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
4789 <div class="doc_text">
4792 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
4793 integer bit width. Not all targets support all bit widths however.
4795 declare i8 @llvm.ctlz.i8 (i8 <src>)
4796 declare i16 @llvm.ctlz.i16(i16 <src>)
4797 declare i32 @llvm.ctlz.i32(i32 <src>)
4798 declare i64 @llvm.ctlz.i64(i64 <src>)
4799 declare i256 @llvm.ctlz.i256(i256 <src>)
4805 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
4806 leading zeros in a variable.
4812 The only argument is the value to be counted. The argument may be of any
4813 integer type. The return type must match the argument type.
4819 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
4820 in a variable. If the src == 0 then the result is the size in bits of the type
4821 of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.
4827 <!-- _______________________________________________________________________ -->
4828 <div class="doc_subsubsection">
4829 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
4832 <div class="doc_text">
4835 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
4836 integer bit width. Not all targets support all bit widths however.
4838 declare i8 @llvm.cttz.i8 (i8 <src>)
4839 declare i16 @llvm.cttz.i16(i16 <src>)
4840 declare i32 @llvm.cttz.i32(i32 <src>)
4841 declare i64 @llvm.cttz.i64(i64 <src>)
4842 declare i256 @llvm.cttz.i256(i256 <src>)
4848 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
4855 The only argument is the value to be counted. The argument may be of any
4856 integer type. The return type must match the argument type.
4862 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
4863 in a variable. If the src == 0 then the result is the size in bits of the type
4864 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
4868 <!-- _______________________________________________________________________ -->
4869 <div class="doc_subsubsection">
4870 <a name="int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic</a>
4873 <div class="doc_text">
4876 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.select</tt>
4877 on any integer bit width.
4879 declare i17 @llvm.part.select.i17 (i17 %val, i32 %loBit, i32 %hiBit)
4880 declare i29 @llvm.part.select.i29 (i29 %val, i32 %loBit, i32 %hiBit)
4884 <p>The '<tt>llvm.part.select</tt>' family of intrinsic functions selects a
4885 range of bits from an integer value and returns them in the same bit width as
4886 the original value.</p>
4889 <p>The first argument, <tt>%val</tt> and the result may be integer types of
4890 any bit width but they must have the same bit width. The second and third
4891 arguments must be <tt>i32</tt> type since they specify only a bit index.</p>
4894 <p>The operation of the '<tt>llvm.part.select</tt>' intrinsic has two modes
4895 of operation: forwards and reverse. If <tt>%loBit</tt> is greater than
4896 <tt>%hiBits</tt> then the intrinsic operates in reverse mode. Otherwise it
4897 operates in forward mode.</p>
4898 <p>In forward mode, this intrinsic is the equivalent of shifting <tt>%val</tt>
4899 right by <tt>%loBit</tt> bits and then ANDing it with a mask with
4900 only the <tt>%hiBit - %loBit</tt> bits set, as follows:</p>
4902 <li>The <tt>%val</tt> is shifted right (LSHR) by the number of bits specified
4903 by <tt>%loBits</tt>. This normalizes the value to the low order bits.</li>
4904 <li>The <tt>%loBits</tt> value is subtracted from the <tt>%hiBits</tt> value
4905 to determine the number of bits to retain.</li>
4906 <li>A mask of the retained bits is created by shifting a -1 value.</li>
4907 <li>The mask is ANDed with <tt>%val</tt> to produce the result.
4909 <p>In reverse mode, a similar computation is made except that the bits are
4910 returned in the reverse order. So, for example, if <tt>X</tt> has the value
4911 <tt>i16 0x0ACF (101011001111)</tt> and we apply
4912 <tt>part.select(i16 X, 8, 3)</tt> to it, we get back the value
4913 <tt>i16 0x0026 (000000100110)</tt>.</p>
4916 <div class="doc_subsubsection">
4917 <a name="int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic</a>
4920 <div class="doc_text">
4923 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.set</tt>
4924 on any integer bit width.
4926 declare i17 @llvm.part.set.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
4927 declare i29 @llvm.part.set.i29.i9 (i29 %val, i9 %repl, i32 %lo, i32 %hi)
4931 <p>The '<tt>llvm.part.set</tt>' family of intrinsic functions replaces a range
4932 of bits in an integer value with another integer value. It returns the integer
4933 with the replaced bits.</p>
4936 <p>The first argument, <tt>%val</tt> and the result may be integer types of
4937 any bit width but they must have the same bit width. <tt>%val</tt> is the value
4938 whose bits will be replaced. The second argument, <tt>%repl</tt> may be an
4939 integer of any bit width. The third and fourth arguments must be <tt>i32</tt>
4940 type since they specify only a bit index.</p>
4943 <p>The operation of the '<tt>llvm.part.set</tt>' intrinsic has two modes
4944 of operation: forwards and reverse. If <tt>%lo</tt> is greater than
4945 <tt>%hi</tt> then the intrinsic operates in reverse mode. Otherwise it
4946 operates in forward mode.</p>
4947 <p>For both modes, the <tt>%repl</tt> value is prepared for use by either
4948 truncating it down to the size of the replacement area or zero extending it
4949 up to that size.</p>
4950 <p>In forward mode, the bits between <tt>%lo</tt> and <tt>%hi</tt> (inclusive)
4951 are replaced with corresponding bits from <tt>%repl</tt>. That is the 0th bit
4952 in <tt>%repl</tt> replaces the <tt>%lo</tt>th bit in <tt>%val</tt> and etc. up
4953 to the <tt>%hi</tt>th bit.
4954 <p>In reverse mode, a similar computation is made except that the bits are
4955 reversed. That is, the <tt>0</tt>th bit in <tt>%repl</tt> replaces the
4956 <tt>%hi</tt> bit in <tt>%val</tt> and etc. down to the <tt>%lo</tt>th bit.
4959 llvm.part.set(0xFFFF, 0, 4, 7) -> 0xFF0F
4960 llvm.part.set(0xFFFF, 0, 7, 4) -> 0xFF0F
4961 llvm.part.set(0xFFFF, 1, 7, 4) -> 0xFF8F
4962 llvm.part.set(0xFFFF, F, 8, 3) -> 0xFFE7
4963 llvm.part.set(0xFFFF, 0, 3, 8) -> 0xFE07
4967 <!-- ======================================================================= -->
4968 <div class="doc_subsection">
4969 <a name="int_debugger">Debugger Intrinsics</a>
4972 <div class="doc_text">
4974 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
4975 are described in the <a
4976 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
4977 Debugging</a> document.
4982 <!-- ======================================================================= -->
4983 <div class="doc_subsection">
4984 <a name="int_eh">Exception Handling Intrinsics</a>
4987 <div class="doc_text">
4988 <p> The LLVM exception handling intrinsics (which all start with
4989 <tt>llvm.eh.</tt> prefix), are described in the <a
4990 href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
4991 Handling</a> document. </p>
4994 <!-- ======================================================================= -->
4995 <div class="doc_subsection">
4996 <a name="int_trampoline">Trampoline Intrinsic</a>
4999 <div class="doc_text">
5001 This intrinsic makes it possible to excise one parameter, marked with
5002 the <tt>nest</tt> attribute, from a function. The result is a callable
5003 function pointer lacking the nest parameter - the caller does not need
5004 to provide a value for it. Instead, the value to use is stored in
5005 advance in a "trampoline", a block of memory usually allocated
5006 on the stack, which also contains code to splice the nest value into the
5007 argument list. This is used to implement the GCC nested function address
5011 For example, if the function is
5012 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
5013 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as follows:</p>
5015 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
5016 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
5017 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
5018 %fp = bitcast i8* %p to i32 (i32, i32)*
5020 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
5021 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
5024 <!-- _______________________________________________________________________ -->
5025 <div class="doc_subsubsection">
5026 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
5028 <div class="doc_text">
5031 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
5035 This fills the memory pointed to by <tt>tramp</tt> with code
5036 and returns a function pointer suitable for executing it.
5040 The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
5041 pointers. The <tt>tramp</tt> argument must point to a sufficiently large
5042 and sufficiently aligned block of memory; this memory is written to by the
5043 intrinsic. Note that the size and the alignment are target-specific - LLVM
5044 currently provides no portable way of determining them, so a front-end that
5045 generates this intrinsic needs to have some target-specific knowledge.
5046 The <tt>func</tt> argument must hold a function bitcast to an <tt>i8*</tt>.
5050 The block of memory pointed to by <tt>tramp</tt> is filled with target
5051 dependent code, turning it into a function. A pointer to this function is
5052 returned, but needs to be bitcast to an
5053 <a href="#int_trampoline">appropriate function pointer type</a>
5054 before being called. The new function's signature is the same as that of
5055 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
5056 removed. At most one such <tt>nest</tt> argument is allowed, and it must be
5057 of pointer type. Calling the new function is equivalent to calling
5058 <tt>func</tt> with the same argument list, but with <tt>nval</tt> used for the
5059 missing <tt>nest</tt> argument. If, after calling
5060 <tt>llvm.init.trampoline</tt>, the memory pointed to by <tt>tramp</tt> is
5061 modified, then the effect of any later call to the returned function pointer is
5066 <!-- ======================================================================= -->
5067 <div class="doc_subsection">
5068 <a name="int_general">General Intrinsics</a>
5071 <div class="doc_text">
5072 <p> This class of intrinsics is designed to be generic and has
5073 no specific purpose. </p>
5076 <!-- _______________________________________________________________________ -->
5077 <div class="doc_subsubsection">
5078 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
5081 <div class="doc_text">
5085 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
5091 The '<tt>llvm.var.annotation</tt>' intrinsic
5097 The first argument is a pointer to a value, the second is a pointer to a
5098 global string, the third is a pointer to a global string which is the source
5099 file name, and the last argument is the line number.
5105 This intrinsic allows annotation of local variables with arbitrary strings.
5106 This can be useful for special purpose optimizations that want to look for these
5107 annotations. These have no other defined use, they are ignored by code
5108 generation and optimization.
5111 <!-- _______________________________________________________________________ -->
5112 <div class="doc_subsubsection">
5113 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
5116 <div class="doc_text">
5119 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
5120 any integer bit width.
5123 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
5124 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
5125 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
5126 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
5127 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
5133 The '<tt>llvm.annotation</tt>' intrinsic.
5139 The first argument is an integer value (result of some expression),
5140 the second is a pointer to a global string, the third is a pointer to a global
5141 string which is the source file name, and the last argument is the line number.
5142 It returns the value of the first argument.
5148 This intrinsic allows annotations to be put on arbitrary expressions
5149 with arbitrary strings. This can be useful for special purpose optimizations
5150 that want to look for these annotations. These have no other defined use, they
5151 are ignored by code generation and optimization.
5154 <!-- *********************************************************************** -->
5157 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
5158 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
5159 <a href="http://validator.w3.org/check/referer"><img
5160 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a>
5162 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
5163 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
5164 Last modified: $Date$