1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <title>LLVM Assembly Language Reference Manual</title>
6 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
7 <meta name="author" content="Chris Lattner">
8 <meta name="description"
9 content="LLVM Assembly Language Reference Manual.">
10 <link rel="stylesheet" href="llvm.css" type="text/css">
15 <div class="doc_title"> LLVM Language Reference Manual </div>
17 <li><a href="#abstract">Abstract</a></li>
18 <li><a href="#introduction">Introduction</a></li>
19 <li><a href="#identifiers">Identifiers</a></li>
20 <li><a href="#highlevel">High Level Structure</a>
22 <li><a href="#modulestructure">Module Structure</a></li>
23 <li><a href="#linkage">Linkage Types</a></li>
24 <li><a href="#callingconv">Calling Conventions</a></li>
25 <li><a href="#globalvars">Global Variables</a></li>
26 <li><a href="#functionstructure">Functions</a></li>
27 <li><a href="#aliasstructure">Aliases</a>
28 <li><a href="#paramattrs">Parameter Attributes</a></li>
29 <li><a href="#gc">Garbage Collector Names</a></li>
30 <li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
31 <li><a href="#datalayout">Data Layout</a></li>
34 <li><a href="#typesystem">Type System</a>
36 <li><a href="#t_classifications">Type Classifications</a></li>
37 <li><a href="#t_primitive">Primitive Types</a>
39 <li><a href="#t_floating">Floating Point Types</a></li>
40 <li><a href="#t_void">Void Type</a></li>
41 <li><a href="#t_label">Label Type</a></li>
44 <li><a href="#t_derived">Derived Types</a>
46 <li><a href="#t_integer">Integer Type</a></li>
47 <li><a href="#t_array">Array Type</a></li>
48 <li><a href="#t_function">Function Type</a></li>
49 <li><a href="#t_pointer">Pointer Type</a></li>
50 <li><a href="#t_struct">Structure Type</a></li>
51 <li><a href="#t_pstruct">Packed Structure Type</a></li>
52 <li><a href="#t_vector">Vector Type</a></li>
53 <li><a href="#t_opaque">Opaque Type</a></li>
58 <li><a href="#constants">Constants</a>
60 <li><a href="#simpleconstants">Simple Constants</a>
61 <li><a href="#aggregateconstants">Aggregate Constants</a>
62 <li><a href="#globalconstants">Global Variable and Function Addresses</a>
63 <li><a href="#undefvalues">Undefined Values</a>
64 <li><a href="#constantexprs">Constant Expressions</a>
67 <li><a href="#othervalues">Other Values</a>
69 <li><a href="#inlineasm">Inline Assembler Expressions</a>
72 <li><a href="#instref">Instruction Reference</a>
74 <li><a href="#terminators">Terminator Instructions</a>
76 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
77 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
78 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
79 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
80 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
81 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
84 <li><a href="#binaryops">Binary Operations</a>
86 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
87 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
88 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
89 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
90 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
91 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
92 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
93 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
94 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
97 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
99 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
100 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
101 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
102 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
103 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
104 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
107 <li><a href="#vectorops">Vector Operations</a>
109 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
110 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
111 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
114 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
116 <li><a href="#i_malloc">'<tt>malloc</tt>' Instruction</a></li>
117 <li><a href="#i_free">'<tt>free</tt>' Instruction</a></li>
118 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
119 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
120 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
121 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
124 <li><a href="#convertops">Conversion Operations</a>
126 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
127 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
128 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
129 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
130 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
131 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
132 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
133 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
134 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
135 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
136 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
137 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
139 <li><a href="#otherops">Other Operations</a>
141 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
142 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
143 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
144 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
145 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
146 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
151 <li><a href="#intrinsics">Intrinsic Functions</a>
153 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
155 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
156 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
157 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
160 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
162 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
163 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
164 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
167 <li><a href="#int_codegen">Code Generator Intrinsics</a>
169 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
170 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
171 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
172 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
173 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
174 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
175 <li><a href="#int_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
178 <li><a href="#int_libc">Standard C Library Intrinsics</a>
180 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
181 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
182 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
183 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
184 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
185 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
186 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
187 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
190 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
192 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
193 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
194 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
195 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
196 <li><a href="#int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic </a></li>
197 <li><a href="#int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic </a></li>
200 <li><a href="#int_debugger">Debugger intrinsics</a></li>
201 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
202 <li><a href="#int_trampoline">Trampoline Intrinsic</a>
204 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
207 <li><a href="#int_atomics">Atomic intrinsics</a>
209 <li><a href="#int_memory_barrier"><tt>llvm.memory_barrier</tt></li>
212 <li><a href="#int_general">General intrinsics</a>
214 <li><a href="#int_var_annotation">
215 <tt>llvm.var.annotation</tt>' Intrinsic</a></li>
216 <li><a href="#int_annotation">
217 <tt>llvm.annotation.*</tt>' Intrinsic</a></li>
218 <li><a href="#int_trap">
219 <tt>llvm.trap</tt>' Intrinsic</a></li>
226 <div class="doc_author">
227 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
228 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
231 <!-- *********************************************************************** -->
232 <div class="doc_section"> <a name="abstract">Abstract </a></div>
233 <!-- *********************************************************************** -->
235 <div class="doc_text">
236 <p>This document is a reference manual for the LLVM assembly language.
237 LLVM is an SSA based representation that provides type safety,
238 low-level operations, flexibility, and the capability of representing
239 'all' high-level languages cleanly. It is the common code
240 representation used throughout all phases of the LLVM compilation
244 <!-- *********************************************************************** -->
245 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
246 <!-- *********************************************************************** -->
248 <div class="doc_text">
250 <p>The LLVM code representation is designed to be used in three
251 different forms: as an in-memory compiler IR, as an on-disk bitcode
252 representation (suitable for fast loading by a Just-In-Time compiler),
253 and as a human readable assembly language representation. This allows
254 LLVM to provide a powerful intermediate representation for efficient
255 compiler transformations and analysis, while providing a natural means
256 to debug and visualize the transformations. The three different forms
257 of LLVM are all equivalent. This document describes the human readable
258 representation and notation.</p>
260 <p>The LLVM representation aims to be light-weight and low-level
261 while being expressive, typed, and extensible at the same time. It
262 aims to be a "universal IR" of sorts, by being at a low enough level
263 that high-level ideas may be cleanly mapped to it (similar to how
264 microprocessors are "universal IR's", allowing many source languages to
265 be mapped to them). By providing type information, LLVM can be used as
266 the target of optimizations: for example, through pointer analysis, it
267 can be proven that a C automatic variable is never accessed outside of
268 the current function... allowing it to be promoted to a simple SSA
269 value instead of a memory location.</p>
273 <!-- _______________________________________________________________________ -->
274 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
276 <div class="doc_text">
278 <p>It is important to note that this document describes 'well formed'
279 LLVM assembly language. There is a difference between what the parser
280 accepts and what is considered 'well formed'. For example, the
281 following instruction is syntactically okay, but not well formed:</p>
283 <div class="doc_code">
285 %x = <a href="#i_add">add</a> i32 1, %x
289 <p>...because the definition of <tt>%x</tt> does not dominate all of
290 its uses. The LLVM infrastructure provides a verification pass that may
291 be used to verify that an LLVM module is well formed. This pass is
292 automatically run by the parser after parsing input assembly and by
293 the optimizer before it outputs bitcode. The violations pointed out
294 by the verifier pass indicate bugs in transformation passes or input to
298 <!-- Describe the typesetting conventions here. -->
300 <!-- *********************************************************************** -->
301 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
302 <!-- *********************************************************************** -->
304 <div class="doc_text">
306 <p>LLVM identifiers come in two basic types: global and local. Global
307 identifiers (functions, global variables) begin with the @ character. Local
308 identifiers (register names, types) begin with the % character. Additionally,
309 there are three different formats for identifiers, for different purposes:
312 <li>Named values are represented as a string of characters with their prefix.
313 For example, %foo, @DivisionByZero, %a.really.long.identifier. The actual
314 regular expression used is '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
315 Identifiers which require other characters in their names can be surrounded
316 with quotes. In this way, anything except a <tt>"</tt> character can
317 be used in a named value.</li>
319 <li>Unnamed values are represented as an unsigned numeric value with their
320 prefix. For example, %12, @2, %44.</li>
322 <li>Constants, which are described in a <a href="#constants">section about
323 constants</a>, below.</li>
326 <p>LLVM requires that values start with a prefix for two reasons: Compilers
327 don't need to worry about name clashes with reserved words, and the set of
328 reserved words may be expanded in the future without penalty. Additionally,
329 unnamed identifiers allow a compiler to quickly come up with a temporary
330 variable without having to avoid symbol table conflicts.</p>
332 <p>Reserved words in LLVM are very similar to reserved words in other
333 languages. There are keywords for different opcodes
334 ('<tt><a href="#i_add">add</a></tt>',
335 '<tt><a href="#i_bitcast">bitcast</a></tt>',
336 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
337 href="#t_void">void</a></tt>', '<tt><a href="#t_primitive">i32</a></tt>', etc...),
338 and others. These reserved words cannot conflict with variable names, because
339 none of them start with a prefix character ('%' or '@').</p>
341 <p>Here is an example of LLVM code to multiply the integer variable
342 '<tt>%X</tt>' by 8:</p>
346 <div class="doc_code">
348 %result = <a href="#i_mul">mul</a> i32 %X, 8
352 <p>After strength reduction:</p>
354 <div class="doc_code">
356 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
360 <p>And the hard way:</p>
362 <div class="doc_code">
364 <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
365 <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
366 %result = <a href="#i_add">add</a> i32 %1, %1
370 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
371 important lexical features of LLVM:</p>
375 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
378 <li>Unnamed temporaries are created when the result of a computation is not
379 assigned to a named value.</li>
381 <li>Unnamed temporaries are numbered sequentially</li>
385 <p>...and it also shows a convention that we follow in this document. When
386 demonstrating instructions, we will follow an instruction with a comment that
387 defines the type and name of value produced. Comments are shown in italic
392 <!-- *********************************************************************** -->
393 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
394 <!-- *********************************************************************** -->
396 <!-- ======================================================================= -->
397 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
400 <div class="doc_text">
402 <p>LLVM programs are composed of "Module"s, each of which is a
403 translation unit of the input programs. Each module consists of
404 functions, global variables, and symbol table entries. Modules may be
405 combined together with the LLVM linker, which merges function (and
406 global variable) definitions, resolves forward declarations, and merges
407 symbol table entries. Here is an example of the "hello world" module:</p>
409 <div class="doc_code">
410 <pre><i>; Declare the string constant as a global constant...</i>
411 <a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a
412 href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
414 <i>; External declaration of the puts function</i>
415 <a href="#functionstructure">declare</a> i32 @puts(i8 *) <i>; i32(i8 *)* </i>
417 <i>; Definition of main function</i>
418 define i32 @main() { <i>; i32()* </i>
419 <i>; Convert [13x i8 ]* to i8 *...</i>
421 href="#i_getelementptr">getelementptr</a> [13 x i8 ]* @.LC0, i64 0, i64 0 <i>; i8 *</i>
423 <i>; Call puts function to write out the string to stdout...</i>
425 href="#i_call">call</a> i32 @puts(i8 * %cast210) <i>; i32</i>
427 href="#i_ret">ret</a> i32 0<br>}<br>
431 <p>This example is made up of a <a href="#globalvars">global variable</a>
432 named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
433 function, and a <a href="#functionstructure">function definition</a>
434 for "<tt>main</tt>".</p>
436 <p>In general, a module is made up of a list of global values,
437 where both functions and global variables are global values. Global values are
438 represented by a pointer to a memory location (in this case, a pointer to an
439 array of char, and a pointer to a function), and have one of the following <a
440 href="#linkage">linkage types</a>.</p>
444 <!-- ======================================================================= -->
445 <div class="doc_subsection">
446 <a name="linkage">Linkage Types</a>
449 <div class="doc_text">
452 All Global Variables and Functions have one of the following types of linkage:
457 <dt><tt><b><a name="linkage_internal">internal</a></b></tt> </dt>
459 <dd>Global values with internal linkage are only directly accessible by
460 objects in the current module. In particular, linking code into a module with
461 an internal global value may cause the internal to be renamed as necessary to
462 avoid collisions. Because the symbol is internal to the module, all
463 references can be updated. This corresponds to the notion of the
464 '<tt>static</tt>' keyword in C.
467 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
469 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
470 the same name when linkage occurs. This is typically used to implement
471 inline functions, templates, or other code which must be generated in each
472 translation unit that uses it. Unreferenced <tt>linkonce</tt> globals are
473 allowed to be discarded.
476 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
478 <dd>"<tt>weak</tt>" linkage is exactly the same as <tt>linkonce</tt> linkage,
479 except that unreferenced <tt>weak</tt> globals may not be discarded. This is
480 used for globals that may be emitted in multiple translation units, but that
481 are not guaranteed to be emitted into every translation unit that uses them.
482 One example of this are common globals in C, such as "<tt>int X;</tt>" at
486 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
488 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
489 pointer to array type. When two global variables with appending linkage are
490 linked together, the two global arrays are appended together. This is the
491 LLVM, typesafe, equivalent of having the system linker append together
492 "sections" with identical names when .o files are linked.
495 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt>: </dt>
496 <dd>The semantics of this linkage follow the ELF model: the symbol is weak
497 until linked, if not linked, the symbol becomes null instead of being an
501 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
503 <dd>If none of the above identifiers are used, the global is externally
504 visible, meaning that it participates in linkage and can be used to resolve
505 external symbol references.
510 The next two types of linkage are targeted for Microsoft Windows platform
511 only. They are designed to support importing (exporting) symbols from (to)
516 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt>: </dt>
518 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
519 or variable via a global pointer to a pointer that is set up by the DLL
520 exporting the symbol. On Microsoft Windows targets, the pointer name is
521 formed by combining <code>_imp__</code> and the function or variable name.
524 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt>: </dt>
526 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
527 pointer to a pointer in a DLL, so that it can be referenced with the
528 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
529 name is formed by combining <code>_imp__</code> and the function or variable
535 <p><a name="linkage_external"></a>For example, since the "<tt>.LC0</tt>"
536 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
537 variable and was linked with this one, one of the two would be renamed,
538 preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
539 external (i.e., lacking any linkage declarations), they are accessible
540 outside of the current module.</p>
541 <p>It is illegal for a function <i>declaration</i>
542 to have any linkage type other than "externally visible", <tt>dllimport</tt>,
543 or <tt>extern_weak</tt>.</p>
544 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt> and <tt>weak</tt>
548 <!-- ======================================================================= -->
549 <div class="doc_subsection">
550 <a name="callingconv">Calling Conventions</a>
553 <div class="doc_text">
555 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
556 and <a href="#i_invoke">invokes</a> can all have an optional calling convention
557 specified for the call. The calling convention of any pair of dynamic
558 caller/callee must match, or the behavior of the program is undefined. The
559 following calling conventions are supported by LLVM, and more may be added in
563 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
565 <dd>This calling convention (the default if no other calling convention is
566 specified) matches the target C calling conventions. This calling convention
567 supports varargs function calls and tolerates some mismatch in the declared
568 prototype and implemented declaration of the function (as does normal C).
571 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
573 <dd>This calling convention attempts to make calls as fast as possible
574 (e.g. by passing things in registers). This calling convention allows the
575 target to use whatever tricks it wants to produce fast code for the target,
576 without having to conform to an externally specified ABI. Implementations of
577 this convention should allow arbitrary tail call optimization to be supported.
578 This calling convention does not support varargs and requires the prototype of
579 all callees to exactly match the prototype of the function definition.
582 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
584 <dd>This calling convention attempts to make code in the caller as efficient
585 as possible under the assumption that the call is not commonly executed. As
586 such, these calls often preserve all registers so that the call does not break
587 any live ranges in the caller side. This calling convention does not support
588 varargs and requires the prototype of all callees to exactly match the
589 prototype of the function definition.
592 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
594 <dd>Any calling convention may be specified by number, allowing
595 target-specific calling conventions to be used. Target specific calling
596 conventions start at 64.
600 <p>More calling conventions can be added/defined on an as-needed basis, to
601 support pascal conventions or any other well-known target-independent
606 <!-- ======================================================================= -->
607 <div class="doc_subsection">
608 <a name="visibility">Visibility Styles</a>
611 <div class="doc_text">
614 All Global Variables and Functions have one of the following visibility styles:
618 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
620 <dd>On ELF, default visibility means that the declaration is visible to other
621 modules and, in shared libraries, means that the declared entity may be
622 overridden. On Darwin, default visibility means that the declaration is
623 visible to other modules. Default visibility corresponds to "external
624 linkage" in the language.
627 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
629 <dd>Two declarations of an object with hidden visibility refer to the same
630 object if they are in the same shared object. Usually, hidden visibility
631 indicates that the symbol will not be placed into the dynamic symbol table,
632 so no other module (executable or shared library) can reference it
636 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
638 <dd>On ELF, protected visibility indicates that the symbol will be placed in
639 the dynamic symbol table, but that references within the defining module will
640 bind to the local symbol. That is, the symbol cannot be overridden by another
647 <!-- ======================================================================= -->
648 <div class="doc_subsection">
649 <a name="globalvars">Global Variables</a>
652 <div class="doc_text">
654 <p>Global variables define regions of memory allocated at compilation time
655 instead of run-time. Global variables may optionally be initialized, may have
656 an explicit section to be placed in, and may have an optional explicit alignment
657 specified. A variable may be defined as "thread_local", which means that it
658 will not be shared by threads (each thread will have a separated copy of the
659 variable). A variable may be defined as a global "constant," which indicates
660 that the contents of the variable will <b>never</b> be modified (enabling better
661 optimization, allowing the global data to be placed in the read-only section of
662 an executable, etc). Note that variables that need runtime initialization
663 cannot be marked "constant" as there is a store to the variable.</p>
666 LLVM explicitly allows <em>declarations</em> of global variables to be marked
667 constant, even if the final definition of the global is not. This capability
668 can be used to enable slightly better optimization of the program, but requires
669 the language definition to guarantee that optimizations based on the
670 'constantness' are valid for the translation units that do not include the
674 <p>As SSA values, global variables define pointer values that are in
675 scope (i.e. they dominate) all basic blocks in the program. Global
676 variables always define a pointer to their "content" type because they
677 describe a region of memory, and all memory objects in LLVM are
678 accessed through pointers.</p>
680 <p>A global variable may be declared to reside in a target-specifc numbered
681 address space. For targets that support them, address spaces may affect how
682 optimizations are performed and/or what target instructions are used to access
683 the variable. The default address space is zero. The address space qualifier
684 must precede any other attributes.</p>
686 <p>LLVM allows an explicit section to be specified for globals. If the target
687 supports it, it will emit globals to the section specified.</p>
689 <p>An explicit alignment may be specified for a global. If not present, or if
690 the alignment is set to zero, the alignment of the global is set by the target
691 to whatever it feels convenient. If an explicit alignment is specified, the
692 global is forced to have at least that much alignment. All alignments must be
695 <p>For example, the following defines a global in a numbered address space with
696 an initializer, section, and alignment:</p>
698 <div class="doc_code">
700 @G = constant float 1.0 addrspace(5), section "foo", align 4
707 <!-- ======================================================================= -->
708 <div class="doc_subsection">
709 <a name="functionstructure">Functions</a>
712 <div class="doc_text">
714 <p>LLVM function definitions consist of the "<tt>define</tt>" keyord,
715 an optional <a href="#linkage">linkage type</a>, an optional
716 <a href="#visibility">visibility style</a>, an optional
717 <a href="#callingconv">calling convention</a>, a return type, an optional
718 <a href="#paramattrs">parameter attribute</a> for the return type, a function
719 name, a (possibly empty) argument list (each with optional
720 <a href="#paramattrs">parameter attributes</a>), an optional section, an
721 optional alignment, an optional <a href="#gc">garbage collector name</a>, an
722 opening curly brace, a list of basic blocks, and a closing curly brace.
724 LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
725 optional <a href="#linkage">linkage type</a>, an optional
726 <a href="#visibility">visibility style</a>, an optional
727 <a href="#callingconv">calling convention</a>, a return type, an optional
728 <a href="#paramattrs">parameter attribute</a> for the return type, a function
729 name, a possibly empty list of arguments, an optional alignment, and an optional
730 <a href="#gc">garbage collector name</a>.</p>
732 <p>A function definition contains a list of basic blocks, forming the CFG for
733 the function. Each basic block may optionally start with a label (giving the
734 basic block a symbol table entry), contains a list of instructions, and ends
735 with a <a href="#terminators">terminator</a> instruction (such as a branch or
736 function return).</p>
738 <p>The first basic block in a function is special in two ways: it is immediately
739 executed on entrance to the function, and it is not allowed to have predecessor
740 basic blocks (i.e. there can not be any branches to the entry block of a
741 function). Because the block can have no predecessors, it also cannot have any
742 <a href="#i_phi">PHI nodes</a>.</p>
744 <p>LLVM allows an explicit section to be specified for functions. If the target
745 supports it, it will emit functions to the section specified.</p>
747 <p>An explicit alignment may be specified for a function. If not present, or if
748 the alignment is set to zero, the alignment of the function is set by the target
749 to whatever it feels convenient. If an explicit alignment is specified, the
750 function is forced to have at least that much alignment. All alignments must be
756 <!-- ======================================================================= -->
757 <div class="doc_subsection">
758 <a name="aliasstructure">Aliases</a>
760 <div class="doc_text">
761 <p>Aliases act as "second name" for the aliasee value (which can be either
762 function or global variable or bitcast of global value). Aliases may have an
763 optional <a href="#linkage">linkage type</a>, and an
764 optional <a href="#visibility">visibility style</a>.</p>
768 <div class="doc_code">
770 @<Name> = [Linkage] [Visibility] alias <AliaseeTy> @<Aliasee>
778 <!-- ======================================================================= -->
779 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
780 <div class="doc_text">
781 <p>The return type and each parameter of a function type may have a set of
782 <i>parameter attributes</i> associated with them. Parameter attributes are
783 used to communicate additional information about the result or parameters of
784 a function. Parameter attributes are considered to be part of the function,
785 not of the function type, so functions with different parameter attributes
786 can have the same function type.</p>
788 <p>Parameter attributes are simple keywords that follow the type specified. If
789 multiple parameter attributes are needed, they are space separated. For
792 <div class="doc_code">
794 declare i32 @printf(i8* noalias , ...) nounwind
795 declare i32 @atoi(i8*) nounwind readonly
799 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
800 <tt>readonly</tt>) come immediately after the argument list.</p>
802 <p>Currently, only the following parameter attributes are defined:</p>
804 <dt><tt>zeroext</tt></dt>
805 <dd>This indicates that the parameter should be zero extended just before
806 a call to this function.</dd>
808 <dt><tt>signext</tt></dt>
809 <dd>This indicates that the parameter should be sign extended just before
810 a call to this function.</dd>
812 <dt><tt>inreg</tt></dt>
813 <dd>This indicates that the parameter should be placed in register (if
814 possible) during assembling function call. Support for this attribute is
817 <dt><tt>byval</tt></dt>
818 <dd>This indicates that the pointer parameter should really be passed by
819 value to the function. The attribute implies that a hidden copy of the
820 pointee is made between the caller and the callee, so the callee is unable
821 to modify the value in the callee. This attribute is only valid on llvm
822 pointer arguments. It is generally used to pass structs and arrays by
823 value, but is also valid on scalars (even though this is silly).</dd>
825 <dt><tt>sret</tt></dt>
826 <dd>This indicates that the parameter specifies the address of a structure
827 that is the return value of the function in the source program.</dd>
829 <dt><tt>noalias</tt></dt>
830 <dd>This indicates that the parameter does not alias any global or any other
831 parameter. The caller is responsible for ensuring that this is the case,
832 usually by placing the value in a stack allocation.</dd>
834 <dt><tt>noreturn</tt></dt>
835 <dd>This function attribute indicates that the function never returns. This
836 indicates to LLVM that every call to this function should be treated as if
837 an <tt>unreachable</tt> instruction immediately followed the call.</dd>
839 <dt><tt>nounwind</tt></dt>
840 <dd>This function attribute indicates that the function type does not use
841 the unwind instruction and does not allow stack unwinding to propagate
844 <dt><tt>nest</tt></dt>
845 <dd>This indicates that the parameter can be excised using the
846 <a href="#int_trampoline">trampoline intrinsics</a>.</dd>
847 <dt><tt>readonly</tt></dt>
848 <dd>This function attribute indicates that the function has no side-effects
849 except for producing a return value or throwing an exception. The value
850 returned must only depend on the function arguments and/or global variables.
851 It may use values obtained by dereferencing pointers.</dd>
852 <dt><tt>readnone</tt></dt>
853 <dd>A <tt>readnone</tt> function has the same restrictions as a <tt>readonly</tt>
854 function, but in addition it is not allowed to dereference any pointer arguments
860 <!-- ======================================================================= -->
861 <div class="doc_subsection">
862 <a name="gc">Garbage Collector Names</a>
865 <div class="doc_text">
866 <p>Each function may specify a garbage collector name, which is simply a
869 <div class="doc_code"><pre
870 >define void @f() gc "name" { ...</pre></div>
872 <p>The compiler declares the supported values of <i>name</i>. Specifying a
873 collector which will cause the compiler to alter its output in order to support
874 the named garbage collection algorithm.</p>
877 <!-- ======================================================================= -->
878 <div class="doc_subsection">
879 <a name="moduleasm">Module-Level Inline Assembly</a>
882 <div class="doc_text">
884 Modules may contain "module-level inline asm" blocks, which corresponds to the
885 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
886 LLVM and treated as a single unit, but may be separated in the .ll file if
887 desired. The syntax is very simple:
890 <div class="doc_code">
892 module asm "inline asm code goes here"
893 module asm "more can go here"
897 <p>The strings can contain any character by escaping non-printable characters.
898 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
903 The inline asm code is simply printed to the machine code .s file when
904 assembly code is generated.
908 <!-- ======================================================================= -->
909 <div class="doc_subsection">
910 <a name="datalayout">Data Layout</a>
913 <div class="doc_text">
914 <p>A module may specify a target specific data layout string that specifies how
915 data is to be laid out in memory. The syntax for the data layout is simply:</p>
916 <pre> target datalayout = "<i>layout specification</i>"</pre>
917 <p>The <i>layout specification</i> consists of a list of specifications
918 separated by the minus sign character ('-'). Each specification starts with a
919 letter and may include other information after the letter to define some
920 aspect of the data layout. The specifications accepted are as follows: </p>
923 <dd>Specifies that the target lays out data in big-endian form. That is, the
924 bits with the most significance have the lowest address location.</dd>
926 <dd>Specifies that hte target lays out data in little-endian form. That is,
927 the bits with the least significance have the lowest address location.</dd>
928 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
929 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
930 <i>preferred</i> alignments. All sizes are in bits. Specifying the <i>pref</i>
931 alignment is optional. If omitted, the preceding <tt>:</tt> should be omitted
933 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
934 <dd>This specifies the alignment for an integer type of a given bit
935 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
936 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
937 <dd>This specifies the alignment for a vector type of a given bit
939 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
940 <dd>This specifies the alignment for a floating point type of a given bit
941 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
943 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
944 <dd>This specifies the alignment for an aggregate type of a given bit
947 <p>When constructing the data layout for a given target, LLVM starts with a
948 default set of specifications which are then (possibly) overriden by the
949 specifications in the <tt>datalayout</tt> keyword. The default specifications
950 are given in this list:</p>
952 <li><tt>E</tt> - big endian</li>
953 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
954 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
955 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
956 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
957 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
958 <li><tt>i64:32:64</tt> - i64 has abi alignment of 32-bits but preferred
959 alignment of 64-bits</li>
960 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
961 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
962 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
963 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
964 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
966 <p>When llvm is determining the alignment for a given type, it uses the
969 <li>If the type sought is an exact match for one of the specifications, that
970 specification is used.</li>
971 <li>If no match is found, and the type sought is an integer type, then the
972 smallest integer type that is larger than the bitwidth of the sought type is
973 used. If none of the specifications are larger than the bitwidth then the the
974 largest integer type is used. For example, given the default specifications
975 above, the i7 type will use the alignment of i8 (next largest) while both
976 i65 and i256 will use the alignment of i64 (largest specified).</li>
977 <li>If no match is found, and the type sought is a vector type, then the
978 largest vector type that is smaller than the sought vector type will be used
979 as a fall back. This happens because <128 x double> can be implemented in
980 terms of 64 <2 x double>, for example.</li>
984 <!-- *********************************************************************** -->
985 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
986 <!-- *********************************************************************** -->
988 <div class="doc_text">
990 <p>The LLVM type system is one of the most important features of the
991 intermediate representation. Being typed enables a number of
992 optimizations to be performed on the IR directly, without having to do
993 extra analyses on the side before the transformation. A strong type
994 system makes it easier to read the generated code and enables novel
995 analyses and transformations that are not feasible to perform on normal
996 three address code representations.</p>
1000 <!-- ======================================================================= -->
1001 <div class="doc_subsection"> <a name="t_classifications">Type
1002 Classifications</a> </div>
1003 <div class="doc_text">
1004 <p>The types fall into a few useful
1005 classifications:</p>
1007 <table border="1" cellspacing="0" cellpadding="4">
1009 <tr><th>Classification</th><th>Types</th></tr>
1011 <td><a href="#t_integer">integer</a></td>
1012 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1015 <td><a href="#t_floating">floating point</a></td>
1016 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1019 <td><a name="t_firstclass">first class</a></td>
1020 <td><a href="#t_integer">integer</a>,
1021 <a href="#t_floating">floating point</a>,
1022 <a href="#t_pointer">pointer</a>,
1023 <a href="#t_vector">vector</a>
1027 <td><a href="#t_primitive">primitive</a></td>
1028 <td><a href="#t_label">label</a>,
1029 <a href="#t_void">void</a>,
1030 <a href="#t_integer">integer</a>,
1031 <a href="#t_floating">floating point</a>.</td>
1034 <td><a href="#t_derived">derived</a></td>
1035 <td><a href="#t_integer">integer</a>,
1036 <a href="#t_array">array</a>,
1037 <a href="#t_function">function</a>,
1038 <a href="#t_pointer">pointer</a>,
1039 <a href="#t_struct">structure</a>,
1040 <a href="#t_pstruct">packed structure</a>,
1041 <a href="#t_vector">vector</a>,
1042 <a href="#t_opaque">opaque</a>.
1047 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
1048 most important. Values of these types are the only ones which can be
1049 produced by instructions, passed as arguments, or used as operands to
1050 instructions. This means that all structures and arrays must be
1051 manipulated either by pointer or by component.</p>
1054 <!-- ======================================================================= -->
1055 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
1057 <div class="doc_text">
1058 <p>The primitive types are the fundamental building blocks of the LLVM
1063 <!-- _______________________________________________________________________ -->
1064 <div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1066 <div class="doc_text">
1069 <tr><th>Type</th><th>Description</th></tr>
1070 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1071 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1072 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1073 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1074 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1079 <!-- _______________________________________________________________________ -->
1080 <div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1082 <div class="doc_text">
1084 <p>The void type does not represent any value and has no size.</p>
1093 <!-- _______________________________________________________________________ -->
1094 <div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1096 <div class="doc_text">
1098 <p>The label type represents code labels.</p>
1108 <!-- ======================================================================= -->
1109 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1111 <div class="doc_text">
1113 <p>The real power in LLVM comes from the derived types in the system.
1114 This is what allows a programmer to represent arrays, functions,
1115 pointers, and other useful types. Note that these derived types may be
1116 recursive: For example, it is possible to have a two dimensional array.</p>
1120 <!-- _______________________________________________________________________ -->
1121 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1123 <div class="doc_text">
1126 <p>The integer type is a very simple derived type that simply specifies an
1127 arbitrary bit width for the integer type desired. Any bit width from 1 bit to
1128 2^23-1 (about 8 million) can be specified.</p>
1136 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1140 <table class="layout">
1143 <td><tt>i1</tt></td>
1144 <td>a single-bit integer.</td>
1146 <td><tt>i32</tt></td>
1147 <td>a 32-bit integer.</td>
1149 <td><tt>i1942652</tt></td>
1150 <td>a really big integer of over 1 million bits.</td>
1156 <!-- _______________________________________________________________________ -->
1157 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1159 <div class="doc_text">
1163 <p>The array type is a very simple derived type that arranges elements
1164 sequentially in memory. The array type requires a size (number of
1165 elements) and an underlying data type.</p>
1170 [<# elements> x <elementtype>]
1173 <p>The number of elements is a constant integer value; elementtype may
1174 be any type with a size.</p>
1177 <table class="layout">
1179 <td class="left"><tt>[40 x i32]</tt></td>
1180 <td class="left">Array of 40 32-bit integer values.</td>
1183 <td class="left"><tt>[41 x i32]</tt></td>
1184 <td class="left">Array of 41 32-bit integer values.</td>
1187 <td class="left"><tt>[4 x i8]</tt></td>
1188 <td class="left">Array of 4 8-bit integer values.</td>
1191 <p>Here are some examples of multidimensional arrays:</p>
1192 <table class="layout">
1194 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1195 <td class="left">3x4 array of 32-bit integer values.</td>
1198 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1199 <td class="left">12x10 array of single precision floating point values.</td>
1202 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1203 <td class="left">2x3x4 array of 16-bit integer values.</td>
1207 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1208 length array. Normally, accesses past the end of an array are undefined in
1209 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
1210 As a special case, however, zero length arrays are recognized to be variable
1211 length. This allows implementation of 'pascal style arrays' with the LLVM
1212 type "{ i32, [0 x float]}", for example.</p>
1216 <!-- _______________________________________________________________________ -->
1217 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1218 <div class="doc_text">
1220 <p>The function type can be thought of as a function signature. It
1221 consists of a return type and a list of formal parameter types.
1222 Function types are usually used to build virtual function tables
1223 (which are structures of pointers to functions), for indirect function
1224 calls, and when defining a function.</p>
1226 The return type of a function type cannot be an aggregate type.
1229 <pre> <returntype> (<parameter list>)<br></pre>
1230 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1231 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1232 which indicates that the function takes a variable number of arguments.
1233 Variable argument functions can access their arguments with the <a
1234 href="#int_varargs">variable argument handling intrinsic</a> functions.</p>
1236 <table class="layout">
1238 <td class="left"><tt>i32 (i32)</tt></td>
1239 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1241 </tr><tr class="layout">
1242 <td class="left"><tt>float (i16 signext, i32 *) *
1244 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1245 an <tt>i16</tt> that should be sign extended and a
1246 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1249 </tr><tr class="layout">
1250 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1251 <td class="left">A vararg function that takes at least one
1252 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1253 which returns an integer. This is the signature for <tt>printf</tt> in
1260 <!-- _______________________________________________________________________ -->
1261 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1262 <div class="doc_text">
1264 <p>The structure type is used to represent a collection of data members
1265 together in memory. The packing of the field types is defined to match
1266 the ABI of the underlying processor. The elements of a structure may
1267 be any type that has a size.</p>
1268 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1269 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1270 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1273 <pre> { <type list> }<br></pre>
1275 <table class="layout">
1277 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1278 <td class="left">A triple of three <tt>i32</tt> values</td>
1279 </tr><tr class="layout">
1280 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1281 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1282 second element is a <a href="#t_pointer">pointer</a> to a
1283 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1284 an <tt>i32</tt>.</td>
1289 <!-- _______________________________________________________________________ -->
1290 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1292 <div class="doc_text">
1294 <p>The packed structure type is used to represent a collection of data members
1295 together in memory. There is no padding between fields. Further, the alignment
1296 of a packed structure is 1 byte. The elements of a packed structure may
1297 be any type that has a size.</p>
1298 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1299 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1300 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1303 <pre> < { <type list> } > <br></pre>
1305 <table class="layout">
1307 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1308 <td class="left">A triple of three <tt>i32</tt> values</td>
1309 </tr><tr class="layout">
1310 <td class="left"><tt>< { float, i32 (i32)* } ></tt></td>
1311 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1312 second element is a <a href="#t_pointer">pointer</a> to a
1313 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1314 an <tt>i32</tt>.</td>
1319 <!-- _______________________________________________________________________ -->
1320 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1321 <div class="doc_text">
1323 <p>As in many languages, the pointer type represents a pointer or
1324 reference to another object, which must live in memory. Pointer types may have
1325 an optional address space attribute defining the target-specific numbered
1326 address space where the pointed-to object resides. The default address space is
1329 <pre> <type> *<br></pre>
1331 <table class="layout">
1333 <td class="left"><tt>[4x i32]*</tt></td>
1334 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1335 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1338 <td class="left"><tt>i32 (i32 *) *</tt></td>
1339 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
1340 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1344 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1345 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1346 that resides in address space #5.</td>
1351 <!-- _______________________________________________________________________ -->
1352 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1353 <div class="doc_text">
1357 <p>A vector type is a simple derived type that represents a vector
1358 of elements. Vector types are used when multiple primitive data
1359 are operated in parallel using a single instruction (SIMD).
1360 A vector type requires a size (number of
1361 elements) and an underlying primitive data type. Vectors must have a power
1362 of two length (1, 2, 4, 8, 16 ...). Vector types are
1363 considered <a href="#t_firstclass">first class</a>.</p>
1368 < <# elements> x <elementtype> >
1371 <p>The number of elements is a constant integer value; elementtype may
1372 be any integer or floating point type.</p>
1376 <table class="layout">
1378 <td class="left"><tt><4 x i32></tt></td>
1379 <td class="left">Vector of 4 32-bit integer values.</td>
1382 <td class="left"><tt><8 x float></tt></td>
1383 <td class="left">Vector of 8 32-bit floating-point values.</td>
1386 <td class="left"><tt><2 x i64></tt></td>
1387 <td class="left">Vector of 2 64-bit integer values.</td>
1392 <!-- _______________________________________________________________________ -->
1393 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1394 <div class="doc_text">
1398 <p>Opaque types are used to represent unknown types in the system. This
1399 corresponds (for example) to the C notion of a forward declared structure type.
1400 In LLVM, opaque types can eventually be resolved to any type (not just a
1401 structure type).</p>
1411 <table class="layout">
1413 <td class="left"><tt>opaque</tt></td>
1414 <td class="left">An opaque type.</td>
1420 <!-- *********************************************************************** -->
1421 <div class="doc_section"> <a name="constants">Constants</a> </div>
1422 <!-- *********************************************************************** -->
1424 <div class="doc_text">
1426 <p>LLVM has several different basic types of constants. This section describes
1427 them all and their syntax.</p>
1431 <!-- ======================================================================= -->
1432 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1434 <div class="doc_text">
1437 <dt><b>Boolean constants</b></dt>
1439 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1440 constants of the <tt><a href="#t_primitive">i1</a></tt> type.
1443 <dt><b>Integer constants</b></dt>
1445 <dd>Standard integers (such as '4') are constants of the <a
1446 href="#t_integer">integer</a> type. Negative numbers may be used with
1450 <dt><b>Floating point constants</b></dt>
1452 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1453 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1454 notation (see below). Floating point constants must have a <a
1455 href="#t_floating">floating point</a> type. </dd>
1457 <dt><b>Null pointer constants</b></dt>
1459 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1460 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1464 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1465 of floating point constants. For example, the form '<tt>double
1466 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1467 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1468 (and the only time that they are generated by the disassembler) is when a
1469 floating point constant must be emitted but it cannot be represented as a
1470 decimal floating point number. For example, NaN's, infinities, and other
1471 special values are represented in their IEEE hexadecimal format so that
1472 assembly and disassembly do not cause any bits to change in the constants.</p>
1476 <!-- ======================================================================= -->
1477 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1480 <div class="doc_text">
1481 <p>Aggregate constants arise from aggregation of simple constants
1482 and smaller aggregate constants.</p>
1485 <dt><b>Structure constants</b></dt>
1487 <dd>Structure constants are represented with notation similar to structure
1488 type definitions (a comma separated list of elements, surrounded by braces
1489 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
1490 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>". Structure constants
1491 must have <a href="#t_struct">structure type</a>, and the number and
1492 types of elements must match those specified by the type.
1495 <dt><b>Array constants</b></dt>
1497 <dd>Array constants are represented with notation similar to array type
1498 definitions (a comma separated list of elements, surrounded by square brackets
1499 (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74 ]</tt>". Array
1500 constants must have <a href="#t_array">array type</a>, and the number and
1501 types of elements must match those specified by the type.
1504 <dt><b>Vector constants</b></dt>
1506 <dd>Vector constants are represented with notation similar to vector type
1507 definitions (a comma separated list of elements, surrounded by
1508 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32 42,
1509 i32 11, i32 74, i32 100 ></tt>". Vector constants must have <a
1510 href="#t_vector">vector type</a>, and the number and types of elements must
1511 match those specified by the type.
1514 <dt><b>Zero initialization</b></dt>
1516 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1517 value to zero of <em>any</em> type, including scalar and aggregate types.
1518 This is often used to avoid having to print large zero initializers (e.g. for
1519 large arrays) and is always exactly equivalent to using explicit zero
1526 <!-- ======================================================================= -->
1527 <div class="doc_subsection">
1528 <a name="globalconstants">Global Variable and Function Addresses</a>
1531 <div class="doc_text">
1533 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1534 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1535 constants. These constants are explicitly referenced when the <a
1536 href="#identifiers">identifier for the global</a> is used and always have <a
1537 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1540 <div class="doc_code">
1544 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1550 <!-- ======================================================================= -->
1551 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1552 <div class="doc_text">
1553 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1554 no specific value. Undefined values may be of any type and be used anywhere
1555 a constant is permitted.</p>
1557 <p>Undefined values indicate to the compiler that the program is well defined
1558 no matter what value is used, giving the compiler more freedom to optimize.
1562 <!-- ======================================================================= -->
1563 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1566 <div class="doc_text">
1568 <p>Constant expressions are used to allow expressions involving other constants
1569 to be used as constants. Constant expressions may be of any <a
1570 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1571 that does not have side effects (e.g. load and call are not supported). The
1572 following is the syntax for constant expressions:</p>
1575 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1576 <dd>Truncate a constant to another type. The bit size of CST must be larger
1577 than the bit size of TYPE. Both types must be integers.</dd>
1579 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1580 <dd>Zero extend a constant to another type. The bit size of CST must be
1581 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1583 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1584 <dd>Sign extend a constant to another type. The bit size of CST must be
1585 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1587 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1588 <dd>Truncate a floating point constant to another floating point type. The
1589 size of CST must be larger than the size of TYPE. Both types must be
1590 floating point.</dd>
1592 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1593 <dd>Floating point extend a constant to another type. The size of CST must be
1594 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1596 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
1597 <dd>Convert a floating point constant to the corresponding unsigned integer
1598 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1599 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1600 of the same number of elements. If the value won't fit in the integer type,
1601 the results are undefined.</dd>
1603 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
1604 <dd>Convert a floating point constant to the corresponding signed integer
1605 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1606 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1607 of the same number of elements. If the value won't fit in the integer type,
1608 the results are undefined.</dd>
1610 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
1611 <dd>Convert an unsigned integer constant to the corresponding floating point
1612 constant. TYPE must be a scalar or vector floating point type. CST must be of
1613 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1614 of the same number of elements. If the value won't fit in the floating point
1615 type, the results are undefined.</dd>
1617 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
1618 <dd>Convert a signed integer constant to the corresponding floating point
1619 constant. TYPE must be a scalar or vector floating point type. CST must be of
1620 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1621 of the same number of elements. If the value won't fit in the floating point
1622 type, the results are undefined.</dd>
1624 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
1625 <dd>Convert a pointer typed constant to the corresponding integer constant
1626 TYPE must be an integer type. CST must be of pointer type. The CST value is
1627 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1629 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
1630 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1631 pointer type. CST must be of integer type. The CST value is zero extended,
1632 truncated, or unchanged to make it fit in a pointer size. This one is
1633 <i>really</i> dangerous!</dd>
1635 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
1636 <dd>Convert a constant, CST, to another TYPE. The size of CST and TYPE must be
1637 identical (same number of bits). The conversion is done as if the CST value
1638 was stored to memory and read back as TYPE. In other words, no bits change
1639 with this operator, just the type. This can be used for conversion of
1640 vector types to any other type, as long as they have the same bit width. For
1641 pointers it is only valid to cast to another pointer type.
1644 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1646 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1647 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1648 instruction, the index list may have zero or more indexes, which are required
1649 to make sense for the type of "CSTPTR".</dd>
1651 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1653 <dd>Perform the <a href="#i_select">select operation</a> on
1656 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
1657 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
1659 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
1660 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
1662 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1664 <dd>Perform the <a href="#i_extractelement">extractelement
1665 operation</a> on constants.
1667 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1669 <dd>Perform the <a href="#i_insertelement">insertelement
1670 operation</a> on constants.</dd>
1673 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
1675 <dd>Perform the <a href="#i_shufflevector">shufflevector
1676 operation</a> on constants.</dd>
1678 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1680 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1681 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1682 binary</a> operations. The constraints on operands are the same as those for
1683 the corresponding instruction (e.g. no bitwise operations on floating point
1684 values are allowed).</dd>
1688 <!-- *********************************************************************** -->
1689 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1690 <!-- *********************************************************************** -->
1692 <!-- ======================================================================= -->
1693 <div class="doc_subsection">
1694 <a name="inlineasm">Inline Assembler Expressions</a>
1697 <div class="doc_text">
1700 LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
1701 Module-Level Inline Assembly</a>) through the use of a special value. This
1702 value represents the inline assembler as a string (containing the instructions
1703 to emit), a list of operand constraints (stored as a string), and a flag that
1704 indicates whether or not the inline asm expression has side effects. An example
1705 inline assembler expression is:
1708 <div class="doc_code">
1710 i32 (i32) asm "bswap $0", "=r,r"
1715 Inline assembler expressions may <b>only</b> be used as the callee operand of
1716 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
1719 <div class="doc_code">
1721 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
1726 Inline asms with side effects not visible in the constraint list must be marked
1727 as having side effects. This is done through the use of the
1728 '<tt>sideeffect</tt>' keyword, like so:
1731 <div class="doc_code">
1733 call void asm sideeffect "eieio", ""()
1737 <p>TODO: The format of the asm and constraints string still need to be
1738 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1739 need to be documented).
1744 <!-- *********************************************************************** -->
1745 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1746 <!-- *********************************************************************** -->
1748 <div class="doc_text">
1750 <p>The LLVM instruction set consists of several different
1751 classifications of instructions: <a href="#terminators">terminator
1752 instructions</a>, <a href="#binaryops">binary instructions</a>,
1753 <a href="#bitwiseops">bitwise binary instructions</a>, <a
1754 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1755 instructions</a>.</p>
1759 <!-- ======================================================================= -->
1760 <div class="doc_subsection"> <a name="terminators">Terminator
1761 Instructions</a> </div>
1763 <div class="doc_text">
1765 <p>As mentioned <a href="#functionstructure">previously</a>, every
1766 basic block in a program ends with a "Terminator" instruction, which
1767 indicates which block should be executed after the current block is
1768 finished. These terminator instructions typically yield a '<tt>void</tt>'
1769 value: they produce control flow, not values (the one exception being
1770 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1771 <p>There are six different terminator instructions: the '<a
1772 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1773 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1774 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1775 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1776 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1780 <!-- _______________________________________________________________________ -->
1781 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1782 Instruction</a> </div>
1783 <div class="doc_text">
1785 <pre> ret <type> <value> <i>; Return a value from a non-void function</i>
1786 ret void <i>; Return from void function</i>
1789 <p>The '<tt>ret</tt>' instruction is used to return control flow (and a
1790 value) from a function back to the caller.</p>
1791 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
1792 returns a value and then causes control flow, and one that just causes
1793 control flow to occur.</p>
1795 <p>The '<tt>ret</tt>' instruction may return any '<a
1796 href="#t_firstclass">first class</a>' type. Notice that a function is
1797 not <a href="#wellformed">well formed</a> if there exists a '<tt>ret</tt>'
1798 instruction inside of the function that returns a value that does not
1799 match the return type of the function.</p>
1801 <p>When the '<tt>ret</tt>' instruction is executed, control flow
1802 returns back to the calling function's context. If the caller is a "<a
1803 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1804 the instruction after the call. If the caller was an "<a
1805 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1806 at the beginning of the "normal" destination block. If the instruction
1807 returns a value, that value shall set the call or invoke instruction's
1810 <pre> ret i32 5 <i>; Return an integer value of 5</i>
1811 ret void <i>; Return from a void function</i>
1814 <!-- _______________________________________________________________________ -->
1815 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1816 <div class="doc_text">
1818 <pre> br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1821 <p>The '<tt>br</tt>' instruction is used to cause control flow to
1822 transfer to a different basic block in the current function. There are
1823 two forms of this instruction, corresponding to a conditional branch
1824 and an unconditional branch.</p>
1826 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1827 single '<tt>i1</tt>' value and two '<tt>label</tt>' values. The
1828 unconditional form of the '<tt>br</tt>' instruction takes a single
1829 '<tt>label</tt>' value as a target.</p>
1831 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
1832 argument is evaluated. If the value is <tt>true</tt>, control flows
1833 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1834 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1836 <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
1837 href="#i_ret">ret</a> i32 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> i32 0<br></pre>
1839 <!-- _______________________________________________________________________ -->
1840 <div class="doc_subsubsection">
1841 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
1844 <div class="doc_text">
1848 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
1853 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
1854 several different places. It is a generalization of the '<tt>br</tt>'
1855 instruction, allowing a branch to occur to one of many possible
1861 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
1862 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
1863 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
1864 table is not allowed to contain duplicate constant entries.</p>
1868 <p>The <tt>switch</tt> instruction specifies a table of values and
1869 destinations. When the '<tt>switch</tt>' instruction is executed, this
1870 table is searched for the given value. If the value is found, control flow is
1871 transfered to the corresponding destination; otherwise, control flow is
1872 transfered to the default destination.</p>
1874 <h5>Implementation:</h5>
1876 <p>Depending on properties of the target machine and the particular
1877 <tt>switch</tt> instruction, this instruction may be code generated in different
1878 ways. For example, it could be generated as a series of chained conditional
1879 branches or with a lookup table.</p>
1884 <i>; Emulate a conditional br instruction</i>
1885 %Val = <a href="#i_zext">zext</a> i1 %value to i32
1886 switch i32 %Val, label %truedest [i32 0, label %falsedest ]
1888 <i>; Emulate an unconditional br instruction</i>
1889 switch i32 0, label %dest [ ]
1891 <i>; Implement a jump table:</i>
1892 switch i32 %val, label %otherwise [ i32 0, label %onzero
1894 i32 2, label %ontwo ]
1898 <!-- _______________________________________________________________________ -->
1899 <div class="doc_subsubsection">
1900 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
1903 <div class="doc_text">
1908 <result> = invoke [<a href="#callingconv">cconv</a>] <ptr to function ty> %<function ptr val>(<function args>)
1909 to label <normal label> unwind label <exception label>
1914 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
1915 function, with the possibility of control flow transfer to either the
1916 '<tt>normal</tt>' label or the
1917 '<tt>exception</tt>' label. If the callee function returns with the
1918 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
1919 "normal" label. If the callee (or any indirect callees) returns with the "<a
1920 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
1921 continued at the dynamically nearest "exception" label.</p>
1925 <p>This instruction requires several arguments:</p>
1929 The optional "cconv" marker indicates which <a href="#callingconv">calling
1930 convention</a> the call should use. If none is specified, the call defaults
1931 to using C calling conventions.
1933 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
1934 function value being invoked. In most cases, this is a direct function
1935 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
1936 an arbitrary pointer to function value.
1939 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
1940 function to be invoked. </li>
1942 <li>'<tt>function args</tt>': argument list whose types match the function
1943 signature argument types. If the function signature indicates the function
1944 accepts a variable number of arguments, the extra arguments can be
1947 <li>'<tt>normal label</tt>': the label reached when the called function
1948 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
1950 <li>'<tt>exception label</tt>': the label reached when a callee returns with
1951 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
1957 <p>This instruction is designed to operate as a standard '<tt><a
1958 href="#i_call">call</a></tt>' instruction in most regards. The primary
1959 difference is that it establishes an association with a label, which is used by
1960 the runtime library to unwind the stack.</p>
1962 <p>This instruction is used in languages with destructors to ensure that proper
1963 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
1964 exception. Additionally, this is important for implementation of
1965 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
1969 %retval = invoke i32 %Test(i32 15) to label %Continue
1970 unwind label %TestCleanup <i>; {i32}:retval set</i>
1971 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Test(i32 15) to label %Continue
1972 unwind label %TestCleanup <i>; {i32}:retval set</i>
1977 <!-- _______________________________________________________________________ -->
1979 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
1980 Instruction</a> </div>
1982 <div class="doc_text">
1991 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
1992 at the first callee in the dynamic call stack which used an <a
1993 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
1994 primarily used to implement exception handling.</p>
1998 <p>The '<tt>unwind</tt>' intrinsic causes execution of the current function to
1999 immediately halt. The dynamic call stack is then searched for the first <a
2000 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
2001 execution continues at the "exceptional" destination block specified by the
2002 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
2003 dynamic call chain, undefined behavior results.</p>
2006 <!-- _______________________________________________________________________ -->
2008 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
2009 Instruction</a> </div>
2011 <div class="doc_text">
2020 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
2021 instruction is used to inform the optimizer that a particular portion of the
2022 code is not reachable. This can be used to indicate that the code after a
2023 no-return function cannot be reached, and other facts.</p>
2027 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
2032 <!-- ======================================================================= -->
2033 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
2034 <div class="doc_text">
2035 <p>Binary operators are used to do most of the computation in a
2036 program. They require two operands, execute an operation on them, and
2037 produce a single value. The operands might represent
2038 multiple data, as is the case with the <a href="#t_vector">vector</a> data type.
2039 The result value of a binary operator is not
2040 necessarily the same type as its operands.</p>
2041 <p>There are several different binary operators:</p>
2043 <!-- _______________________________________________________________________ -->
2044 <div class="doc_subsubsection"> <a name="i_add">'<tt>add</tt>'
2045 Instruction</a> </div>
2046 <div class="doc_text">
2048 <pre> <result> = add <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2051 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
2053 <p>The two arguments to the '<tt>add</tt>' instruction must be either <a
2054 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a> values.
2055 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2056 Both arguments must have identical types.</p>
2058 <p>The value produced is the integer or floating point sum of the two
2060 <p>If an integer sum has unsigned overflow, the result returned is the
2061 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2063 <p>Because LLVM integers use a two's complement representation, this
2064 instruction is appropriate for both signed and unsigned integers.</p>
2066 <pre> <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
2069 <!-- _______________________________________________________________________ -->
2070 <div class="doc_subsubsection"> <a name="i_sub">'<tt>sub</tt>'
2071 Instruction</a> </div>
2072 <div class="doc_text">
2074 <pre> <result> = sub <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2077 <p>The '<tt>sub</tt>' instruction returns the difference of its two
2079 <p>Note that the '<tt>sub</tt>' instruction is used to represent the '<tt>neg</tt>'
2080 instruction present in most other intermediate representations.</p>
2082 <p>The two arguments to the '<tt>sub</tt>' instruction must be either <a
2083 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
2085 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2086 Both arguments must have identical types.</p>
2088 <p>The value produced is the integer or floating point difference of
2089 the two operands.</p>
2090 <p>If an integer difference has unsigned overflow, the result returned is the
2091 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2093 <p>Because LLVM integers use a two's complement representation, this
2094 instruction is appropriate for both signed and unsigned integers.</p>
2097 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
2098 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
2101 <!-- _______________________________________________________________________ -->
2102 <div class="doc_subsubsection"> <a name="i_mul">'<tt>mul</tt>'
2103 Instruction</a> </div>
2104 <div class="doc_text">
2106 <pre> <result> = mul <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2109 <p>The '<tt>mul</tt>' instruction returns the product of its two
2112 <p>The two arguments to the '<tt>mul</tt>' instruction must be either <a
2113 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
2115 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2116 Both arguments must have identical types.</p>
2118 <p>The value produced is the integer or floating point product of the
2120 <p>If the result of an integer multiplication has unsigned overflow,
2121 the result returned is the mathematical result modulo
2122 2<sup>n</sup>, where n is the bit width of the result.</p>
2123 <p>Because LLVM integers use a two's complement representation, and the
2124 result is the same width as the operands, this instruction returns the
2125 correct result for both signed and unsigned integers. If a full product
2126 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands
2127 should be sign-extended or zero-extended as appropriate to the
2128 width of the full product.</p>
2130 <pre> <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
2133 <!-- _______________________________________________________________________ -->
2134 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
2136 <div class="doc_text">
2138 <pre> <result> = udiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2141 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two
2144 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
2145 <a href="#t_integer">integer</a> values. Both arguments must have identical
2146 types. This instruction can also take <a href="#t_vector">vector</a> versions
2147 of the values in which case the elements must be integers.</p>
2149 <p>The value produced is the unsigned integer quotient of the two operands.</p>
2150 <p>Note that unsigned integer division and signed integer division are distinct
2151 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
2152 <p>Division by zero leads to undefined behavior.</p>
2154 <pre> <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2157 <!-- _______________________________________________________________________ -->
2158 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
2160 <div class="doc_text">
2162 <pre> <result> = sdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2165 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
2168 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
2169 <a href="#t_integer">integer</a> values. Both arguments must have identical
2170 types. This instruction can also take <a href="#t_vector">vector</a> versions
2171 of the values in which case the elements must be integers.</p>
2173 <p>The value produced is the signed integer quotient of the two operands.</p>
2174 <p>Note that signed integer division and unsigned integer division are distinct
2175 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
2176 <p>Division by zero leads to undefined behavior. Overflow also leads to
2177 undefined behavior; this is a rare case, but can occur, for example,
2178 by doing a 32-bit division of -2147483648 by -1.</p>
2180 <pre> <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2183 <!-- _______________________________________________________________________ -->
2184 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
2185 Instruction</a> </div>
2186 <div class="doc_text">
2188 <pre> <result> = fdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2191 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
2194 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
2195 <a href="#t_floating">floating point</a> values. Both arguments must have
2196 identical types. This instruction can also take <a href="#t_vector">vector</a>
2197 versions of floating point values.</p>
2199 <p>The value produced is the floating point quotient of the two operands.</p>
2201 <pre> <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
2204 <!-- _______________________________________________________________________ -->
2205 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
2207 <div class="doc_text">
2209 <pre> <result> = urem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2212 <p>The '<tt>urem</tt>' instruction returns the remainder from the
2213 unsigned division of its two arguments.</p>
2215 <p>The two arguments to the '<tt>urem</tt>' instruction must be
2216 <a href="#t_integer">integer</a> values. Both arguments must have identical
2217 types. This instruction can also take <a href="#t_vector">vector</a> versions
2218 of the values in which case the elements must be integers.</p>
2220 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
2221 This instruction always performs an unsigned division to get the remainder,
2222 regardless of whether the arguments are unsigned or not.</p>
2223 <p>Note that unsigned integer remainder and signed integer remainder are
2224 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
2225 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
2227 <pre> <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2231 <!-- _______________________________________________________________________ -->
2232 <div class="doc_subsubsection"> <a name="i_srem">'<tt>srem</tt>'
2233 Instruction</a> </div>
2234 <div class="doc_text">
2236 <pre> <result> = srem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2239 <p>The '<tt>srem</tt>' instruction returns the remainder from the
2240 signed division of its two operands. This instruction can also take
2241 <a href="#t_vector">vector</a> versions of the values in which case
2242 the elements must be integers.</p>
2245 <p>The two arguments to the '<tt>srem</tt>' instruction must be
2246 <a href="#t_integer">integer</a> values. Both arguments must have identical
2249 <p>This instruction returns the <i>remainder</i> of a division (where the result
2250 has the same sign as the dividend, <tt>var1</tt>), not the <i>modulo</i>
2251 operator (where the result has the same sign as the divisor, <tt>var2</tt>) of
2252 a value. For more information about the difference, see <a
2253 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
2254 Math Forum</a>. For a table of how this is implemented in various languages,
2255 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
2256 Wikipedia: modulo operation</a>.</p>
2257 <p>Note that signed integer remainder and unsigned integer remainder are
2258 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
2259 <p>Taking the remainder of a division by zero leads to undefined behavior.
2260 Overflow also leads to undefined behavior; this is a rare case, but can occur,
2261 for example, by taking the remainder of a 32-bit division of -2147483648 by -1.
2262 (The remainder doesn't actually overflow, but this rule lets srem be
2263 implemented using instructions that return both the result of the division
2264 and the remainder.)</p>
2266 <pre> <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2270 <!-- _______________________________________________________________________ -->
2271 <div class="doc_subsubsection"> <a name="i_frem">'<tt>frem</tt>'
2272 Instruction</a> </div>
2273 <div class="doc_text">
2275 <pre> <result> = frem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2278 <p>The '<tt>frem</tt>' instruction returns the remainder from the
2279 division of its two operands.</p>
2281 <p>The two arguments to the '<tt>frem</tt>' instruction must be
2282 <a href="#t_floating">floating point</a> values. Both arguments must have
2283 identical types. This instruction can also take <a href="#t_vector">vector</a>
2284 versions of floating point values.</p>
2286 <p>This instruction returns the <i>remainder</i> of a division.</p>
2288 <pre> <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
2292 <!-- ======================================================================= -->
2293 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
2294 Operations</a> </div>
2295 <div class="doc_text">
2296 <p>Bitwise binary operators are used to do various forms of
2297 bit-twiddling in a program. They are generally very efficient
2298 instructions and can commonly be strength reduced from other
2299 instructions. They require two operands, execute an operation on them,
2300 and produce a single value. The resulting value of the bitwise binary
2301 operators is always the same type as its first operand.</p>
2304 <!-- _______________________________________________________________________ -->
2305 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
2306 Instruction</a> </div>
2307 <div class="doc_text">
2309 <pre> <result> = shl <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2314 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2315 the left a specified number of bits.</p>
2319 <p>Both arguments to the '<tt>shl</tt>' instruction must be the same <a
2320 href="#t_integer">integer</a> type.</p>
2324 <p>The value produced is <tt>var1</tt> * 2<sup><tt>var2</tt></sup>. If
2325 <tt>var2</tt> is (statically or dynamically) equal to or larger than the number
2326 of bits in <tt>var1</tt>, the result is undefined.</p>
2328 <h5>Example:</h5><pre>
2329 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
2330 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
2331 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
2332 <result> = shl i32 1, 32 <i>; undefined</i>
2335 <!-- _______________________________________________________________________ -->
2336 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
2337 Instruction</a> </div>
2338 <div class="doc_text">
2340 <pre> <result> = lshr <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2344 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
2345 operand shifted to the right a specified number of bits with zero fill.</p>
2348 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
2349 <a href="#t_integer">integer</a> type.</p>
2353 <p>This instruction always performs a logical shift right operation. The most
2354 significant bits of the result will be filled with zero bits after the
2355 shift. If <tt>var2</tt> is (statically or dynamically) equal to or larger than
2356 the number of bits in <tt>var1</tt>, the result is undefined.</p>
2360 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
2361 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
2362 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
2363 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
2364 <result> = lshr i32 1, 32 <i>; undefined</i>
2368 <!-- _______________________________________________________________________ -->
2369 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
2370 Instruction</a> </div>
2371 <div class="doc_text">
2374 <pre> <result> = ashr <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2378 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
2379 operand shifted to the right a specified number of bits with sign extension.</p>
2382 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
2383 <a href="#t_integer">integer</a> type.</p>
2386 <p>This instruction always performs an arithmetic shift right operation,
2387 The most significant bits of the result will be filled with the sign bit
2388 of <tt>var1</tt>. If <tt>var2</tt> is (statically or dynamically) equal to or
2389 larger than the number of bits in <tt>var1</tt>, the result is undefined.
2394 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
2395 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
2396 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
2397 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
2398 <result> = ashr i32 1, 32 <i>; undefined</i>
2402 <!-- _______________________________________________________________________ -->
2403 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
2404 Instruction</a> </div>
2405 <div class="doc_text">
2407 <pre> <result> = and <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2410 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
2411 its two operands.</p>
2413 <p>The two arguments to the '<tt>and</tt>' instruction must be <a
2414 href="#t_integer">integer</a> values. Both arguments must have
2415 identical types.</p>
2417 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
2419 <div style="align: center">
2420 <table border="1" cellspacing="0" cellpadding="4">
2451 <pre> <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
2452 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
2453 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
2456 <!-- _______________________________________________________________________ -->
2457 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
2458 <div class="doc_text">
2460 <pre> <result> = or <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2463 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
2464 or of its two operands.</p>
2466 <p>The two arguments to the '<tt>or</tt>' instruction must be <a
2467 href="#t_integer">integer</a> values. Both arguments must have
2468 identical types.</p>
2470 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
2472 <div style="align: center">
2473 <table border="1" cellspacing="0" cellpadding="4">
2504 <pre> <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
2505 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
2506 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
2509 <!-- _______________________________________________________________________ -->
2510 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
2511 Instruction</a> </div>
2512 <div class="doc_text">
2514 <pre> <result> = xor <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2517 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
2518 or of its two operands. The <tt>xor</tt> is used to implement the
2519 "one's complement" operation, which is the "~" operator in C.</p>
2521 <p>The two arguments to the '<tt>xor</tt>' instruction must be <a
2522 href="#t_integer">integer</a> values. Both arguments must have
2523 identical types.</p>
2525 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
2527 <div style="align: center">
2528 <table border="1" cellspacing="0" cellpadding="4">
2560 <pre> <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
2561 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
2562 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
2563 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
2567 <!-- ======================================================================= -->
2568 <div class="doc_subsection">
2569 <a name="vectorops">Vector Operations</a>
2572 <div class="doc_text">
2574 <p>LLVM supports several instructions to represent vector operations in a
2575 target-independent manner. These instructions cover the element-access and
2576 vector-specific operations needed to process vectors effectively. While LLVM
2577 does directly support these vector operations, many sophisticated algorithms
2578 will want to use target-specific intrinsics to take full advantage of a specific
2583 <!-- _______________________________________________________________________ -->
2584 <div class="doc_subsubsection">
2585 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2588 <div class="doc_text">
2593 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
2599 The '<tt>extractelement</tt>' instruction extracts a single scalar
2600 element from a vector at a specified index.
2607 The first operand of an '<tt>extractelement</tt>' instruction is a
2608 value of <a href="#t_vector">vector</a> type. The second operand is
2609 an index indicating the position from which to extract the element.
2610 The index may be a variable.</p>
2615 The result is a scalar of the same type as the element type of
2616 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2617 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2618 results are undefined.
2624 %result = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
2629 <!-- _______________________________________________________________________ -->
2630 <div class="doc_subsubsection">
2631 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
2634 <div class="doc_text">
2639 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
2645 The '<tt>insertelement</tt>' instruction inserts a scalar
2646 element into a vector at a specified index.
2653 The first operand of an '<tt>insertelement</tt>' instruction is a
2654 value of <a href="#t_vector">vector</a> type. The second operand is a
2655 scalar value whose type must equal the element type of the first
2656 operand. The third operand is an index indicating the position at
2657 which to insert the value. The index may be a variable.</p>
2662 The result is a vector of the same type as <tt>val</tt>. Its
2663 element values are those of <tt>val</tt> except at position
2664 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
2665 exceeds the length of <tt>val</tt>, the results are undefined.
2671 %result = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
2675 <!-- _______________________________________________________________________ -->
2676 <div class="doc_subsubsection">
2677 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
2680 <div class="doc_text">
2685 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <n x i32> <mask> <i>; yields <n x <ty>></i>
2691 The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
2692 from two input vectors, returning a vector of the same type.
2698 The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
2699 with types that match each other and types that match the result of the
2700 instruction. The third argument is a shuffle mask, which has the same number
2701 of elements as the other vector type, but whose element type is always 'i32'.
2705 The shuffle mask operand is required to be a constant vector with either
2706 constant integer or undef values.
2712 The elements of the two input vectors are numbered from left to right across
2713 both of the vectors. The shuffle mask operand specifies, for each element of
2714 the result vector, which element of the two input registers the result element
2715 gets. The element selector may be undef (meaning "don't care") and the second
2716 operand may be undef if performing a shuffle from only one vector.
2722 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
2723 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
2724 %result = shufflevector <4 x i32> %v1, <4 x i32> undef,
2725 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
2730 <!-- ======================================================================= -->
2731 <div class="doc_subsection">
2732 <a name="memoryops">Memory Access and Addressing Operations</a>
2735 <div class="doc_text">
2737 <p>A key design point of an SSA-based representation is how it
2738 represents memory. In LLVM, no memory locations are in SSA form, which
2739 makes things very simple. This section describes how to read, write,
2740 allocate, and free memory in LLVM.</p>
2744 <!-- _______________________________________________________________________ -->
2745 <div class="doc_subsubsection">
2746 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
2749 <div class="doc_text">
2754 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2759 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
2760 heap and returns a pointer to it. The object is always allocated in the generic
2761 address space (address space zero).</p>
2765 <p>The '<tt>malloc</tt>' instruction allocates
2766 <tt>sizeof(<type>)*NumElements</tt>
2767 bytes of memory from the operating system and returns a pointer of the
2768 appropriate type to the program. If "NumElements" is specified, it is the
2769 number of elements allocated, otherwise "NumElements" is defaulted to be one.
2770 If an alignment is specified, the value result of the allocation is guaranteed to
2771 be aligned to at least that boundary. If not specified, or if zero, the target can
2772 choose to align the allocation on any convenient boundary.</p>
2774 <p>'<tt>type</tt>' must be a sized type.</p>
2778 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
2779 a pointer is returned.</p>
2784 %array = malloc [4 x i8 ] <i>; yields {[%4 x i8]*}:array</i>
2786 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
2787 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
2788 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
2789 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
2790 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
2794 <!-- _______________________________________________________________________ -->
2795 <div class="doc_subsubsection">
2796 <a name="i_free">'<tt>free</tt>' Instruction</a>
2799 <div class="doc_text">
2804 free <type> <value> <i>; yields {void}</i>
2809 <p>The '<tt>free</tt>' instruction returns memory back to the unused
2810 memory heap to be reallocated in the future.</p>
2814 <p>'<tt>value</tt>' shall be a pointer value that points to a value
2815 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
2820 <p>Access to the memory pointed to by the pointer is no longer defined
2821 after this instruction executes.</p>
2826 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
2827 free [4 x i8]* %array
2831 <!-- _______________________________________________________________________ -->
2832 <div class="doc_subsubsection">
2833 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
2836 <div class="doc_text">
2841 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2846 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
2847 currently executing function, to be automatically released when this function
2848 returns to its caller. The object is always allocated in the generic address
2849 space (address space zero).</p>
2853 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
2854 bytes of memory on the runtime stack, returning a pointer of the
2855 appropriate type to the program. If "NumElements" is specified, it is the
2856 number of elements allocated, otherwise "NumElements" is defaulted to be one.
2857 If an alignment is specified, the value result of the allocation is guaranteed
2858 to be aligned to at least that boundary. If not specified, or if zero, the target
2859 can choose to align the allocation on any convenient boundary.</p>
2861 <p>'<tt>type</tt>' may be any sized type.</p>
2865 <p>Memory is allocated; a pointer is returned. '<tt>alloca</tt>'d
2866 memory is automatically released when the function returns. The '<tt>alloca</tt>'
2867 instruction is commonly used to represent automatic variables that must
2868 have an address available. When the function returns (either with the <tt><a
2869 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
2870 instructions), the memory is reclaimed.</p>
2875 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
2876 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
2877 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
2878 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
2882 <!-- _______________________________________________________________________ -->
2883 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
2884 Instruction</a> </div>
2885 <div class="doc_text">
2887 <pre> <result> = load <ty>* <pointer>[, align <alignment>]<br> <result> = volatile load <ty>* <pointer>[, align <alignment>]<br></pre>
2889 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
2891 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
2892 address from which to load. The pointer must point to a <a
2893 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
2894 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
2895 the number or order of execution of this <tt>load</tt> with other
2896 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
2899 The optional "align" argument specifies the alignment of the operation
2900 (that is, the alignment of the memory address). A value of 0 or an
2901 omitted "align" argument means that the operation has the preferential
2902 alignment for the target. It is the responsibility of the code emitter
2903 to ensure that the alignment information is correct. Overestimating
2904 the alignment results in an undefined behavior. Underestimating the
2905 alignment may produce less efficient code. An alignment of 1 is always
2909 <p>The location of memory pointed to is loaded.</p>
2911 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
2913 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
2914 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
2917 <!-- _______________________________________________________________________ -->
2918 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
2919 Instruction</a> </div>
2920 <div class="doc_text">
2922 <pre> store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
2923 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
2926 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
2928 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
2929 to store and an address at which to store it. The type of the '<tt><pointer></tt>'
2930 operand must be a pointer to the type of the '<tt><value></tt>'
2931 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
2932 optimizer is not allowed to modify the number or order of execution of
2933 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
2934 href="#i_store">store</a></tt> instructions.</p>
2936 The optional "align" argument specifies the alignment of the operation
2937 (that is, the alignment of the memory address). A value of 0 or an
2938 omitted "align" argument means that the operation has the preferential
2939 alignment for the target. It is the responsibility of the code emitter
2940 to ensure that the alignment information is correct. Overestimating
2941 the alignment results in an undefined behavior. Underestimating the
2942 alignment may produce less efficient code. An alignment of 1 is always
2946 <p>The contents of memory are updated to contain '<tt><value></tt>'
2947 at the location specified by the '<tt><pointer></tt>' operand.</p>
2949 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
2950 store i32 3, i32* %ptr <i>; yields {void}</i>
2951 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
2955 <!-- _______________________________________________________________________ -->
2956 <div class="doc_subsubsection">
2957 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
2960 <div class="doc_text">
2963 <result> = getelementptr <ty>* <ptrval>{, <ty> <idx>}*
2969 The '<tt>getelementptr</tt>' instruction is used to get the address of a
2970 subelement of an aggregate data structure.</p>
2974 <p>This instruction takes a list of integer operands that indicate what
2975 elements of the aggregate object to index to. The actual types of the arguments
2976 provided depend on the type of the first pointer argument. The
2977 '<tt>getelementptr</tt>' instruction is used to index down through the type
2978 levels of a structure or to a specific index in an array. When indexing into a
2979 structure, only <tt>i32</tt> integer constants are allowed. When indexing
2980 into an array or pointer, only integers of 32 or 64 bits are allowed, and will
2981 be sign extended to 64-bit values.</p>
2983 <p>For example, let's consider a C code fragment and how it gets
2984 compiled to LLVM:</p>
2986 <div class="doc_code">
2999 int *foo(struct ST *s) {
3000 return &s[1].Z.B[5][13];
3005 <p>The LLVM code generated by the GCC frontend is:</p>
3007 <div class="doc_code">
3009 %RT = type { i8 , [10 x [20 x i32]], i8 }
3010 %ST = type { i32, double, %RT }
3012 define i32* %foo(%ST* %s) {
3014 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
3022 <p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
3023 on the pointer type that is being indexed into. <a href="#t_pointer">Pointer</a>
3024 and <a href="#t_array">array</a> types can use a 32-bit or 64-bit
3025 <a href="#t_integer">integer</a> type but the value will always be sign extended
3026 to 64-bits. <a href="#t_struct">Structure</a> types require <tt>i32</tt>
3027 <b>constants</b>.</p>
3029 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
3030 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
3031 }</tt>' type, a structure. The second index indexes into the third element of
3032 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
3033 i8 }</tt>' type, another structure. The third index indexes into the second
3034 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
3035 array. The two dimensions of the array are subscripted into, yielding an
3036 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
3037 to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
3039 <p>Note that it is perfectly legal to index partially through a
3040 structure, returning a pointer to an inner element. Because of this,
3041 the LLVM code for the given testcase is equivalent to:</p>
3044 define i32* %foo(%ST* %s) {
3045 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
3046 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
3047 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
3048 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
3049 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
3054 <p>Note that it is undefined to access an array out of bounds: array and
3055 pointer indexes must always be within the defined bounds of the array type.
3056 The one exception for this rules is zero length arrays. These arrays are
3057 defined to be accessible as variable length arrays, which requires access
3058 beyond the zero'th element.</p>
3060 <p>The getelementptr instruction is often confusing. For some more insight
3061 into how it works, see <a href="GetElementPtr.html">the getelementptr
3067 <i>; yields [12 x i8]*:aptr</i>
3068 %aptr = getelementptr {i32, [12 x i8]}* %sptr, i64 0, i32 1
3072 <!-- ======================================================================= -->
3073 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
3075 <div class="doc_text">
3076 <p>The instructions in this category are the conversion instructions (casting)
3077 which all take a single operand and a type. They perform various bit conversions
3081 <!-- _______________________________________________________________________ -->
3082 <div class="doc_subsubsection">
3083 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
3085 <div class="doc_text">
3089 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
3094 The '<tt>trunc</tt>' instruction truncates its operand to the type <tt>ty2</tt>.
3099 The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
3100 be an <a href="#t_integer">integer</a> type, and a type that specifies the size
3101 and type of the result, which must be an <a href="#t_integer">integer</a>
3102 type. The bit size of <tt>value</tt> must be larger than the bit size of
3103 <tt>ty2</tt>. Equal sized types are not allowed.</p>
3107 The '<tt>trunc</tt>' instruction truncates the high order bits in <tt>value</tt>
3108 and converts the remaining bits to <tt>ty2</tt>. Since the source size must be
3109 larger than the destination size, <tt>trunc</tt> cannot be a <i>no-op cast</i>.
3110 It will always truncate bits.</p>
3114 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
3115 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
3116 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
3120 <!-- _______________________________________________________________________ -->
3121 <div class="doc_subsubsection">
3122 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
3124 <div class="doc_text">
3128 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
3132 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
3137 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
3138 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3139 also be of <a href="#t_integer">integer</a> type. The bit size of the
3140 <tt>value</tt> must be smaller than the bit size of the destination type,
3144 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
3145 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
3147 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
3151 %X = zext i32 257 to i64 <i>; yields i64:257</i>
3152 %Y = zext i1 true to i32 <i>; yields i32:1</i>
3156 <!-- _______________________________________________________________________ -->
3157 <div class="doc_subsubsection">
3158 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
3160 <div class="doc_text">
3164 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
3168 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
3172 The '<tt>sext</tt>' instruction takes a value to cast, which must be of
3173 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3174 also be of <a href="#t_integer">integer</a> type. The bit size of the
3175 <tt>value</tt> must be smaller than the bit size of the destination type,
3180 The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
3181 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size of
3182 the type <tt>ty2</tt>.</p>
3184 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
3188 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
3189 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
3193 <!-- _______________________________________________________________________ -->
3194 <div class="doc_subsubsection">
3195 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
3198 <div class="doc_text">
3203 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
3207 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
3212 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
3213 point</a> value to cast and a <a href="#t_floating">floating point</a> type to
3214 cast it to. The size of <tt>value</tt> must be larger than the size of
3215 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
3216 <i>no-op cast</i>.</p>
3219 <p> The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
3220 <a href="#t_floating">floating point</a> type to a smaller
3221 <a href="#t_floating">floating point</a> type. If the value cannot fit within
3222 the destination type, <tt>ty2</tt>, then the results are undefined.</p>
3226 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
3227 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
3231 <!-- _______________________________________________________________________ -->
3232 <div class="doc_subsubsection">
3233 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
3235 <div class="doc_text">
3239 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
3243 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
3244 floating point value.</p>
3247 <p>The '<tt>fpext</tt>' instruction takes a
3248 <a href="#t_floating">floating point</a> <tt>value</tt> to cast,
3249 and a <a href="#t_floating">floating point</a> type to cast it to. The source
3250 type must be smaller than the destination type.</p>
3253 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
3254 <a href="#t_floating">floating point</a> type to a larger
3255 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
3256 used to make a <i>no-op cast</i> because it always changes bits. Use
3257 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
3261 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
3262 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
3266 <!-- _______________________________________________________________________ -->
3267 <div class="doc_subsubsection">
3268 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
3270 <div class="doc_text">
3274 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
3278 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
3279 unsigned integer equivalent of type <tt>ty2</tt>.
3283 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
3284 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3285 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3286 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3287 vector integer type with the same number of elements as <tt>ty</tt></p>
3290 <p> The '<tt>fptoui</tt>' instruction converts its
3291 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3292 towards zero) unsigned integer value. If the value cannot fit in <tt>ty2</tt>,
3293 the results are undefined.</p>
3297 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
3298 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
3299 %X = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
3303 <!-- _______________________________________________________________________ -->
3304 <div class="doc_subsubsection">
3305 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
3307 <div class="doc_text">
3311 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
3315 <p>The '<tt>fptosi</tt>' instruction converts
3316 <a href="#t_floating">floating point</a> <tt>value</tt> to type <tt>ty2</tt>.
3320 <p> The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
3321 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3322 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3323 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3324 vector integer type with the same number of elements as <tt>ty</tt></p>
3327 <p>The '<tt>fptosi</tt>' instruction converts its
3328 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3329 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
3330 the results are undefined.</p>
3334 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
3335 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
3336 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
3340 <!-- _______________________________________________________________________ -->
3341 <div class="doc_subsubsection">
3342 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
3344 <div class="doc_text">
3348 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3352 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
3353 integer and converts that value to the <tt>ty2</tt> type.</p>
3356 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
3357 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3358 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3359 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3360 floating point type with the same number of elements as <tt>ty</tt></p>
3363 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
3364 integer quantity and converts it to the corresponding floating point value. If
3365 the value cannot fit in the floating point value, the results are undefined.</p>
3369 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
3370 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
3374 <!-- _______________________________________________________________________ -->
3375 <div class="doc_subsubsection">
3376 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
3378 <div class="doc_text">
3382 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3386 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed
3387 integer and converts that value to the <tt>ty2</tt> type.</p>
3390 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
3391 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3392 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3393 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3394 floating point type with the same number of elements as <tt>ty</tt></p>
3397 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed
3398 integer quantity and converts it to the corresponding floating point value. If
3399 the value cannot fit in the floating point value, the results are undefined.</p>
3403 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
3404 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
3408 <!-- _______________________________________________________________________ -->
3409 <div class="doc_subsubsection">
3410 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
3412 <div class="doc_text">
3416 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
3420 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
3421 the integer type <tt>ty2</tt>.</p>
3424 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
3425 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
3426 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.
3429 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
3430 <tt>ty2</tt> by interpreting the pointer value as an integer and either
3431 truncating or zero extending that value to the size of the integer type. If
3432 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
3433 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
3434 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
3439 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
3440 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
3444 <!-- _______________________________________________________________________ -->
3445 <div class="doc_subsubsection">
3446 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
3448 <div class="doc_text">
3452 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
3456 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to
3457 a pointer type, <tt>ty2</tt>.</p>
3460 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
3461 value to cast, and a type to cast it to, which must be a
3462 <a href="#t_pointer">pointer</a> type.
3465 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
3466 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
3467 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
3468 size of a pointer then a truncation is done. If <tt>value</tt> is smaller than
3469 the size of a pointer then a zero extension is done. If they are the same size,
3470 nothing is done (<i>no-op cast</i>).</p>
3474 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
3475 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
3476 %Y = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
3480 <!-- _______________________________________________________________________ -->
3481 <div class="doc_subsubsection">
3482 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
3484 <div class="doc_text">
3488 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
3492 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3493 <tt>ty2</tt> without changing any bits.</p>
3496 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be
3497 a first class value, and a type to cast it to, which must also be a <a
3498 href="#t_firstclass">first class</a> type. The bit sizes of <tt>value</tt>
3499 and the destination type, <tt>ty2</tt>, must be identical. If the source
3500 type is a pointer, the destination type must also be a pointer.</p>
3503 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3504 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
3505 this conversion. The conversion is done as if the <tt>value</tt> had been
3506 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only be
3507 converted to other pointer types with this instruction. To convert pointers to
3508 other types, use the <a href="#i_inttoptr">inttoptr</a> or
3509 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
3513 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
3514 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
3515 %Z = bitcast <2xint> %V to i64; <i>; yields i64: %V</i>
3519 <!-- ======================================================================= -->
3520 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
3521 <div class="doc_text">
3522 <p>The instructions in this category are the "miscellaneous"
3523 instructions, which defy better classification.</p>
3526 <!-- _______________________________________________________________________ -->
3527 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
3529 <div class="doc_text">
3531 <pre> <result> = icmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3534 <p>The '<tt>icmp</tt>' instruction returns a boolean value based on comparison
3535 of its two integer operands.</p>
3537 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
3538 the condition code indicating the kind of comparison to perform. It is not
3539 a value, just a keyword. The possible condition code are:
3541 <li><tt>eq</tt>: equal</li>
3542 <li><tt>ne</tt>: not equal </li>
3543 <li><tt>ugt</tt>: unsigned greater than</li>
3544 <li><tt>uge</tt>: unsigned greater or equal</li>
3545 <li><tt>ult</tt>: unsigned less than</li>
3546 <li><tt>ule</tt>: unsigned less or equal</li>
3547 <li><tt>sgt</tt>: signed greater than</li>
3548 <li><tt>sge</tt>: signed greater or equal</li>
3549 <li><tt>slt</tt>: signed less than</li>
3550 <li><tt>sle</tt>: signed less or equal</li>
3552 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
3553 <a href="#t_pointer">pointer</a> typed. They must also be identical types.</p>
3555 <p>The '<tt>icmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3556 the condition code given as <tt>cond</tt>. The comparison performed always
3557 yields a <a href="#t_primitive">i1</a> result, as follows:
3559 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
3560 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3562 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
3563 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3564 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
3565 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3566 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
3567 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3568 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
3569 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3570 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
3571 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3572 <li><tt>sgt</tt>: interprets the operands as signed values and yields
3573 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3574 <li><tt>sge</tt>: interprets the operands as signed values and yields
3575 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3576 <li><tt>slt</tt>: interprets the operands as signed values and yields
3577 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3578 <li><tt>sle</tt>: interprets the operands as signed values and yields
3579 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3581 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
3582 values are compared as if they were integers.</p>
3585 <pre> <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
3586 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
3587 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
3588 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
3589 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
3590 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
3594 <!-- _______________________________________________________________________ -->
3595 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
3597 <div class="doc_text">
3599 <pre> <result> = fcmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3602 <p>The '<tt>fcmp</tt>' instruction returns a boolean value based on comparison
3603 of its floating point operands.</p>
3605 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
3606 the condition code indicating the kind of comparison to perform. It is not
3607 a value, just a keyword. The possible condition code are:
3609 <li><tt>false</tt>: no comparison, always returns false</li>
3610 <li><tt>oeq</tt>: ordered and equal</li>
3611 <li><tt>ogt</tt>: ordered and greater than </li>
3612 <li><tt>oge</tt>: ordered and greater than or equal</li>
3613 <li><tt>olt</tt>: ordered and less than </li>
3614 <li><tt>ole</tt>: ordered and less than or equal</li>
3615 <li><tt>one</tt>: ordered and not equal</li>
3616 <li><tt>ord</tt>: ordered (no nans)</li>
3617 <li><tt>ueq</tt>: unordered or equal</li>
3618 <li><tt>ugt</tt>: unordered or greater than </li>
3619 <li><tt>uge</tt>: unordered or greater than or equal</li>
3620 <li><tt>ult</tt>: unordered or less than </li>
3621 <li><tt>ule</tt>: unordered or less than or equal</li>
3622 <li><tt>une</tt>: unordered or not equal</li>
3623 <li><tt>uno</tt>: unordered (either nans)</li>
3624 <li><tt>true</tt>: no comparison, always returns true</li>
3626 <p><i>Ordered</i> means that neither operand is a QNAN while
3627 <i>unordered</i> means that either operand may be a QNAN.</p>
3628 <p>The <tt>val1</tt> and <tt>val2</tt> arguments must be
3629 <a href="#t_floating">floating point</a> typed. They must have identical
3632 <p>The '<tt>fcmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3633 the condition code given as <tt>cond</tt>. The comparison performed always
3634 yields a <a href="#t_primitive">i1</a> result, as follows:
3636 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
3637 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3638 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3639 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3640 <tt>var1</tt> is greather than <tt>var2</tt>.</li>
3641 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3642 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3643 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3644 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3645 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3646 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3647 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3648 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3649 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
3650 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
3651 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3652 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
3653 <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3654 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
3655 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3656 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
3657 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3658 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
3659 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3660 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
3661 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3662 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
3663 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
3667 <pre> <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
3668 <result> = icmp one float 4.0, 5.0 <i>; yields: result=true</i>
3669 <result> = icmp olt float 4.0, 5.0 <i>; yields: result=true</i>
3670 <result> = icmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
3674 <!-- _______________________________________________________________________ -->
3675 <div class="doc_subsubsection"> <a name="i_phi">'<tt>phi</tt>'
3676 Instruction</a> </div>
3677 <div class="doc_text">
3679 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
3681 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
3682 the SSA graph representing the function.</p>
3684 <p>The type of the incoming values is specified with the first type
3685 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
3686 as arguments, with one pair for each predecessor basic block of the
3687 current block. Only values of <a href="#t_firstclass">first class</a>
3688 type may be used as the value arguments to the PHI node. Only labels
3689 may be used as the label arguments.</p>
3690 <p>There must be no non-phi instructions between the start of a basic
3691 block and the PHI instructions: i.e. PHI instructions must be first in
3694 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
3695 specified by the pair corresponding to the predecessor basic block that executed
3696 just prior to the current block.</p>
3698 <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>
3701 <!-- _______________________________________________________________________ -->
3702 <div class="doc_subsubsection">
3703 <a name="i_select">'<tt>select</tt>' Instruction</a>
3706 <div class="doc_text">
3711 <result> = select i1 <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
3717 The '<tt>select</tt>' instruction is used to choose one value based on a
3718 condition, without branching.
3725 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.
3731 If the boolean condition evaluates to true, the instruction returns the first
3732 value argument; otherwise, it returns the second value argument.
3738 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
3743 <!-- _______________________________________________________________________ -->
3744 <div class="doc_subsubsection">
3745 <a name="i_call">'<tt>call</tt>' Instruction</a>
3748 <div class="doc_text">
3752 <result> = [tail] call [<a href="#callingconv">cconv</a>] <ty> [<fnty>*] <fnptrval>(<param list>)
3757 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
3761 <p>This instruction requires several arguments:</p>
3765 <p>The optional "tail" marker indicates whether the callee function accesses
3766 any allocas or varargs in the caller. If the "tail" marker is present, the
3767 function call is eligible for tail call optimization. Note that calls may
3768 be marked "tail" even if they do not occur before a <a
3769 href="#i_ret"><tt>ret</tt></a> instruction.
3772 <p>The optional "cconv" marker indicates which <a href="#callingconv">calling
3773 convention</a> the call should use. If none is specified, the call defaults
3774 to using C calling conventions.
3777 <p>'<tt>ty</tt>': the type of the call instruction itself which is also
3778 the type of the return value. Functions that return no value are marked
3779 <tt><a href="#t_void">void</a></tt>.</p>
3782 <p>'<tt>fnty</tt>': shall be the signature of the pointer to function
3783 value being invoked. The argument types must match the types implied by
3784 this signature. This type can be omitted if the function is not varargs
3785 and if the function type does not return a pointer to a function.</p>
3788 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
3789 be invoked. In most cases, this is a direct function invocation, but
3790 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
3791 to function value.</p>
3794 <p>'<tt>function args</tt>': argument list whose types match the
3795 function signature argument types. All arguments must be of
3796 <a href="#t_firstclass">first class</a> type. If the function signature
3797 indicates the function accepts a variable number of arguments, the extra
3798 arguments can be specified.</p>
3804 <p>The '<tt>call</tt>' instruction is used to cause control flow to
3805 transfer to a specified function, with its incoming arguments bound to
3806 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
3807 instruction in the called function, control flow continues with the
3808 instruction after the function call, and the return value of the
3809 function is bound to the result argument. This is a simpler case of
3810 the <a href="#i_invoke">invoke</a> instruction.</p>
3815 %retval = call i32 @test(i32 %argc)
3816 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42);
3817 %X = tail call i32 @foo()
3818 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo()
3819 %Z = call void %foo(i8 97 signext)
3824 <!-- _______________________________________________________________________ -->
3825 <div class="doc_subsubsection">
3826 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
3829 <div class="doc_text">
3834 <resultval> = va_arg <va_list*> <arglist>, <argty>
3839 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
3840 the "variable argument" area of a function call. It is used to implement the
3841 <tt>va_arg</tt> macro in C.</p>
3845 <p>This instruction takes a <tt>va_list*</tt> value and the type of
3846 the argument. It returns a value of the specified argument type and
3847 increments the <tt>va_list</tt> to point to the next argument. The
3848 actual type of <tt>va_list</tt> is target specific.</p>
3852 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
3853 type from the specified <tt>va_list</tt> and causes the
3854 <tt>va_list</tt> to point to the next argument. For more information,
3855 see the variable argument handling <a href="#int_varargs">Intrinsic
3858 <p>It is legal for this instruction to be called in a function which does not
3859 take a variable number of arguments, for example, the <tt>vfprintf</tt>
3862 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
3863 href="#intrinsics">intrinsic function</a> because it takes a type as an
3868 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
3872 <!-- *********************************************************************** -->
3873 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
3874 <!-- *********************************************************************** -->
3876 <div class="doc_text">
3878 <p>LLVM supports the notion of an "intrinsic function". These functions have
3879 well known names and semantics and are required to follow certain restrictions.
3880 Overall, these intrinsics represent an extension mechanism for the LLVM
3881 language that does not require changing all of the transformations in LLVM when
3882 adding to the language (or the bitcode reader/writer, the parser, etc...).</p>
3884 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
3885 prefix is reserved in LLVM for intrinsic names; thus, function names may not
3886 begin with this prefix. Intrinsic functions must always be external functions:
3887 you cannot define the body of intrinsic functions. Intrinsic functions may
3888 only be used in call or invoke instructions: it is illegal to take the address
3889 of an intrinsic function. Additionally, because intrinsic functions are part
3890 of the LLVM language, it is required if any are added that they be documented
3893 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
3894 a family of functions that perform the same operation but on different data
3895 types. Because LLVM can represent over 8 million different integer types,
3896 overloading is used commonly to allow an intrinsic function to operate on any
3897 integer type. One or more of the argument types or the result type can be
3898 overloaded to accept any integer type. Argument types may also be defined as
3899 exactly matching a previous argument's type or the result type. This allows an
3900 intrinsic function which accepts multiple arguments, but needs all of them to
3901 be of the same type, to only be overloaded with respect to a single argument or
3904 <p>Overloaded intrinsics will have the names of its overloaded argument types
3905 encoded into its function name, each preceded by a period. Only those types
3906 which are overloaded result in a name suffix. Arguments whose type is matched
3907 against another type do not. For example, the <tt>llvm.ctpop</tt> function can
3908 take an integer of any width and returns an integer of exactly the same integer
3909 width. This leads to a family of functions such as
3910 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29 %val)</tt>.
3911 Only one type, the return type, is overloaded, and only one type suffix is
3912 required. Because the argument's type is matched against the return type, it
3913 does not require its own name suffix.</p>
3915 <p>To learn how to add an intrinsic function, please see the
3916 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.
3921 <!-- ======================================================================= -->
3922 <div class="doc_subsection">
3923 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
3926 <div class="doc_text">
3928 <p>Variable argument support is defined in LLVM with the <a
3929 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
3930 intrinsic functions. These functions are related to the similarly
3931 named macros defined in the <tt><stdarg.h></tt> header file.</p>
3933 <p>All of these functions operate on arguments that use a
3934 target-specific value type "<tt>va_list</tt>". The LLVM assembly
3935 language reference manual does not define what this type is, so all
3936 transformations should be prepared to handle these functions regardless of
3939 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
3940 instruction and the variable argument handling intrinsic functions are
3943 <div class="doc_code">
3945 define i32 @test(i32 %X, ...) {
3946 ; Initialize variable argument processing
3948 %ap2 = bitcast i8** %ap to i8*
3949 call void @llvm.va_start(i8* %ap2)
3951 ; Read a single integer argument
3952 %tmp = va_arg i8** %ap, i32
3954 ; Demonstrate usage of llvm.va_copy and llvm.va_end
3956 %aq2 = bitcast i8** %aq to i8*
3957 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
3958 call void @llvm.va_end(i8* %aq2)
3960 ; Stop processing of arguments.
3961 call void @llvm.va_end(i8* %ap2)
3965 declare void @llvm.va_start(i8*)
3966 declare void @llvm.va_copy(i8*, i8*)
3967 declare void @llvm.va_end(i8*)
3973 <!-- _______________________________________________________________________ -->
3974 <div class="doc_subsubsection">
3975 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
3979 <div class="doc_text">
3981 <pre> declare void %llvm.va_start(i8* <arglist>)<br></pre>
3983 <P>The '<tt>llvm.va_start</tt>' intrinsic initializes
3984 <tt>*<arglist></tt> for subsequent use by <tt><a
3985 href="#i_va_arg">va_arg</a></tt>.</p>
3989 <P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
3993 <P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
3994 macro available in C. In a target-dependent way, it initializes the
3995 <tt>va_list</tt> element to which the argument points, so that the next call to
3996 <tt>va_arg</tt> will produce the first variable argument passed to the function.
3997 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
3998 last argument of the function as the compiler can figure that out.</p>
4002 <!-- _______________________________________________________________________ -->
4003 <div class="doc_subsubsection">
4004 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
4007 <div class="doc_text">
4009 <pre> declare void @llvm.va_end(i8* <arglist>)<br></pre>
4012 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
4013 which has been initialized previously with <tt><a href="#int_va_start">llvm.va_start</a></tt>
4014 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
4018 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
4022 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
4023 macro available in C. In a target-dependent way, it destroys the
4024 <tt>va_list</tt> element to which the argument points. Calls to <a
4025 href="#int_va_start"><tt>llvm.va_start</tt></a> and <a href="#int_va_copy">
4026 <tt>llvm.va_copy</tt></a> must be matched exactly with calls to
4027 <tt>llvm.va_end</tt>.</p>
4031 <!-- _______________________________________________________________________ -->
4032 <div class="doc_subsubsection">
4033 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
4036 <div class="doc_text">
4041 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
4046 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
4047 from the source argument list to the destination argument list.</p>
4051 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
4052 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
4057 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
4058 macro available in C. In a target-dependent way, it copies the source
4059 <tt>va_list</tt> element into the destination <tt>va_list</tt> element. This
4060 intrinsic is necessary because the <tt><a href="#int_va_start">
4061 llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
4062 example, memory allocation.</p>
4066 <!-- ======================================================================= -->
4067 <div class="doc_subsection">
4068 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
4071 <div class="doc_text">
4074 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
4075 Collection</a> requires the implementation and generation of these intrinsics.
4076 These intrinsics allow identification of <a href="#int_gcroot">GC roots on the
4077 stack</a>, as well as garbage collector implementations that require <a
4078 href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a> barriers.
4079 Front-ends for type-safe garbage collected languages should generate these
4080 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
4081 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
4084 <p>The garbage collection intrinsics only operate on objects in the generic
4085 address space (address space zero).</p>
4089 <!-- _______________________________________________________________________ -->
4090 <div class="doc_subsubsection">
4091 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
4094 <div class="doc_text">
4099 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
4104 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
4105 the code generator, and allows some metadata to be associated with it.</p>
4109 <p>The first argument specifies the address of a stack object that contains the
4110 root pointer. The second pointer (which must be either a constant or a global
4111 value address) contains the meta-data to be associated with the root.</p>
4115 <p>At runtime, a call to this intrinsics stores a null pointer into the "ptrloc"
4116 location. At compile-time, the code generator generates information to allow
4117 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
4118 intrinsic may only be used in a function which <a href="#gc">specifies a GC
4124 <!-- _______________________________________________________________________ -->
4125 <div class="doc_subsubsection">
4126 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
4129 <div class="doc_text">
4134 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
4139 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
4140 locations, allowing garbage collector implementations that require read
4145 <p>The second argument is the address to read from, which should be an address
4146 allocated from the garbage collector. The first object is a pointer to the
4147 start of the referenced object, if needed by the language runtime (otherwise
4152 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
4153 instruction, but may be replaced with substantially more complex code by the
4154 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
4155 may only be used in a function which <a href="#gc">specifies a GC
4161 <!-- _______________________________________________________________________ -->
4162 <div class="doc_subsubsection">
4163 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
4166 <div class="doc_text">
4171 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
4176 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
4177 locations, allowing garbage collector implementations that require write
4178 barriers (such as generational or reference counting collectors).</p>
4182 <p>The first argument is the reference to store, the second is the start of the
4183 object to store it to, and the third is the address of the field of Obj to
4184 store to. If the runtime does not require a pointer to the object, Obj may be
4189 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
4190 instruction, but may be replaced with substantially more complex code by the
4191 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
4192 may only be used in a function which <a href="#gc">specifies a GC
4199 <!-- ======================================================================= -->
4200 <div class="doc_subsection">
4201 <a name="int_codegen">Code Generator Intrinsics</a>
4204 <div class="doc_text">
4206 These intrinsics are provided by LLVM to expose special features that may only
4207 be implemented with code generator support.
4212 <!-- _______________________________________________________________________ -->
4213 <div class="doc_subsubsection">
4214 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
4217 <div class="doc_text">
4221 declare i8 *@llvm.returnaddress(i32 <level>)
4227 The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
4228 target-specific value indicating the return address of the current function
4229 or one of its callers.
4235 The argument to this intrinsic indicates which function to return the address
4236 for. Zero indicates the calling function, one indicates its caller, etc. The
4237 argument is <b>required</b> to be a constant integer value.
4243 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
4244 the return address of the specified call frame, or zero if it cannot be
4245 identified. The value returned by this intrinsic is likely to be incorrect or 0
4246 for arguments other than zero, so it should only be used for debugging purposes.
4250 Note that calling this intrinsic does not prevent function inlining or other
4251 aggressive transformations, so the value returned may not be that of the obvious
4252 source-language caller.
4257 <!-- _______________________________________________________________________ -->
4258 <div class="doc_subsubsection">
4259 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
4262 <div class="doc_text">
4266 declare i8 *@llvm.frameaddress(i32 <level>)
4272 The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
4273 target-specific frame pointer value for the specified stack frame.
4279 The argument to this intrinsic indicates which function to return the frame
4280 pointer for. Zero indicates the calling function, one indicates its caller,
4281 etc. The argument is <b>required</b> to be a constant integer value.
4287 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
4288 the frame address of the specified call frame, or zero if it cannot be
4289 identified. The value returned by this intrinsic is likely to be incorrect or 0
4290 for arguments other than zero, so it should only be used for debugging purposes.
4294 Note that calling this intrinsic does not prevent function inlining or other
4295 aggressive transformations, so the value returned may not be that of the obvious
4296 source-language caller.
4300 <!-- _______________________________________________________________________ -->
4301 <div class="doc_subsubsection">
4302 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
4305 <div class="doc_text">
4309 declare i8 *@llvm.stacksave()
4315 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
4316 the function stack, for use with <a href="#int_stackrestore">
4317 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
4318 features like scoped automatic variable sized arrays in C99.
4324 This intrinsic returns a opaque pointer value that can be passed to <a
4325 href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
4326 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
4327 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
4328 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
4329 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
4330 that were allocated after the <tt>llvm.stacksave</tt> was executed.
4335 <!-- _______________________________________________________________________ -->
4336 <div class="doc_subsubsection">
4337 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
4340 <div class="doc_text">
4344 declare void @llvm.stackrestore(i8 * %ptr)
4350 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
4351 the function stack to the state it was in when the corresponding <a
4352 href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
4353 useful for implementing language features like scoped automatic variable sized
4360 See the description for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.
4366 <!-- _______________________________________________________________________ -->
4367 <div class="doc_subsubsection">
4368 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
4371 <div class="doc_text">
4375 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>)
4382 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
4383 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
4385 effect on the behavior of the program but can change its performance
4392 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
4393 determining if the fetch should be for a read (0) or write (1), and
4394 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
4395 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
4396 <tt>locality</tt> arguments must be constant integers.
4402 This intrinsic does not modify the behavior of the program. In particular,
4403 prefetches cannot trap and do not produce a value. On targets that support this
4404 intrinsic, the prefetch can provide hints to the processor cache for better
4410 <!-- _______________________________________________________________________ -->
4411 <div class="doc_subsubsection">
4412 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
4415 <div class="doc_text">
4419 declare void @llvm.pcmarker(i32 <id>)
4426 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
4428 code to simulators and other tools. The method is target specific, but it is
4429 expected that the marker will use exported symbols to transmit the PC of the marker.
4430 The marker makes no guarantees that it will remain with any specific instruction
4431 after optimizations. It is possible that the presence of a marker will inhibit
4432 optimizations. The intended use is to be inserted after optimizations to allow
4433 correlations of simulation runs.
4439 <tt>id</tt> is a numerical id identifying the marker.
4445 This intrinsic does not modify the behavior of the program. Backends that do not
4446 support this intrinisic may ignore it.
4451 <!-- _______________________________________________________________________ -->
4452 <div class="doc_subsubsection">
4453 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
4456 <div class="doc_text">
4460 declare i64 @llvm.readcyclecounter( )
4467 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
4468 counter register (or similar low latency, high accuracy clocks) on those targets
4469 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
4470 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
4471 should only be used for small timings.
4477 When directly supported, reading the cycle counter should not modify any memory.
4478 Implementations are allowed to either return a application specific value or a
4479 system wide value. On backends without support, this is lowered to a constant 0.
4484 <!-- ======================================================================= -->
4485 <div class="doc_subsection">
4486 <a name="int_libc">Standard C Library Intrinsics</a>
4489 <div class="doc_text">
4491 LLVM provides intrinsics for a few important standard C library functions.
4492 These intrinsics allow source-language front-ends to pass information about the
4493 alignment of the pointer arguments to the code generator, providing opportunity
4494 for more efficient code generation.
4499 <!-- _______________________________________________________________________ -->
4500 <div class="doc_subsubsection">
4501 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
4504 <div class="doc_text">
4508 declare void @llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
4509 i32 <len>, i32 <align>)
4510 declare void @llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
4511 i64 <len>, i32 <align>)
4517 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4518 location to the destination location.
4522 Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
4523 intrinsics do not return a value, and takes an extra alignment argument.
4529 The first argument is a pointer to the destination, the second is a pointer to
4530 the source. The third argument is an integer argument
4531 specifying the number of bytes to copy, and the fourth argument is the alignment
4532 of the source and destination locations.
4536 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4537 the caller guarantees that both the source and destination pointers are aligned
4544 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4545 location to the destination location, which are not allowed to overlap. It
4546 copies "len" bytes of memory over. If the argument is known to be aligned to
4547 some boundary, this can be specified as the fourth argument, otherwise it should
4553 <!-- _______________________________________________________________________ -->
4554 <div class="doc_subsubsection">
4555 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
4558 <div class="doc_text">
4562 declare void @llvm.memmove.i32(i8 * <dest>, i8 * <src>,
4563 i32 <len>, i32 <align>)
4564 declare void @llvm.memmove.i64(i8 * <dest>, i8 * <src>,
4565 i64 <len>, i32 <align>)
4571 The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
4572 location to the destination location. It is similar to the
4573 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to overlap.
4577 Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
4578 intrinsics do not return a value, and takes an extra alignment argument.
4584 The first argument is a pointer to the destination, the second is a pointer to
4585 the source. The third argument is an integer argument
4586 specifying the number of bytes to copy, and the fourth argument is the alignment
4587 of the source and destination locations.
4591 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4592 the caller guarantees that the source and destination pointers are aligned to
4599 The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
4600 location to the destination location, which may overlap. It
4601 copies "len" bytes of memory over. If the argument is known to be aligned to
4602 some boundary, this can be specified as the fourth argument, otherwise it should
4608 <!-- _______________________________________________________________________ -->
4609 <div class="doc_subsubsection">
4610 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
4613 <div class="doc_text">
4617 declare void @llvm.memset.i32(i8 * <dest>, i8 <val>,
4618 i32 <len>, i32 <align>)
4619 declare void @llvm.memset.i64(i8 * <dest>, i8 <val>,
4620 i64 <len>, i32 <align>)
4626 The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
4631 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
4632 does not return a value, and takes an extra alignment argument.
4638 The first argument is a pointer to the destination to fill, the second is the
4639 byte value to fill it with, the third argument is an integer
4640 argument specifying the number of bytes to fill, and the fourth argument is the
4641 known alignment of destination location.
4645 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4646 the caller guarantees that the destination pointer is aligned to that boundary.
4652 The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
4654 destination location. If the argument is known to be aligned to some boundary,
4655 this can be specified as the fourth argument, otherwise it should be set to 0 or
4661 <!-- _______________________________________________________________________ -->
4662 <div class="doc_subsubsection">
4663 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
4666 <div class="doc_text">
4669 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
4670 floating point or vector of floating point type. Not all targets support all
4673 declare float @llvm.sqrt.f32(float %Val)
4674 declare double @llvm.sqrt.f64(double %Val)
4675 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
4676 declare fp128 @llvm.sqrt.f128(fp128 %Val)
4677 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
4683 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
4684 returning the same value as the libm '<tt>sqrt</tt>' functions would. Unlike
4685 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
4686 negative numbers other than -0.0 (which allows for better optimization, because
4687 there is no need to worry about errno being set). <tt>llvm.sqrt(-0.0)</tt> is
4688 defined to return -0.0 like IEEE sqrt.
4694 The argument and return value are floating point numbers of the same type.
4700 This function returns the sqrt of the specified operand if it is a nonnegative
4701 floating point number.
4705 <!-- _______________________________________________________________________ -->
4706 <div class="doc_subsubsection">
4707 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
4710 <div class="doc_text">
4713 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
4714 floating point or vector of floating point type. Not all targets support all
4717 declare float @llvm.powi.f32(float %Val, i32 %power)
4718 declare double @llvm.powi.f64(double %Val, i32 %power)
4719 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
4720 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
4721 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
4727 The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
4728 specified (positive or negative) power. The order of evaluation of
4729 multiplications is not defined. When a vector of floating point type is
4730 used, the second argument remains a scalar integer value.
4736 The second argument is an integer power, and the first is a value to raise to
4743 This function returns the first value raised to the second power with an
4744 unspecified sequence of rounding operations.</p>
4747 <!-- _______________________________________________________________________ -->
4748 <div class="doc_subsubsection">
4749 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
4752 <div class="doc_text">
4755 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
4756 floating point or vector of floating point type. Not all targets support all
4759 declare float @llvm.sin.f32(float %Val)
4760 declare double @llvm.sin.f64(double %Val)
4761 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
4762 declare fp128 @llvm.sin.f128(fp128 %Val)
4763 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
4769 The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.
4775 The argument and return value are floating point numbers of the same type.
4781 This function returns the sine of the specified operand, returning the
4782 same values as the libm <tt>sin</tt> functions would, and handles error
4783 conditions in the same way.</p>
4786 <!-- _______________________________________________________________________ -->
4787 <div class="doc_subsubsection">
4788 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
4791 <div class="doc_text">
4794 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
4795 floating point or vector of floating point type. Not all targets support all
4798 declare float @llvm.cos.f32(float %Val)
4799 declare double @llvm.cos.f64(double %Val)
4800 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
4801 declare fp128 @llvm.cos.f128(fp128 %Val)
4802 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
4808 The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.
4814 The argument and return value are floating point numbers of the same type.
4820 This function returns the cosine of the specified operand, returning the
4821 same values as the libm <tt>cos</tt> functions would, and handles error
4822 conditions in the same way.</p>
4825 <!-- _______________________________________________________________________ -->
4826 <div class="doc_subsubsection">
4827 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
4830 <div class="doc_text">
4833 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
4834 floating point or vector of floating point type. Not all targets support all
4837 declare float @llvm.pow.f32(float %Val, float %Power)
4838 declare double @llvm.pow.f64(double %Val, double %Power)
4839 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
4840 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
4841 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
4847 The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
4848 specified (positive or negative) power.
4854 The second argument is a floating point power, and the first is a value to
4855 raise to that power.
4861 This function returns the first value raised to the second power,
4863 same values as the libm <tt>pow</tt> functions would, and handles error
4864 conditions in the same way.</p>
4868 <!-- ======================================================================= -->
4869 <div class="doc_subsection">
4870 <a name="int_manip">Bit Manipulation Intrinsics</a>
4873 <div class="doc_text">
4875 LLVM provides intrinsics for a few important bit manipulation operations.
4876 These allow efficient code generation for some algorithms.
4881 <!-- _______________________________________________________________________ -->
4882 <div class="doc_subsubsection">
4883 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
4886 <div class="doc_text">
4889 <p>This is an overloaded intrinsic function. You can use bswap on any integer
4890 type that is an even number of bytes (i.e. BitWidth % 16 == 0).
4892 declare i16 @llvm.bswap.i16(i16 <id>)
4893 declare i32 @llvm.bswap.i32(i32 <id>)
4894 declare i64 @llvm.bswap.i64(i64 <id>)
4900 The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
4901 values with an even number of bytes (positive multiple of 16 bits). These are
4902 useful for performing operations on data that is not in the target's native
4909 The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
4910 and low byte of the input i16 swapped. Similarly, the <tt>llvm.bswap.i32</tt>
4911 intrinsic returns an i32 value that has the four bytes of the input i32
4912 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the returned
4913 i32 will have its bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i48</tt>,
4914 <tt>llvm.bswap.i64</tt> and other intrinsics extend this concept to
4915 additional even-byte lengths (6 bytes, 8 bytes and more, respectively).
4920 <!-- _______________________________________________________________________ -->
4921 <div class="doc_subsubsection">
4922 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
4925 <div class="doc_text">
4928 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
4929 width. Not all targets support all bit widths however.
4931 declare i8 @llvm.ctpop.i8 (i8 <src>)
4932 declare i16 @llvm.ctpop.i16(i16 <src>)
4933 declare i32 @llvm.ctpop.i32(i32 <src>)
4934 declare i64 @llvm.ctpop.i64(i64 <src>)
4935 declare i256 @llvm.ctpop.i256(i256 <src>)
4941 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
4948 The only argument is the value to be counted. The argument may be of any
4949 integer type. The return type must match the argument type.
4955 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
4959 <!-- _______________________________________________________________________ -->
4960 <div class="doc_subsubsection">
4961 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
4964 <div class="doc_text">
4967 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
4968 integer bit width. Not all targets support all bit widths however.
4970 declare i8 @llvm.ctlz.i8 (i8 <src>)
4971 declare i16 @llvm.ctlz.i16(i16 <src>)
4972 declare i32 @llvm.ctlz.i32(i32 <src>)
4973 declare i64 @llvm.ctlz.i64(i64 <src>)
4974 declare i256 @llvm.ctlz.i256(i256 <src>)
4980 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
4981 leading zeros in a variable.
4987 The only argument is the value to be counted. The argument may be of any
4988 integer type. The return type must match the argument type.
4994 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
4995 in a variable. If the src == 0 then the result is the size in bits of the type
4996 of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.
5002 <!-- _______________________________________________________________________ -->
5003 <div class="doc_subsubsection">
5004 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
5007 <div class="doc_text">
5010 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
5011 integer bit width. Not all targets support all bit widths however.
5013 declare i8 @llvm.cttz.i8 (i8 <src>)
5014 declare i16 @llvm.cttz.i16(i16 <src>)
5015 declare i32 @llvm.cttz.i32(i32 <src>)
5016 declare i64 @llvm.cttz.i64(i64 <src>)
5017 declare i256 @llvm.cttz.i256(i256 <src>)
5023 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
5030 The only argument is the value to be counted. The argument may be of any
5031 integer type. The return type must match the argument type.
5037 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
5038 in a variable. If the src == 0 then the result is the size in bits of the type
5039 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
5043 <!-- _______________________________________________________________________ -->
5044 <div class="doc_subsubsection">
5045 <a name="int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic</a>
5048 <div class="doc_text">
5051 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.select</tt>
5052 on any integer bit width.
5054 declare i17 @llvm.part.select.i17 (i17 %val, i32 %loBit, i32 %hiBit)
5055 declare i29 @llvm.part.select.i29 (i29 %val, i32 %loBit, i32 %hiBit)
5059 <p>The '<tt>llvm.part.select</tt>' family of intrinsic functions selects a
5060 range of bits from an integer value and returns them in the same bit width as
5061 the original value.</p>
5064 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5065 any bit width but they must have the same bit width. The second and third
5066 arguments must be <tt>i32</tt> type since they specify only a bit index.</p>
5069 <p>The operation of the '<tt>llvm.part.select</tt>' intrinsic has two modes
5070 of operation: forwards and reverse. If <tt>%loBit</tt> is greater than
5071 <tt>%hiBits</tt> then the intrinsic operates in reverse mode. Otherwise it
5072 operates in forward mode.</p>
5073 <p>In forward mode, this intrinsic is the equivalent of shifting <tt>%val</tt>
5074 right by <tt>%loBit</tt> bits and then ANDing it with a mask with
5075 only the <tt>%hiBit - %loBit</tt> bits set, as follows:</p>
5077 <li>The <tt>%val</tt> is shifted right (LSHR) by the number of bits specified
5078 by <tt>%loBits</tt>. This normalizes the value to the low order bits.</li>
5079 <li>The <tt>%loBits</tt> value is subtracted from the <tt>%hiBits</tt> value
5080 to determine the number of bits to retain.</li>
5081 <li>A mask of the retained bits is created by shifting a -1 value.</li>
5082 <li>The mask is ANDed with <tt>%val</tt> to produce the result.
5084 <p>In reverse mode, a similar computation is made except that the bits are
5085 returned in the reverse order. So, for example, if <tt>X</tt> has the value
5086 <tt>i16 0x0ACF (101011001111)</tt> and we apply
5087 <tt>part.select(i16 X, 8, 3)</tt> to it, we get back the value
5088 <tt>i16 0x0026 (000000100110)</tt>.</p>
5091 <div class="doc_subsubsection">
5092 <a name="int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic</a>
5095 <div class="doc_text">
5098 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.set</tt>
5099 on any integer bit width.
5101 declare i17 @llvm.part.set.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
5102 declare i29 @llvm.part.set.i29.i9 (i29 %val, i9 %repl, i32 %lo, i32 %hi)
5106 <p>The '<tt>llvm.part.set</tt>' family of intrinsic functions replaces a range
5107 of bits in an integer value with another integer value. It returns the integer
5108 with the replaced bits.</p>
5111 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5112 any bit width but they must have the same bit width. <tt>%val</tt> is the value
5113 whose bits will be replaced. The second argument, <tt>%repl</tt> may be an
5114 integer of any bit width. The third and fourth arguments must be <tt>i32</tt>
5115 type since they specify only a bit index.</p>
5118 <p>The operation of the '<tt>llvm.part.set</tt>' intrinsic has two modes
5119 of operation: forwards and reverse. If <tt>%lo</tt> is greater than
5120 <tt>%hi</tt> then the intrinsic operates in reverse mode. Otherwise it
5121 operates in forward mode.</p>
5122 <p>For both modes, the <tt>%repl</tt> value is prepared for use by either
5123 truncating it down to the size of the replacement area or zero extending it
5124 up to that size.</p>
5125 <p>In forward mode, the bits between <tt>%lo</tt> and <tt>%hi</tt> (inclusive)
5126 are replaced with corresponding bits from <tt>%repl</tt>. That is the 0th bit
5127 in <tt>%repl</tt> replaces the <tt>%lo</tt>th bit in <tt>%val</tt> and etc. up
5128 to the <tt>%hi</tt>th bit.
5129 <p>In reverse mode, a similar computation is made except that the bits are
5130 reversed. That is, the <tt>0</tt>th bit in <tt>%repl</tt> replaces the
5131 <tt>%hi</tt> bit in <tt>%val</tt> and etc. down to the <tt>%lo</tt>th bit.
5134 llvm.part.set(0xFFFF, 0, 4, 7) -> 0xFF0F
5135 llvm.part.set(0xFFFF, 0, 7, 4) -> 0xFF0F
5136 llvm.part.set(0xFFFF, 1, 7, 4) -> 0xFF8F
5137 llvm.part.set(0xFFFF, F, 8, 3) -> 0xFFE7
5138 llvm.part.set(0xFFFF, 0, 3, 8) -> 0xFE07
5142 <!-- ======================================================================= -->
5143 <div class="doc_subsection">
5144 <a name="int_debugger">Debugger Intrinsics</a>
5147 <div class="doc_text">
5149 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
5150 are described in the <a
5151 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
5152 Debugging</a> document.
5157 <!-- ======================================================================= -->
5158 <div class="doc_subsection">
5159 <a name="int_eh">Exception Handling Intrinsics</a>
5162 <div class="doc_text">
5163 <p> The LLVM exception handling intrinsics (which all start with
5164 <tt>llvm.eh.</tt> prefix), are described in the <a
5165 href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
5166 Handling</a> document. </p>
5169 <!-- ======================================================================= -->
5170 <div class="doc_subsection">
5171 <a name="int_trampoline">Trampoline Intrinsic</a>
5174 <div class="doc_text">
5176 This intrinsic makes it possible to excise one parameter, marked with
5177 the <tt>nest</tt> attribute, from a function. The result is a callable
5178 function pointer lacking the nest parameter - the caller does not need
5179 to provide a value for it. Instead, the value to use is stored in
5180 advance in a "trampoline", a block of memory usually allocated
5181 on the stack, which also contains code to splice the nest value into the
5182 argument list. This is used to implement the GCC nested function address
5186 For example, if the function is
5187 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
5188 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as follows:</p>
5190 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
5191 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
5192 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
5193 %fp = bitcast i8* %p to i32 (i32, i32)*
5195 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
5196 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
5199 <!-- _______________________________________________________________________ -->
5200 <div class="doc_subsubsection">
5201 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
5203 <div class="doc_text">
5206 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
5210 This fills the memory pointed to by <tt>tramp</tt> with code
5211 and returns a function pointer suitable for executing it.
5215 The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
5216 pointers. The <tt>tramp</tt> argument must point to a sufficiently large
5217 and sufficiently aligned block of memory; this memory is written to by the
5218 intrinsic. Note that the size and the alignment are target-specific - LLVM
5219 currently provides no portable way of determining them, so a front-end that
5220 generates this intrinsic needs to have some target-specific knowledge.
5221 The <tt>func</tt> argument must hold a function bitcast to an <tt>i8*</tt>.
5225 The block of memory pointed to by <tt>tramp</tt> is filled with target
5226 dependent code, turning it into a function. A pointer to this function is
5227 returned, but needs to be bitcast to an
5228 <a href="#int_trampoline">appropriate function pointer type</a>
5229 before being called. The new function's signature is the same as that of
5230 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
5231 removed. At most one such <tt>nest</tt> argument is allowed, and it must be
5232 of pointer type. Calling the new function is equivalent to calling
5233 <tt>func</tt> with the same argument list, but with <tt>nval</tt> used for the
5234 missing <tt>nest</tt> argument. If, after calling
5235 <tt>llvm.init.trampoline</tt>, the memory pointed to by <tt>tramp</tt> is
5236 modified, then the effect of any later call to the returned function pointer is
5241 <!-- ======================================================================= -->
5242 <div class="doc_subsection">
5243 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
5246 <div class="doc_text">
5248 These intrinsic functions expand the "universal IR" of LLVM to represent
5249 hardware constructs for atomic operations and memory synchronization. This
5250 provides an interface to the hardware, not an interface to the programmer. It
5251 is aimed at a low enough level to allow any programming models or APIs which
5252 need atomic behaviors to map cleanly onto it. It is also modeled primarily on
5253 hardware behavior. Just as hardware provides a "universal IR" for source
5254 languages, it also provides a starting point for developing a "universal"
5255 atomic operation and synchronization IR.
5258 These do <em>not</em> form an API such as high-level threading libraries,
5259 software transaction memory systems, atomic primitives, and intrinsic
5260 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
5261 application libraries. The hardware interface provided by LLVM should allow
5262 a clean implementation of all of these APIs and parallel programming models.
5263 No one model or paradigm should be selected above others unless the hardware
5264 itself ubiquitously does so.
5269 <!-- _______________________________________________________________________ -->
5270 <div class="doc_subsubsection">
5271 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
5273 <div class="doc_text">
5276 declare void @llvm.memory.barrier( i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>,
5282 The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
5283 specific pairs of memory access types.
5287 The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
5288 The first four arguments enables a specific barrier as listed below. The fith
5289 argument specifies that the barrier applies to io or device or uncached memory.
5293 <li><tt>ll</tt>: load-load barrier</li>
5294 <li><tt>ls</tt>: load-store barrier</li>
5295 <li><tt>sl</tt>: store-load barrier</li>
5296 <li><tt>ss</tt>: store-store barrier</li>
5297 <li><tt>device</tt>: barrier applies to device and uncached memory also.
5301 This intrinsic causes the system to enforce some ordering constraints upon
5302 the loads and stores of the program. This barrier does not indicate
5303 <em>when</em> any events will occur, it only enforces an <em>order</em> in
5304 which they occur. For any of the specified pairs of load and store operations
5305 (f.ex. load-load, or store-load), all of the first operations preceding the
5306 barrier will complete before any of the second operations succeeding the
5307 barrier begin. Specifically the semantics for each pairing is as follows:
5310 <li><tt>ll</tt>: All loads before the barrier must complete before any load
5311 after the barrier begins.</li>
5313 <li><tt>ls</tt>: All loads before the barrier must complete before any
5314 store after the barrier begins.</li>
5315 <li><tt>ss</tt>: All stores before the barrier must complete before any
5316 store after the barrier begins.</li>
5317 <li><tt>sl</tt>: All stores before the barrier must complete before any
5318 load after the barrier begins.</li>
5321 These semantics are applied with a logical "and" behavior when more than one
5322 is enabled in a single memory barrier intrinsic.
5325 Backends may implement stronger barriers than those requested when they do not
5326 support as fine grained a barrier as requested. Some architectures do not
5327 need all types of barriers and on such architectures, these become noops.
5334 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
5335 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
5336 <i>; guarantee the above finishes</i>
5337 store i32 8, %ptr <i>; before this begins</i>
5342 <!-- ======================================================================= -->
5343 <div class="doc_subsection">
5344 <a name="int_general">General Intrinsics</a>
5347 <div class="doc_text">
5348 <p> This class of intrinsics is designed to be generic and has
5349 no specific purpose. </p>
5352 <!-- _______________________________________________________________________ -->
5353 <div class="doc_subsubsection">
5354 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
5357 <div class="doc_text">
5361 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
5367 The '<tt>llvm.var.annotation</tt>' intrinsic
5373 The first argument is a pointer to a value, the second is a pointer to a
5374 global string, the third is a pointer to a global string which is the source
5375 file name, and the last argument is the line number.
5381 This intrinsic allows annotation of local variables with arbitrary strings.
5382 This can be useful for special purpose optimizations that want to look for these
5383 annotations. These have no other defined use, they are ignored by code
5384 generation and optimization.
5388 <!-- _______________________________________________________________________ -->
5389 <div class="doc_subsubsection">
5390 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
5393 <div class="doc_text">
5396 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
5397 any integer bit width.
5400 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
5401 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
5402 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
5403 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
5404 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
5410 The '<tt>llvm.annotation</tt>' intrinsic.
5416 The first argument is an integer value (result of some expression),
5417 the second is a pointer to a global string, the third is a pointer to a global
5418 string which is the source file name, and the last argument is the line number.
5419 It returns the value of the first argument.
5425 This intrinsic allows annotations to be put on arbitrary expressions
5426 with arbitrary strings. This can be useful for special purpose optimizations
5427 that want to look for these annotations. These have no other defined use, they
5428 are ignored by code generation and optimization.
5431 <!-- _______________________________________________________________________ -->
5432 <div class="doc_subsubsection">
5433 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
5436 <div class="doc_text">
5440 declare void @llvm.trap()
5446 The '<tt>llvm.trap</tt>' intrinsic
5458 This intrinsics is lowered to the target dependent trap instruction. If the
5459 target does not have a trap instruction, this intrinsic will be lowered to the
5460 call of the abort() function.
5464 <!-- *********************************************************************** -->
5467 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
5468 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
5469 <a href="http://validator.w3.org/check/referer"><img
5470 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
5472 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
5473 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
5474 Last modified: $Date$