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 pointer parameter specifies the address of a
827 structure that is the return value of the function in the source program.
828 May only be applied to the first parameter.</dd>
830 <dt><tt>noalias</tt></dt>
831 <dd>This indicates that the parameter does not alias any global or any other
832 parameter. The caller is responsible for ensuring that this is the case,
833 usually by placing the value in a stack allocation.</dd>
835 <dt><tt>noreturn</tt></dt>
836 <dd>This function attribute indicates that the function never returns. This
837 indicates to LLVM that every call to this function should be treated as if
838 an <tt>unreachable</tt> instruction immediately followed the call.</dd>
840 <dt><tt>nounwind</tt></dt>
841 <dd>This function attribute indicates that the function type does not use
842 the unwind instruction and does not allow stack unwinding to propagate
845 <dt><tt>nest</tt></dt>
846 <dd>This indicates that the parameter can be excised using the
847 <a href="#int_trampoline">trampoline intrinsics</a>.</dd>
848 <dt><tt>readonly</tt></dt>
849 <dd>This function attribute indicates that the function has no side-effects
850 except for producing a return value or throwing an exception. The value
851 returned must only depend on the function arguments and/or global variables.
852 It may use values obtained by dereferencing pointers.</dd>
853 <dt><tt>readnone</tt></dt>
854 <dd>A <tt>readnone</tt> function has the same restrictions as a <tt>readonly</tt>
855 function, but in addition it is not allowed to dereference any pointer arguments
861 <!-- ======================================================================= -->
862 <div class="doc_subsection">
863 <a name="gc">Garbage Collector Names</a>
866 <div class="doc_text">
867 <p>Each function may specify a garbage collector name, which is simply a
870 <div class="doc_code"><pre
871 >define void @f() gc "name" { ...</pre></div>
873 <p>The compiler declares the supported values of <i>name</i>. Specifying a
874 collector which will cause the compiler to alter its output in order to support
875 the named garbage collection algorithm.</p>
878 <!-- ======================================================================= -->
879 <div class="doc_subsection">
880 <a name="moduleasm">Module-Level Inline Assembly</a>
883 <div class="doc_text">
885 Modules may contain "module-level inline asm" blocks, which corresponds to the
886 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
887 LLVM and treated as a single unit, but may be separated in the .ll file if
888 desired. The syntax is very simple:
891 <div class="doc_code">
893 module asm "inline asm code goes here"
894 module asm "more can go here"
898 <p>The strings can contain any character by escaping non-printable characters.
899 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
904 The inline asm code is simply printed to the machine code .s file when
905 assembly code is generated.
909 <!-- ======================================================================= -->
910 <div class="doc_subsection">
911 <a name="datalayout">Data Layout</a>
914 <div class="doc_text">
915 <p>A module may specify a target specific data layout string that specifies how
916 data is to be laid out in memory. The syntax for the data layout is simply:</p>
917 <pre> target datalayout = "<i>layout specification</i>"</pre>
918 <p>The <i>layout specification</i> consists of a list of specifications
919 separated by the minus sign character ('-'). Each specification starts with a
920 letter and may include other information after the letter to define some
921 aspect of the data layout. The specifications accepted are as follows: </p>
924 <dd>Specifies that the target lays out data in big-endian form. That is, the
925 bits with the most significance have the lowest address location.</dd>
927 <dd>Specifies that hte target lays out data in little-endian form. That is,
928 the bits with the least significance have the lowest address location.</dd>
929 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
930 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
931 <i>preferred</i> alignments. All sizes are in bits. Specifying the <i>pref</i>
932 alignment is optional. If omitted, the preceding <tt>:</tt> should be omitted
934 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
935 <dd>This specifies the alignment for an integer type of a given bit
936 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
937 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
938 <dd>This specifies the alignment for a vector type of a given bit
940 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
941 <dd>This specifies the alignment for a floating point type of a given bit
942 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
944 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
945 <dd>This specifies the alignment for an aggregate type of a given bit
948 <p>When constructing the data layout for a given target, LLVM starts with a
949 default set of specifications which are then (possibly) overriden by the
950 specifications in the <tt>datalayout</tt> keyword. The default specifications
951 are given in this list:</p>
953 <li><tt>E</tt> - big endian</li>
954 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
955 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
956 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
957 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
958 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
959 <li><tt>i64:32:64</tt> - i64 has abi alignment of 32-bits but preferred
960 alignment of 64-bits</li>
961 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
962 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
963 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
964 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
965 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
967 <p>When llvm is determining the alignment for a given type, it uses the
970 <li>If the type sought is an exact match for one of the specifications, that
971 specification is used.</li>
972 <li>If no match is found, and the type sought is an integer type, then the
973 smallest integer type that is larger than the bitwidth of the sought type is
974 used. If none of the specifications are larger than the bitwidth then the the
975 largest integer type is used. For example, given the default specifications
976 above, the i7 type will use the alignment of i8 (next largest) while both
977 i65 and i256 will use the alignment of i64 (largest specified).</li>
978 <li>If no match is found, and the type sought is a vector type, then the
979 largest vector type that is smaller than the sought vector type will be used
980 as a fall back. This happens because <128 x double> can be implemented in
981 terms of 64 <2 x double>, for example.</li>
985 <!-- *********************************************************************** -->
986 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
987 <!-- *********************************************************************** -->
989 <div class="doc_text">
991 <p>The LLVM type system is one of the most important features of the
992 intermediate representation. Being typed enables a number of
993 optimizations to be performed on the IR directly, without having to do
994 extra analyses on the side before the transformation. A strong type
995 system makes it easier to read the generated code and enables novel
996 analyses and transformations that are not feasible to perform on normal
997 three address code representations.</p>
1001 <!-- ======================================================================= -->
1002 <div class="doc_subsection"> <a name="t_classifications">Type
1003 Classifications</a> </div>
1004 <div class="doc_text">
1005 <p>The types fall into a few useful
1006 classifications:</p>
1008 <table border="1" cellspacing="0" cellpadding="4">
1010 <tr><th>Classification</th><th>Types</th></tr>
1012 <td><a href="#t_integer">integer</a></td>
1013 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1016 <td><a href="#t_floating">floating point</a></td>
1017 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1020 <td><a name="t_firstclass">first class</a></td>
1021 <td><a href="#t_integer">integer</a>,
1022 <a href="#t_floating">floating point</a>,
1023 <a href="#t_pointer">pointer</a>,
1024 <a href="#t_vector">vector</a>
1028 <td><a href="#t_primitive">primitive</a></td>
1029 <td><a href="#t_label">label</a>,
1030 <a href="#t_void">void</a>,
1031 <a href="#t_integer">integer</a>,
1032 <a href="#t_floating">floating point</a>.</td>
1035 <td><a href="#t_derived">derived</a></td>
1036 <td><a href="#t_integer">integer</a>,
1037 <a href="#t_array">array</a>,
1038 <a href="#t_function">function</a>,
1039 <a href="#t_pointer">pointer</a>,
1040 <a href="#t_struct">structure</a>,
1041 <a href="#t_pstruct">packed structure</a>,
1042 <a href="#t_vector">vector</a>,
1043 <a href="#t_opaque">opaque</a>.
1048 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
1049 most important. Values of these types are the only ones which can be
1050 produced by instructions, passed as arguments, or used as operands to
1051 instructions. This means that all structures and arrays must be
1052 manipulated either by pointer or by component.</p>
1055 <!-- ======================================================================= -->
1056 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
1058 <div class="doc_text">
1059 <p>The primitive types are the fundamental building blocks of the LLVM
1064 <!-- _______________________________________________________________________ -->
1065 <div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1067 <div class="doc_text">
1070 <tr><th>Type</th><th>Description</th></tr>
1071 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1072 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1073 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1074 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1075 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1080 <!-- _______________________________________________________________________ -->
1081 <div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1083 <div class="doc_text">
1085 <p>The void type does not represent any value and has no size.</p>
1094 <!-- _______________________________________________________________________ -->
1095 <div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1097 <div class="doc_text">
1099 <p>The label type represents code labels.</p>
1109 <!-- ======================================================================= -->
1110 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1112 <div class="doc_text">
1114 <p>The real power in LLVM comes from the derived types in the system.
1115 This is what allows a programmer to represent arrays, functions,
1116 pointers, and other useful types. Note that these derived types may be
1117 recursive: For example, it is possible to have a two dimensional array.</p>
1121 <!-- _______________________________________________________________________ -->
1122 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1124 <div class="doc_text">
1127 <p>The integer type is a very simple derived type that simply specifies an
1128 arbitrary bit width for the integer type desired. Any bit width from 1 bit to
1129 2^23-1 (about 8 million) can be specified.</p>
1137 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1141 <table class="layout">
1144 <td><tt>i1</tt></td>
1145 <td>a single-bit integer.</td>
1147 <td><tt>i32</tt></td>
1148 <td>a 32-bit integer.</td>
1150 <td><tt>i1942652</tt></td>
1151 <td>a really big integer of over 1 million bits.</td>
1157 <!-- _______________________________________________________________________ -->
1158 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1160 <div class="doc_text">
1164 <p>The array type is a very simple derived type that arranges elements
1165 sequentially in memory. The array type requires a size (number of
1166 elements) and an underlying data type.</p>
1171 [<# elements> x <elementtype>]
1174 <p>The number of elements is a constant integer value; elementtype may
1175 be any type with a size.</p>
1178 <table class="layout">
1180 <td class="left"><tt>[40 x i32]</tt></td>
1181 <td class="left">Array of 40 32-bit integer values.</td>
1184 <td class="left"><tt>[41 x i32]</tt></td>
1185 <td class="left">Array of 41 32-bit integer values.</td>
1188 <td class="left"><tt>[4 x i8]</tt></td>
1189 <td class="left">Array of 4 8-bit integer values.</td>
1192 <p>Here are some examples of multidimensional arrays:</p>
1193 <table class="layout">
1195 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1196 <td class="left">3x4 array of 32-bit integer values.</td>
1199 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1200 <td class="left">12x10 array of single precision floating point values.</td>
1203 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1204 <td class="left">2x3x4 array of 16-bit integer values.</td>
1208 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1209 length array. Normally, accesses past the end of an array are undefined in
1210 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
1211 As a special case, however, zero length arrays are recognized to be variable
1212 length. This allows implementation of 'pascal style arrays' with the LLVM
1213 type "{ i32, [0 x float]}", for example.</p>
1217 <!-- _______________________________________________________________________ -->
1218 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1219 <div class="doc_text">
1221 <p>The function type can be thought of as a function signature. It
1222 consists of a return type and a list of formal parameter types.
1223 Function types are usually used to build virtual function tables
1224 (which are structures of pointers to functions), for indirect function
1225 calls, and when defining a function.</p>
1227 The return type of a function type cannot be an aggregate type.
1230 <pre> <returntype> (<parameter list>)<br></pre>
1231 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1232 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1233 which indicates that the function takes a variable number of arguments.
1234 Variable argument functions can access their arguments with the <a
1235 href="#int_varargs">variable argument handling intrinsic</a> functions.</p>
1237 <table class="layout">
1239 <td class="left"><tt>i32 (i32)</tt></td>
1240 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1242 </tr><tr class="layout">
1243 <td class="left"><tt>float (i16 signext, i32 *) *
1245 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1246 an <tt>i16</tt> that should be sign extended and a
1247 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1250 </tr><tr class="layout">
1251 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1252 <td class="left">A vararg function that takes at least one
1253 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1254 which returns an integer. This is the signature for <tt>printf</tt> in
1261 <!-- _______________________________________________________________________ -->
1262 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1263 <div class="doc_text">
1265 <p>The structure type is used to represent a collection of data members
1266 together in memory. The packing of the field types is defined to match
1267 the ABI of the underlying processor. The elements of a structure may
1268 be any type that has a size.</p>
1269 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1270 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1271 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1274 <pre> { <type list> }<br></pre>
1276 <table class="layout">
1278 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1279 <td class="left">A triple of three <tt>i32</tt> values</td>
1280 </tr><tr class="layout">
1281 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1282 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1283 second element is a <a href="#t_pointer">pointer</a> to a
1284 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1285 an <tt>i32</tt>.</td>
1290 <!-- _______________________________________________________________________ -->
1291 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1293 <div class="doc_text">
1295 <p>The packed structure type is used to represent a collection of data members
1296 together in memory. There is no padding between fields. Further, the alignment
1297 of a packed structure is 1 byte. The elements of a packed structure may
1298 be any type that has a size.</p>
1299 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1300 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1301 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1304 <pre> < { <type list> } > <br></pre>
1306 <table class="layout">
1308 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1309 <td class="left">A triple of three <tt>i32</tt> values</td>
1310 </tr><tr class="layout">
1311 <td class="left"><tt>< { float, i32 (i32)* } ></tt></td>
1312 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1313 second element is a <a href="#t_pointer">pointer</a> to a
1314 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1315 an <tt>i32</tt>.</td>
1320 <!-- _______________________________________________________________________ -->
1321 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1322 <div class="doc_text">
1324 <p>As in many languages, the pointer type represents a pointer or
1325 reference to another object, which must live in memory. Pointer types may have
1326 an optional address space attribute defining the target-specific numbered
1327 address space where the pointed-to object resides. The default address space is
1330 <pre> <type> *<br></pre>
1332 <table class="layout">
1334 <td class="left"><tt>[4x i32]*</tt></td>
1335 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1336 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1339 <td class="left"><tt>i32 (i32 *) *</tt></td>
1340 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
1341 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1345 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1346 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1347 that resides in address space #5.</td>
1352 <!-- _______________________________________________________________________ -->
1353 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1354 <div class="doc_text">
1358 <p>A vector type is a simple derived type that represents a vector
1359 of elements. Vector types are used when multiple primitive data
1360 are operated in parallel using a single instruction (SIMD).
1361 A vector type requires a size (number of
1362 elements) and an underlying primitive data type. Vectors must have a power
1363 of two length (1, 2, 4, 8, 16 ...). Vector types are
1364 considered <a href="#t_firstclass">first class</a>.</p>
1369 < <# elements> x <elementtype> >
1372 <p>The number of elements is a constant integer value; elementtype may
1373 be any integer or floating point type.</p>
1377 <table class="layout">
1379 <td class="left"><tt><4 x i32></tt></td>
1380 <td class="left">Vector of 4 32-bit integer values.</td>
1383 <td class="left"><tt><8 x float></tt></td>
1384 <td class="left">Vector of 8 32-bit floating-point values.</td>
1387 <td class="left"><tt><2 x i64></tt></td>
1388 <td class="left">Vector of 2 64-bit integer values.</td>
1393 <!-- _______________________________________________________________________ -->
1394 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1395 <div class="doc_text">
1399 <p>Opaque types are used to represent unknown types in the system. This
1400 corresponds (for example) to the C notion of a forward declared structure type.
1401 In LLVM, opaque types can eventually be resolved to any type (not just a
1402 structure type).</p>
1412 <table class="layout">
1414 <td class="left"><tt>opaque</tt></td>
1415 <td class="left">An opaque type.</td>
1421 <!-- *********************************************************************** -->
1422 <div class="doc_section"> <a name="constants">Constants</a> </div>
1423 <!-- *********************************************************************** -->
1425 <div class="doc_text">
1427 <p>LLVM has several different basic types of constants. This section describes
1428 them all and their syntax.</p>
1432 <!-- ======================================================================= -->
1433 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1435 <div class="doc_text">
1438 <dt><b>Boolean constants</b></dt>
1440 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1441 constants of the <tt><a href="#t_primitive">i1</a></tt> type.
1444 <dt><b>Integer constants</b></dt>
1446 <dd>Standard integers (such as '4') are constants of the <a
1447 href="#t_integer">integer</a> type. Negative numbers may be used with
1451 <dt><b>Floating point constants</b></dt>
1453 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1454 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1455 notation (see below). Floating point constants must have a <a
1456 href="#t_floating">floating point</a> type. </dd>
1458 <dt><b>Null pointer constants</b></dt>
1460 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1461 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1465 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1466 of floating point constants. For example, the form '<tt>double
1467 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1468 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1469 (and the only time that they are generated by the disassembler) is when a
1470 floating point constant must be emitted but it cannot be represented as a
1471 decimal floating point number. For example, NaN's, infinities, and other
1472 special values are represented in their IEEE hexadecimal format so that
1473 assembly and disassembly do not cause any bits to change in the constants.</p>
1477 <!-- ======================================================================= -->
1478 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1481 <div class="doc_text">
1482 <p>Aggregate constants arise from aggregation of simple constants
1483 and smaller aggregate constants.</p>
1486 <dt><b>Structure constants</b></dt>
1488 <dd>Structure constants are represented with notation similar to structure
1489 type definitions (a comma separated list of elements, surrounded by braces
1490 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
1491 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>". Structure constants
1492 must have <a href="#t_struct">structure type</a>, and the number and
1493 types of elements must match those specified by the type.
1496 <dt><b>Array constants</b></dt>
1498 <dd>Array constants are represented with notation similar to array type
1499 definitions (a comma separated list of elements, surrounded by square brackets
1500 (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74 ]</tt>". Array
1501 constants must have <a href="#t_array">array type</a>, and the number and
1502 types of elements must match those specified by the type.
1505 <dt><b>Vector constants</b></dt>
1507 <dd>Vector constants are represented with notation similar to vector type
1508 definitions (a comma separated list of elements, surrounded by
1509 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32 42,
1510 i32 11, i32 74, i32 100 ></tt>". Vector constants must have <a
1511 href="#t_vector">vector type</a>, and the number and types of elements must
1512 match those specified by the type.
1515 <dt><b>Zero initialization</b></dt>
1517 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1518 value to zero of <em>any</em> type, including scalar and aggregate types.
1519 This is often used to avoid having to print large zero initializers (e.g. for
1520 large arrays) and is always exactly equivalent to using explicit zero
1527 <!-- ======================================================================= -->
1528 <div class="doc_subsection">
1529 <a name="globalconstants">Global Variable and Function Addresses</a>
1532 <div class="doc_text">
1534 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1535 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1536 constants. These constants are explicitly referenced when the <a
1537 href="#identifiers">identifier for the global</a> is used and always have <a
1538 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1541 <div class="doc_code">
1545 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1551 <!-- ======================================================================= -->
1552 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1553 <div class="doc_text">
1554 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1555 no specific value. Undefined values may be of any type and be used anywhere
1556 a constant is permitted.</p>
1558 <p>Undefined values indicate to the compiler that the program is well defined
1559 no matter what value is used, giving the compiler more freedom to optimize.
1563 <!-- ======================================================================= -->
1564 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1567 <div class="doc_text">
1569 <p>Constant expressions are used to allow expressions involving other constants
1570 to be used as constants. Constant expressions may be of any <a
1571 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1572 that does not have side effects (e.g. load and call are not supported). The
1573 following is the syntax for constant expressions:</p>
1576 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1577 <dd>Truncate a constant to another type. The bit size of CST must be larger
1578 than the bit size of TYPE. Both types must be integers.</dd>
1580 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1581 <dd>Zero extend a constant to another type. The bit size of CST must be
1582 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1584 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1585 <dd>Sign extend a constant to another type. The bit size of CST must be
1586 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1588 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1589 <dd>Truncate a floating point constant to another floating point type. The
1590 size of CST must be larger than the size of TYPE. Both types must be
1591 floating point.</dd>
1593 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1594 <dd>Floating point extend a constant to another type. The size of CST must be
1595 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1597 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
1598 <dd>Convert a floating point constant to the corresponding unsigned integer
1599 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1600 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1601 of the same number of elements. If the value won't fit in the integer type,
1602 the results are undefined.</dd>
1604 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
1605 <dd>Convert a floating point constant to the corresponding signed integer
1606 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1607 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1608 of the same number of elements. If the value won't fit in the integer type,
1609 the results are undefined.</dd>
1611 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
1612 <dd>Convert an unsigned integer constant to the corresponding floating point
1613 constant. TYPE must be a scalar or vector floating point type. CST must be of
1614 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1615 of the same number of elements. If the value won't fit in the floating point
1616 type, the results are undefined.</dd>
1618 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
1619 <dd>Convert a signed integer constant to the corresponding floating point
1620 constant. TYPE must be a scalar or vector floating point type. CST must be of
1621 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1622 of the same number of elements. If the value won't fit in the floating point
1623 type, the results are undefined.</dd>
1625 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
1626 <dd>Convert a pointer typed constant to the corresponding integer constant
1627 TYPE must be an integer type. CST must be of pointer type. The CST value is
1628 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1630 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
1631 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1632 pointer type. CST must be of integer type. The CST value is zero extended,
1633 truncated, or unchanged to make it fit in a pointer size. This one is
1634 <i>really</i> dangerous!</dd>
1636 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
1637 <dd>Convert a constant, CST, to another TYPE. The size of CST and TYPE must be
1638 identical (same number of bits). The conversion is done as if the CST value
1639 was stored to memory and read back as TYPE. In other words, no bits change
1640 with this operator, just the type. This can be used for conversion of
1641 vector types to any other type, as long as they have the same bit width. For
1642 pointers it is only valid to cast to another pointer type.
1645 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1647 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1648 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1649 instruction, the index list may have zero or more indexes, which are required
1650 to make sense for the type of "CSTPTR".</dd>
1652 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1654 <dd>Perform the <a href="#i_select">select operation</a> on
1657 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
1658 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
1660 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
1661 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
1663 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1665 <dd>Perform the <a href="#i_extractelement">extractelement
1666 operation</a> on constants.
1668 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1670 <dd>Perform the <a href="#i_insertelement">insertelement
1671 operation</a> on constants.</dd>
1674 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
1676 <dd>Perform the <a href="#i_shufflevector">shufflevector
1677 operation</a> on constants.</dd>
1679 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1681 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1682 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1683 binary</a> operations. The constraints on operands are the same as those for
1684 the corresponding instruction (e.g. no bitwise operations on floating point
1685 values are allowed).</dd>
1689 <!-- *********************************************************************** -->
1690 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1691 <!-- *********************************************************************** -->
1693 <!-- ======================================================================= -->
1694 <div class="doc_subsection">
1695 <a name="inlineasm">Inline Assembler Expressions</a>
1698 <div class="doc_text">
1701 LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
1702 Module-Level Inline Assembly</a>) through the use of a special value. This
1703 value represents the inline assembler as a string (containing the instructions
1704 to emit), a list of operand constraints (stored as a string), and a flag that
1705 indicates whether or not the inline asm expression has side effects. An example
1706 inline assembler expression is:
1709 <div class="doc_code">
1711 i32 (i32) asm "bswap $0", "=r,r"
1716 Inline assembler expressions may <b>only</b> be used as the callee operand of
1717 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
1720 <div class="doc_code">
1722 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
1727 Inline asms with side effects not visible in the constraint list must be marked
1728 as having side effects. This is done through the use of the
1729 '<tt>sideeffect</tt>' keyword, like so:
1732 <div class="doc_code">
1734 call void asm sideeffect "eieio", ""()
1738 <p>TODO: The format of the asm and constraints string still need to be
1739 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1740 need to be documented).
1745 <!-- *********************************************************************** -->
1746 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1747 <!-- *********************************************************************** -->
1749 <div class="doc_text">
1751 <p>The LLVM instruction set consists of several different
1752 classifications of instructions: <a href="#terminators">terminator
1753 instructions</a>, <a href="#binaryops">binary instructions</a>,
1754 <a href="#bitwiseops">bitwise binary instructions</a>, <a
1755 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1756 instructions</a>.</p>
1760 <!-- ======================================================================= -->
1761 <div class="doc_subsection"> <a name="terminators">Terminator
1762 Instructions</a> </div>
1764 <div class="doc_text">
1766 <p>As mentioned <a href="#functionstructure">previously</a>, every
1767 basic block in a program ends with a "Terminator" instruction, which
1768 indicates which block should be executed after the current block is
1769 finished. These terminator instructions typically yield a '<tt>void</tt>'
1770 value: they produce control flow, not values (the one exception being
1771 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1772 <p>There are six different terminator instructions: the '<a
1773 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1774 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1775 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1776 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1777 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1781 <!-- _______________________________________________________________________ -->
1782 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1783 Instruction</a> </div>
1784 <div class="doc_text">
1786 <pre> ret <type> <value> <i>; Return a value from a non-void function</i>
1787 ret void <i>; Return from void function</i>
1790 <p>The '<tt>ret</tt>' instruction is used to return control flow (and a
1791 value) from a function back to the caller.</p>
1792 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
1793 returns a value and then causes control flow, and one that just causes
1794 control flow to occur.</p>
1796 <p>The '<tt>ret</tt>' instruction may return any '<a
1797 href="#t_firstclass">first class</a>' type. Notice that a function is
1798 not <a href="#wellformed">well formed</a> if there exists a '<tt>ret</tt>'
1799 instruction inside of the function that returns a value that does not
1800 match the return type of the function.</p>
1802 <p>When the '<tt>ret</tt>' instruction is executed, control flow
1803 returns back to the calling function's context. If the caller is a "<a
1804 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1805 the instruction after the call. If the caller was an "<a
1806 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1807 at the beginning of the "normal" destination block. If the instruction
1808 returns a value, that value shall set the call or invoke instruction's
1811 <pre> ret i32 5 <i>; Return an integer value of 5</i>
1812 ret void <i>; Return from a void function</i>
1815 <!-- _______________________________________________________________________ -->
1816 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1817 <div class="doc_text">
1819 <pre> br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1822 <p>The '<tt>br</tt>' instruction is used to cause control flow to
1823 transfer to a different basic block in the current function. There are
1824 two forms of this instruction, corresponding to a conditional branch
1825 and an unconditional branch.</p>
1827 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1828 single '<tt>i1</tt>' value and two '<tt>label</tt>' values. The
1829 unconditional form of the '<tt>br</tt>' instruction takes a single
1830 '<tt>label</tt>' value as a target.</p>
1832 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
1833 argument is evaluated. If the value is <tt>true</tt>, control flows
1834 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1835 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1837 <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
1838 href="#i_ret">ret</a> i32 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> i32 0<br></pre>
1840 <!-- _______________________________________________________________________ -->
1841 <div class="doc_subsubsection">
1842 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
1845 <div class="doc_text">
1849 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
1854 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
1855 several different places. It is a generalization of the '<tt>br</tt>'
1856 instruction, allowing a branch to occur to one of many possible
1862 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
1863 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
1864 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
1865 table is not allowed to contain duplicate constant entries.</p>
1869 <p>The <tt>switch</tt> instruction specifies a table of values and
1870 destinations. When the '<tt>switch</tt>' instruction is executed, this
1871 table is searched for the given value. If the value is found, control flow is
1872 transfered to the corresponding destination; otherwise, control flow is
1873 transfered to the default destination.</p>
1875 <h5>Implementation:</h5>
1877 <p>Depending on properties of the target machine and the particular
1878 <tt>switch</tt> instruction, this instruction may be code generated in different
1879 ways. For example, it could be generated as a series of chained conditional
1880 branches or with a lookup table.</p>
1885 <i>; Emulate a conditional br instruction</i>
1886 %Val = <a href="#i_zext">zext</a> i1 %value to i32
1887 switch i32 %Val, label %truedest [i32 0, label %falsedest ]
1889 <i>; Emulate an unconditional br instruction</i>
1890 switch i32 0, label %dest [ ]
1892 <i>; Implement a jump table:</i>
1893 switch i32 %val, label %otherwise [ i32 0, label %onzero
1895 i32 2, label %ontwo ]
1899 <!-- _______________________________________________________________________ -->
1900 <div class="doc_subsubsection">
1901 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
1904 <div class="doc_text">
1909 <result> = invoke [<a href="#callingconv">cconv</a>] <ptr to function ty> %<function ptr val>(<function args>)
1910 to label <normal label> unwind label <exception label>
1915 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
1916 function, with the possibility of control flow transfer to either the
1917 '<tt>normal</tt>' label or the
1918 '<tt>exception</tt>' label. If the callee function returns with the
1919 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
1920 "normal" label. If the callee (or any indirect callees) returns with the "<a
1921 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
1922 continued at the dynamically nearest "exception" label.</p>
1926 <p>This instruction requires several arguments:</p>
1930 The optional "cconv" marker indicates which <a href="#callingconv">calling
1931 convention</a> the call should use. If none is specified, the call defaults
1932 to using C calling conventions.
1934 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
1935 function value being invoked. In most cases, this is a direct function
1936 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
1937 an arbitrary pointer to function value.
1940 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
1941 function to be invoked. </li>
1943 <li>'<tt>function args</tt>': argument list whose types match the function
1944 signature argument types. If the function signature indicates the function
1945 accepts a variable number of arguments, the extra arguments can be
1948 <li>'<tt>normal label</tt>': the label reached when the called function
1949 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
1951 <li>'<tt>exception label</tt>': the label reached when a callee returns with
1952 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
1958 <p>This instruction is designed to operate as a standard '<tt><a
1959 href="#i_call">call</a></tt>' instruction in most regards. The primary
1960 difference is that it establishes an association with a label, which is used by
1961 the runtime library to unwind the stack.</p>
1963 <p>This instruction is used in languages with destructors to ensure that proper
1964 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
1965 exception. Additionally, this is important for implementation of
1966 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
1970 %retval = invoke i32 %Test(i32 15) to label %Continue
1971 unwind label %TestCleanup <i>; {i32}:retval set</i>
1972 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Test(i32 15) to label %Continue
1973 unwind label %TestCleanup <i>; {i32}:retval set</i>
1978 <!-- _______________________________________________________________________ -->
1980 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
1981 Instruction</a> </div>
1983 <div class="doc_text">
1992 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
1993 at the first callee in the dynamic call stack which used an <a
1994 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
1995 primarily used to implement exception handling.</p>
1999 <p>The '<tt>unwind</tt>' intrinsic causes execution of the current function to
2000 immediately halt. The dynamic call stack is then searched for the first <a
2001 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
2002 execution continues at the "exceptional" destination block specified by the
2003 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
2004 dynamic call chain, undefined behavior results.</p>
2007 <!-- _______________________________________________________________________ -->
2009 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
2010 Instruction</a> </div>
2012 <div class="doc_text">
2021 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
2022 instruction is used to inform the optimizer that a particular portion of the
2023 code is not reachable. This can be used to indicate that the code after a
2024 no-return function cannot be reached, and other facts.</p>
2028 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
2033 <!-- ======================================================================= -->
2034 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
2035 <div class="doc_text">
2036 <p>Binary operators are used to do most of the computation in a
2037 program. They require two operands, execute an operation on them, and
2038 produce a single value. The operands might represent
2039 multiple data, as is the case with the <a href="#t_vector">vector</a> data type.
2040 The result value of a binary operator is not
2041 necessarily the same type as its operands.</p>
2042 <p>There are several different binary operators:</p>
2044 <!-- _______________________________________________________________________ -->
2045 <div class="doc_subsubsection"> <a name="i_add">'<tt>add</tt>'
2046 Instruction</a> </div>
2047 <div class="doc_text">
2049 <pre> <result> = add <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2052 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
2054 <p>The two arguments to the '<tt>add</tt>' instruction must be either <a
2055 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a> values.
2056 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2057 Both arguments must have identical types.</p>
2059 <p>The value produced is the integer or floating point sum of the two
2061 <p>If an integer sum has unsigned overflow, the result returned is the
2062 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2064 <p>Because LLVM integers use a two's complement representation, this
2065 instruction is appropriate for both signed and unsigned integers.</p>
2067 <pre> <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
2070 <!-- _______________________________________________________________________ -->
2071 <div class="doc_subsubsection"> <a name="i_sub">'<tt>sub</tt>'
2072 Instruction</a> </div>
2073 <div class="doc_text">
2075 <pre> <result> = sub <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2078 <p>The '<tt>sub</tt>' instruction returns the difference of its two
2080 <p>Note that the '<tt>sub</tt>' instruction is used to represent the '<tt>neg</tt>'
2081 instruction present in most other intermediate representations.</p>
2083 <p>The two arguments to the '<tt>sub</tt>' instruction must be either <a
2084 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
2086 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2087 Both arguments must have identical types.</p>
2089 <p>The value produced is the integer or floating point difference of
2090 the two operands.</p>
2091 <p>If an integer difference has unsigned overflow, the result returned is the
2092 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2094 <p>Because LLVM integers use a two's complement representation, this
2095 instruction is appropriate for both signed and unsigned integers.</p>
2098 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
2099 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
2102 <!-- _______________________________________________________________________ -->
2103 <div class="doc_subsubsection"> <a name="i_mul">'<tt>mul</tt>'
2104 Instruction</a> </div>
2105 <div class="doc_text">
2107 <pre> <result> = mul <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2110 <p>The '<tt>mul</tt>' instruction returns the product of its two
2113 <p>The two arguments to the '<tt>mul</tt>' instruction must be either <a
2114 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
2116 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2117 Both arguments must have identical types.</p>
2119 <p>The value produced is the integer or floating point product of the
2121 <p>If the result of an integer multiplication has unsigned overflow,
2122 the result returned is the mathematical result modulo
2123 2<sup>n</sup>, where n is the bit width of the result.</p>
2124 <p>Because LLVM integers use a two's complement representation, and the
2125 result is the same width as the operands, this instruction returns the
2126 correct result for both signed and unsigned integers. If a full product
2127 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands
2128 should be sign-extended or zero-extended as appropriate to the
2129 width of the full product.</p>
2131 <pre> <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
2134 <!-- _______________________________________________________________________ -->
2135 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
2137 <div class="doc_text">
2139 <pre> <result> = udiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2142 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two
2145 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
2146 <a href="#t_integer">integer</a> values. Both arguments must have identical
2147 types. This instruction can also take <a href="#t_vector">vector</a> versions
2148 of the values in which case the elements must be integers.</p>
2150 <p>The value produced is the unsigned integer quotient of the two operands.</p>
2151 <p>Note that unsigned integer division and signed integer division are distinct
2152 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
2153 <p>Division by zero leads to undefined behavior.</p>
2155 <pre> <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2158 <!-- _______________________________________________________________________ -->
2159 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
2161 <div class="doc_text">
2163 <pre> <result> = sdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2166 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
2169 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
2170 <a href="#t_integer">integer</a> values. Both arguments must have identical
2171 types. This instruction can also take <a href="#t_vector">vector</a> versions
2172 of the values in which case the elements must be integers.</p>
2174 <p>The value produced is the signed integer quotient of the two operands.</p>
2175 <p>Note that signed integer division and unsigned integer division are distinct
2176 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
2177 <p>Division by zero leads to undefined behavior. Overflow also leads to
2178 undefined behavior; this is a rare case, but can occur, for example,
2179 by doing a 32-bit division of -2147483648 by -1.</p>
2181 <pre> <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2184 <!-- _______________________________________________________________________ -->
2185 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
2186 Instruction</a> </div>
2187 <div class="doc_text">
2189 <pre> <result> = fdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2192 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
2195 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
2196 <a href="#t_floating">floating point</a> values. Both arguments must have
2197 identical types. This instruction can also take <a href="#t_vector">vector</a>
2198 versions of floating point values.</p>
2200 <p>The value produced is the floating point quotient of the two operands.</p>
2202 <pre> <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
2205 <!-- _______________________________________________________________________ -->
2206 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
2208 <div class="doc_text">
2210 <pre> <result> = urem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2213 <p>The '<tt>urem</tt>' instruction returns the remainder from the
2214 unsigned division of its two arguments.</p>
2216 <p>The two arguments to the '<tt>urem</tt>' instruction must be
2217 <a href="#t_integer">integer</a> values. Both arguments must have identical
2218 types. This instruction can also take <a href="#t_vector">vector</a> versions
2219 of the values in which case the elements must be integers.</p>
2221 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
2222 This instruction always performs an unsigned division to get the remainder,
2223 regardless of whether the arguments are unsigned or not.</p>
2224 <p>Note that unsigned integer remainder and signed integer remainder are
2225 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
2226 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
2228 <pre> <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2232 <!-- _______________________________________________________________________ -->
2233 <div class="doc_subsubsection"> <a name="i_srem">'<tt>srem</tt>'
2234 Instruction</a> </div>
2235 <div class="doc_text">
2237 <pre> <result> = srem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2240 <p>The '<tt>srem</tt>' instruction returns the remainder from the
2241 signed division of its two operands. This instruction can also take
2242 <a href="#t_vector">vector</a> versions of the values in which case
2243 the elements must be integers.</p>
2246 <p>The two arguments to the '<tt>srem</tt>' instruction must be
2247 <a href="#t_integer">integer</a> values. Both arguments must have identical
2250 <p>This instruction returns the <i>remainder</i> of a division (where the result
2251 has the same sign as the dividend, <tt>var1</tt>), not the <i>modulo</i>
2252 operator (where the result has the same sign as the divisor, <tt>var2</tt>) of
2253 a value. For more information about the difference, see <a
2254 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
2255 Math Forum</a>. For a table of how this is implemented in various languages,
2256 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
2257 Wikipedia: modulo operation</a>.</p>
2258 <p>Note that signed integer remainder and unsigned integer remainder are
2259 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
2260 <p>Taking the remainder of a division by zero leads to undefined behavior.
2261 Overflow also leads to undefined behavior; this is a rare case, but can occur,
2262 for example, by taking the remainder of a 32-bit division of -2147483648 by -1.
2263 (The remainder doesn't actually overflow, but this rule lets srem be
2264 implemented using instructions that return both the result of the division
2265 and the remainder.)</p>
2267 <pre> <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2271 <!-- _______________________________________________________________________ -->
2272 <div class="doc_subsubsection"> <a name="i_frem">'<tt>frem</tt>'
2273 Instruction</a> </div>
2274 <div class="doc_text">
2276 <pre> <result> = frem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2279 <p>The '<tt>frem</tt>' instruction returns the remainder from the
2280 division of its two operands.</p>
2282 <p>The two arguments to the '<tt>frem</tt>' instruction must be
2283 <a href="#t_floating">floating point</a> values. Both arguments must have
2284 identical types. This instruction can also take <a href="#t_vector">vector</a>
2285 versions of floating point values.</p>
2287 <p>This instruction returns the <i>remainder</i> of a division.</p>
2289 <pre> <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
2293 <!-- ======================================================================= -->
2294 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
2295 Operations</a> </div>
2296 <div class="doc_text">
2297 <p>Bitwise binary operators are used to do various forms of
2298 bit-twiddling in a program. They are generally very efficient
2299 instructions and can commonly be strength reduced from other
2300 instructions. They require two operands, execute an operation on them,
2301 and produce a single value. The resulting value of the bitwise binary
2302 operators is always the same type as its first operand.</p>
2305 <!-- _______________________________________________________________________ -->
2306 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
2307 Instruction</a> </div>
2308 <div class="doc_text">
2310 <pre> <result> = shl <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2315 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2316 the left a specified number of bits.</p>
2320 <p>Both arguments to the '<tt>shl</tt>' instruction must be the same <a
2321 href="#t_integer">integer</a> type.</p>
2325 <p>The value produced is <tt>var1</tt> * 2<sup><tt>var2</tt></sup>. If
2326 <tt>var2</tt> is (statically or dynamically) equal to or larger than the number
2327 of bits in <tt>var1</tt>, the result is undefined.</p>
2329 <h5>Example:</h5><pre>
2330 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
2331 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
2332 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
2333 <result> = shl i32 1, 32 <i>; undefined</i>
2336 <!-- _______________________________________________________________________ -->
2337 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
2338 Instruction</a> </div>
2339 <div class="doc_text">
2341 <pre> <result> = lshr <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2345 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
2346 operand shifted to the right a specified number of bits with zero fill.</p>
2349 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
2350 <a href="#t_integer">integer</a> type.</p>
2354 <p>This instruction always performs a logical shift right operation. The most
2355 significant bits of the result will be filled with zero bits after the
2356 shift. If <tt>var2</tt> is (statically or dynamically) equal to or larger than
2357 the number of bits in <tt>var1</tt>, the result is undefined.</p>
2361 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
2362 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
2363 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
2364 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
2365 <result> = lshr i32 1, 32 <i>; undefined</i>
2369 <!-- _______________________________________________________________________ -->
2370 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
2371 Instruction</a> </div>
2372 <div class="doc_text">
2375 <pre> <result> = ashr <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2379 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
2380 operand shifted to the right a specified number of bits with sign extension.</p>
2383 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
2384 <a href="#t_integer">integer</a> type.</p>
2387 <p>This instruction always performs an arithmetic shift right operation,
2388 The most significant bits of the result will be filled with the sign bit
2389 of <tt>var1</tt>. If <tt>var2</tt> is (statically or dynamically) equal to or
2390 larger than the number of bits in <tt>var1</tt>, the result is undefined.
2395 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
2396 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
2397 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
2398 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
2399 <result> = ashr i32 1, 32 <i>; undefined</i>
2403 <!-- _______________________________________________________________________ -->
2404 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
2405 Instruction</a> </div>
2406 <div class="doc_text">
2408 <pre> <result> = and <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2411 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
2412 its two operands.</p>
2414 <p>The two arguments to the '<tt>and</tt>' instruction must be <a
2415 href="#t_integer">integer</a> values. Both arguments must have
2416 identical types.</p>
2418 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
2420 <div style="align: center">
2421 <table border="1" cellspacing="0" cellpadding="4">
2452 <pre> <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
2453 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
2454 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
2457 <!-- _______________________________________________________________________ -->
2458 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
2459 <div class="doc_text">
2461 <pre> <result> = or <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2464 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
2465 or of its two operands.</p>
2467 <p>The two arguments to the '<tt>or</tt>' instruction must be <a
2468 href="#t_integer">integer</a> values. Both arguments must have
2469 identical types.</p>
2471 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
2473 <div style="align: center">
2474 <table border="1" cellspacing="0" cellpadding="4">
2505 <pre> <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
2506 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
2507 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
2510 <!-- _______________________________________________________________________ -->
2511 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
2512 Instruction</a> </div>
2513 <div class="doc_text">
2515 <pre> <result> = xor <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2518 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
2519 or of its two operands. The <tt>xor</tt> is used to implement the
2520 "one's complement" operation, which is the "~" operator in C.</p>
2522 <p>The two arguments to the '<tt>xor</tt>' instruction must be <a
2523 href="#t_integer">integer</a> values. Both arguments must have
2524 identical types.</p>
2526 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
2528 <div style="align: center">
2529 <table border="1" cellspacing="0" cellpadding="4">
2561 <pre> <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
2562 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
2563 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
2564 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
2568 <!-- ======================================================================= -->
2569 <div class="doc_subsection">
2570 <a name="vectorops">Vector Operations</a>
2573 <div class="doc_text">
2575 <p>LLVM supports several instructions to represent vector operations in a
2576 target-independent manner. These instructions cover the element-access and
2577 vector-specific operations needed to process vectors effectively. While LLVM
2578 does directly support these vector operations, many sophisticated algorithms
2579 will want to use target-specific intrinsics to take full advantage of a specific
2584 <!-- _______________________________________________________________________ -->
2585 <div class="doc_subsubsection">
2586 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2589 <div class="doc_text">
2594 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
2600 The '<tt>extractelement</tt>' instruction extracts a single scalar
2601 element from a vector at a specified index.
2608 The first operand of an '<tt>extractelement</tt>' instruction is a
2609 value of <a href="#t_vector">vector</a> type. The second operand is
2610 an index indicating the position from which to extract the element.
2611 The index may be a variable.</p>
2616 The result is a scalar of the same type as the element type of
2617 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2618 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2619 results are undefined.
2625 %result = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
2630 <!-- _______________________________________________________________________ -->
2631 <div class="doc_subsubsection">
2632 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
2635 <div class="doc_text">
2640 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
2646 The '<tt>insertelement</tt>' instruction inserts a scalar
2647 element into a vector at a specified index.
2654 The first operand of an '<tt>insertelement</tt>' instruction is a
2655 value of <a href="#t_vector">vector</a> type. The second operand is a
2656 scalar value whose type must equal the element type of the first
2657 operand. The third operand is an index indicating the position at
2658 which to insert the value. The index may be a variable.</p>
2663 The result is a vector of the same type as <tt>val</tt>. Its
2664 element values are those of <tt>val</tt> except at position
2665 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
2666 exceeds the length of <tt>val</tt>, the results are undefined.
2672 %result = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
2676 <!-- _______________________________________________________________________ -->
2677 <div class="doc_subsubsection">
2678 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
2681 <div class="doc_text">
2686 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <n x i32> <mask> <i>; yields <n x <ty>></i>
2692 The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
2693 from two input vectors, returning a vector of the same type.
2699 The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
2700 with types that match each other and types that match the result of the
2701 instruction. The third argument is a shuffle mask, which has the same number
2702 of elements as the other vector type, but whose element type is always 'i32'.
2706 The shuffle mask operand is required to be a constant vector with either
2707 constant integer or undef values.
2713 The elements of the two input vectors are numbered from left to right across
2714 both of the vectors. The shuffle mask operand specifies, for each element of
2715 the result vector, which element of the two input registers the result element
2716 gets. The element selector may be undef (meaning "don't care") and the second
2717 operand may be undef if performing a shuffle from only one vector.
2723 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
2724 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
2725 %result = shufflevector <4 x i32> %v1, <4 x i32> undef,
2726 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
2731 <!-- ======================================================================= -->
2732 <div class="doc_subsection">
2733 <a name="memoryops">Memory Access and Addressing Operations</a>
2736 <div class="doc_text">
2738 <p>A key design point of an SSA-based representation is how it
2739 represents memory. In LLVM, no memory locations are in SSA form, which
2740 makes things very simple. This section describes how to read, write,
2741 allocate, and free memory in LLVM.</p>
2745 <!-- _______________________________________________________________________ -->
2746 <div class="doc_subsubsection">
2747 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
2750 <div class="doc_text">
2755 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2760 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
2761 heap and returns a pointer to it. The object is always allocated in the generic
2762 address space (address space zero).</p>
2766 <p>The '<tt>malloc</tt>' instruction allocates
2767 <tt>sizeof(<type>)*NumElements</tt>
2768 bytes of memory from the operating system and returns a pointer of the
2769 appropriate type to the program. If "NumElements" is specified, it is the
2770 number of elements allocated, otherwise "NumElements" is defaulted to be one.
2771 If an alignment is specified, the value result of the allocation is guaranteed to
2772 be aligned to at least that boundary. If not specified, or if zero, the target can
2773 choose to align the allocation on any convenient boundary.</p>
2775 <p>'<tt>type</tt>' must be a sized type.</p>
2779 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
2780 a pointer is returned.</p>
2785 %array = malloc [4 x i8 ] <i>; yields {[%4 x i8]*}:array</i>
2787 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
2788 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
2789 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
2790 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
2791 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
2795 <!-- _______________________________________________________________________ -->
2796 <div class="doc_subsubsection">
2797 <a name="i_free">'<tt>free</tt>' Instruction</a>
2800 <div class="doc_text">
2805 free <type> <value> <i>; yields {void}</i>
2810 <p>The '<tt>free</tt>' instruction returns memory back to the unused
2811 memory heap to be reallocated in the future.</p>
2815 <p>'<tt>value</tt>' shall be a pointer value that points to a value
2816 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
2821 <p>Access to the memory pointed to by the pointer is no longer defined
2822 after this instruction executes.</p>
2827 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
2828 free [4 x i8]* %array
2832 <!-- _______________________________________________________________________ -->
2833 <div class="doc_subsubsection">
2834 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
2837 <div class="doc_text">
2842 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2847 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
2848 currently executing function, to be automatically released when this function
2849 returns to its caller. The object is always allocated in the generic address
2850 space (address space zero).</p>
2854 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
2855 bytes of memory on the runtime stack, returning a pointer of the
2856 appropriate type to the program. If "NumElements" is specified, it is the
2857 number of elements allocated, otherwise "NumElements" is defaulted to be one.
2858 If an alignment is specified, the value result of the allocation is guaranteed
2859 to be aligned to at least that boundary. If not specified, or if zero, the target
2860 can choose to align the allocation on any convenient boundary.</p>
2862 <p>'<tt>type</tt>' may be any sized type.</p>
2866 <p>Memory is allocated; a pointer is returned. '<tt>alloca</tt>'d
2867 memory is automatically released when the function returns. The '<tt>alloca</tt>'
2868 instruction is commonly used to represent automatic variables that must
2869 have an address available. When the function returns (either with the <tt><a
2870 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
2871 instructions), the memory is reclaimed.</p>
2876 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
2877 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
2878 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
2879 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
2883 <!-- _______________________________________________________________________ -->
2884 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
2885 Instruction</a> </div>
2886 <div class="doc_text">
2888 <pre> <result> = load <ty>* <pointer>[, align <alignment>]<br> <result> = volatile load <ty>* <pointer>[, align <alignment>]<br></pre>
2890 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
2892 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
2893 address from which to load. The pointer must point to a <a
2894 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
2895 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
2896 the number or order of execution of this <tt>load</tt> with other
2897 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
2900 The optional "align" argument specifies the alignment of the operation
2901 (that is, the alignment of the memory address). A value of 0 or an
2902 omitted "align" argument means that the operation has the preferential
2903 alignment for the target. It is the responsibility of the code emitter
2904 to ensure that the alignment information is correct. Overestimating
2905 the alignment results in an undefined behavior. Underestimating the
2906 alignment may produce less efficient code. An alignment of 1 is always
2910 <p>The location of memory pointed to is loaded.</p>
2912 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
2914 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
2915 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
2918 <!-- _______________________________________________________________________ -->
2919 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
2920 Instruction</a> </div>
2921 <div class="doc_text">
2923 <pre> store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
2924 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
2927 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
2929 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
2930 to store and an address at which to store it. The type of the '<tt><pointer></tt>'
2931 operand must be a pointer to the type of the '<tt><value></tt>'
2932 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
2933 optimizer is not allowed to modify the number or order of execution of
2934 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
2935 href="#i_store">store</a></tt> instructions.</p>
2937 The optional "align" argument specifies the alignment of the operation
2938 (that is, the alignment of the memory address). A value of 0 or an
2939 omitted "align" argument means that the operation has the preferential
2940 alignment for the target. It is the responsibility of the code emitter
2941 to ensure that the alignment information is correct. Overestimating
2942 the alignment results in an undefined behavior. Underestimating the
2943 alignment may produce less efficient code. An alignment of 1 is always
2947 <p>The contents of memory are updated to contain '<tt><value></tt>'
2948 at the location specified by the '<tt><pointer></tt>' operand.</p>
2950 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
2951 store i32 3, i32* %ptr <i>; yields {void}</i>
2952 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
2956 <!-- _______________________________________________________________________ -->
2957 <div class="doc_subsubsection">
2958 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
2961 <div class="doc_text">
2964 <result> = getelementptr <ty>* <ptrval>{, <ty> <idx>}*
2970 The '<tt>getelementptr</tt>' instruction is used to get the address of a
2971 subelement of an aggregate data structure.</p>
2975 <p>This instruction takes a list of integer operands that indicate what
2976 elements of the aggregate object to index to. The actual types of the arguments
2977 provided depend on the type of the first pointer argument. The
2978 '<tt>getelementptr</tt>' instruction is used to index down through the type
2979 levels of a structure or to a specific index in an array. When indexing into a
2980 structure, only <tt>i32</tt> integer constants are allowed. When indexing
2981 into an array or pointer, only integers of 32 or 64 bits are allowed, and will
2982 be sign extended to 64-bit values.</p>
2984 <p>For example, let's consider a C code fragment and how it gets
2985 compiled to LLVM:</p>
2987 <div class="doc_code">
3000 int *foo(struct ST *s) {
3001 return &s[1].Z.B[5][13];
3006 <p>The LLVM code generated by the GCC frontend is:</p>
3008 <div class="doc_code">
3010 %RT = type { i8 , [10 x [20 x i32]], i8 }
3011 %ST = type { i32, double, %RT }
3013 define i32* %foo(%ST* %s) {
3015 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
3023 <p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
3024 on the pointer type that is being indexed into. <a href="#t_pointer">Pointer</a>
3025 and <a href="#t_array">array</a> types can use a 32-bit or 64-bit
3026 <a href="#t_integer">integer</a> type but the value will always be sign extended
3027 to 64-bits. <a href="#t_struct">Structure</a> types require <tt>i32</tt>
3028 <b>constants</b>.</p>
3030 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
3031 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
3032 }</tt>' type, a structure. The second index indexes into the third element of
3033 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
3034 i8 }</tt>' type, another structure. The third index indexes into the second
3035 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
3036 array. The two dimensions of the array are subscripted into, yielding an
3037 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
3038 to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
3040 <p>Note that it is perfectly legal to index partially through a
3041 structure, returning a pointer to an inner element. Because of this,
3042 the LLVM code for the given testcase is equivalent to:</p>
3045 define i32* %foo(%ST* %s) {
3046 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
3047 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
3048 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
3049 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
3050 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
3055 <p>Note that it is undefined to access an array out of bounds: array and
3056 pointer indexes must always be within the defined bounds of the array type.
3057 The one exception for this rules is zero length arrays. These arrays are
3058 defined to be accessible as variable length arrays, which requires access
3059 beyond the zero'th element.</p>
3061 <p>The getelementptr instruction is often confusing. For some more insight
3062 into how it works, see <a href="GetElementPtr.html">the getelementptr
3068 <i>; yields [12 x i8]*:aptr</i>
3069 %aptr = getelementptr {i32, [12 x i8]}* %sptr, i64 0, i32 1
3073 <!-- ======================================================================= -->
3074 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
3076 <div class="doc_text">
3077 <p>The instructions in this category are the conversion instructions (casting)
3078 which all take a single operand and a type. They perform various bit conversions
3082 <!-- _______________________________________________________________________ -->
3083 <div class="doc_subsubsection">
3084 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
3086 <div class="doc_text">
3090 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
3095 The '<tt>trunc</tt>' instruction truncates its operand to the type <tt>ty2</tt>.
3100 The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
3101 be an <a href="#t_integer">integer</a> type, and a type that specifies the size
3102 and type of the result, which must be an <a href="#t_integer">integer</a>
3103 type. The bit size of <tt>value</tt> must be larger than the bit size of
3104 <tt>ty2</tt>. Equal sized types are not allowed.</p>
3108 The '<tt>trunc</tt>' instruction truncates the high order bits in <tt>value</tt>
3109 and converts the remaining bits to <tt>ty2</tt>. Since the source size must be
3110 larger than the destination size, <tt>trunc</tt> cannot be a <i>no-op cast</i>.
3111 It will always truncate bits.</p>
3115 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
3116 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
3117 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
3121 <!-- _______________________________________________________________________ -->
3122 <div class="doc_subsubsection">
3123 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
3125 <div class="doc_text">
3129 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
3133 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
3138 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
3139 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3140 also be of <a href="#t_integer">integer</a> type. The bit size of the
3141 <tt>value</tt> must be smaller than the bit size of the destination type,
3145 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
3146 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
3148 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
3152 %X = zext i32 257 to i64 <i>; yields i64:257</i>
3153 %Y = zext i1 true to i32 <i>; yields i32:1</i>
3157 <!-- _______________________________________________________________________ -->
3158 <div class="doc_subsubsection">
3159 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
3161 <div class="doc_text">
3165 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
3169 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
3173 The '<tt>sext</tt>' instruction takes a value to cast, which must be of
3174 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3175 also be of <a href="#t_integer">integer</a> type. The bit size of the
3176 <tt>value</tt> must be smaller than the bit size of the destination type,
3181 The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
3182 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size of
3183 the type <tt>ty2</tt>.</p>
3185 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
3189 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
3190 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
3194 <!-- _______________________________________________________________________ -->
3195 <div class="doc_subsubsection">
3196 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
3199 <div class="doc_text">
3204 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
3208 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
3213 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
3214 point</a> value to cast and a <a href="#t_floating">floating point</a> type to
3215 cast it to. The size of <tt>value</tt> must be larger than the size of
3216 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
3217 <i>no-op cast</i>.</p>
3220 <p> The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
3221 <a href="#t_floating">floating point</a> type to a smaller
3222 <a href="#t_floating">floating point</a> type. If the value cannot fit within
3223 the destination type, <tt>ty2</tt>, then the results are undefined.</p>
3227 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
3228 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
3232 <!-- _______________________________________________________________________ -->
3233 <div class="doc_subsubsection">
3234 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
3236 <div class="doc_text">
3240 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
3244 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
3245 floating point value.</p>
3248 <p>The '<tt>fpext</tt>' instruction takes a
3249 <a href="#t_floating">floating point</a> <tt>value</tt> to cast,
3250 and a <a href="#t_floating">floating point</a> type to cast it to. The source
3251 type must be smaller than the destination type.</p>
3254 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
3255 <a href="#t_floating">floating point</a> type to a larger
3256 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
3257 used to make a <i>no-op cast</i> because it always changes bits. Use
3258 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
3262 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
3263 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
3267 <!-- _______________________________________________________________________ -->
3268 <div class="doc_subsubsection">
3269 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
3271 <div class="doc_text">
3275 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
3279 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
3280 unsigned integer equivalent of type <tt>ty2</tt>.
3284 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
3285 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3286 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3287 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3288 vector integer type with the same number of elements as <tt>ty</tt></p>
3291 <p> The '<tt>fptoui</tt>' instruction converts its
3292 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3293 towards zero) unsigned integer value. If the value cannot fit in <tt>ty2</tt>,
3294 the results are undefined.</p>
3298 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
3299 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
3300 %X = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
3304 <!-- _______________________________________________________________________ -->
3305 <div class="doc_subsubsection">
3306 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
3308 <div class="doc_text">
3312 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
3316 <p>The '<tt>fptosi</tt>' instruction converts
3317 <a href="#t_floating">floating point</a> <tt>value</tt> to type <tt>ty2</tt>.
3321 <p> The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
3322 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3323 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3324 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3325 vector integer type with the same number of elements as <tt>ty</tt></p>
3328 <p>The '<tt>fptosi</tt>' instruction converts its
3329 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3330 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
3331 the results are undefined.</p>
3335 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
3336 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
3337 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
3341 <!-- _______________________________________________________________________ -->
3342 <div class="doc_subsubsection">
3343 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
3345 <div class="doc_text">
3349 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3353 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
3354 integer and converts that value to the <tt>ty2</tt> type.</p>
3357 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
3358 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3359 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3360 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3361 floating point type with the same number of elements as <tt>ty</tt></p>
3364 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
3365 integer quantity and converts it to the corresponding floating point value. If
3366 the value cannot fit in the floating point value, the results are undefined.</p>
3370 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
3371 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
3375 <!-- _______________________________________________________________________ -->
3376 <div class="doc_subsubsection">
3377 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
3379 <div class="doc_text">
3383 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3387 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed
3388 integer and converts that value to the <tt>ty2</tt> type.</p>
3391 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
3392 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3393 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3394 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3395 floating point type with the same number of elements as <tt>ty</tt></p>
3398 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed
3399 integer quantity and converts it to the corresponding floating point value. If
3400 the value cannot fit in the floating point value, the results are undefined.</p>
3404 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
3405 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
3409 <!-- _______________________________________________________________________ -->
3410 <div class="doc_subsubsection">
3411 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
3413 <div class="doc_text">
3417 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
3421 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
3422 the integer type <tt>ty2</tt>.</p>
3425 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
3426 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
3427 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.
3430 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
3431 <tt>ty2</tt> by interpreting the pointer value as an integer and either
3432 truncating or zero extending that value to the size of the integer type. If
3433 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
3434 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
3435 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
3440 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
3441 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
3445 <!-- _______________________________________________________________________ -->
3446 <div class="doc_subsubsection">
3447 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
3449 <div class="doc_text">
3453 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
3457 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to
3458 a pointer type, <tt>ty2</tt>.</p>
3461 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
3462 value to cast, and a type to cast it to, which must be a
3463 <a href="#t_pointer">pointer</a> type.
3466 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
3467 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
3468 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
3469 size of a pointer then a truncation is done. If <tt>value</tt> is smaller than
3470 the size of a pointer then a zero extension is done. If they are the same size,
3471 nothing is done (<i>no-op cast</i>).</p>
3475 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
3476 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
3477 %Y = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
3481 <!-- _______________________________________________________________________ -->
3482 <div class="doc_subsubsection">
3483 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
3485 <div class="doc_text">
3489 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
3493 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3494 <tt>ty2</tt> without changing any bits.</p>
3497 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be
3498 a first class value, and a type to cast it to, which must also be a <a
3499 href="#t_firstclass">first class</a> type. The bit sizes of <tt>value</tt>
3500 and the destination type, <tt>ty2</tt>, must be identical. If the source
3501 type is a pointer, the destination type must also be a pointer.</p>
3504 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3505 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
3506 this conversion. The conversion is done as if the <tt>value</tt> had been
3507 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only be
3508 converted to other pointer types with this instruction. To convert pointers to
3509 other types, use the <a href="#i_inttoptr">inttoptr</a> or
3510 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
3514 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
3515 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
3516 %Z = bitcast <2xint> %V to i64; <i>; yields i64: %V</i>
3520 <!-- ======================================================================= -->
3521 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
3522 <div class="doc_text">
3523 <p>The instructions in this category are the "miscellaneous"
3524 instructions, which defy better classification.</p>
3527 <!-- _______________________________________________________________________ -->
3528 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
3530 <div class="doc_text">
3532 <pre> <result> = icmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3535 <p>The '<tt>icmp</tt>' instruction returns a boolean value based on comparison
3536 of its two integer operands.</p>
3538 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
3539 the condition code indicating the kind of comparison to perform. It is not
3540 a value, just a keyword. The possible condition code are:
3542 <li><tt>eq</tt>: equal</li>
3543 <li><tt>ne</tt>: not equal </li>
3544 <li><tt>ugt</tt>: unsigned greater than</li>
3545 <li><tt>uge</tt>: unsigned greater or equal</li>
3546 <li><tt>ult</tt>: unsigned less than</li>
3547 <li><tt>ule</tt>: unsigned less or equal</li>
3548 <li><tt>sgt</tt>: signed greater than</li>
3549 <li><tt>sge</tt>: signed greater or equal</li>
3550 <li><tt>slt</tt>: signed less than</li>
3551 <li><tt>sle</tt>: signed less or equal</li>
3553 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
3554 <a href="#t_pointer">pointer</a> typed. They must also be identical types.</p>
3556 <p>The '<tt>icmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3557 the condition code given as <tt>cond</tt>. The comparison performed always
3558 yields a <a href="#t_primitive">i1</a> result, as follows:
3560 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
3561 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3563 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
3564 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3565 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
3566 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3567 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
3568 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3569 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
3570 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3571 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
3572 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3573 <li><tt>sgt</tt>: interprets the operands as signed values and yields
3574 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3575 <li><tt>sge</tt>: interprets the operands as signed values and yields
3576 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3577 <li><tt>slt</tt>: interprets the operands as signed values and yields
3578 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3579 <li><tt>sle</tt>: interprets the operands as signed values and yields
3580 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3582 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
3583 values are compared as if they were integers.</p>
3586 <pre> <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
3587 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
3588 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
3589 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
3590 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
3591 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
3595 <!-- _______________________________________________________________________ -->
3596 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
3598 <div class="doc_text">
3600 <pre> <result> = fcmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3603 <p>The '<tt>fcmp</tt>' instruction returns a boolean value based on comparison
3604 of its floating point operands.</p>
3606 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
3607 the condition code indicating the kind of comparison to perform. It is not
3608 a value, just a keyword. The possible condition code are:
3610 <li><tt>false</tt>: no comparison, always returns false</li>
3611 <li><tt>oeq</tt>: ordered and equal</li>
3612 <li><tt>ogt</tt>: ordered and greater than </li>
3613 <li><tt>oge</tt>: ordered and greater than or equal</li>
3614 <li><tt>olt</tt>: ordered and less than </li>
3615 <li><tt>ole</tt>: ordered and less than or equal</li>
3616 <li><tt>one</tt>: ordered and not equal</li>
3617 <li><tt>ord</tt>: ordered (no nans)</li>
3618 <li><tt>ueq</tt>: unordered or equal</li>
3619 <li><tt>ugt</tt>: unordered or greater than </li>
3620 <li><tt>uge</tt>: unordered or greater than or equal</li>
3621 <li><tt>ult</tt>: unordered or less than </li>
3622 <li><tt>ule</tt>: unordered or less than or equal</li>
3623 <li><tt>une</tt>: unordered or not equal</li>
3624 <li><tt>uno</tt>: unordered (either nans)</li>
3625 <li><tt>true</tt>: no comparison, always returns true</li>
3627 <p><i>Ordered</i> means that neither operand is a QNAN while
3628 <i>unordered</i> means that either operand may be a QNAN.</p>
3629 <p>The <tt>val1</tt> and <tt>val2</tt> arguments must be
3630 <a href="#t_floating">floating point</a> typed. They must have identical
3633 <p>The '<tt>fcmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3634 the condition code given as <tt>cond</tt>. The comparison performed always
3635 yields a <a href="#t_primitive">i1</a> result, as follows:
3637 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
3638 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3639 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3640 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3641 <tt>var1</tt> is greather than <tt>var2</tt>.</li>
3642 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3643 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3644 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3645 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3646 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3647 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3648 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3649 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3650 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
3651 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
3652 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3653 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
3654 <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3655 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
3656 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3657 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
3658 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3659 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
3660 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3661 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
3662 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3663 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
3664 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
3668 <pre> <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
3669 <result> = icmp one float 4.0, 5.0 <i>; yields: result=true</i>
3670 <result> = icmp olt float 4.0, 5.0 <i>; yields: result=true</i>
3671 <result> = icmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
3675 <!-- _______________________________________________________________________ -->
3676 <div class="doc_subsubsection"> <a name="i_phi">'<tt>phi</tt>'
3677 Instruction</a> </div>
3678 <div class="doc_text">
3680 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
3682 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
3683 the SSA graph representing the function.</p>
3685 <p>The type of the incoming values is specified with the first type
3686 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
3687 as arguments, with one pair for each predecessor basic block of the
3688 current block. Only values of <a href="#t_firstclass">first class</a>
3689 type may be used as the value arguments to the PHI node. Only labels
3690 may be used as the label arguments.</p>
3691 <p>There must be no non-phi instructions between the start of a basic
3692 block and the PHI instructions: i.e. PHI instructions must be first in
3695 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
3696 specified by the pair corresponding to the predecessor basic block that executed
3697 just prior to the current block.</p>
3699 <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>
3702 <!-- _______________________________________________________________________ -->
3703 <div class="doc_subsubsection">
3704 <a name="i_select">'<tt>select</tt>' Instruction</a>
3707 <div class="doc_text">
3712 <result> = select i1 <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
3718 The '<tt>select</tt>' instruction is used to choose one value based on a
3719 condition, without branching.
3726 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.
3732 If the boolean condition evaluates to true, the instruction returns the first
3733 value argument; otherwise, it returns the second value argument.
3739 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
3744 <!-- _______________________________________________________________________ -->
3745 <div class="doc_subsubsection">
3746 <a name="i_call">'<tt>call</tt>' Instruction</a>
3749 <div class="doc_text">
3753 <result> = [tail] call [<a href="#callingconv">cconv</a>] <ty> [<fnty>*] <fnptrval>(<param list>)
3758 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
3762 <p>This instruction requires several arguments:</p>
3766 <p>The optional "tail" marker indicates whether the callee function accesses
3767 any allocas or varargs in the caller. If the "tail" marker is present, the
3768 function call is eligible for tail call optimization. Note that calls may
3769 be marked "tail" even if they do not occur before a <a
3770 href="#i_ret"><tt>ret</tt></a> instruction.
3773 <p>The optional "cconv" marker indicates which <a href="#callingconv">calling
3774 convention</a> the call should use. If none is specified, the call defaults
3775 to using C calling conventions.
3778 <p>'<tt>ty</tt>': the type of the call instruction itself which is also
3779 the type of the return value. Functions that return no value are marked
3780 <tt><a href="#t_void">void</a></tt>.</p>
3783 <p>'<tt>fnty</tt>': shall be the signature of the pointer to function
3784 value being invoked. The argument types must match the types implied by
3785 this signature. This type can be omitted if the function is not varargs
3786 and if the function type does not return a pointer to a function.</p>
3789 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
3790 be invoked. In most cases, this is a direct function invocation, but
3791 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
3792 to function value.</p>
3795 <p>'<tt>function args</tt>': argument list whose types match the
3796 function signature argument types. All arguments must be of
3797 <a href="#t_firstclass">first class</a> type. If the function signature
3798 indicates the function accepts a variable number of arguments, the extra
3799 arguments can be specified.</p>
3805 <p>The '<tt>call</tt>' instruction is used to cause control flow to
3806 transfer to a specified function, with its incoming arguments bound to
3807 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
3808 instruction in the called function, control flow continues with the
3809 instruction after the function call, and the return value of the
3810 function is bound to the result argument. This is a simpler case of
3811 the <a href="#i_invoke">invoke</a> instruction.</p>
3816 %retval = call i32 @test(i32 %argc)
3817 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42);
3818 %X = tail call i32 @foo()
3819 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo()
3820 %Z = call void %foo(i8 97 signext)
3825 <!-- _______________________________________________________________________ -->
3826 <div class="doc_subsubsection">
3827 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
3830 <div class="doc_text">
3835 <resultval> = va_arg <va_list*> <arglist>, <argty>
3840 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
3841 the "variable argument" area of a function call. It is used to implement the
3842 <tt>va_arg</tt> macro in C.</p>
3846 <p>This instruction takes a <tt>va_list*</tt> value and the type of
3847 the argument. It returns a value of the specified argument type and
3848 increments the <tt>va_list</tt> to point to the next argument. The
3849 actual type of <tt>va_list</tt> is target specific.</p>
3853 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
3854 type from the specified <tt>va_list</tt> and causes the
3855 <tt>va_list</tt> to point to the next argument. For more information,
3856 see the variable argument handling <a href="#int_varargs">Intrinsic
3859 <p>It is legal for this instruction to be called in a function which does not
3860 take a variable number of arguments, for example, the <tt>vfprintf</tt>
3863 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
3864 href="#intrinsics">intrinsic function</a> because it takes a type as an
3869 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
3873 <!-- *********************************************************************** -->
3874 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
3875 <!-- *********************************************************************** -->
3877 <div class="doc_text">
3879 <p>LLVM supports the notion of an "intrinsic function". These functions have
3880 well known names and semantics and are required to follow certain restrictions.
3881 Overall, these intrinsics represent an extension mechanism for the LLVM
3882 language that does not require changing all of the transformations in LLVM when
3883 adding to the language (or the bitcode reader/writer, the parser, etc...).</p>
3885 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
3886 prefix is reserved in LLVM for intrinsic names; thus, function names may not
3887 begin with this prefix. Intrinsic functions must always be external functions:
3888 you cannot define the body of intrinsic functions. Intrinsic functions may
3889 only be used in call or invoke instructions: it is illegal to take the address
3890 of an intrinsic function. Additionally, because intrinsic functions are part
3891 of the LLVM language, it is required if any are added that they be documented
3894 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
3895 a family of functions that perform the same operation but on different data
3896 types. Because LLVM can represent over 8 million different integer types,
3897 overloading is used commonly to allow an intrinsic function to operate on any
3898 integer type. One or more of the argument types or the result type can be
3899 overloaded to accept any integer type. Argument types may also be defined as
3900 exactly matching a previous argument's type or the result type. This allows an
3901 intrinsic function which accepts multiple arguments, but needs all of them to
3902 be of the same type, to only be overloaded with respect to a single argument or
3905 <p>Overloaded intrinsics will have the names of its overloaded argument types
3906 encoded into its function name, each preceded by a period. Only those types
3907 which are overloaded result in a name suffix. Arguments whose type is matched
3908 against another type do not. For example, the <tt>llvm.ctpop</tt> function can
3909 take an integer of any width and returns an integer of exactly the same integer
3910 width. This leads to a family of functions such as
3911 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29 %val)</tt>.
3912 Only one type, the return type, is overloaded, and only one type suffix is
3913 required. Because the argument's type is matched against the return type, it
3914 does not require its own name suffix.</p>
3916 <p>To learn how to add an intrinsic function, please see the
3917 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.
3922 <!-- ======================================================================= -->
3923 <div class="doc_subsection">
3924 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
3927 <div class="doc_text">
3929 <p>Variable argument support is defined in LLVM with the <a
3930 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
3931 intrinsic functions. These functions are related to the similarly
3932 named macros defined in the <tt><stdarg.h></tt> header file.</p>
3934 <p>All of these functions operate on arguments that use a
3935 target-specific value type "<tt>va_list</tt>". The LLVM assembly
3936 language reference manual does not define what this type is, so all
3937 transformations should be prepared to handle these functions regardless of
3940 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
3941 instruction and the variable argument handling intrinsic functions are
3944 <div class="doc_code">
3946 define i32 @test(i32 %X, ...) {
3947 ; Initialize variable argument processing
3949 %ap2 = bitcast i8** %ap to i8*
3950 call void @llvm.va_start(i8* %ap2)
3952 ; Read a single integer argument
3953 %tmp = va_arg i8** %ap, i32
3955 ; Demonstrate usage of llvm.va_copy and llvm.va_end
3957 %aq2 = bitcast i8** %aq to i8*
3958 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
3959 call void @llvm.va_end(i8* %aq2)
3961 ; Stop processing of arguments.
3962 call void @llvm.va_end(i8* %ap2)
3966 declare void @llvm.va_start(i8*)
3967 declare void @llvm.va_copy(i8*, i8*)
3968 declare void @llvm.va_end(i8*)
3974 <!-- _______________________________________________________________________ -->
3975 <div class="doc_subsubsection">
3976 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
3980 <div class="doc_text">
3982 <pre> declare void %llvm.va_start(i8* <arglist>)<br></pre>
3984 <P>The '<tt>llvm.va_start</tt>' intrinsic initializes
3985 <tt>*<arglist></tt> for subsequent use by <tt><a
3986 href="#i_va_arg">va_arg</a></tt>.</p>
3990 <P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
3994 <P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
3995 macro available in C. In a target-dependent way, it initializes the
3996 <tt>va_list</tt> element to which the argument points, so that the next call to
3997 <tt>va_arg</tt> will produce the first variable argument passed to the function.
3998 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
3999 last argument of the function as the compiler can figure that out.</p>
4003 <!-- _______________________________________________________________________ -->
4004 <div class="doc_subsubsection">
4005 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
4008 <div class="doc_text">
4010 <pre> declare void @llvm.va_end(i8* <arglist>)<br></pre>
4013 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
4014 which has been initialized previously with <tt><a href="#int_va_start">llvm.va_start</a></tt>
4015 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
4019 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
4023 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
4024 macro available in C. In a target-dependent way, it destroys the
4025 <tt>va_list</tt> element to which the argument points. Calls to <a
4026 href="#int_va_start"><tt>llvm.va_start</tt></a> and <a href="#int_va_copy">
4027 <tt>llvm.va_copy</tt></a> must be matched exactly with calls to
4028 <tt>llvm.va_end</tt>.</p>
4032 <!-- _______________________________________________________________________ -->
4033 <div class="doc_subsubsection">
4034 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
4037 <div class="doc_text">
4042 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
4047 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
4048 from the source argument list to the destination argument list.</p>
4052 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
4053 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
4058 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
4059 macro available in C. In a target-dependent way, it copies the source
4060 <tt>va_list</tt> element into the destination <tt>va_list</tt> element. This
4061 intrinsic is necessary because the <tt><a href="#int_va_start">
4062 llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
4063 example, memory allocation.</p>
4067 <!-- ======================================================================= -->
4068 <div class="doc_subsection">
4069 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
4072 <div class="doc_text">
4075 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
4076 Collection</a> requires the implementation and generation of these intrinsics.
4077 These intrinsics allow identification of <a href="#int_gcroot">GC roots on the
4078 stack</a>, as well as garbage collector implementations that require <a
4079 href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a> barriers.
4080 Front-ends for type-safe garbage collected languages should generate these
4081 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
4082 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
4085 <p>The garbage collection intrinsics only operate on objects in the generic
4086 address space (address space zero).</p>
4090 <!-- _______________________________________________________________________ -->
4091 <div class="doc_subsubsection">
4092 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
4095 <div class="doc_text">
4100 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
4105 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
4106 the code generator, and allows some metadata to be associated with it.</p>
4110 <p>The first argument specifies the address of a stack object that contains the
4111 root pointer. The second pointer (which must be either a constant or a global
4112 value address) contains the meta-data to be associated with the root.</p>
4116 <p>At runtime, a call to this intrinsics stores a null pointer into the "ptrloc"
4117 location. At compile-time, the code generator generates information to allow
4118 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
4119 intrinsic may only be used in a function which <a href="#gc">specifies a GC
4125 <!-- _______________________________________________________________________ -->
4126 <div class="doc_subsubsection">
4127 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
4130 <div class="doc_text">
4135 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
4140 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
4141 locations, allowing garbage collector implementations that require read
4146 <p>The second argument is the address to read from, which should be an address
4147 allocated from the garbage collector. The first object is a pointer to the
4148 start of the referenced object, if needed by the language runtime (otherwise
4153 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
4154 instruction, but may be replaced with substantially more complex code by the
4155 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
4156 may only be used in a function which <a href="#gc">specifies a GC
4162 <!-- _______________________________________________________________________ -->
4163 <div class="doc_subsubsection">
4164 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
4167 <div class="doc_text">
4172 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
4177 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
4178 locations, allowing garbage collector implementations that require write
4179 barriers (such as generational or reference counting collectors).</p>
4183 <p>The first argument is the reference to store, the second is the start of the
4184 object to store it to, and the third is the address of the field of Obj to
4185 store to. If the runtime does not require a pointer to the object, Obj may be
4190 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
4191 instruction, but may be replaced with substantially more complex code by the
4192 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
4193 may only be used in a function which <a href="#gc">specifies a GC
4200 <!-- ======================================================================= -->
4201 <div class="doc_subsection">
4202 <a name="int_codegen">Code Generator Intrinsics</a>
4205 <div class="doc_text">
4207 These intrinsics are provided by LLVM to expose special features that may only
4208 be implemented with code generator support.
4213 <!-- _______________________________________________________________________ -->
4214 <div class="doc_subsubsection">
4215 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
4218 <div class="doc_text">
4222 declare i8 *@llvm.returnaddress(i32 <level>)
4228 The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
4229 target-specific value indicating the return address of the current function
4230 or one of its callers.
4236 The argument to this intrinsic indicates which function to return the address
4237 for. Zero indicates the calling function, one indicates its caller, etc. The
4238 argument is <b>required</b> to be a constant integer value.
4244 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
4245 the return address of the specified call frame, or zero if it cannot be
4246 identified. The value returned by this intrinsic is likely to be incorrect or 0
4247 for arguments other than zero, so it should only be used for debugging purposes.
4251 Note that calling this intrinsic does not prevent function inlining or other
4252 aggressive transformations, so the value returned may not be that of the obvious
4253 source-language caller.
4258 <!-- _______________________________________________________________________ -->
4259 <div class="doc_subsubsection">
4260 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
4263 <div class="doc_text">
4267 declare i8 *@llvm.frameaddress(i32 <level>)
4273 The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
4274 target-specific frame pointer value for the specified stack frame.
4280 The argument to this intrinsic indicates which function to return the frame
4281 pointer for. Zero indicates the calling function, one indicates its caller,
4282 etc. The argument is <b>required</b> to be a constant integer value.
4288 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
4289 the frame address of the specified call frame, or zero if it cannot be
4290 identified. The value returned by this intrinsic is likely to be incorrect or 0
4291 for arguments other than zero, so it should only be used for debugging purposes.
4295 Note that calling this intrinsic does not prevent function inlining or other
4296 aggressive transformations, so the value returned may not be that of the obvious
4297 source-language caller.
4301 <!-- _______________________________________________________________________ -->
4302 <div class="doc_subsubsection">
4303 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
4306 <div class="doc_text">
4310 declare i8 *@llvm.stacksave()
4316 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
4317 the function stack, for use with <a href="#int_stackrestore">
4318 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
4319 features like scoped automatic variable sized arrays in C99.
4325 This intrinsic returns a opaque pointer value that can be passed to <a
4326 href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
4327 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
4328 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
4329 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
4330 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
4331 that were allocated after the <tt>llvm.stacksave</tt> was executed.
4336 <!-- _______________________________________________________________________ -->
4337 <div class="doc_subsubsection">
4338 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
4341 <div class="doc_text">
4345 declare void @llvm.stackrestore(i8 * %ptr)
4351 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
4352 the function stack to the state it was in when the corresponding <a
4353 href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
4354 useful for implementing language features like scoped automatic variable sized
4361 See the description for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.
4367 <!-- _______________________________________________________________________ -->
4368 <div class="doc_subsubsection">
4369 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
4372 <div class="doc_text">
4376 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>)
4383 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
4384 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
4386 effect on the behavior of the program but can change its performance
4393 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
4394 determining if the fetch should be for a read (0) or write (1), and
4395 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
4396 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
4397 <tt>locality</tt> arguments must be constant integers.
4403 This intrinsic does not modify the behavior of the program. In particular,
4404 prefetches cannot trap and do not produce a value. On targets that support this
4405 intrinsic, the prefetch can provide hints to the processor cache for better
4411 <!-- _______________________________________________________________________ -->
4412 <div class="doc_subsubsection">
4413 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
4416 <div class="doc_text">
4420 declare void @llvm.pcmarker(i32 <id>)
4427 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
4429 code to simulators and other tools. The method is target specific, but it is
4430 expected that the marker will use exported symbols to transmit the PC of the marker.
4431 The marker makes no guarantees that it will remain with any specific instruction
4432 after optimizations. It is possible that the presence of a marker will inhibit
4433 optimizations. The intended use is to be inserted after optimizations to allow
4434 correlations of simulation runs.
4440 <tt>id</tt> is a numerical id identifying the marker.
4446 This intrinsic does not modify the behavior of the program. Backends that do not
4447 support this intrinisic may ignore it.
4452 <!-- _______________________________________________________________________ -->
4453 <div class="doc_subsubsection">
4454 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
4457 <div class="doc_text">
4461 declare i64 @llvm.readcyclecounter( )
4468 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
4469 counter register (or similar low latency, high accuracy clocks) on those targets
4470 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
4471 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
4472 should only be used for small timings.
4478 When directly supported, reading the cycle counter should not modify any memory.
4479 Implementations are allowed to either return a application specific value or a
4480 system wide value. On backends without support, this is lowered to a constant 0.
4485 <!-- ======================================================================= -->
4486 <div class="doc_subsection">
4487 <a name="int_libc">Standard C Library Intrinsics</a>
4490 <div class="doc_text">
4492 LLVM provides intrinsics for a few important standard C library functions.
4493 These intrinsics allow source-language front-ends to pass information about the
4494 alignment of the pointer arguments to the code generator, providing opportunity
4495 for more efficient code generation.
4500 <!-- _______________________________________________________________________ -->
4501 <div class="doc_subsubsection">
4502 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
4505 <div class="doc_text">
4509 declare void @llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
4510 i32 <len>, i32 <align>)
4511 declare void @llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
4512 i64 <len>, i32 <align>)
4518 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4519 location to the destination location.
4523 Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
4524 intrinsics do not return a value, and takes an extra alignment argument.
4530 The first argument is a pointer to the destination, the second is a pointer to
4531 the source. The third argument is an integer argument
4532 specifying the number of bytes to copy, and the fourth argument is the alignment
4533 of the source and destination locations.
4537 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4538 the caller guarantees that both the source and destination pointers are aligned
4545 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4546 location to the destination location, which are not allowed to overlap. It
4547 copies "len" bytes of memory over. If the argument is known to be aligned to
4548 some boundary, this can be specified as the fourth argument, otherwise it should
4554 <!-- _______________________________________________________________________ -->
4555 <div class="doc_subsubsection">
4556 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
4559 <div class="doc_text">
4563 declare void @llvm.memmove.i32(i8 * <dest>, i8 * <src>,
4564 i32 <len>, i32 <align>)
4565 declare void @llvm.memmove.i64(i8 * <dest>, i8 * <src>,
4566 i64 <len>, i32 <align>)
4572 The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
4573 location to the destination location. It is similar to the
4574 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to overlap.
4578 Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
4579 intrinsics do not return a value, and takes an extra alignment argument.
4585 The first argument is a pointer to the destination, the second is a pointer to
4586 the source. The third argument is an integer argument
4587 specifying the number of bytes to copy, and the fourth argument is the alignment
4588 of the source and destination locations.
4592 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4593 the caller guarantees that the source and destination pointers are aligned to
4600 The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
4601 location to the destination location, which may overlap. It
4602 copies "len" bytes of memory over. If the argument is known to be aligned to
4603 some boundary, this can be specified as the fourth argument, otherwise it should
4609 <!-- _______________________________________________________________________ -->
4610 <div class="doc_subsubsection">
4611 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
4614 <div class="doc_text">
4618 declare void @llvm.memset.i32(i8 * <dest>, i8 <val>,
4619 i32 <len>, i32 <align>)
4620 declare void @llvm.memset.i64(i8 * <dest>, i8 <val>,
4621 i64 <len>, i32 <align>)
4627 The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
4632 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
4633 does not return a value, and takes an extra alignment argument.
4639 The first argument is a pointer to the destination to fill, the second is the
4640 byte value to fill it with, the third argument is an integer
4641 argument specifying the number of bytes to fill, and the fourth argument is the
4642 known alignment of destination location.
4646 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4647 the caller guarantees that the destination pointer is aligned to that boundary.
4653 The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
4655 destination location. If the argument is known to be aligned to some boundary,
4656 this can be specified as the fourth argument, otherwise it should be set to 0 or
4662 <!-- _______________________________________________________________________ -->
4663 <div class="doc_subsubsection">
4664 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
4667 <div class="doc_text">
4670 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
4671 floating point or vector of floating point type. Not all targets support all
4674 declare float @llvm.sqrt.f32(float %Val)
4675 declare double @llvm.sqrt.f64(double %Val)
4676 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
4677 declare fp128 @llvm.sqrt.f128(fp128 %Val)
4678 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
4684 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
4685 returning the same value as the libm '<tt>sqrt</tt>' functions would. Unlike
4686 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
4687 negative numbers other than -0.0 (which allows for better optimization, because
4688 there is no need to worry about errno being set). <tt>llvm.sqrt(-0.0)</tt> is
4689 defined to return -0.0 like IEEE sqrt.
4695 The argument and return value are floating point numbers of the same type.
4701 This function returns the sqrt of the specified operand if it is a nonnegative
4702 floating point number.
4706 <!-- _______________________________________________________________________ -->
4707 <div class="doc_subsubsection">
4708 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
4711 <div class="doc_text">
4714 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
4715 floating point or vector of floating point type. Not all targets support all
4718 declare float @llvm.powi.f32(float %Val, i32 %power)
4719 declare double @llvm.powi.f64(double %Val, i32 %power)
4720 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
4721 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
4722 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
4728 The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
4729 specified (positive or negative) power. The order of evaluation of
4730 multiplications is not defined. When a vector of floating point type is
4731 used, the second argument remains a scalar integer value.
4737 The second argument is an integer power, and the first is a value to raise to
4744 This function returns the first value raised to the second power with an
4745 unspecified sequence of rounding operations.</p>
4748 <!-- _______________________________________________________________________ -->
4749 <div class="doc_subsubsection">
4750 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
4753 <div class="doc_text">
4756 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
4757 floating point or vector of floating point type. Not all targets support all
4760 declare float @llvm.sin.f32(float %Val)
4761 declare double @llvm.sin.f64(double %Val)
4762 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
4763 declare fp128 @llvm.sin.f128(fp128 %Val)
4764 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
4770 The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.
4776 The argument and return value are floating point numbers of the same type.
4782 This function returns the sine of the specified operand, returning the
4783 same values as the libm <tt>sin</tt> functions would, and handles error
4784 conditions in the same way.</p>
4787 <!-- _______________________________________________________________________ -->
4788 <div class="doc_subsubsection">
4789 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
4792 <div class="doc_text">
4795 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
4796 floating point or vector of floating point type. Not all targets support all
4799 declare float @llvm.cos.f32(float %Val)
4800 declare double @llvm.cos.f64(double %Val)
4801 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
4802 declare fp128 @llvm.cos.f128(fp128 %Val)
4803 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
4809 The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.
4815 The argument and return value are floating point numbers of the same type.
4821 This function returns the cosine of the specified operand, returning the
4822 same values as the libm <tt>cos</tt> functions would, and handles error
4823 conditions in the same way.</p>
4826 <!-- _______________________________________________________________________ -->
4827 <div class="doc_subsubsection">
4828 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
4831 <div class="doc_text">
4834 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
4835 floating point or vector of floating point type. Not all targets support all
4838 declare float @llvm.pow.f32(float %Val, float %Power)
4839 declare double @llvm.pow.f64(double %Val, double %Power)
4840 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
4841 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
4842 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
4848 The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
4849 specified (positive or negative) power.
4855 The second argument is a floating point power, and the first is a value to
4856 raise to that power.
4862 This function returns the first value raised to the second power,
4864 same values as the libm <tt>pow</tt> functions would, and handles error
4865 conditions in the same way.</p>
4869 <!-- ======================================================================= -->
4870 <div class="doc_subsection">
4871 <a name="int_manip">Bit Manipulation Intrinsics</a>
4874 <div class="doc_text">
4876 LLVM provides intrinsics for a few important bit manipulation operations.
4877 These allow efficient code generation for some algorithms.
4882 <!-- _______________________________________________________________________ -->
4883 <div class="doc_subsubsection">
4884 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
4887 <div class="doc_text">
4890 <p>This is an overloaded intrinsic function. You can use bswap on any integer
4891 type that is an even number of bytes (i.e. BitWidth % 16 == 0).
4893 declare i16 @llvm.bswap.i16(i16 <id>)
4894 declare i32 @llvm.bswap.i32(i32 <id>)
4895 declare i64 @llvm.bswap.i64(i64 <id>)
4901 The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
4902 values with an even number of bytes (positive multiple of 16 bits). These are
4903 useful for performing operations on data that is not in the target's native
4910 The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
4911 and low byte of the input i16 swapped. Similarly, the <tt>llvm.bswap.i32</tt>
4912 intrinsic returns an i32 value that has the four bytes of the input i32
4913 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the returned
4914 i32 will have its bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i48</tt>,
4915 <tt>llvm.bswap.i64</tt> and other intrinsics extend this concept to
4916 additional even-byte lengths (6 bytes, 8 bytes and more, respectively).
4921 <!-- _______________________________________________________________________ -->
4922 <div class="doc_subsubsection">
4923 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
4926 <div class="doc_text">
4929 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
4930 width. Not all targets support all bit widths however.
4932 declare i8 @llvm.ctpop.i8 (i8 <src>)
4933 declare i16 @llvm.ctpop.i16(i16 <src>)
4934 declare i32 @llvm.ctpop.i32(i32 <src>)
4935 declare i64 @llvm.ctpop.i64(i64 <src>)
4936 declare i256 @llvm.ctpop.i256(i256 <src>)
4942 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
4949 The only argument is the value to be counted. The argument may be of any
4950 integer type. The return type must match the argument type.
4956 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
4960 <!-- _______________________________________________________________________ -->
4961 <div class="doc_subsubsection">
4962 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
4965 <div class="doc_text">
4968 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
4969 integer bit width. Not all targets support all bit widths however.
4971 declare i8 @llvm.ctlz.i8 (i8 <src>)
4972 declare i16 @llvm.ctlz.i16(i16 <src>)
4973 declare i32 @llvm.ctlz.i32(i32 <src>)
4974 declare i64 @llvm.ctlz.i64(i64 <src>)
4975 declare i256 @llvm.ctlz.i256(i256 <src>)
4981 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
4982 leading zeros in a variable.
4988 The only argument is the value to be counted. The argument may be of any
4989 integer type. The return type must match the argument type.
4995 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
4996 in a variable. If the src == 0 then the result is the size in bits of the type
4997 of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.
5003 <!-- _______________________________________________________________________ -->
5004 <div class="doc_subsubsection">
5005 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
5008 <div class="doc_text">
5011 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
5012 integer bit width. Not all targets support all bit widths however.
5014 declare i8 @llvm.cttz.i8 (i8 <src>)
5015 declare i16 @llvm.cttz.i16(i16 <src>)
5016 declare i32 @llvm.cttz.i32(i32 <src>)
5017 declare i64 @llvm.cttz.i64(i64 <src>)
5018 declare i256 @llvm.cttz.i256(i256 <src>)
5024 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
5031 The only argument is the value to be counted. The argument may be of any
5032 integer type. The return type must match the argument type.
5038 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
5039 in a variable. If the src == 0 then the result is the size in bits of the type
5040 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
5044 <!-- _______________________________________________________________________ -->
5045 <div class="doc_subsubsection">
5046 <a name="int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic</a>
5049 <div class="doc_text">
5052 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.select</tt>
5053 on any integer bit width.
5055 declare i17 @llvm.part.select.i17 (i17 %val, i32 %loBit, i32 %hiBit)
5056 declare i29 @llvm.part.select.i29 (i29 %val, i32 %loBit, i32 %hiBit)
5060 <p>The '<tt>llvm.part.select</tt>' family of intrinsic functions selects a
5061 range of bits from an integer value and returns them in the same bit width as
5062 the original value.</p>
5065 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5066 any bit width but they must have the same bit width. The second and third
5067 arguments must be <tt>i32</tt> type since they specify only a bit index.</p>
5070 <p>The operation of the '<tt>llvm.part.select</tt>' intrinsic has two modes
5071 of operation: forwards and reverse. If <tt>%loBit</tt> is greater than
5072 <tt>%hiBits</tt> then the intrinsic operates in reverse mode. Otherwise it
5073 operates in forward mode.</p>
5074 <p>In forward mode, this intrinsic is the equivalent of shifting <tt>%val</tt>
5075 right by <tt>%loBit</tt> bits and then ANDing it with a mask with
5076 only the <tt>%hiBit - %loBit</tt> bits set, as follows:</p>
5078 <li>The <tt>%val</tt> is shifted right (LSHR) by the number of bits specified
5079 by <tt>%loBits</tt>. This normalizes the value to the low order bits.</li>
5080 <li>The <tt>%loBits</tt> value is subtracted from the <tt>%hiBits</tt> value
5081 to determine the number of bits to retain.</li>
5082 <li>A mask of the retained bits is created by shifting a -1 value.</li>
5083 <li>The mask is ANDed with <tt>%val</tt> to produce the result.
5085 <p>In reverse mode, a similar computation is made except that the bits are
5086 returned in the reverse order. So, for example, if <tt>X</tt> has the value
5087 <tt>i16 0x0ACF (101011001111)</tt> and we apply
5088 <tt>part.select(i16 X, 8, 3)</tt> to it, we get back the value
5089 <tt>i16 0x0026 (000000100110)</tt>.</p>
5092 <div class="doc_subsubsection">
5093 <a name="int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic</a>
5096 <div class="doc_text">
5099 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.set</tt>
5100 on any integer bit width.
5102 declare i17 @llvm.part.set.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
5103 declare i29 @llvm.part.set.i29.i9 (i29 %val, i9 %repl, i32 %lo, i32 %hi)
5107 <p>The '<tt>llvm.part.set</tt>' family of intrinsic functions replaces a range
5108 of bits in an integer value with another integer value. It returns the integer
5109 with the replaced bits.</p>
5112 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5113 any bit width but they must have the same bit width. <tt>%val</tt> is the value
5114 whose bits will be replaced. The second argument, <tt>%repl</tt> may be an
5115 integer of any bit width. The third and fourth arguments must be <tt>i32</tt>
5116 type since they specify only a bit index.</p>
5119 <p>The operation of the '<tt>llvm.part.set</tt>' intrinsic has two modes
5120 of operation: forwards and reverse. If <tt>%lo</tt> is greater than
5121 <tt>%hi</tt> then the intrinsic operates in reverse mode. Otherwise it
5122 operates in forward mode.</p>
5123 <p>For both modes, the <tt>%repl</tt> value is prepared for use by either
5124 truncating it down to the size of the replacement area or zero extending it
5125 up to that size.</p>
5126 <p>In forward mode, the bits between <tt>%lo</tt> and <tt>%hi</tt> (inclusive)
5127 are replaced with corresponding bits from <tt>%repl</tt>. That is the 0th bit
5128 in <tt>%repl</tt> replaces the <tt>%lo</tt>th bit in <tt>%val</tt> and etc. up
5129 to the <tt>%hi</tt>th bit.
5130 <p>In reverse mode, a similar computation is made except that the bits are
5131 reversed. That is, the <tt>0</tt>th bit in <tt>%repl</tt> replaces the
5132 <tt>%hi</tt> bit in <tt>%val</tt> and etc. down to the <tt>%lo</tt>th bit.
5135 llvm.part.set(0xFFFF, 0, 4, 7) -> 0xFF0F
5136 llvm.part.set(0xFFFF, 0, 7, 4) -> 0xFF0F
5137 llvm.part.set(0xFFFF, 1, 7, 4) -> 0xFF8F
5138 llvm.part.set(0xFFFF, F, 8, 3) -> 0xFFE7
5139 llvm.part.set(0xFFFF, 0, 3, 8) -> 0xFE07
5143 <!-- ======================================================================= -->
5144 <div class="doc_subsection">
5145 <a name="int_debugger">Debugger Intrinsics</a>
5148 <div class="doc_text">
5150 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
5151 are described in the <a
5152 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
5153 Debugging</a> document.
5158 <!-- ======================================================================= -->
5159 <div class="doc_subsection">
5160 <a name="int_eh">Exception Handling Intrinsics</a>
5163 <div class="doc_text">
5164 <p> The LLVM exception handling intrinsics (which all start with
5165 <tt>llvm.eh.</tt> prefix), are described in the <a
5166 href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
5167 Handling</a> document. </p>
5170 <!-- ======================================================================= -->
5171 <div class="doc_subsection">
5172 <a name="int_trampoline">Trampoline Intrinsic</a>
5175 <div class="doc_text">
5177 This intrinsic makes it possible to excise one parameter, marked with
5178 the <tt>nest</tt> attribute, from a function. The result is a callable
5179 function pointer lacking the nest parameter - the caller does not need
5180 to provide a value for it. Instead, the value to use is stored in
5181 advance in a "trampoline", a block of memory usually allocated
5182 on the stack, which also contains code to splice the nest value into the
5183 argument list. This is used to implement the GCC nested function address
5187 For example, if the function is
5188 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
5189 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as follows:</p>
5191 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
5192 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
5193 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
5194 %fp = bitcast i8* %p to i32 (i32, i32)*
5196 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
5197 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
5200 <!-- _______________________________________________________________________ -->
5201 <div class="doc_subsubsection">
5202 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
5204 <div class="doc_text">
5207 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
5211 This fills the memory pointed to by <tt>tramp</tt> with code
5212 and returns a function pointer suitable for executing it.
5216 The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
5217 pointers. The <tt>tramp</tt> argument must point to a sufficiently large
5218 and sufficiently aligned block of memory; this memory is written to by the
5219 intrinsic. Note that the size and the alignment are target-specific - LLVM
5220 currently provides no portable way of determining them, so a front-end that
5221 generates this intrinsic needs to have some target-specific knowledge.
5222 The <tt>func</tt> argument must hold a function bitcast to an <tt>i8*</tt>.
5226 The block of memory pointed to by <tt>tramp</tt> is filled with target
5227 dependent code, turning it into a function. A pointer to this function is
5228 returned, but needs to be bitcast to an
5229 <a href="#int_trampoline">appropriate function pointer type</a>
5230 before being called. The new function's signature is the same as that of
5231 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
5232 removed. At most one such <tt>nest</tt> argument is allowed, and it must be
5233 of pointer type. Calling the new function is equivalent to calling
5234 <tt>func</tt> with the same argument list, but with <tt>nval</tt> used for the
5235 missing <tt>nest</tt> argument. If, after calling
5236 <tt>llvm.init.trampoline</tt>, the memory pointed to by <tt>tramp</tt> is
5237 modified, then the effect of any later call to the returned function pointer is
5242 <!-- ======================================================================= -->
5243 <div class="doc_subsection">
5244 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
5247 <div class="doc_text">
5249 These intrinsic functions expand the "universal IR" of LLVM to represent
5250 hardware constructs for atomic operations and memory synchronization. This
5251 provides an interface to the hardware, not an interface to the programmer. It
5252 is aimed at a low enough level to allow any programming models or APIs which
5253 need atomic behaviors to map cleanly onto it. It is also modeled primarily on
5254 hardware behavior. Just as hardware provides a "universal IR" for source
5255 languages, it also provides a starting point for developing a "universal"
5256 atomic operation and synchronization IR.
5259 These do <em>not</em> form an API such as high-level threading libraries,
5260 software transaction memory systems, atomic primitives, and intrinsic
5261 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
5262 application libraries. The hardware interface provided by LLVM should allow
5263 a clean implementation of all of these APIs and parallel programming models.
5264 No one model or paradigm should be selected above others unless the hardware
5265 itself ubiquitously does so.
5270 <!-- _______________________________________________________________________ -->
5271 <div class="doc_subsubsection">
5272 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
5274 <div class="doc_text">
5277 declare void @llvm.memory.barrier( i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>,
5283 The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
5284 specific pairs of memory access types.
5288 The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
5289 The first four arguments enables a specific barrier as listed below. The fith
5290 argument specifies that the barrier applies to io or device or uncached memory.
5294 <li><tt>ll</tt>: load-load barrier</li>
5295 <li><tt>ls</tt>: load-store barrier</li>
5296 <li><tt>sl</tt>: store-load barrier</li>
5297 <li><tt>ss</tt>: store-store barrier</li>
5298 <li><tt>device</tt>: barrier applies to device and uncached memory also.
5302 This intrinsic causes the system to enforce some ordering constraints upon
5303 the loads and stores of the program. This barrier does not indicate
5304 <em>when</em> any events will occur, it only enforces an <em>order</em> in
5305 which they occur. For any of the specified pairs of load and store operations
5306 (f.ex. load-load, or store-load), all of the first operations preceding the
5307 barrier will complete before any of the second operations succeeding the
5308 barrier begin. Specifically the semantics for each pairing is as follows:
5311 <li><tt>ll</tt>: All loads before the barrier must complete before any load
5312 after the barrier begins.</li>
5314 <li><tt>ls</tt>: All loads before the barrier must complete before any
5315 store after the barrier begins.</li>
5316 <li><tt>ss</tt>: All stores before the barrier must complete before any
5317 store after the barrier begins.</li>
5318 <li><tt>sl</tt>: All stores before the barrier must complete before any
5319 load after the barrier begins.</li>
5322 These semantics are applied with a logical "and" behavior when more than one
5323 is enabled in a single memory barrier intrinsic.
5326 Backends may implement stronger barriers than those requested when they do not
5327 support as fine grained a barrier as requested. Some architectures do not
5328 need all types of barriers and on such architectures, these become noops.
5335 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
5336 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
5337 <i>; guarantee the above finishes</i>
5338 store i32 8, %ptr <i>; before this begins</i>
5343 <!-- ======================================================================= -->
5344 <div class="doc_subsection">
5345 <a name="int_general">General Intrinsics</a>
5348 <div class="doc_text">
5349 <p> This class of intrinsics is designed to be generic and has
5350 no specific purpose. </p>
5353 <!-- _______________________________________________________________________ -->
5354 <div class="doc_subsubsection">
5355 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
5358 <div class="doc_text">
5362 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
5368 The '<tt>llvm.var.annotation</tt>' intrinsic
5374 The first argument is a pointer to a value, the second is a pointer to a
5375 global string, the third is a pointer to a global string which is the source
5376 file name, and the last argument is the line number.
5382 This intrinsic allows annotation of local variables with arbitrary strings.
5383 This can be useful for special purpose optimizations that want to look for these
5384 annotations. These have no other defined use, they are ignored by code
5385 generation and optimization.
5389 <!-- _______________________________________________________________________ -->
5390 <div class="doc_subsubsection">
5391 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
5394 <div class="doc_text">
5397 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
5398 any integer bit width.
5401 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
5402 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
5403 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
5404 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
5405 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
5411 The '<tt>llvm.annotation</tt>' intrinsic.
5417 The first argument is an integer value (result of some expression),
5418 the second is a pointer to a global string, the third is a pointer to a global
5419 string which is the source file name, and the last argument is the line number.
5420 It returns the value of the first argument.
5426 This intrinsic allows annotations to be put on arbitrary expressions
5427 with arbitrary strings. This can be useful for special purpose optimizations
5428 that want to look for these annotations. These have no other defined use, they
5429 are ignored by code generation and optimization.
5432 <!-- _______________________________________________________________________ -->
5433 <div class="doc_subsubsection">
5434 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
5437 <div class="doc_text">
5441 declare void @llvm.trap()
5447 The '<tt>llvm.trap</tt>' intrinsic
5459 This intrinsics is lowered to the target dependent trap instruction. If the
5460 target does not have a trap instruction, this intrinsic will be lowered to the
5461 call of the abort() function.
5465 <!-- *********************************************************************** -->
5468 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
5469 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
5470 <a href="http://validator.w3.org/check/referer"><img
5471 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
5473 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
5474 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
5475 Last modified: $Date$