1 <!DOCTYPE HTML PUBLIC "-//W3C//DTD HTML 4.01//EN"
2 "http://www.w3.org/TR/html4/strict.dtd">
5 <title>LLVM Assembly Language Reference Manual</title>
6 <meta http-equiv="Content-Type" content="text/html; charset=utf-8">
7 <meta name="author" content="Chris Lattner">
8 <meta name="description"
9 content="LLVM Assembly Language Reference Manual.">
10 <link rel="stylesheet" href="llvm.css" type="text/css">
15 <div class="doc_title"> LLVM Language Reference Manual </div>
17 <li><a href="#abstract">Abstract</a></li>
18 <li><a href="#introduction">Introduction</a></li>
19 <li><a href="#identifiers">Identifiers</a></li>
20 <li><a href="#highlevel">High Level Structure</a>
22 <li><a href="#modulestructure">Module Structure</a></li>
23 <li><a href="#linkage">Linkage Types</a></li>
24 <li><a href="#callingconv">Calling Conventions</a></li>
25 <li><a href="#globalvars">Global Variables</a></li>
26 <li><a href="#functionstructure">Functions</a></li>
27 <li><a href="#aliasstructure">Aliases</a>
28 <li><a href="#paramattrs">Parameter Attributes</a></li>
29 <li><a href="#gc">Garbage Collector Names</a></li>
30 <li><a href="#moduleasm">Module-Level Inline Assembly</a></li>
31 <li><a href="#datalayout">Data Layout</a></li>
34 <li><a href="#typesystem">Type System</a>
36 <li><a href="#t_classifications">Type Classifications</a></li>
37 <li><a href="#t_primitive">Primitive Types</a>
39 <li><a href="#t_floating">Floating Point Types</a></li>
40 <li><a href="#t_void">Void Type</a></li>
41 <li><a href="#t_label">Label Type</a></li>
44 <li><a href="#t_derived">Derived Types</a>
46 <li><a href="#t_integer">Integer Type</a></li>
47 <li><a href="#t_array">Array Type</a></li>
48 <li><a href="#t_function">Function Type</a></li>
49 <li><a href="#t_pointer">Pointer Type</a></li>
50 <li><a href="#t_struct">Structure Type</a></li>
51 <li><a href="#t_pstruct">Packed Structure Type</a></li>
52 <li><a href="#t_vector">Vector Type</a></li>
53 <li><a href="#t_opaque">Opaque Type</a></li>
58 <li><a href="#constants">Constants</a>
60 <li><a href="#simpleconstants">Simple Constants</a>
61 <li><a href="#aggregateconstants">Aggregate Constants</a>
62 <li><a href="#globalconstants">Global Variable and Function Addresses</a>
63 <li><a href="#undefvalues">Undefined Values</a>
64 <li><a href="#constantexprs">Constant Expressions</a>
67 <li><a href="#othervalues">Other Values</a>
69 <li><a href="#inlineasm">Inline Assembler Expressions</a>
72 <li><a href="#instref">Instruction Reference</a>
74 <li><a href="#terminators">Terminator Instructions</a>
76 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
77 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
78 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
79 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
80 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
81 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
84 <li><a href="#binaryops">Binary Operations</a>
86 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
87 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
88 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
89 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
90 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
91 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
92 <li><a href="#i_urem">'<tt>urem</tt>' Instruction</a></li>
93 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
94 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
97 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
99 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
100 <li><a href="#i_lshr">'<tt>lshr</tt>' Instruction</a></li>
101 <li><a href="#i_ashr">'<tt>ashr</tt>' Instruction</a></li>
102 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
103 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
104 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
107 <li><a href="#vectorops">Vector Operations</a>
109 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
110 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
111 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
114 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
116 <li><a href="#i_malloc">'<tt>malloc</tt>' Instruction</a></li>
117 <li><a href="#i_free">'<tt>free</tt>' Instruction</a></li>
118 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
119 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
120 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
121 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
124 <li><a href="#convertops">Conversion Operations</a>
126 <li><a href="#i_trunc">'<tt>trunc .. to</tt>' Instruction</a></li>
127 <li><a href="#i_zext">'<tt>zext .. to</tt>' Instruction</a></li>
128 <li><a href="#i_sext">'<tt>sext .. to</tt>' Instruction</a></li>
129 <li><a href="#i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a></li>
130 <li><a href="#i_fpext">'<tt>fpext .. to</tt>' Instruction</a></li>
131 <li><a href="#i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a></li>
132 <li><a href="#i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a></li>
133 <li><a href="#i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a></li>
134 <li><a href="#i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a></li>
135 <li><a href="#i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a></li>
136 <li><a href="#i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a></li>
137 <li><a href="#i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a></li>
139 <li><a href="#otherops">Other Operations</a>
141 <li><a href="#i_icmp">'<tt>icmp</tt>' Instruction</a></li>
142 <li><a href="#i_fcmp">'<tt>fcmp</tt>' Instruction</a></li>
143 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
144 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
145 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
146 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
147 <li><a href="#i_getresult">'<tt>getresult</tt>' Instruction</a></li>
152 <li><a href="#intrinsics">Intrinsic Functions</a>
154 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
156 <li><a href="#int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
157 <li><a href="#int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
158 <li><a href="#int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
161 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
163 <li><a href="#int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
164 <li><a href="#int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
165 <li><a href="#int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
168 <li><a href="#int_codegen">Code Generator Intrinsics</a>
170 <li><a href="#int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
171 <li><a href="#int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
172 <li><a href="#int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
173 <li><a href="#int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
174 <li><a href="#int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
175 <li><a href="#int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
176 <li><a href="#int_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
179 <li><a href="#int_libc">Standard C Library Intrinsics</a>
181 <li><a href="#int_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
182 <li><a href="#int_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
183 <li><a href="#int_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
184 <li><a href="#int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
185 <li><a href="#int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
186 <li><a href="#int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a></li>
187 <li><a href="#int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a></li>
188 <li><a href="#int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a></li>
191 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
193 <li><a href="#int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
194 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
195 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
196 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
197 <li><a href="#int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic </a></li>
198 <li><a href="#int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic </a></li>
201 <li><a href="#int_debugger">Debugger intrinsics</a></li>
202 <li><a href="#int_eh">Exception Handling intrinsics</a></li>
203 <li><a href="#int_trampoline">Trampoline Intrinsic</a>
205 <li><a href="#int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a></li>
208 <li><a href="#int_atomics">Atomic intrinsics</a>
210 <li><a href="#int_memory_barrier"><tt>llvm.memory_barrier</tt></a></li>
211 <li><a href="#int_atomic_lcs"><tt>llvm.atomic.lcs</tt></a></li>
212 <li><a href="#int_atomic_las"><tt>llvm.atomic.las</tt></a></li>
213 <li><a href="#int_atomic_swap"><tt>llvm.atomic.swap</tt></a></li>
216 <li><a href="#int_general">General intrinsics</a>
218 <li><a href="#int_var_annotation">
219 <tt>llvm.var.annotation</tt>' Intrinsic</a></li>
220 <li><a href="#int_annotation">
221 <tt>llvm.annotation.*</tt>' Intrinsic</a></li>
222 <li><a href="#int_trap">
223 <tt>llvm.trap</tt>' Intrinsic</a></li>
230 <div class="doc_author">
231 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
232 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
235 <!-- *********************************************************************** -->
236 <div class="doc_section"> <a name="abstract">Abstract </a></div>
237 <!-- *********************************************************************** -->
239 <div class="doc_text">
240 <p>This document is a reference manual for the LLVM assembly language.
241 LLVM is an SSA based representation that provides type safety,
242 low-level operations, flexibility, and the capability of representing
243 'all' high-level languages cleanly. It is the common code
244 representation used throughout all phases of the LLVM compilation
248 <!-- *********************************************************************** -->
249 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
250 <!-- *********************************************************************** -->
252 <div class="doc_text">
254 <p>The LLVM code representation is designed to be used in three
255 different forms: as an in-memory compiler IR, as an on-disk bitcode
256 representation (suitable for fast loading by a Just-In-Time compiler),
257 and as a human readable assembly language representation. This allows
258 LLVM to provide a powerful intermediate representation for efficient
259 compiler transformations and analysis, while providing a natural means
260 to debug and visualize the transformations. The three different forms
261 of LLVM are all equivalent. This document describes the human readable
262 representation and notation.</p>
264 <p>The LLVM representation aims to be light-weight and low-level
265 while being expressive, typed, and extensible at the same time. It
266 aims to be a "universal IR" of sorts, by being at a low enough level
267 that high-level ideas may be cleanly mapped to it (similar to how
268 microprocessors are "universal IR's", allowing many source languages to
269 be mapped to them). By providing type information, LLVM can be used as
270 the target of optimizations: for example, through pointer analysis, it
271 can be proven that a C automatic variable is never accessed outside of
272 the current function... allowing it to be promoted to a simple SSA
273 value instead of a memory location.</p>
277 <!-- _______________________________________________________________________ -->
278 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
280 <div class="doc_text">
282 <p>It is important to note that this document describes 'well formed'
283 LLVM assembly language. There is a difference between what the parser
284 accepts and what is considered 'well formed'. For example, the
285 following instruction is syntactically okay, but not well formed:</p>
287 <div class="doc_code">
289 %x = <a href="#i_add">add</a> i32 1, %x
293 <p>...because the definition of <tt>%x</tt> does not dominate all of
294 its uses. The LLVM infrastructure provides a verification pass that may
295 be used to verify that an LLVM module is well formed. This pass is
296 automatically run by the parser after parsing input assembly and by
297 the optimizer before it outputs bitcode. The violations pointed out
298 by the verifier pass indicate bugs in transformation passes or input to
302 <!-- Describe the typesetting conventions here. -->
304 <!-- *********************************************************************** -->
305 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
306 <!-- *********************************************************************** -->
308 <div class="doc_text">
310 <p>LLVM identifiers come in two basic types: global and local. Global
311 identifiers (functions, global variables) begin with the @ character. Local
312 identifiers (register names, types) begin with the % character. Additionally,
313 there are three different formats for identifiers, for different purposes:
316 <li>Named values are represented as a string of characters with their prefix.
317 For example, %foo, @DivisionByZero, %a.really.long.identifier. The actual
318 regular expression used is '<tt>[%@][a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
319 Identifiers which require other characters in their names can be surrounded
320 with quotes. In this way, anything except a <tt>"</tt> character can
321 be used in a named value.</li>
323 <li>Unnamed values are represented as an unsigned numeric value with their
324 prefix. For example, %12, @2, %44.</li>
326 <li>Constants, which are described in a <a href="#constants">section about
327 constants</a>, below.</li>
330 <p>LLVM requires that values start with a prefix for two reasons: Compilers
331 don't need to worry about name clashes with reserved words, and the set of
332 reserved words may be expanded in the future without penalty. Additionally,
333 unnamed identifiers allow a compiler to quickly come up with a temporary
334 variable without having to avoid symbol table conflicts.</p>
336 <p>Reserved words in LLVM are very similar to reserved words in other
337 languages. There are keywords for different opcodes
338 ('<tt><a href="#i_add">add</a></tt>',
339 '<tt><a href="#i_bitcast">bitcast</a></tt>',
340 '<tt><a href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
341 href="#t_void">void</a></tt>', '<tt><a href="#t_primitive">i32</a></tt>', etc...),
342 and others. These reserved words cannot conflict with variable names, because
343 none of them start with a prefix character ('%' or '@').</p>
345 <p>Here is an example of LLVM code to multiply the integer variable
346 '<tt>%X</tt>' by 8:</p>
350 <div class="doc_code">
352 %result = <a href="#i_mul">mul</a> i32 %X, 8
356 <p>After strength reduction:</p>
358 <div class="doc_code">
360 %result = <a href="#i_shl">shl</a> i32 %X, i8 3
364 <p>And the hard way:</p>
366 <div class="doc_code">
368 <a href="#i_add">add</a> i32 %X, %X <i>; yields {i32}:%0</i>
369 <a href="#i_add">add</a> i32 %0, %0 <i>; yields {i32}:%1</i>
370 %result = <a href="#i_add">add</a> i32 %1, %1
374 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
375 important lexical features of LLVM:</p>
379 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
382 <li>Unnamed temporaries are created when the result of a computation is not
383 assigned to a named value.</li>
385 <li>Unnamed temporaries are numbered sequentially</li>
389 <p>...and it also shows a convention that we follow in this document. When
390 demonstrating instructions, we will follow an instruction with a comment that
391 defines the type and name of value produced. Comments are shown in italic
396 <!-- *********************************************************************** -->
397 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
398 <!-- *********************************************************************** -->
400 <!-- ======================================================================= -->
401 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
404 <div class="doc_text">
406 <p>LLVM programs are composed of "Module"s, each of which is a
407 translation unit of the input programs. Each module consists of
408 functions, global variables, and symbol table entries. Modules may be
409 combined together with the LLVM linker, which merges function (and
410 global variable) definitions, resolves forward declarations, and merges
411 symbol table entries. Here is an example of the "hello world" module:</p>
413 <div class="doc_code">
414 <pre><i>; Declare the string constant as a global constant...</i>
415 <a href="#identifiers">@.LC0</a> = <a href="#linkage_internal">internal</a> <a
416 href="#globalvars">constant</a> <a href="#t_array">[13 x i8]</a> c"hello world\0A\00" <i>; [13 x i8]*</i>
418 <i>; External declaration of the puts function</i>
419 <a href="#functionstructure">declare</a> i32 @puts(i8 *) <i>; i32(i8 *)* </i>
421 <i>; Definition of main function</i>
422 define i32 @main() { <i>; i32()* </i>
423 <i>; Convert [13x i8 ]* to i8 *...</i>
425 href="#i_getelementptr">getelementptr</a> [13 x i8 ]* @.LC0, i64 0, i64 0 <i>; i8 *</i>
427 <i>; Call puts function to write out the string to stdout...</i>
429 href="#i_call">call</a> i32 @puts(i8 * %cast210) <i>; i32</i>
431 href="#i_ret">ret</a> i32 0<br>}<br>
435 <p>This example is made up of a <a href="#globalvars">global variable</a>
436 named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
437 function, and a <a href="#functionstructure">function definition</a>
438 for "<tt>main</tt>".</p>
440 <p>In general, a module is made up of a list of global values,
441 where both functions and global variables are global values. Global values are
442 represented by a pointer to a memory location (in this case, a pointer to an
443 array of char, and a pointer to a function), and have one of the following <a
444 href="#linkage">linkage types</a>.</p>
448 <!-- ======================================================================= -->
449 <div class="doc_subsection">
450 <a name="linkage">Linkage Types</a>
453 <div class="doc_text">
456 All Global Variables and Functions have one of the following types of linkage:
461 <dt><tt><b><a name="linkage_internal">internal</a></b></tt> </dt>
463 <dd>Global values with internal linkage are only directly accessible by
464 objects in the current module. In particular, linking code into a module with
465 an internal global value may cause the internal to be renamed as necessary to
466 avoid collisions. Because the symbol is internal to the module, all
467 references can be updated. This corresponds to the notion of the
468 '<tt>static</tt>' keyword in C.
471 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
473 <dd>Globals with "<tt>linkonce</tt>" linkage are merged with other globals of
474 the same name when linkage occurs. This is typically used to implement
475 inline functions, templates, or other code which must be generated in each
476 translation unit that uses it. Unreferenced <tt>linkonce</tt> globals are
477 allowed to be discarded.
480 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
482 <dd>"<tt>weak</tt>" linkage is exactly the same as <tt>linkonce</tt> linkage,
483 except that unreferenced <tt>weak</tt> globals may not be discarded. This is
484 used for globals that may be emitted in multiple translation units, but that
485 are not guaranteed to be emitted into every translation unit that uses them.
486 One example of this are common globals in C, such as "<tt>int X;</tt>" at
490 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
492 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
493 pointer to array type. When two global variables with appending linkage are
494 linked together, the two global arrays are appended together. This is the
495 LLVM, typesafe, equivalent of having the system linker append together
496 "sections" with identical names when .o files are linked.
499 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt>: </dt>
500 <dd>The semantics of this linkage follow the ELF model: the symbol is weak
501 until linked, if not linked, the symbol becomes null instead of being an
505 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
507 <dd>If none of the above identifiers are used, the global is externally
508 visible, meaning that it participates in linkage and can be used to resolve
509 external symbol references.
514 The next two types of linkage are targeted for Microsoft Windows platform
515 only. They are designed to support importing (exporting) symbols from (to)
520 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt>: </dt>
522 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
523 or variable via a global pointer to a pointer that is set up by the DLL
524 exporting the symbol. On Microsoft Windows targets, the pointer name is
525 formed by combining <code>_imp__</code> and the function or variable name.
528 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt>: </dt>
530 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
531 pointer to a pointer in a DLL, so that it can be referenced with the
532 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
533 name is formed by combining <code>_imp__</code> and the function or variable
539 <p><a name="linkage_external"></a>For example, since the "<tt>.LC0</tt>"
540 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
541 variable and was linked with this one, one of the two would be renamed,
542 preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
543 external (i.e., lacking any linkage declarations), they are accessible
544 outside of the current module.</p>
545 <p>It is illegal for a function <i>declaration</i>
546 to have any linkage type other than "externally visible", <tt>dllimport</tt>,
547 or <tt>extern_weak</tt>.</p>
548 <p>Aliases can have only <tt>external</tt>, <tt>internal</tt> and <tt>weak</tt>
552 <!-- ======================================================================= -->
553 <div class="doc_subsection">
554 <a name="callingconv">Calling Conventions</a>
557 <div class="doc_text">
559 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
560 and <a href="#i_invoke">invokes</a> can all have an optional calling convention
561 specified for the call. The calling convention of any pair of dynamic
562 caller/callee must match, or the behavior of the program is undefined. The
563 following calling conventions are supported by LLVM, and more may be added in
567 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
569 <dd>This calling convention (the default if no other calling convention is
570 specified) matches the target C calling conventions. This calling convention
571 supports varargs function calls and tolerates some mismatch in the declared
572 prototype and implemented declaration of the function (as does normal C).
575 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
577 <dd>This calling convention attempts to make calls as fast as possible
578 (e.g. by passing things in registers). This calling convention allows the
579 target to use whatever tricks it wants to produce fast code for the target,
580 without having to conform to an externally specified ABI. Implementations of
581 this convention should allow arbitrary tail call optimization to be supported.
582 This calling convention does not support varargs and requires the prototype of
583 all callees to exactly match the prototype of the function definition.
586 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
588 <dd>This calling convention attempts to make code in the caller as efficient
589 as possible under the assumption that the call is not commonly executed. As
590 such, these calls often preserve all registers so that the call does not break
591 any live ranges in the caller side. This calling convention does not support
592 varargs and requires the prototype of all callees to exactly match the
593 prototype of the function definition.
596 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
598 <dd>Any calling convention may be specified by number, allowing
599 target-specific calling conventions to be used. Target specific calling
600 conventions start at 64.
604 <p>More calling conventions can be added/defined on an as-needed basis, to
605 support pascal conventions or any other well-known target-independent
610 <!-- ======================================================================= -->
611 <div class="doc_subsection">
612 <a name="visibility">Visibility Styles</a>
615 <div class="doc_text">
618 All Global Variables and Functions have one of the following visibility styles:
622 <dt><b>"<tt>default</tt>" - Default style</b>:</dt>
624 <dd>On ELF, default visibility means that the declaration is visible to other
625 modules and, in shared libraries, means that the declared entity may be
626 overridden. On Darwin, default visibility means that the declaration is
627 visible to other modules. Default visibility corresponds to "external
628 linkage" in the language.
631 <dt><b>"<tt>hidden</tt>" - Hidden style</b>:</dt>
633 <dd>Two declarations of an object with hidden visibility refer to the same
634 object if they are in the same shared object. Usually, hidden visibility
635 indicates that the symbol will not be placed into the dynamic symbol table,
636 so no other module (executable or shared library) can reference it
640 <dt><b>"<tt>protected</tt>" - Protected style</b>:</dt>
642 <dd>On ELF, protected visibility indicates that the symbol will be placed in
643 the dynamic symbol table, but that references within the defining module will
644 bind to the local symbol. That is, the symbol cannot be overridden by another
651 <!-- ======================================================================= -->
652 <div class="doc_subsection">
653 <a name="globalvars">Global Variables</a>
656 <div class="doc_text">
658 <p>Global variables define regions of memory allocated at compilation time
659 instead of run-time. Global variables may optionally be initialized, may have
660 an explicit section to be placed in, and may have an optional explicit alignment
661 specified. A variable may be defined as "thread_local", which means that it
662 will not be shared by threads (each thread will have a separated copy of the
663 variable). A variable may be defined as a global "constant," which indicates
664 that the contents of the variable will <b>never</b> be modified (enabling better
665 optimization, allowing the global data to be placed in the read-only section of
666 an executable, etc). Note that variables that need runtime initialization
667 cannot be marked "constant" as there is a store to the variable.</p>
670 LLVM explicitly allows <em>declarations</em> of global variables to be marked
671 constant, even if the final definition of the global is not. This capability
672 can be used to enable slightly better optimization of the program, but requires
673 the language definition to guarantee that optimizations based on the
674 'constantness' are valid for the translation units that do not include the
678 <p>As SSA values, global variables define pointer values that are in
679 scope (i.e. they dominate) all basic blocks in the program. Global
680 variables always define a pointer to their "content" type because they
681 describe a region of memory, and all memory objects in LLVM are
682 accessed through pointers.</p>
684 <p>A global variable may be declared to reside in a target-specifc numbered
685 address space. For targets that support them, address spaces may affect how
686 optimizations are performed and/or what target instructions are used to access
687 the variable. The default address space is zero. The address space qualifier
688 must precede any other attributes.</p>
690 <p>LLVM allows an explicit section to be specified for globals. If the target
691 supports it, it will emit globals to the section specified.</p>
693 <p>An explicit alignment may be specified for a global. If not present, or if
694 the alignment is set to zero, the alignment of the global is set by the target
695 to whatever it feels convenient. If an explicit alignment is specified, the
696 global is forced to have at least that much alignment. All alignments must be
699 <p>For example, the following defines a global in a numbered address space with
700 an initializer, section, and alignment:</p>
702 <div class="doc_code">
704 @G = constant float 1.0 addrspace(5), section "foo", align 4
711 <!-- ======================================================================= -->
712 <div class="doc_subsection">
713 <a name="functionstructure">Functions</a>
716 <div class="doc_text">
718 <p>LLVM function definitions consist of the "<tt>define</tt>" keyord,
719 an optional <a href="#linkage">linkage type</a>, an optional
720 <a href="#visibility">visibility style</a>, an optional
721 <a href="#callingconv">calling convention</a>, a return type, an optional
722 <a href="#paramattrs">parameter attribute</a> for the return type, a function
723 name, a (possibly empty) argument list (each with optional
724 <a href="#paramattrs">parameter attributes</a>), an optional section, an
725 optional alignment, an optional <a href="#gc">garbage collector name</a>, an
726 opening curly brace, a list of basic blocks, and a closing curly brace.
728 LLVM function declarations consist of the "<tt>declare</tt>" keyword, an
729 optional <a href="#linkage">linkage type</a>, an optional
730 <a href="#visibility">visibility style</a>, an optional
731 <a href="#callingconv">calling convention</a>, a return type, an optional
732 <a href="#paramattrs">parameter attribute</a> for the return type, a function
733 name, a possibly empty list of arguments, an optional alignment, and an optional
734 <a href="#gc">garbage collector name</a>.</p>
736 <p>A function definition contains a list of basic blocks, forming the CFG for
737 the function. Each basic block may optionally start with a label (giving the
738 basic block a symbol table entry), contains a list of instructions, and ends
739 with a <a href="#terminators">terminator</a> instruction (such as a branch or
740 function return).</p>
742 <p>The first basic block in a function is special in two ways: it is immediately
743 executed on entrance to the function, and it is not allowed to have predecessor
744 basic blocks (i.e. there can not be any branches to the entry block of a
745 function). Because the block can have no predecessors, it also cannot have any
746 <a href="#i_phi">PHI nodes</a>.</p>
748 <p>LLVM allows an explicit section to be specified for functions. If the target
749 supports it, it will emit functions to the section specified.</p>
751 <p>An explicit alignment may be specified for a function. If not present, or if
752 the alignment is set to zero, the alignment of the function is set by the target
753 to whatever it feels convenient. If an explicit alignment is specified, the
754 function is forced to have at least that much alignment. All alignments must be
760 <!-- ======================================================================= -->
761 <div class="doc_subsection">
762 <a name="aliasstructure">Aliases</a>
764 <div class="doc_text">
765 <p>Aliases act as "second name" for the aliasee value (which can be either
766 function, global variable, another alias or bitcast of global value). Aliases
767 may have an optional <a href="#linkage">linkage type</a>, and an
768 optional <a href="#visibility">visibility style</a>.</p>
772 <div class="doc_code">
774 @<Name> = [Linkage] [Visibility] alias <AliaseeTy> @<Aliasee>
782 <!-- ======================================================================= -->
783 <div class="doc_subsection"><a name="paramattrs">Parameter Attributes</a></div>
784 <div class="doc_text">
785 <p>The return type and each parameter of a function type may have a set of
786 <i>parameter attributes</i> associated with them. Parameter attributes are
787 used to communicate additional information about the result or parameters of
788 a function. Parameter attributes are considered to be part of the function,
789 not of the function type, so functions with different parameter attributes
790 can have the same function type.</p>
792 <p>Parameter attributes are simple keywords that follow the type specified. If
793 multiple parameter attributes are needed, they are space separated. For
796 <div class="doc_code">
798 declare i32 @printf(i8* noalias , ...) nounwind
799 declare i32 @atoi(i8*) nounwind readonly
803 <p>Note that any attributes for the function result (<tt>nounwind</tt>,
804 <tt>readonly</tt>) come immediately after the argument list.</p>
806 <p>Currently, only the following parameter attributes are defined:</p>
808 <dt><tt>zeroext</tt></dt>
809 <dd>This indicates that the parameter should be zero extended just before
810 a call to this function.</dd>
812 <dt><tt>signext</tt></dt>
813 <dd>This indicates that the parameter should be sign extended just before
814 a call to this function.</dd>
816 <dt><tt>inreg</tt></dt>
817 <dd>This indicates that the parameter should be placed in register (if
818 possible) during assembling function call. Support for this attribute is
821 <dt><tt>byval</tt></dt>
822 <dd>This indicates that the pointer parameter should really be passed by
823 value to the function. The attribute implies that a hidden copy of the
824 pointee is made between the caller and the callee, so the callee is unable
825 to modify the value in the callee. This attribute is only valid on llvm
826 pointer arguments. It is generally used to pass structs and arrays by
827 value, but is also valid on scalars (even though this is silly).</dd>
829 <dt><tt>sret</tt></dt>
830 <dd>This indicates that the pointer parameter specifies the address of a
831 structure that is the return value of the function in the source program.
832 Loads and stores to the structure are assumed not to trap.
833 May only be applied to the first parameter.</dd>
835 <dt><tt>noalias</tt></dt>
836 <dd>This indicates that the parameter does not alias any global or any other
837 parameter. The caller is responsible for ensuring that this is the case,
838 usually by placing the value in a stack allocation.</dd>
840 <dt><tt>noreturn</tt></dt>
841 <dd>This function attribute indicates that the function never returns. This
842 indicates to LLVM that every call to this function should be treated as if
843 an <tt>unreachable</tt> instruction immediately followed the call.</dd>
845 <dt><tt>nounwind</tt></dt>
846 <dd>This function attribute indicates that no exceptions unwind out of the
847 function. Usually this is because the function makes no use of exceptions,
848 but it may also be that the function catches any exceptions thrown when
851 <dt><tt>nest</tt></dt>
852 <dd>This indicates that the parameter can be excised using the
853 <a href="#int_trampoline">trampoline intrinsics</a>.</dd>
854 <dt><tt>readonly</tt></dt>
855 <dd>This function attribute indicates that the function has no side-effects
856 except for producing a return value or throwing an exception. The value
857 returned must only depend on the function arguments and/or global variables.
858 It may use values obtained by dereferencing pointers.</dd>
859 <dt><tt>readnone</tt></dt>
860 <dd>A <tt>readnone</tt> function has the same restrictions as a <tt>readonly</tt>
861 function, but in addition it is not allowed to dereference any pointer arguments
867 <!-- ======================================================================= -->
868 <div class="doc_subsection">
869 <a name="gc">Garbage Collector Names</a>
872 <div class="doc_text">
873 <p>Each function may specify a garbage collector name, which is simply a
876 <div class="doc_code"><pre
877 >define void @f() gc "name" { ...</pre></div>
879 <p>The compiler declares the supported values of <i>name</i>. Specifying a
880 collector which will cause the compiler to alter its output in order to support
881 the named garbage collection algorithm.</p>
884 <!-- ======================================================================= -->
885 <div class="doc_subsection">
886 <a name="moduleasm">Module-Level Inline Assembly</a>
889 <div class="doc_text">
891 Modules may contain "module-level inline asm" blocks, which corresponds to the
892 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
893 LLVM and treated as a single unit, but may be separated in the .ll file if
894 desired. The syntax is very simple:
897 <div class="doc_code">
899 module asm "inline asm code goes here"
900 module asm "more can go here"
904 <p>The strings can contain any character by escaping non-printable characters.
905 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
910 The inline asm code is simply printed to the machine code .s file when
911 assembly code is generated.
915 <!-- ======================================================================= -->
916 <div class="doc_subsection">
917 <a name="datalayout">Data Layout</a>
920 <div class="doc_text">
921 <p>A module may specify a target specific data layout string that specifies how
922 data is to be laid out in memory. The syntax for the data layout is simply:</p>
923 <pre> target datalayout = "<i>layout specification</i>"</pre>
924 <p>The <i>layout specification</i> consists of a list of specifications
925 separated by the minus sign character ('-'). Each specification starts with a
926 letter and may include other information after the letter to define some
927 aspect of the data layout. The specifications accepted are as follows: </p>
930 <dd>Specifies that the target lays out data in big-endian form. That is, the
931 bits with the most significance have the lowest address location.</dd>
933 <dd>Specifies that hte target lays out data in little-endian form. That is,
934 the bits with the least significance have the lowest address location.</dd>
935 <dt><tt>p:<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
936 <dd>This specifies the <i>size</i> of a pointer and its <i>abi</i> and
937 <i>preferred</i> alignments. All sizes are in bits. Specifying the <i>pref</i>
938 alignment is optional. If omitted, the preceding <tt>:</tt> should be omitted
940 <dt><tt>i<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
941 <dd>This specifies the alignment for an integer type of a given bit
942 <i>size</i>. The value of <i>size</i> must be in the range [1,2^23).</dd>
943 <dt><tt>v<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
944 <dd>This specifies the alignment for a vector type of a given bit
946 <dt><tt>f<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
947 <dd>This specifies the alignment for a floating point type of a given bit
948 <i>size</i>. The value of <i>size</i> must be either 32 (float) or 64
950 <dt><tt>a<i>size</i>:<i>abi</i>:<i>pref</i></tt></dt>
951 <dd>This specifies the alignment for an aggregate type of a given bit
954 <p>When constructing the data layout for a given target, LLVM starts with a
955 default set of specifications which are then (possibly) overriden by the
956 specifications in the <tt>datalayout</tt> keyword. The default specifications
957 are given in this list:</p>
959 <li><tt>E</tt> - big endian</li>
960 <li><tt>p:32:64:64</tt> - 32-bit pointers with 64-bit alignment</li>
961 <li><tt>i1:8:8</tt> - i1 is 8-bit (byte) aligned</li>
962 <li><tt>i8:8:8</tt> - i8 is 8-bit (byte) aligned</li>
963 <li><tt>i16:16:16</tt> - i16 is 16-bit aligned</li>
964 <li><tt>i32:32:32</tt> - i32 is 32-bit aligned</li>
965 <li><tt>i64:32:64</tt> - i64 has abi alignment of 32-bits but preferred
966 alignment of 64-bits</li>
967 <li><tt>f32:32:32</tt> - float is 32-bit aligned</li>
968 <li><tt>f64:64:64</tt> - double is 64-bit aligned</li>
969 <li><tt>v64:64:64</tt> - 64-bit vector is 64-bit aligned</li>
970 <li><tt>v128:128:128</tt> - 128-bit vector is 128-bit aligned</li>
971 <li><tt>a0:0:1</tt> - aggregates are 8-bit aligned</li>
973 <p>When llvm is determining the alignment for a given type, it uses the
976 <li>If the type sought is an exact match for one of the specifications, that
977 specification is used.</li>
978 <li>If no match is found, and the type sought is an integer type, then the
979 smallest integer type that is larger than the bitwidth of the sought type is
980 used. If none of the specifications are larger than the bitwidth then the the
981 largest integer type is used. For example, given the default specifications
982 above, the i7 type will use the alignment of i8 (next largest) while both
983 i65 and i256 will use the alignment of i64 (largest specified).</li>
984 <li>If no match is found, and the type sought is a vector type, then the
985 largest vector type that is smaller than the sought vector type will be used
986 as a fall back. This happens because <128 x double> can be implemented in
987 terms of 64 <2 x double>, for example.</li>
991 <!-- *********************************************************************** -->
992 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
993 <!-- *********************************************************************** -->
995 <div class="doc_text">
997 <p>The LLVM type system is one of the most important features of the
998 intermediate representation. Being typed enables a number of
999 optimizations to be performed on the IR directly, without having to do
1000 extra analyses on the side before the transformation. A strong type
1001 system makes it easier to read the generated code and enables novel
1002 analyses and transformations that are not feasible to perform on normal
1003 three address code representations.</p>
1007 <!-- ======================================================================= -->
1008 <div class="doc_subsection"> <a name="t_classifications">Type
1009 Classifications</a> </div>
1010 <div class="doc_text">
1011 <p>The types fall into a few useful
1012 classifications:</p>
1014 <table border="1" cellspacing="0" cellpadding="4">
1016 <tr><th>Classification</th><th>Types</th></tr>
1018 <td><a href="#t_integer">integer</a></td>
1019 <td><tt>i1, i2, i3, ... i8, ... i16, ... i32, ... i64, ... </tt></td>
1022 <td><a href="#t_floating">floating point</a></td>
1023 <td><tt>float, double, x86_fp80, fp128, ppc_fp128</tt></td>
1026 <td><a name="t_firstclass">first class</a></td>
1027 <td><a href="#t_integer">integer</a>,
1028 <a href="#t_floating">floating point</a>,
1029 <a href="#t_pointer">pointer</a>,
1030 <a href="#t_vector">vector</a>
1034 <td><a href="#t_primitive">primitive</a></td>
1035 <td><a href="#t_label">label</a>,
1036 <a href="#t_void">void</a>,
1037 <a href="#t_integer">integer</a>,
1038 <a href="#t_floating">floating point</a>.</td>
1041 <td><a href="#t_derived">derived</a></td>
1042 <td><a href="#t_integer">integer</a>,
1043 <a href="#t_array">array</a>,
1044 <a href="#t_function">function</a>,
1045 <a href="#t_pointer">pointer</a>,
1046 <a href="#t_struct">structure</a>,
1047 <a href="#t_pstruct">packed structure</a>,
1048 <a href="#t_vector">vector</a>,
1049 <a href="#t_opaque">opaque</a>.
1054 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
1055 most important. Values of these types are the only ones which can be
1056 produced by instructions, passed as arguments, or used as operands to
1057 instructions. This means that all structures and arrays must be
1058 manipulated either by pointer or by component.</p>
1061 <!-- ======================================================================= -->
1062 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
1064 <div class="doc_text">
1065 <p>The primitive types are the fundamental building blocks of the LLVM
1070 <!-- _______________________________________________________________________ -->
1071 <div class="doc_subsubsection"> <a name="t_floating">Floating Point Types</a> </div>
1073 <div class="doc_text">
1076 <tr><th>Type</th><th>Description</th></tr>
1077 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
1078 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
1079 <tr><td><tt>fp128</tt></td><td>128-bit floating point value (112-bit mantissa)</td></tr>
1080 <tr><td><tt>x86_fp80</tt></td><td>80-bit floating point value (X87)</td></tr>
1081 <tr><td><tt>ppc_fp128</tt></td><td>128-bit floating point value (two 64-bits)</td></tr>
1086 <!-- _______________________________________________________________________ -->
1087 <div class="doc_subsubsection"> <a name="t_void">Void Type</a> </div>
1089 <div class="doc_text">
1091 <p>The void type does not represent any value and has no size.</p>
1100 <!-- _______________________________________________________________________ -->
1101 <div class="doc_subsubsection"> <a name="t_label">Label Type</a> </div>
1103 <div class="doc_text">
1105 <p>The label type represents code labels.</p>
1115 <!-- ======================================================================= -->
1116 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
1118 <div class="doc_text">
1120 <p>The real power in LLVM comes from the derived types in the system.
1121 This is what allows a programmer to represent arrays, functions,
1122 pointers, and other useful types. Note that these derived types may be
1123 recursive: For example, it is possible to have a two dimensional array.</p>
1127 <!-- _______________________________________________________________________ -->
1128 <div class="doc_subsubsection"> <a name="t_integer">Integer Type</a> </div>
1130 <div class="doc_text">
1133 <p>The integer type is a very simple derived type that simply specifies an
1134 arbitrary bit width for the integer type desired. Any bit width from 1 bit to
1135 2^23-1 (about 8 million) can be specified.</p>
1143 <p>The number of bits the integer will occupy is specified by the <tt>N</tt>
1147 <table class="layout">
1150 <td><tt>i1</tt></td>
1151 <td>a single-bit integer.</td>
1153 <td><tt>i32</tt></td>
1154 <td>a 32-bit integer.</td>
1156 <td><tt>i1942652</tt></td>
1157 <td>a really big integer of over 1 million bits.</td>
1163 <!-- _______________________________________________________________________ -->
1164 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
1166 <div class="doc_text">
1170 <p>The array type is a very simple derived type that arranges elements
1171 sequentially in memory. The array type requires a size (number of
1172 elements) and an underlying data type.</p>
1177 [<# elements> x <elementtype>]
1180 <p>The number of elements is a constant integer value; elementtype may
1181 be any type with a size.</p>
1184 <table class="layout">
1186 <td class="left"><tt>[40 x i32]</tt></td>
1187 <td class="left">Array of 40 32-bit integer values.</td>
1190 <td class="left"><tt>[41 x i32]</tt></td>
1191 <td class="left">Array of 41 32-bit integer values.</td>
1194 <td class="left"><tt>[4 x i8]</tt></td>
1195 <td class="left">Array of 4 8-bit integer values.</td>
1198 <p>Here are some examples of multidimensional arrays:</p>
1199 <table class="layout">
1201 <td class="left"><tt>[3 x [4 x i32]]</tt></td>
1202 <td class="left">3x4 array of 32-bit integer values.</td>
1205 <td class="left"><tt>[12 x [10 x float]]</tt></td>
1206 <td class="left">12x10 array of single precision floating point values.</td>
1209 <td class="left"><tt>[2 x [3 x [4 x i16]]]</tt></td>
1210 <td class="left">2x3x4 array of 16-bit integer values.</td>
1214 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
1215 length array. Normally, accesses past the end of an array are undefined in
1216 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
1217 As a special case, however, zero length arrays are recognized to be variable
1218 length. This allows implementation of 'pascal style arrays' with the LLVM
1219 type "{ i32, [0 x float]}", for example.</p>
1223 <!-- _______________________________________________________________________ -->
1224 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
1225 <div class="doc_text">
1227 <p>The function type can be thought of as a function signature. It
1228 consists of a return type and a list of formal parameter types. The
1229 return type of a function type is a scalar type or a void type or a struct type.
1230 If the return type is a struct type then all struct elements must be of first
1231 class types. Function types are usually used to build virtual function tables
1232 (which are structures of pointers to functions), for indirect function
1233 calls, and when defining a function.</p>
1236 <pre> <returntype list> (<parameter list>)<br></pre>
1237 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
1238 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
1239 which indicates that the function takes a variable number of arguments.
1240 Variable argument functions can access their arguments with the <a
1241 href="#int_varargs">variable argument handling intrinsic</a> functions.
1242 '<tt><returntype list></tt>' is a comma-separated list of
1243 <a href="#t_firstclass">first class</a> type specifiers.</p>
1245 <table class="layout">
1247 <td class="left"><tt>i32 (i32)</tt></td>
1248 <td class="left">function taking an <tt>i32</tt>, returning an <tt>i32</tt>
1250 </tr><tr class="layout">
1251 <td class="left"><tt>float (i16 signext, i32 *) *
1253 <td class="left"><a href="#t_pointer">Pointer</a> to a function that takes
1254 an <tt>i16</tt> that should be sign extended and a
1255 <a href="#t_pointer">pointer</a> to <tt>i32</tt>, returning
1258 </tr><tr class="layout">
1259 <td class="left"><tt>i32 (i8*, ...)</tt></td>
1260 <td class="left">A vararg function that takes at least one
1261 <a href="#t_pointer">pointer</a> to <tt>i8 </tt> (char in C),
1262 which returns an integer. This is the signature for <tt>printf</tt> in
1265 </tr><tr class="layout">
1266 <td class="left"><tt>{i32, i32} (i32)</tt></td>
1267 <td class="left">A function taking an <tt>i32></tt>, returning two
1268 <tt> i32 </tt> values as an aggregate of type <tt>{ i32, i32 }</tt>
1274 <!-- _______________________________________________________________________ -->
1275 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
1276 <div class="doc_text">
1278 <p>The structure type is used to represent a collection of data members
1279 together in memory. The packing of the field types is defined to match
1280 the ABI of the underlying processor. The elements of a structure may
1281 be any type that has a size.</p>
1282 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1283 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1284 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1287 <pre> { <type list> }<br></pre>
1289 <table class="layout">
1291 <td class="left"><tt>{ i32, i32, i32 }</tt></td>
1292 <td class="left">A triple of three <tt>i32</tt> values</td>
1293 </tr><tr class="layout">
1294 <td class="left"><tt>{ float, i32 (i32) * }</tt></td>
1295 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1296 second element is a <a href="#t_pointer">pointer</a> to a
1297 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1298 an <tt>i32</tt>.</td>
1303 <!-- _______________________________________________________________________ -->
1304 <div class="doc_subsubsection"> <a name="t_pstruct">Packed Structure Type</a>
1306 <div class="doc_text">
1308 <p>The packed structure type is used to represent a collection of data members
1309 together in memory. There is no padding between fields. Further, the alignment
1310 of a packed structure is 1 byte. The elements of a packed structure may
1311 be any type that has a size.</p>
1312 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
1313 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
1314 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
1317 <pre> < { <type list> } > <br></pre>
1319 <table class="layout">
1321 <td class="left"><tt>< { i32, i32, i32 } ></tt></td>
1322 <td class="left">A triple of three <tt>i32</tt> values</td>
1323 </tr><tr class="layout">
1324 <td class="left"><tt>< { float, i32 (i32)* } ></tt></td>
1325 <td class="left">A pair, where the first element is a <tt>float</tt> and the
1326 second element is a <a href="#t_pointer">pointer</a> to a
1327 <a href="#t_function">function</a> that takes an <tt>i32</tt>, returning
1328 an <tt>i32</tt>.</td>
1333 <!-- _______________________________________________________________________ -->
1334 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
1335 <div class="doc_text">
1337 <p>As in many languages, the pointer type represents a pointer or
1338 reference to another object, which must live in memory. Pointer types may have
1339 an optional address space attribute defining the target-specific numbered
1340 address space where the pointed-to object resides. The default address space is
1343 <pre> <type> *<br></pre>
1345 <table class="layout">
1347 <td class="left"><tt>[4x i32]*</tt></td>
1348 <td class="left">A <a href="#t_pointer">pointer</a> to <a
1349 href="#t_array">array</a> of four <tt>i32</tt> values.</td>
1352 <td class="left"><tt>i32 (i32 *) *</tt></td>
1353 <td class="left"> A <a href="#t_pointer">pointer</a> to a <a
1354 href="#t_function">function</a> that takes an <tt>i32*</tt>, returning an
1358 <td class="left"><tt>i32 addrspace(5)*</tt></td>
1359 <td class="left">A <a href="#t_pointer">pointer</a> to an <tt>i32</tt> value
1360 that resides in address space #5.</td>
1365 <!-- _______________________________________________________________________ -->
1366 <div class="doc_subsubsection"> <a name="t_vector">Vector Type</a> </div>
1367 <div class="doc_text">
1371 <p>A vector type is a simple derived type that represents a vector
1372 of elements. Vector types are used when multiple primitive data
1373 are operated in parallel using a single instruction (SIMD).
1374 A vector type requires a size (number of
1375 elements) and an underlying primitive data type. Vectors must have a power
1376 of two length (1, 2, 4, 8, 16 ...). Vector types are
1377 considered <a href="#t_firstclass">first class</a>.</p>
1382 < <# elements> x <elementtype> >
1385 <p>The number of elements is a constant integer value; elementtype may
1386 be any integer or floating point type.</p>
1390 <table class="layout">
1392 <td class="left"><tt><4 x i32></tt></td>
1393 <td class="left">Vector of 4 32-bit integer values.</td>
1396 <td class="left"><tt><8 x float></tt></td>
1397 <td class="left">Vector of 8 32-bit floating-point values.</td>
1400 <td class="left"><tt><2 x i64></tt></td>
1401 <td class="left">Vector of 2 64-bit integer values.</td>
1406 <!-- _______________________________________________________________________ -->
1407 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
1408 <div class="doc_text">
1412 <p>Opaque types are used to represent unknown types in the system. This
1413 corresponds (for example) to the C notion of a forward declared structure type.
1414 In LLVM, opaque types can eventually be resolved to any type (not just a
1415 structure type).</p>
1425 <table class="layout">
1427 <td class="left"><tt>opaque</tt></td>
1428 <td class="left">An opaque type.</td>
1434 <!-- *********************************************************************** -->
1435 <div class="doc_section"> <a name="constants">Constants</a> </div>
1436 <!-- *********************************************************************** -->
1438 <div class="doc_text">
1440 <p>LLVM has several different basic types of constants. This section describes
1441 them all and their syntax.</p>
1445 <!-- ======================================================================= -->
1446 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1448 <div class="doc_text">
1451 <dt><b>Boolean constants</b></dt>
1453 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1454 constants of the <tt><a href="#t_primitive">i1</a></tt> type.
1457 <dt><b>Integer constants</b></dt>
1459 <dd>Standard integers (such as '4') are constants of the <a
1460 href="#t_integer">integer</a> type. Negative numbers may be used with
1464 <dt><b>Floating point constants</b></dt>
1466 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1467 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1468 notation (see below). The assembler requires the exact decimal value of
1469 a floating-point constant. For example, the assembler accepts 1.25 but
1470 rejects 1.3 because 1.3 is a repeating decimal in binary. Floating point
1471 constants must have a <a href="#t_floating">floating point</a> type. </dd>
1473 <dt><b>Null pointer constants</b></dt>
1475 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1476 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1480 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1481 of floating point constants. For example, the form '<tt>double
1482 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1483 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1484 (and the only time that they are generated by the disassembler) is when a
1485 floating point constant must be emitted but it cannot be represented as a
1486 decimal floating point number. For example, NaN's, infinities, and other
1487 special values are represented in their IEEE hexadecimal format so that
1488 assembly and disassembly do not cause any bits to change in the constants.</p>
1492 <!-- ======================================================================= -->
1493 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1496 <div class="doc_text">
1497 <p>Aggregate constants arise from aggregation of simple constants
1498 and smaller aggregate constants.</p>
1501 <dt><b>Structure constants</b></dt>
1503 <dd>Structure constants are represented with notation similar to structure
1504 type definitions (a comma separated list of elements, surrounded by braces
1505 (<tt>{}</tt>)). For example: "<tt>{ i32 4, float 17.0, i32* @G }</tt>",
1506 where "<tt>@G</tt>" is declared as "<tt>@G = external global i32</tt>". Structure constants
1507 must have <a href="#t_struct">structure type</a>, and the number and
1508 types of elements must match those specified by the type.
1511 <dt><b>Array constants</b></dt>
1513 <dd>Array constants are represented with notation similar to array type
1514 definitions (a comma separated list of elements, surrounded by square brackets
1515 (<tt>[]</tt>)). For example: "<tt>[ i32 42, i32 11, i32 74 ]</tt>". Array
1516 constants must have <a href="#t_array">array type</a>, and the number and
1517 types of elements must match those specified by the type.
1520 <dt><b>Vector constants</b></dt>
1522 <dd>Vector constants are represented with notation similar to vector type
1523 definitions (a comma separated list of elements, surrounded by
1524 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< i32 42,
1525 i32 11, i32 74, i32 100 ></tt>". Vector constants must have <a
1526 href="#t_vector">vector type</a>, and the number and types of elements must
1527 match those specified by the type.
1530 <dt><b>Zero initialization</b></dt>
1532 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1533 value to zero of <em>any</em> type, including scalar and aggregate types.
1534 This is often used to avoid having to print large zero initializers (e.g. for
1535 large arrays) and is always exactly equivalent to using explicit zero
1542 <!-- ======================================================================= -->
1543 <div class="doc_subsection">
1544 <a name="globalconstants">Global Variable and Function Addresses</a>
1547 <div class="doc_text">
1549 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1550 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1551 constants. These constants are explicitly referenced when the <a
1552 href="#identifiers">identifier for the global</a> is used and always have <a
1553 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1556 <div class="doc_code">
1560 @Z = global [2 x i32*] [ i32* @X, i32* @Y ]
1566 <!-- ======================================================================= -->
1567 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1568 <div class="doc_text">
1569 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1570 no specific value. Undefined values may be of any type and be used anywhere
1571 a constant is permitted.</p>
1573 <p>Undefined values indicate to the compiler that the program is well defined
1574 no matter what value is used, giving the compiler more freedom to optimize.
1578 <!-- ======================================================================= -->
1579 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1582 <div class="doc_text">
1584 <p>Constant expressions are used to allow expressions involving other constants
1585 to be used as constants. Constant expressions may be of any <a
1586 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1587 that does not have side effects (e.g. load and call are not supported). The
1588 following is the syntax for constant expressions:</p>
1591 <dt><b><tt>trunc ( CST to TYPE )</tt></b></dt>
1592 <dd>Truncate a constant to another type. The bit size of CST must be larger
1593 than the bit size of TYPE. Both types must be integers.</dd>
1595 <dt><b><tt>zext ( CST to TYPE )</tt></b></dt>
1596 <dd>Zero extend a constant to another type. The bit size of CST must be
1597 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1599 <dt><b><tt>sext ( CST to TYPE )</tt></b></dt>
1600 <dd>Sign extend a constant to another type. The bit size of CST must be
1601 smaller or equal to the bit size of TYPE. Both types must be integers.</dd>
1603 <dt><b><tt>fptrunc ( CST to TYPE )</tt></b></dt>
1604 <dd>Truncate a floating point constant to another floating point type. The
1605 size of CST must be larger than the size of TYPE. Both types must be
1606 floating point.</dd>
1608 <dt><b><tt>fpext ( CST to TYPE )</tt></b></dt>
1609 <dd>Floating point extend a constant to another type. The size of CST must be
1610 smaller or equal to the size of TYPE. Both types must be floating point.</dd>
1612 <dt><b><tt>fptoui ( CST to TYPE )</tt></b></dt>
1613 <dd>Convert a floating point constant to the corresponding unsigned integer
1614 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1615 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1616 of the same number of elements. If the value won't fit in the integer type,
1617 the results are undefined.</dd>
1619 <dt><b><tt>fptosi ( CST to TYPE )</tt></b></dt>
1620 <dd>Convert a floating point constant to the corresponding signed integer
1621 constant. TYPE must be a scalar or vector integer type. CST must be of scalar
1622 or vector floating point type. Both CST and TYPE must be scalars, or vectors
1623 of the same number of elements. If the value won't fit in the integer type,
1624 the results are undefined.</dd>
1626 <dt><b><tt>uitofp ( CST to TYPE )</tt></b></dt>
1627 <dd>Convert an unsigned integer constant to the corresponding floating point
1628 constant. TYPE must be a scalar or vector floating point type. CST must be of
1629 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1630 of the same number of elements. If the value won't fit in the floating point
1631 type, the results are undefined.</dd>
1633 <dt><b><tt>sitofp ( CST to TYPE )</tt></b></dt>
1634 <dd>Convert a signed integer constant to the corresponding floating point
1635 constant. TYPE must be a scalar or vector floating point type. CST must be of
1636 scalar or vector integer type. Both CST and TYPE must be scalars, or vectors
1637 of the same number of elements. If the value won't fit in the floating point
1638 type, the results are undefined.</dd>
1640 <dt><b><tt>ptrtoint ( CST to TYPE )</tt></b></dt>
1641 <dd>Convert a pointer typed constant to the corresponding integer constant
1642 TYPE must be an integer type. CST must be of pointer type. The CST value is
1643 zero extended, truncated, or unchanged to make it fit in TYPE.</dd>
1645 <dt><b><tt>inttoptr ( CST to TYPE )</tt></b></dt>
1646 <dd>Convert a integer constant to a pointer constant. TYPE must be a
1647 pointer type. CST must be of integer type. The CST value is zero extended,
1648 truncated, or unchanged to make it fit in a pointer size. This one is
1649 <i>really</i> dangerous!</dd>
1651 <dt><b><tt>bitcast ( CST to TYPE )</tt></b></dt>
1652 <dd>Convert a constant, CST, to another TYPE. The size of CST and TYPE must be
1653 identical (same number of bits). The conversion is done as if the CST value
1654 was stored to memory and read back as TYPE. In other words, no bits change
1655 with this operator, just the type. This can be used for conversion of
1656 vector types to any other type, as long as they have the same bit width. For
1657 pointers it is only valid to cast to another pointer type.
1660 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1662 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1663 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1664 instruction, the index list may have zero or more indexes, which are required
1665 to make sense for the type of "CSTPTR".</dd>
1667 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1669 <dd>Perform the <a href="#i_select">select operation</a> on
1672 <dt><b><tt>icmp COND ( VAL1, VAL2 )</tt></b></dt>
1673 <dd>Performs the <a href="#i_icmp">icmp operation</a> on constants.</dd>
1675 <dt><b><tt>fcmp COND ( VAL1, VAL2 )</tt></b></dt>
1676 <dd>Performs the <a href="#i_fcmp">fcmp operation</a> on constants.</dd>
1678 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1680 <dd>Perform the <a href="#i_extractelement">extractelement
1681 operation</a> on constants.
1683 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1685 <dd>Perform the <a href="#i_insertelement">insertelement
1686 operation</a> on constants.</dd>
1689 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
1691 <dd>Perform the <a href="#i_shufflevector">shufflevector
1692 operation</a> on constants.</dd>
1694 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1696 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1697 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1698 binary</a> operations. The constraints on operands are the same as those for
1699 the corresponding instruction (e.g. no bitwise operations on floating point
1700 values are allowed).</dd>
1704 <!-- *********************************************************************** -->
1705 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1706 <!-- *********************************************************************** -->
1708 <!-- ======================================================================= -->
1709 <div class="doc_subsection">
1710 <a name="inlineasm">Inline Assembler Expressions</a>
1713 <div class="doc_text">
1716 LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
1717 Module-Level Inline Assembly</a>) through the use of a special value. This
1718 value represents the inline assembler as a string (containing the instructions
1719 to emit), a list of operand constraints (stored as a string), and a flag that
1720 indicates whether or not the inline asm expression has side effects. An example
1721 inline assembler expression is:
1724 <div class="doc_code">
1726 i32 (i32) asm "bswap $0", "=r,r"
1731 Inline assembler expressions may <b>only</b> be used as the callee operand of
1732 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
1735 <div class="doc_code">
1737 %X = call i32 asm "<a href="#int_bswap">bswap</a> $0", "=r,r"(i32 %Y)
1742 Inline asms with side effects not visible in the constraint list must be marked
1743 as having side effects. This is done through the use of the
1744 '<tt>sideeffect</tt>' keyword, like so:
1747 <div class="doc_code">
1749 call void asm sideeffect "eieio", ""()
1753 <p>TODO: The format of the asm and constraints string still need to be
1754 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1755 need to be documented).
1760 <!-- *********************************************************************** -->
1761 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1762 <!-- *********************************************************************** -->
1764 <div class="doc_text">
1766 <p>The LLVM instruction set consists of several different
1767 classifications of instructions: <a href="#terminators">terminator
1768 instructions</a>, <a href="#binaryops">binary instructions</a>,
1769 <a href="#bitwiseops">bitwise binary instructions</a>, <a
1770 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1771 instructions</a>.</p>
1775 <!-- ======================================================================= -->
1776 <div class="doc_subsection"> <a name="terminators">Terminator
1777 Instructions</a> </div>
1779 <div class="doc_text">
1781 <p>As mentioned <a href="#functionstructure">previously</a>, every
1782 basic block in a program ends with a "Terminator" instruction, which
1783 indicates which block should be executed after the current block is
1784 finished. These terminator instructions typically yield a '<tt>void</tt>'
1785 value: they produce control flow, not values (the one exception being
1786 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1787 <p>There are six different terminator instructions: the '<a
1788 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1789 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1790 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1791 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1792 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1796 <!-- _______________________________________________________________________ -->
1797 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1798 Instruction</a> </div>
1799 <div class="doc_text">
1801 <pre> ret <type> <value> <i>; Return a value from a non-void function</i>
1802 ret void <i>; Return from void function</i>
1803 ret <type> <value>, <type> <value> <i>; Return two values from a non-void function </i>
1806 <p>The '<tt>ret</tt>' instruction is used to return control flow (and a
1807 value) from a function back to the caller.</p>
1808 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
1809 returns a value and then causes control flow, and one that just causes
1810 control flow to occur.</p>
1812 <p>The '<tt>ret</tt>' instruction may return one or multiple values. The
1813 type of each return value must be a '<a href="#t_firstclass">first class</a>'
1814 type. Note that a function is not <a href="#wellformed">well formed</a>
1815 if there exists a '<tt>ret</tt>' instruction inside of the function that
1816 returns values that do not match the return type of the function.</p>
1818 <p>When the '<tt>ret</tt>' instruction is executed, control flow
1819 returns back to the calling function's context. If the caller is a "<a
1820 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1821 the instruction after the call. If the caller was an "<a
1822 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1823 at the beginning of the "normal" destination block. If the instruction
1824 returns a value, that value shall set the call or invoke instruction's
1825 return value. If the instruction returns multiple values then these
1826 values can only be accessed through a '<a href="#i_getresult"><tt>getresult</tt>
1827 </a>' instruction.</p>
1829 <pre> ret i32 5 <i>; Return an integer value of 5</i>
1830 ret void <i>; Return from a void function</i>
1831 ret i32 4, i8 2 <i>; Return two values 4 and 2 </i>
1834 <!-- _______________________________________________________________________ -->
1835 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1836 <div class="doc_text">
1838 <pre> br i1 <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1841 <p>The '<tt>br</tt>' instruction is used to cause control flow to
1842 transfer to a different basic block in the current function. There are
1843 two forms of this instruction, corresponding to a conditional branch
1844 and an unconditional branch.</p>
1846 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1847 single '<tt>i1</tt>' value and two '<tt>label</tt>' values. The
1848 unconditional form of the '<tt>br</tt>' instruction takes a single
1849 '<tt>label</tt>' value as a target.</p>
1851 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>i1</tt>'
1852 argument is evaluated. If the value is <tt>true</tt>, control flows
1853 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1854 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1856 <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
1857 href="#i_ret">ret</a> i32 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> i32 0<br></pre>
1859 <!-- _______________________________________________________________________ -->
1860 <div class="doc_subsubsection">
1861 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
1864 <div class="doc_text">
1868 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
1873 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
1874 several different places. It is a generalization of the '<tt>br</tt>'
1875 instruction, allowing a branch to occur to one of many possible
1881 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
1882 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
1883 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
1884 table is not allowed to contain duplicate constant entries.</p>
1888 <p>The <tt>switch</tt> instruction specifies a table of values and
1889 destinations. When the '<tt>switch</tt>' instruction is executed, this
1890 table is searched for the given value. If the value is found, control flow is
1891 transfered to the corresponding destination; otherwise, control flow is
1892 transfered to the default destination.</p>
1894 <h5>Implementation:</h5>
1896 <p>Depending on properties of the target machine and the particular
1897 <tt>switch</tt> instruction, this instruction may be code generated in different
1898 ways. For example, it could be generated as a series of chained conditional
1899 branches or with a lookup table.</p>
1904 <i>; Emulate a conditional br instruction</i>
1905 %Val = <a href="#i_zext">zext</a> i1 %value to i32
1906 switch i32 %Val, label %truedest [i32 0, label %falsedest ]
1908 <i>; Emulate an unconditional br instruction</i>
1909 switch i32 0, label %dest [ ]
1911 <i>; Implement a jump table:</i>
1912 switch i32 %val, label %otherwise [ i32 0, label %onzero
1914 i32 2, label %ontwo ]
1918 <!-- _______________________________________________________________________ -->
1919 <div class="doc_subsubsection">
1920 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
1923 <div class="doc_text">
1928 <result> = invoke [<a href="#callingconv">cconv</a>] <ptr to function ty> <function ptr val>(<function args>)
1929 to label <normal label> unwind label <exception label>
1934 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
1935 function, with the possibility of control flow transfer to either the
1936 '<tt>normal</tt>' label or the
1937 '<tt>exception</tt>' label. If the callee function returns with the
1938 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
1939 "normal" label. If the callee (or any indirect callees) returns with the "<a
1940 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
1941 continued at the dynamically nearest "exception" label. If the callee function
1942 returns multiple values then individual return values are only accessible through
1943 a '<tt><a href="#i_getresult">getresult</a></tt>' instruction.</p>
1947 <p>This instruction requires several arguments:</p>
1951 The optional "cconv" marker indicates which <a href="#callingconv">calling
1952 convention</a> the call should use. If none is specified, the call defaults
1953 to using C calling conventions.
1955 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
1956 function value being invoked. In most cases, this is a direct function
1957 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
1958 an arbitrary pointer to function value.
1961 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
1962 function to be invoked. </li>
1964 <li>'<tt>function args</tt>': argument list whose types match the function
1965 signature argument types. If the function signature indicates the function
1966 accepts a variable number of arguments, the extra arguments can be
1969 <li>'<tt>normal label</tt>': the label reached when the called function
1970 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
1972 <li>'<tt>exception label</tt>': the label reached when a callee returns with
1973 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
1979 <p>This instruction is designed to operate as a standard '<tt><a
1980 href="#i_call">call</a></tt>' instruction in most regards. The primary
1981 difference is that it establishes an association with a label, which is used by
1982 the runtime library to unwind the stack.</p>
1984 <p>This instruction is used in languages with destructors to ensure that proper
1985 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
1986 exception. Additionally, this is important for implementation of
1987 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
1991 %retval = invoke i32 @Test(i32 15) to label %Continue
1992 unwind label %TestCleanup <i>; {i32}:retval set</i>
1993 %retval = invoke <a href="#callingconv">coldcc</a> i32 %Testfnptr(i32 15) to label %Continue
1994 unwind label %TestCleanup <i>; {i32}:retval set</i>
1999 <!-- _______________________________________________________________________ -->
2001 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
2002 Instruction</a> </div>
2004 <div class="doc_text">
2013 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
2014 at the first callee in the dynamic call stack which used an <a
2015 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
2016 primarily used to implement exception handling.</p>
2020 <p>The '<tt>unwind</tt>' intrinsic causes execution of the current function to
2021 immediately halt. The dynamic call stack is then searched for the first <a
2022 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
2023 execution continues at the "exceptional" destination block specified by the
2024 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
2025 dynamic call chain, undefined behavior results.</p>
2028 <!-- _______________________________________________________________________ -->
2030 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
2031 Instruction</a> </div>
2033 <div class="doc_text">
2042 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
2043 instruction is used to inform the optimizer that a particular portion of the
2044 code is not reachable. This can be used to indicate that the code after a
2045 no-return function cannot be reached, and other facts.</p>
2049 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
2054 <!-- ======================================================================= -->
2055 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
2056 <div class="doc_text">
2057 <p>Binary operators are used to do most of the computation in a
2058 program. They require two operands of the same type, execute an operation on them, and
2059 produce a single value. The operands might represent
2060 multiple data, as is the case with the <a href="#t_vector">vector</a> data type.
2061 The result value has the same type as its operands.</p>
2062 <p>There are several different binary operators:</p>
2064 <!-- _______________________________________________________________________ -->
2065 <div class="doc_subsubsection"> <a name="i_add">'<tt>add</tt>'
2066 Instruction</a> </div>
2067 <div class="doc_text">
2069 <pre> <result> = add <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2072 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
2074 <p>The two arguments to the '<tt>add</tt>' instruction must be either <a
2075 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a> values.
2076 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2077 Both arguments must have identical types.</p>
2079 <p>The value produced is the integer or floating point sum of the two
2081 <p>If an integer sum has unsigned overflow, the result returned is the
2082 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2084 <p>Because LLVM integers use a two's complement representation, this
2085 instruction is appropriate for both signed and unsigned integers.</p>
2087 <pre> <result> = add i32 4, %var <i>; yields {i32}:result = 4 + %var</i>
2090 <!-- _______________________________________________________________________ -->
2091 <div class="doc_subsubsection"> <a name="i_sub">'<tt>sub</tt>'
2092 Instruction</a> </div>
2093 <div class="doc_text">
2095 <pre> <result> = sub <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2098 <p>The '<tt>sub</tt>' instruction returns the difference of its two
2100 <p>Note that the '<tt>sub</tt>' instruction is used to represent the '<tt>neg</tt>'
2101 instruction present in most other intermediate representations.</p>
2103 <p>The two arguments to the '<tt>sub</tt>' instruction must be either <a
2104 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
2106 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2107 Both arguments must have identical types.</p>
2109 <p>The value produced is the integer or floating point difference of
2110 the two operands.</p>
2111 <p>If an integer difference has unsigned overflow, the result returned is the
2112 mathematical result modulo 2<sup>n</sup>, where n is the bit width of
2114 <p>Because LLVM integers use a two's complement representation, this
2115 instruction is appropriate for both signed and unsigned integers.</p>
2118 <result> = sub i32 4, %var <i>; yields {i32}:result = 4 - %var</i>
2119 <result> = sub i32 0, %val <i>; yields {i32}:result = -%var</i>
2122 <!-- _______________________________________________________________________ -->
2123 <div class="doc_subsubsection"> <a name="i_mul">'<tt>mul</tt>'
2124 Instruction</a> </div>
2125 <div class="doc_text">
2127 <pre> <result> = mul <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2130 <p>The '<tt>mul</tt>' instruction returns the product of its two
2133 <p>The two arguments to the '<tt>mul</tt>' instruction must be either <a
2134 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
2136 This instruction can also take <a href="#t_vector">vector</a> versions of the values.
2137 Both arguments must have identical types.</p>
2139 <p>The value produced is the integer or floating point product of the
2141 <p>If the result of an integer multiplication has unsigned overflow,
2142 the result returned is the mathematical result modulo
2143 2<sup>n</sup>, where n is the bit width of the result.</p>
2144 <p>Because LLVM integers use a two's complement representation, and the
2145 result is the same width as the operands, this instruction returns the
2146 correct result for both signed and unsigned integers. If a full product
2147 (e.g. <tt>i32</tt>x<tt>i32</tt>-><tt>i64</tt>) is needed, the operands
2148 should be sign-extended or zero-extended as appropriate to the
2149 width of the full product.</p>
2151 <pre> <result> = mul i32 4, %var <i>; yields {i32}:result = 4 * %var</i>
2154 <!-- _______________________________________________________________________ -->
2155 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
2157 <div class="doc_text">
2159 <pre> <result> = udiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2162 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two
2165 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
2166 <a href="#t_integer">integer</a> values. Both arguments must have identical
2167 types. This instruction can also take <a href="#t_vector">vector</a> versions
2168 of the values in which case the elements must be integers.</p>
2170 <p>The value produced is the unsigned integer quotient of the two operands.</p>
2171 <p>Note that unsigned integer division and signed integer division are distinct
2172 operations; for signed integer division, use '<tt>sdiv</tt>'.</p>
2173 <p>Division by zero leads to undefined behavior.</p>
2175 <pre> <result> = udiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2178 <!-- _______________________________________________________________________ -->
2179 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
2181 <div class="doc_text">
2183 <pre> <result> = sdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2186 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
2189 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
2190 <a href="#t_integer">integer</a> values. Both arguments must have identical
2191 types. This instruction can also take <a href="#t_vector">vector</a> versions
2192 of the values in which case the elements must be integers.</p>
2194 <p>The value produced is the signed integer quotient of the two operands rounded towards zero.</p>
2195 <p>Note that signed integer division and unsigned integer division are distinct
2196 operations; for unsigned integer division, use '<tt>udiv</tt>'.</p>
2197 <p>Division by zero leads to undefined behavior. Overflow also leads to
2198 undefined behavior; this is a rare case, but can occur, for example,
2199 by doing a 32-bit division of -2147483648 by -1.</p>
2201 <pre> <result> = sdiv i32 4, %var <i>; yields {i32}:result = 4 / %var</i>
2204 <!-- _______________________________________________________________________ -->
2205 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
2206 Instruction</a> </div>
2207 <div class="doc_text">
2209 <pre> <result> = fdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2212 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
2215 <p>The two arguments to the '<tt>fdiv</tt>' instruction must be
2216 <a href="#t_floating">floating point</a> values. Both arguments must have
2217 identical types. This instruction can also take <a href="#t_vector">vector</a>
2218 versions of floating point values.</p>
2220 <p>The value produced is the floating point quotient of the two operands.</p>
2222 <pre> <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
2225 <!-- _______________________________________________________________________ -->
2226 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
2228 <div class="doc_text">
2230 <pre> <result> = urem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2233 <p>The '<tt>urem</tt>' instruction returns the remainder from the
2234 unsigned division of its two arguments.</p>
2236 <p>The two arguments to the '<tt>urem</tt>' instruction must be
2237 <a href="#t_integer">integer</a> values. Both arguments must have identical
2238 types. This instruction can also take <a href="#t_vector">vector</a> versions
2239 of the values in which case the elements must be integers.</p>
2241 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
2242 This instruction always performs an unsigned division to get the remainder.</p>
2243 <p>Note that unsigned integer remainder and signed integer remainder are
2244 distinct operations; for signed integer remainder, use '<tt>srem</tt>'.</p>
2245 <p>Taking the remainder of a division by zero leads to undefined behavior.</p>
2247 <pre> <result> = urem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2251 <!-- _______________________________________________________________________ -->
2252 <div class="doc_subsubsection"> <a name="i_srem">'<tt>srem</tt>'
2253 Instruction</a> </div>
2254 <div class="doc_text">
2256 <pre> <result> = srem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2259 <p>The '<tt>srem</tt>' instruction returns the remainder from the
2260 signed division of its two operands. This instruction can also take
2261 <a href="#t_vector">vector</a> versions of the values in which case
2262 the elements must be integers.</p>
2265 <p>The two arguments to the '<tt>srem</tt>' instruction must be
2266 <a href="#t_integer">integer</a> values. Both arguments must have identical
2269 <p>This instruction returns the <i>remainder</i> of a division (where the result
2270 has the same sign as the dividend, <tt>var1</tt>), not the <i>modulo</i>
2271 operator (where the result has the same sign as the divisor, <tt>var2</tt>) of
2272 a value. For more information about the difference, see <a
2273 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
2274 Math Forum</a>. For a table of how this is implemented in various languages,
2275 please see <a href="http://en.wikipedia.org/wiki/Modulo_operation">
2276 Wikipedia: modulo operation</a>.</p>
2277 <p>Note that signed integer remainder and unsigned integer remainder are
2278 distinct operations; for unsigned integer remainder, use '<tt>urem</tt>'.</p>
2279 <p>Taking the remainder of a division by zero leads to undefined behavior.
2280 Overflow also leads to undefined behavior; this is a rare case, but can occur,
2281 for example, by taking the remainder of a 32-bit division of -2147483648 by -1.
2282 (The remainder doesn't actually overflow, but this rule lets srem be
2283 implemented using instructions that return both the result of the division
2284 and the remainder.)</p>
2286 <pre> <result> = srem i32 4, %var <i>; yields {i32}:result = 4 % %var</i>
2290 <!-- _______________________________________________________________________ -->
2291 <div class="doc_subsubsection"> <a name="i_frem">'<tt>frem</tt>'
2292 Instruction</a> </div>
2293 <div class="doc_text">
2295 <pre> <result> = frem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2298 <p>The '<tt>frem</tt>' instruction returns the remainder from the
2299 division of its two operands.</p>
2301 <p>The two arguments to the '<tt>frem</tt>' instruction must be
2302 <a href="#t_floating">floating point</a> values. Both arguments must have
2303 identical types. This instruction can also take <a href="#t_vector">vector</a>
2304 versions of floating point values.</p>
2306 <p>This instruction returns the <i>remainder</i> of a division.
2307 The remainder has the same sign as the dividend.</p>
2309 <pre> <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
2313 <!-- ======================================================================= -->
2314 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
2315 Operations</a> </div>
2316 <div class="doc_text">
2317 <p>Bitwise binary operators are used to do various forms of
2318 bit-twiddling in a program. They are generally very efficient
2319 instructions and can commonly be strength reduced from other
2320 instructions. They require two operands of the same type, execute an operation on them,
2321 and produce a single value. The resulting value is the same type as its operands.</p>
2324 <!-- _______________________________________________________________________ -->
2325 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
2326 Instruction</a> </div>
2327 <div class="doc_text">
2329 <pre> <result> = shl <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2334 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2335 the left a specified number of bits.</p>
2339 <p>Both arguments to the '<tt>shl</tt>' instruction must be the same <a
2340 href="#t_integer">integer</a> type.</p>
2344 <p>The value produced is <tt>var1</tt> * 2<sup><tt>var2</tt></sup> mod 2<sup>n</sup>,
2345 where n is the width of the result. If <tt>var2</tt> is (statically or dynamically) negative or
2346 equal to or larger than the number of bits in <tt>var1</tt>, the result is undefined.</p>
2348 <h5>Example:</h5><pre>
2349 <result> = shl i32 4, %var <i>; yields {i32}: 4 << %var</i>
2350 <result> = shl i32 4, 2 <i>; yields {i32}: 16</i>
2351 <result> = shl i32 1, 10 <i>; yields {i32}: 1024</i>
2352 <result> = shl i32 1, 32 <i>; undefined</i>
2355 <!-- _______________________________________________________________________ -->
2356 <div class="doc_subsubsection"> <a name="i_lshr">'<tt>lshr</tt>'
2357 Instruction</a> </div>
2358 <div class="doc_text">
2360 <pre> <result> = lshr <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2364 <p>The '<tt>lshr</tt>' instruction (logical shift right) returns the first
2365 operand shifted to the right a specified number of bits with zero fill.</p>
2368 <p>Both arguments to the '<tt>lshr</tt>' instruction must be the same
2369 <a href="#t_integer">integer</a> type.</p>
2373 <p>This instruction always performs a logical shift right operation. The most
2374 significant bits of the result will be filled with zero bits after the
2375 shift. If <tt>var2</tt> is (statically or dynamically) equal to or larger than
2376 the number of bits in <tt>var1</tt>, the result is undefined.</p>
2380 <result> = lshr i32 4, 1 <i>; yields {i32}:result = 2</i>
2381 <result> = lshr i32 4, 2 <i>; yields {i32}:result = 1</i>
2382 <result> = lshr i8 4, 3 <i>; yields {i8}:result = 0</i>
2383 <result> = lshr i8 -2, 1 <i>; yields {i8}:result = 0x7FFFFFFF </i>
2384 <result> = lshr i32 1, 32 <i>; undefined</i>
2388 <!-- _______________________________________________________________________ -->
2389 <div class="doc_subsubsection"> <a name="i_ashr">'<tt>ashr</tt>'
2390 Instruction</a> </div>
2391 <div class="doc_text">
2394 <pre> <result> = ashr <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2398 <p>The '<tt>ashr</tt>' instruction (arithmetic shift right) returns the first
2399 operand shifted to the right a specified number of bits with sign extension.</p>
2402 <p>Both arguments to the '<tt>ashr</tt>' instruction must be the same
2403 <a href="#t_integer">integer</a> type.</p>
2406 <p>This instruction always performs an arithmetic shift right operation,
2407 The most significant bits of the result will be filled with the sign bit
2408 of <tt>var1</tt>. If <tt>var2</tt> is (statically or dynamically) equal to or
2409 larger than the number of bits in <tt>var1</tt>, the result is undefined.
2414 <result> = ashr i32 4, 1 <i>; yields {i32}:result = 2</i>
2415 <result> = ashr i32 4, 2 <i>; yields {i32}:result = 1</i>
2416 <result> = ashr i8 4, 3 <i>; yields {i8}:result = 0</i>
2417 <result> = ashr i8 -2, 1 <i>; yields {i8}:result = -1</i>
2418 <result> = ashr i32 1, 32 <i>; undefined</i>
2422 <!-- _______________________________________________________________________ -->
2423 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
2424 Instruction</a> </div>
2425 <div class="doc_text">
2427 <pre> <result> = and <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2430 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
2431 its two operands.</p>
2433 <p>The two arguments to the '<tt>and</tt>' instruction must be <a
2434 href="#t_integer">integer</a> values. Both arguments must have
2435 identical types.</p>
2437 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
2439 <div style="align: center">
2440 <table border="1" cellspacing="0" cellpadding="4">
2471 <pre> <result> = and i32 4, %var <i>; yields {i32}:result = 4 & %var</i>
2472 <result> = and i32 15, 40 <i>; yields {i32}:result = 8</i>
2473 <result> = and i32 4, 8 <i>; yields {i32}:result = 0</i>
2476 <!-- _______________________________________________________________________ -->
2477 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
2478 <div class="doc_text">
2480 <pre> <result> = or <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2483 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
2484 or of its two operands.</p>
2486 <p>The two arguments to the '<tt>or</tt>' instruction must be <a
2487 href="#t_integer">integer</a> values. Both arguments must have
2488 identical types.</p>
2490 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
2492 <div style="align: center">
2493 <table border="1" cellspacing="0" cellpadding="4">
2524 <pre> <result> = or i32 4, %var <i>; yields {i32}:result = 4 | %var</i>
2525 <result> = or i32 15, 40 <i>; yields {i32}:result = 47</i>
2526 <result> = or i32 4, 8 <i>; yields {i32}:result = 12</i>
2529 <!-- _______________________________________________________________________ -->
2530 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
2531 Instruction</a> </div>
2532 <div class="doc_text">
2534 <pre> <result> = xor <ty> <var1>, <var2> <i>; yields {ty}:result</i>
2537 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
2538 or of its two operands. The <tt>xor</tt> is used to implement the
2539 "one's complement" operation, which is the "~" operator in C.</p>
2541 <p>The two arguments to the '<tt>xor</tt>' instruction must be <a
2542 href="#t_integer">integer</a> values. Both arguments must have
2543 identical types.</p>
2545 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
2547 <div style="align: center">
2548 <table border="1" cellspacing="0" cellpadding="4">
2580 <pre> <result> = xor i32 4, %var <i>; yields {i32}:result = 4 ^ %var</i>
2581 <result> = xor i32 15, 40 <i>; yields {i32}:result = 39</i>
2582 <result> = xor i32 4, 8 <i>; yields {i32}:result = 12</i>
2583 <result> = xor i32 %V, -1 <i>; yields {i32}:result = ~%V</i>
2587 <!-- ======================================================================= -->
2588 <div class="doc_subsection">
2589 <a name="vectorops">Vector Operations</a>
2592 <div class="doc_text">
2594 <p>LLVM supports several instructions to represent vector operations in a
2595 target-independent manner. These instructions cover the element-access and
2596 vector-specific operations needed to process vectors effectively. While LLVM
2597 does directly support these vector operations, many sophisticated algorithms
2598 will want to use target-specific intrinsics to take full advantage of a specific
2603 <!-- _______________________________________________________________________ -->
2604 <div class="doc_subsubsection">
2605 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2608 <div class="doc_text">
2613 <result> = extractelement <n x <ty>> <val>, i32 <idx> <i>; yields <ty></i>
2619 The '<tt>extractelement</tt>' instruction extracts a single scalar
2620 element from a vector at a specified index.
2627 The first operand of an '<tt>extractelement</tt>' instruction is a
2628 value of <a href="#t_vector">vector</a> type. The second operand is
2629 an index indicating the position from which to extract the element.
2630 The index may be a variable.</p>
2635 The result is a scalar of the same type as the element type of
2636 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2637 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2638 results are undefined.
2644 %result = extractelement <4 x i32> %vec, i32 0 <i>; yields i32</i>
2649 <!-- _______________________________________________________________________ -->
2650 <div class="doc_subsubsection">
2651 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
2654 <div class="doc_text">
2659 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, i32 <idx> <i>; yields <n x <ty>></i>
2665 The '<tt>insertelement</tt>' instruction inserts a scalar
2666 element into a vector at a specified index.
2673 The first operand of an '<tt>insertelement</tt>' instruction is a
2674 value of <a href="#t_vector">vector</a> type. The second operand is a
2675 scalar value whose type must equal the element type of the first
2676 operand. The third operand is an index indicating the position at
2677 which to insert the value. The index may be a variable.</p>
2682 The result is a vector of the same type as <tt>val</tt>. Its
2683 element values are those of <tt>val</tt> except at position
2684 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
2685 exceeds the length of <tt>val</tt>, the results are undefined.
2691 %result = insertelement <4 x i32> %vec, i32 1, i32 0 <i>; yields <4 x i32></i>
2695 <!-- _______________________________________________________________________ -->
2696 <div class="doc_subsubsection">
2697 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
2700 <div class="doc_text">
2705 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <n x i32> <mask> <i>; yields <n x <ty>></i>
2711 The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
2712 from two input vectors, returning a vector of the same type.
2718 The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
2719 with types that match each other and types that match the result of the
2720 instruction. The third argument is a shuffle mask, which has the same number
2721 of elements as the other vector type, but whose element type is always 'i32'.
2725 The shuffle mask operand is required to be a constant vector with either
2726 constant integer or undef values.
2732 The elements of the two input vectors are numbered from left to right across
2733 both of the vectors. The shuffle mask operand specifies, for each element of
2734 the result vector, which element of the two input registers the result element
2735 gets. The element selector may be undef (meaning "don't care") and the second
2736 operand may be undef if performing a shuffle from only one vector.
2742 %result = shufflevector <4 x i32> %v1, <4 x i32> %v2,
2743 <4 x i32> <i32 0, i32 4, i32 1, i32 5> <i>; yields <4 x i32></i>
2744 %result = shufflevector <4 x i32> %v1, <4 x i32> undef,
2745 <4 x i32> <i32 0, i32 1, i32 2, i32 3> <i>; yields <4 x i32></i> - Identity shuffle.
2750 <!-- ======================================================================= -->
2751 <div class="doc_subsection">
2752 <a name="memoryops">Memory Access and Addressing Operations</a>
2755 <div class="doc_text">
2757 <p>A key design point of an SSA-based representation is how it
2758 represents memory. In LLVM, no memory locations are in SSA form, which
2759 makes things very simple. This section describes how to read, write,
2760 allocate, and free memory in LLVM.</p>
2764 <!-- _______________________________________________________________________ -->
2765 <div class="doc_subsubsection">
2766 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
2769 <div class="doc_text">
2774 <result> = malloc <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2779 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
2780 heap and returns a pointer to it. The object is always allocated in the generic
2781 address space (address space zero).</p>
2785 <p>The '<tt>malloc</tt>' instruction allocates
2786 <tt>sizeof(<type>)*NumElements</tt>
2787 bytes of memory from the operating system and returns a pointer of the
2788 appropriate type to the program. If "NumElements" is specified, it is the
2789 number of elements allocated, otherwise "NumElements" is defaulted to be one.
2790 If a constant alignment is specified, the value result of the allocation is guaranteed to
2791 be aligned to at least that boundary. If not specified, or if zero, the target can
2792 choose to align the allocation on any convenient boundary.</p>
2794 <p>'<tt>type</tt>' must be a sized type.</p>
2798 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
2799 a pointer is returned. Allocating zero bytes is undefined. The result is null
2800 if there is insufficient memory available.</p>
2805 %array = malloc [4 x i8 ] <i>; yields {[%4 x i8]*}:array</i>
2807 %size = <a href="#i_add">add</a> i32 2, 2 <i>; yields {i32}:size = i32 4</i>
2808 %array1 = malloc i8, i32 4 <i>; yields {i8*}:array1</i>
2809 %array2 = malloc [12 x i8], i32 %size <i>; yields {[12 x i8]*}:array2</i>
2810 %array3 = malloc i32, i32 4, align 1024 <i>; yields {i32*}:array3</i>
2811 %array4 = malloc i32, align 1024 <i>; yields {i32*}:array4</i>
2815 <!-- _______________________________________________________________________ -->
2816 <div class="doc_subsubsection">
2817 <a name="i_free">'<tt>free</tt>' Instruction</a>
2820 <div class="doc_text">
2825 free <type> <value> <i>; yields {void}</i>
2830 <p>The '<tt>free</tt>' instruction returns memory back to the unused
2831 memory heap to be reallocated in the future.</p>
2835 <p>'<tt>value</tt>' shall be a pointer value that points to a value
2836 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
2841 <p>Access to the memory pointed to by the pointer is no longer defined
2842 after this instruction executes. If the pointer is null, the result is
2848 %array = <a href="#i_malloc">malloc</a> [4 x i8] <i>; yields {[4 x i8]*}:array</i>
2849 free [4 x i8]* %array
2853 <!-- _______________________________________________________________________ -->
2854 <div class="doc_subsubsection">
2855 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
2858 <div class="doc_text">
2863 <result> = alloca <type>[, i32 <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2868 <p>The '<tt>alloca</tt>' instruction allocates memory on the stack frame of the
2869 currently executing function, to be automatically released when this function
2870 returns to its caller. The object is always allocated in the generic address
2871 space (address space zero).</p>
2875 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
2876 bytes of memory on the runtime stack, returning a pointer of the
2877 appropriate type to the program. If "NumElements" is specified, it is the
2878 number of elements allocated, otherwise "NumElements" is defaulted to be one.
2879 If a constant alignment is specified, the value result of the allocation is guaranteed
2880 to be aligned to at least that boundary. If not specified, or if zero, the target
2881 can choose to align the allocation on any convenient boundary.</p>
2883 <p>'<tt>type</tt>' may be any sized type.</p>
2887 <p>Memory is allocated; a pointer is returned. '<tt>alloca</tt>'d
2888 memory is automatically released when the function returns. The '<tt>alloca</tt>'
2889 instruction is commonly used to represent automatic variables that must
2890 have an address available. When the function returns (either with the <tt><a
2891 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
2892 instructions), the memory is reclaimed. Allocating zero bytes
2893 is legal, but the result is undefined.</p>
2898 %ptr = alloca i32 <i>; yields {i32*}:ptr</i>
2899 %ptr = alloca i32, i32 4 <i>; yields {i32*}:ptr</i>
2900 %ptr = alloca i32, i32 4, align 1024 <i>; yields {i32*}:ptr</i>
2901 %ptr = alloca i32, align 1024 <i>; yields {i32*}:ptr</i>
2905 <!-- _______________________________________________________________________ -->
2906 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
2907 Instruction</a> </div>
2908 <div class="doc_text">
2910 <pre> <result> = load <ty>* <pointer>[, align <alignment>]<br> <result> = volatile load <ty>* <pointer>[, align <alignment>]<br></pre>
2912 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
2914 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
2915 address from which to load. The pointer must point to a <a
2916 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
2917 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
2918 the number or order of execution of this <tt>load</tt> with other
2919 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
2922 The optional constant "align" argument specifies the alignment of the operation
2923 (that is, the alignment of the memory address). A value of 0 or an
2924 omitted "align" argument means that the operation has the preferential
2925 alignment for the target. It is the responsibility of the code emitter
2926 to ensure that the alignment information is correct. Overestimating
2927 the alignment results in an undefined behavior. Underestimating the
2928 alignment may produce less efficient code. An alignment of 1 is always
2932 <p>The location of memory pointed to is loaded.</p>
2934 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
2936 href="#i_store">store</a> i32 3, i32* %ptr <i>; yields {void}</i>
2937 %val = load i32* %ptr <i>; yields {i32}:val = i32 3</i>
2940 <!-- _______________________________________________________________________ -->
2941 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
2942 Instruction</a> </div>
2943 <div class="doc_text">
2945 <pre> store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
2946 volatile store <ty> <value>, <ty>* <pointer>[, align <alignment>] <i>; yields {void}</i>
2949 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
2951 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
2952 to store and an address at which to store it. The type of the '<tt><pointer></tt>'
2953 operand must be a pointer to the <a href="#t_firstclass">first class</a> type
2954 of the '<tt><value></tt>'
2955 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
2956 optimizer is not allowed to modify the number or order of execution of
2957 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
2958 href="#i_store">store</a></tt> instructions.</p>
2960 The optional constant "align" argument specifies the alignment of the operation
2961 (that is, the alignment of the memory address). A value of 0 or an
2962 omitted "align" argument means that the operation has the preferential
2963 alignment for the target. It is the responsibility of the code emitter
2964 to ensure that the alignment information is correct. Overestimating
2965 the alignment results in an undefined behavior. Underestimating the
2966 alignment may produce less efficient code. An alignment of 1 is always
2970 <p>The contents of memory are updated to contain '<tt><value></tt>'
2971 at the location specified by the '<tt><pointer></tt>' operand.</p>
2973 <pre> %ptr = <a href="#i_alloca">alloca</a> i32 <i>; yields {i32*}:ptr</i>
2974 store i32 3, i32* %ptr <i>; yields {void}</i>
2975 %val = <a href="#i_load">load</a> i32* %ptr <i>; yields {i32}:val = i32 3</i>
2979 <!-- _______________________________________________________________________ -->
2980 <div class="doc_subsubsection">
2981 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
2984 <div class="doc_text">
2987 <result> = getelementptr <ty>* <ptrval>{, <ty> <idx>}*
2993 The '<tt>getelementptr</tt>' instruction is used to get the address of a
2994 subelement of an aggregate data structure.</p>
2998 <p>This instruction takes a list of integer operands that indicate what
2999 elements of the aggregate object to index to. The actual types of the arguments
3000 provided depend on the type of the first pointer argument. The
3001 '<tt>getelementptr</tt>' instruction is used to index down through the type
3002 levels of a structure or to a specific index in an array. When indexing into a
3003 structure, only <tt>i32</tt> integer constants are allowed. When indexing
3004 into an array or pointer, only integers of 32 or 64 bits are allowed, and will
3005 be sign extended to 64-bit values.</p>
3007 <p>For example, let's consider a C code fragment and how it gets
3008 compiled to LLVM:</p>
3010 <div class="doc_code">
3023 int *foo(struct ST *s) {
3024 return &s[1].Z.B[5][13];
3029 <p>The LLVM code generated by the GCC frontend is:</p>
3031 <div class="doc_code">
3033 %RT = type { i8 , [10 x [20 x i32]], i8 }
3034 %ST = type { i32, double, %RT }
3036 define i32* %foo(%ST* %s) {
3038 %reg = getelementptr %ST* %s, i32 1, i32 2, i32 1, i32 5, i32 13
3046 <p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
3047 on the pointer type that is being indexed into. <a href="#t_pointer">Pointer</a>
3048 and <a href="#t_array">array</a> types can use a 32-bit or 64-bit
3049 <a href="#t_integer">integer</a> type but the value will always be sign extended
3050 to 64-bits. <a href="#t_struct">Structure</a> and <a href="#t_pstruct">packed
3051 structure</a> types require <tt>i32</tt> <b>constants</b>.</p>
3053 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
3054 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ i32, double, %RT
3055 }</tt>' type, a structure. The second index indexes into the third element of
3056 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ i8 , [10 x [20 x i32]],
3057 i8 }</tt>' type, another structure. The third index indexes into the second
3058 element of the structure, yielding a '<tt>[10 x [20 x i32]]</tt>' type, an
3059 array. The two dimensions of the array are subscripted into, yielding an
3060 '<tt>i32</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
3061 to this element, thus computing a value of '<tt>i32*</tt>' type.</p>
3063 <p>Note that it is perfectly legal to index partially through a
3064 structure, returning a pointer to an inner element. Because of this,
3065 the LLVM code for the given testcase is equivalent to:</p>
3068 define i32* %foo(%ST* %s) {
3069 %t1 = getelementptr %ST* %s, i32 1 <i>; yields %ST*:%t1</i>
3070 %t2 = getelementptr %ST* %t1, i32 0, i32 2 <i>; yields %RT*:%t2</i>
3071 %t3 = getelementptr %RT* %t2, i32 0, i32 1 <i>; yields [10 x [20 x i32]]*:%t3</i>
3072 %t4 = getelementptr [10 x [20 x i32]]* %t3, i32 0, i32 5 <i>; yields [20 x i32]*:%t4</i>
3073 %t5 = getelementptr [20 x i32]* %t4, i32 0, i32 13 <i>; yields i32*:%t5</i>
3078 <p>Note that it is undefined to access an array out of bounds: array and
3079 pointer indexes must always be within the defined bounds of the array type.
3080 The one exception for this rules is zero length arrays. These arrays are
3081 defined to be accessible as variable length arrays, which requires access
3082 beyond the zero'th element.</p>
3084 <p>The getelementptr instruction is often confusing. For some more insight
3085 into how it works, see <a href="GetElementPtr.html">the getelementptr
3091 <i>; yields [12 x i8]*:aptr</i>
3092 %aptr = getelementptr {i32, [12 x i8]}* %sptr, i64 0, i32 1
3096 <!-- ======================================================================= -->
3097 <div class="doc_subsection"> <a name="convertops">Conversion Operations</a>
3099 <div class="doc_text">
3100 <p>The instructions in this category are the conversion instructions (casting)
3101 which all take a single operand and a type. They perform various bit conversions
3105 <!-- _______________________________________________________________________ -->
3106 <div class="doc_subsubsection">
3107 <a name="i_trunc">'<tt>trunc .. to</tt>' Instruction</a>
3109 <div class="doc_text">
3113 <result> = trunc <ty> <value> to <ty2> <i>; yields ty2</i>
3118 The '<tt>trunc</tt>' instruction truncates its operand to the type <tt>ty2</tt>.
3123 The '<tt>trunc</tt>' instruction takes a <tt>value</tt> to trunc, which must
3124 be an <a href="#t_integer">integer</a> type, and a type that specifies the size
3125 and type of the result, which must be an <a href="#t_integer">integer</a>
3126 type. The bit size of <tt>value</tt> must be larger than the bit size of
3127 <tt>ty2</tt>. Equal sized types are not allowed.</p>
3131 The '<tt>trunc</tt>' instruction truncates the high order bits in <tt>value</tt>
3132 and converts the remaining bits to <tt>ty2</tt>. Since the source size must be
3133 larger than the destination size, <tt>trunc</tt> cannot be a <i>no-op cast</i>.
3134 It will always truncate bits.</p>
3138 %X = trunc i32 257 to i8 <i>; yields i8:1</i>
3139 %Y = trunc i32 123 to i1 <i>; yields i1:true</i>
3140 %Y = trunc i32 122 to i1 <i>; yields i1:false</i>
3144 <!-- _______________________________________________________________________ -->
3145 <div class="doc_subsubsection">
3146 <a name="i_zext">'<tt>zext .. to</tt>' Instruction</a>
3148 <div class="doc_text">
3152 <result> = zext <ty> <value> to <ty2> <i>; yields ty2</i>
3156 <p>The '<tt>zext</tt>' instruction zero extends its operand to type
3161 <p>The '<tt>zext</tt>' instruction takes a value to cast, which must be of
3162 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3163 also be of <a href="#t_integer">integer</a> type. The bit size of the
3164 <tt>value</tt> must be smaller than the bit size of the destination type,
3168 <p>The <tt>zext</tt> fills the high order bits of the <tt>value</tt> with zero
3169 bits until it reaches the size of the destination type, <tt>ty2</tt>.</p>
3171 <p>When zero extending from i1, the result will always be either 0 or 1.</p>
3175 %X = zext i32 257 to i64 <i>; yields i64:257</i>
3176 %Y = zext i1 true to i32 <i>; yields i32:1</i>
3180 <!-- _______________________________________________________________________ -->
3181 <div class="doc_subsubsection">
3182 <a name="i_sext">'<tt>sext .. to</tt>' Instruction</a>
3184 <div class="doc_text">
3188 <result> = sext <ty> <value> to <ty2> <i>; yields ty2</i>
3192 <p>The '<tt>sext</tt>' sign extends <tt>value</tt> to the type <tt>ty2</tt>.</p>
3196 The '<tt>sext</tt>' instruction takes a value to cast, which must be of
3197 <a href="#t_integer">integer</a> type, and a type to cast it to, which must
3198 also be of <a href="#t_integer">integer</a> type. The bit size of the
3199 <tt>value</tt> must be smaller than the bit size of the destination type,
3204 The '<tt>sext</tt>' instruction performs a sign extension by copying the sign
3205 bit (highest order bit) of the <tt>value</tt> until it reaches the bit size of
3206 the type <tt>ty2</tt>.</p>
3208 <p>When sign extending from i1, the extension always results in -1 or 0.</p>
3212 %X = sext i8 -1 to i16 <i>; yields i16 :65535</i>
3213 %Y = sext i1 true to i32 <i>; yields i32:-1</i>
3217 <!-- _______________________________________________________________________ -->
3218 <div class="doc_subsubsection">
3219 <a name="i_fptrunc">'<tt>fptrunc .. to</tt>' Instruction</a>
3222 <div class="doc_text">
3227 <result> = fptrunc <ty> <value> to <ty2> <i>; yields ty2</i>
3231 <p>The '<tt>fptrunc</tt>' instruction truncates <tt>value</tt> to type
3236 <p>The '<tt>fptrunc</tt>' instruction takes a <a href="#t_floating">floating
3237 point</a> value to cast and a <a href="#t_floating">floating point</a> type to
3238 cast it to. The size of <tt>value</tt> must be larger than the size of
3239 <tt>ty2</tt>. This implies that <tt>fptrunc</tt> cannot be used to make a
3240 <i>no-op cast</i>.</p>
3243 <p> The '<tt>fptrunc</tt>' instruction truncates a <tt>value</tt> from a larger
3244 <a href="#t_floating">floating point</a> type to a smaller
3245 <a href="#t_floating">floating point</a> type. If the value cannot fit within
3246 the destination type, <tt>ty2</tt>, then the results are undefined.</p>
3250 %X = fptrunc double 123.0 to float <i>; yields float:123.0</i>
3251 %Y = fptrunc double 1.0E+300 to float <i>; yields undefined</i>
3255 <!-- _______________________________________________________________________ -->
3256 <div class="doc_subsubsection">
3257 <a name="i_fpext">'<tt>fpext .. to</tt>' Instruction</a>
3259 <div class="doc_text">
3263 <result> = fpext <ty> <value> to <ty2> <i>; yields ty2</i>
3267 <p>The '<tt>fpext</tt>' extends a floating point <tt>value</tt> to a larger
3268 floating point value.</p>
3271 <p>The '<tt>fpext</tt>' instruction takes a
3272 <a href="#t_floating">floating point</a> <tt>value</tt> to cast,
3273 and a <a href="#t_floating">floating point</a> type to cast it to. The source
3274 type must be smaller than the destination type.</p>
3277 <p>The '<tt>fpext</tt>' instruction extends the <tt>value</tt> from a smaller
3278 <a href="#t_floating">floating point</a> type to a larger
3279 <a href="#t_floating">floating point</a> type. The <tt>fpext</tt> cannot be
3280 used to make a <i>no-op cast</i> because it always changes bits. Use
3281 <tt>bitcast</tt> to make a <i>no-op cast</i> for a floating point cast.</p>
3285 %X = fpext float 3.1415 to double <i>; yields double:3.1415</i>
3286 %Y = fpext float 1.0 to float <i>; yields float:1.0 (no-op)</i>
3290 <!-- _______________________________________________________________________ -->
3291 <div class="doc_subsubsection">
3292 <a name="i_fptoui">'<tt>fptoui .. to</tt>' Instruction</a>
3294 <div class="doc_text">
3298 <result> = fptoui <ty> <value> to <ty2> <i>; yields ty2</i>
3302 <p>The '<tt>fptoui</tt>' converts a floating point <tt>value</tt> to its
3303 unsigned integer equivalent of type <tt>ty2</tt>.
3307 <p>The '<tt>fptoui</tt>' instruction takes a value to cast, which must be a
3308 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3309 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3310 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3311 vector integer type with the same number of elements as <tt>ty</tt></p>
3314 <p> The '<tt>fptoui</tt>' instruction converts its
3315 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3316 towards zero) unsigned integer value. If the value cannot fit in <tt>ty2</tt>,
3317 the results are undefined.</p>
3321 %X = fptoui double 123.0 to i32 <i>; yields i32:123</i>
3322 %Y = fptoui float 1.0E+300 to i1 <i>; yields undefined:1</i>
3323 %X = fptoui float 1.04E+17 to i8 <i>; yields undefined:1</i>
3327 <!-- _______________________________________________________________________ -->
3328 <div class="doc_subsubsection">
3329 <a name="i_fptosi">'<tt>fptosi .. to</tt>' Instruction</a>
3331 <div class="doc_text">
3335 <result> = fptosi <ty> <value> to <ty2> <i>; yields ty2</i>
3339 <p>The '<tt>fptosi</tt>' instruction converts
3340 <a href="#t_floating">floating point</a> <tt>value</tt> to type <tt>ty2</tt>.
3344 <p> The '<tt>fptosi</tt>' instruction takes a value to cast, which must be a
3345 scalar or vector <a href="#t_floating">floating point</a> value, and a type
3346 to cast it to <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a>
3347 type. If <tt>ty</tt> is a vector floating point type, <tt>ty2</tt> must be a
3348 vector integer type with the same number of elements as <tt>ty</tt></p>
3351 <p>The '<tt>fptosi</tt>' instruction converts its
3352 <a href="#t_floating">floating point</a> operand into the nearest (rounding
3353 towards zero) signed integer value. If the value cannot fit in <tt>ty2</tt>,
3354 the results are undefined.</p>
3358 %X = fptosi double -123.0 to i32 <i>; yields i32:-123</i>
3359 %Y = fptosi float 1.0E-247 to i1 <i>; yields undefined:1</i>
3360 %X = fptosi float 1.04E+17 to i8 <i>; yields undefined:1</i>
3364 <!-- _______________________________________________________________________ -->
3365 <div class="doc_subsubsection">
3366 <a name="i_uitofp">'<tt>uitofp .. to</tt>' Instruction</a>
3368 <div class="doc_text">
3372 <result> = uitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3376 <p>The '<tt>uitofp</tt>' instruction regards <tt>value</tt> as an unsigned
3377 integer and converts that value to the <tt>ty2</tt> type.</p>
3380 <p>The '<tt>uitofp</tt>' instruction takes a value to cast, which must be a
3381 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3382 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3383 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3384 floating point type with the same number of elements as <tt>ty</tt></p>
3387 <p>The '<tt>uitofp</tt>' instruction interprets its operand as an unsigned
3388 integer quantity and converts it to the corresponding floating point value. If
3389 the value cannot fit in the floating point value, the results are undefined.</p>
3393 %X = uitofp i32 257 to float <i>; yields float:257.0</i>
3394 %Y = uitofp i8 -1 to double <i>; yields double:255.0</i>
3398 <!-- _______________________________________________________________________ -->
3399 <div class="doc_subsubsection">
3400 <a name="i_sitofp">'<tt>sitofp .. to</tt>' Instruction</a>
3402 <div class="doc_text">
3406 <result> = sitofp <ty> <value> to <ty2> <i>; yields ty2</i>
3410 <p>The '<tt>sitofp</tt>' instruction regards <tt>value</tt> as a signed
3411 integer and converts that value to the <tt>ty2</tt> type.</p>
3414 <p>The '<tt>sitofp</tt>' instruction takes a value to cast, which must be a
3415 scalar or vector <a href="#t_integer">integer</a> value, and a type to cast it
3416 to <tt>ty2</tt>, which must be an <a href="#t_floating">floating point</a>
3417 type. If <tt>ty</tt> is a vector integer type, <tt>ty2</tt> must be a vector
3418 floating point type with the same number of elements as <tt>ty</tt></p>
3421 <p>The '<tt>sitofp</tt>' instruction interprets its operand as a signed
3422 integer quantity and converts it to the corresponding floating point value. If
3423 the value cannot fit in the floating point value, the results are undefined.</p>
3427 %X = sitofp i32 257 to float <i>; yields float:257.0</i>
3428 %Y = sitofp i8 -1 to double <i>; yields double:-1.0</i>
3432 <!-- _______________________________________________________________________ -->
3433 <div class="doc_subsubsection">
3434 <a name="i_ptrtoint">'<tt>ptrtoint .. to</tt>' Instruction</a>
3436 <div class="doc_text">
3440 <result> = ptrtoint <ty> <value> to <ty2> <i>; yields ty2</i>
3444 <p>The '<tt>ptrtoint</tt>' instruction converts the pointer <tt>value</tt> to
3445 the integer type <tt>ty2</tt>.</p>
3448 <p>The '<tt>ptrtoint</tt>' instruction takes a <tt>value</tt> to cast, which
3449 must be a <a href="#t_pointer">pointer</a> value, and a type to cast it to
3450 <tt>ty2</tt>, which must be an <a href="#t_integer">integer</a> type.
3453 <p>The '<tt>ptrtoint</tt>' instruction converts <tt>value</tt> to integer type
3454 <tt>ty2</tt> by interpreting the pointer value as an integer and either
3455 truncating or zero extending that value to the size of the integer type. If
3456 <tt>value</tt> is smaller than <tt>ty2</tt> then a zero extension is done. If
3457 <tt>value</tt> is larger than <tt>ty2</tt> then a truncation is done. If they
3458 are the same size, then nothing is done (<i>no-op cast</i>) other than a type
3463 %X = ptrtoint i32* %X to i8 <i>; yields truncation on 32-bit architecture</i>
3464 %Y = ptrtoint i32* %x to i64 <i>; yields zero extension on 32-bit architecture</i>
3468 <!-- _______________________________________________________________________ -->
3469 <div class="doc_subsubsection">
3470 <a name="i_inttoptr">'<tt>inttoptr .. to</tt>' Instruction</a>
3472 <div class="doc_text">
3476 <result> = inttoptr <ty> <value> to <ty2> <i>; yields ty2</i>
3480 <p>The '<tt>inttoptr</tt>' instruction converts an integer <tt>value</tt> to
3481 a pointer type, <tt>ty2</tt>.</p>
3484 <p>The '<tt>inttoptr</tt>' instruction takes an <a href="#t_integer">integer</a>
3485 value to cast, and a type to cast it to, which must be a
3486 <a href="#t_pointer">pointer</a> type.
3489 <p>The '<tt>inttoptr</tt>' instruction converts <tt>value</tt> to type
3490 <tt>ty2</tt> by applying either a zero extension or a truncation depending on
3491 the size of the integer <tt>value</tt>. If <tt>value</tt> is larger than the
3492 size of a pointer then a truncation is done. If <tt>value</tt> is smaller than
3493 the size of a pointer then a zero extension is done. If they are the same size,
3494 nothing is done (<i>no-op cast</i>).</p>
3498 %X = inttoptr i32 255 to i32* <i>; yields zero extension on 64-bit architecture</i>
3499 %X = inttoptr i32 255 to i32* <i>; yields no-op on 32-bit architecture</i>
3500 %Y = inttoptr i64 0 to i32* <i>; yields truncation on 32-bit architecture</i>
3504 <!-- _______________________________________________________________________ -->
3505 <div class="doc_subsubsection">
3506 <a name="i_bitcast">'<tt>bitcast .. to</tt>' Instruction</a>
3508 <div class="doc_text">
3512 <result> = bitcast <ty> <value> to <ty2> <i>; yields ty2</i>
3516 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3517 <tt>ty2</tt> without changing any bits.</p>
3520 <p>The '<tt>bitcast</tt>' instruction takes a value to cast, which must be
3521 a first class value, and a type to cast it to, which must also be a <a
3522 href="#t_firstclass">first class</a> type. The bit sizes of <tt>value</tt>
3523 and the destination type, <tt>ty2</tt>, must be identical. If the source
3524 type is a pointer, the destination type must also be a pointer.</p>
3527 <p>The '<tt>bitcast</tt>' instruction converts <tt>value</tt> to type
3528 <tt>ty2</tt>. It is always a <i>no-op cast</i> because no bits change with
3529 this conversion. The conversion is done as if the <tt>value</tt> had been
3530 stored to memory and read back as type <tt>ty2</tt>. Pointer types may only be
3531 converted to other pointer types with this instruction. To convert pointers to
3532 other types, use the <a href="#i_inttoptr">inttoptr</a> or
3533 <a href="#i_ptrtoint">ptrtoint</a> instructions first.</p>
3537 %X = bitcast i8 255 to i8 <i>; yields i8 :-1</i>
3538 %Y = bitcast i32* %x to sint* <i>; yields sint*:%x</i>
3539 %Z = bitcast <2xint> %V to i64; <i>; yields i64: %V</i>
3543 <!-- ======================================================================= -->
3544 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
3545 <div class="doc_text">
3546 <p>The instructions in this category are the "miscellaneous"
3547 instructions, which defy better classification.</p>
3550 <!-- _______________________________________________________________________ -->
3551 <div class="doc_subsubsection"><a name="i_icmp">'<tt>icmp</tt>' Instruction</a>
3553 <div class="doc_text">
3555 <pre> <result> = icmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3558 <p>The '<tt>icmp</tt>' instruction returns a boolean value based on comparison
3559 of its two integer or pointer operands.</p>
3561 <p>The '<tt>icmp</tt>' instruction takes three operands. The first operand is
3562 the condition code indicating the kind of comparison to perform. It is not
3563 a value, just a keyword. The possible condition code are:
3565 <li><tt>eq</tt>: equal</li>
3566 <li><tt>ne</tt>: not equal </li>
3567 <li><tt>ugt</tt>: unsigned greater than</li>
3568 <li><tt>uge</tt>: unsigned greater or equal</li>
3569 <li><tt>ult</tt>: unsigned less than</li>
3570 <li><tt>ule</tt>: unsigned less or equal</li>
3571 <li><tt>sgt</tt>: signed greater than</li>
3572 <li><tt>sge</tt>: signed greater or equal</li>
3573 <li><tt>slt</tt>: signed less than</li>
3574 <li><tt>sle</tt>: signed less or equal</li>
3576 <p>The remaining two arguments must be <a href="#t_integer">integer</a> or
3577 <a href="#t_pointer">pointer</a> typed. They must also be identical types.</p>
3579 <p>The '<tt>icmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3580 the condition code given as <tt>cond</tt>. The comparison performed always
3581 yields a <a href="#t_primitive">i1</a> result, as follows:
3583 <li><tt>eq</tt>: yields <tt>true</tt> if the operands are equal,
3584 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3586 <li><tt>ne</tt>: yields <tt>true</tt> if the operands are unequal,
3587 <tt>false</tt> otherwise. No sign interpretation is necessary or performed.
3588 <li><tt>ugt</tt>: interprets the operands as unsigned values and yields
3589 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3590 <li><tt>uge</tt>: interprets the operands as unsigned values and yields
3591 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3592 <li><tt>ult</tt>: interprets the operands as unsigned values and yields
3593 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3594 <li><tt>ule</tt>: interprets the operands as unsigned values and yields
3595 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3596 <li><tt>sgt</tt>: interprets the operands as signed values and yields
3597 <tt>true</tt> if <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3598 <li><tt>sge</tt>: interprets the operands as signed values and yields
3599 <tt>true</tt> if <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3600 <li><tt>slt</tt>: interprets the operands as signed values and yields
3601 <tt>true</tt> if <tt>var1</tt> is less than <tt>var2</tt>.</li>
3602 <li><tt>sle</tt>: interprets the operands as signed values and yields
3603 <tt>true</tt> if <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3605 <p>If the operands are <a href="#t_pointer">pointer</a> typed, the pointer
3606 values are compared as if they were integers.</p>
3609 <pre> <result> = icmp eq i32 4, 5 <i>; yields: result=false</i>
3610 <result> = icmp ne float* %X, %X <i>; yields: result=false</i>
3611 <result> = icmp ult i16 4, 5 <i>; yields: result=true</i>
3612 <result> = icmp sgt i16 4, 5 <i>; yields: result=false</i>
3613 <result> = icmp ule i16 -4, 5 <i>; yields: result=false</i>
3614 <result> = icmp sge i16 4, 5 <i>; yields: result=false</i>
3618 <!-- _______________________________________________________________________ -->
3619 <div class="doc_subsubsection"><a name="i_fcmp">'<tt>fcmp</tt>' Instruction</a>
3621 <div class="doc_text">
3623 <pre> <result> = fcmp <cond> <ty> <var1>, <var2> <i>; yields {i1}:result</i>
3626 <p>The '<tt>fcmp</tt>' instruction returns a boolean value based on comparison
3627 of its floating point operands.</p>
3629 <p>The '<tt>fcmp</tt>' instruction takes three operands. The first operand is
3630 the condition code indicating the kind of comparison to perform. It is not
3631 a value, just a keyword. The possible condition code are:
3633 <li><tt>false</tt>: no comparison, always returns false</li>
3634 <li><tt>oeq</tt>: ordered and equal</li>
3635 <li><tt>ogt</tt>: ordered and greater than </li>
3636 <li><tt>oge</tt>: ordered and greater than or equal</li>
3637 <li><tt>olt</tt>: ordered and less than </li>
3638 <li><tt>ole</tt>: ordered and less than or equal</li>
3639 <li><tt>one</tt>: ordered and not equal</li>
3640 <li><tt>ord</tt>: ordered (no nans)</li>
3641 <li><tt>ueq</tt>: unordered or equal</li>
3642 <li><tt>ugt</tt>: unordered or greater than </li>
3643 <li><tt>uge</tt>: unordered or greater than or equal</li>
3644 <li><tt>ult</tt>: unordered or less than </li>
3645 <li><tt>ule</tt>: unordered or less than or equal</li>
3646 <li><tt>une</tt>: unordered or not equal</li>
3647 <li><tt>uno</tt>: unordered (either nans)</li>
3648 <li><tt>true</tt>: no comparison, always returns true</li>
3650 <p><i>Ordered</i> means that neither operand is a QNAN while
3651 <i>unordered</i> means that either operand may be a QNAN.</p>
3652 <p>The <tt>val1</tt> and <tt>val2</tt> arguments must be
3653 <a href="#t_floating">floating point</a> typed. They must have identical
3656 <p>The '<tt>fcmp</tt>' compares <tt>var1</tt> and <tt>var2</tt> according to
3657 the condition code given as <tt>cond</tt>. The comparison performed always
3658 yields a <a href="#t_primitive">i1</a> result, as follows:
3660 <li><tt>false</tt>: always yields <tt>false</tt>, regardless of operands.</li>
3661 <li><tt>oeq</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3662 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3663 <li><tt>ogt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3664 <tt>var1</tt> is greather than <tt>var2</tt>.</li>
3665 <li><tt>oge</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3666 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3667 <li><tt>olt</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3668 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3669 <li><tt>ole</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3670 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3671 <li><tt>one</tt>: yields <tt>true</tt> if both operands are not a QNAN and
3672 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3673 <li><tt>ord</tt>: yields <tt>true</tt> if both operands are not a QNAN.</li>
3674 <li><tt>ueq</tt>: yields <tt>true</tt> if either operand is a QNAN or
3675 <tt>var1</tt> is equal to <tt>var2</tt>.</li>
3676 <li><tt>ugt</tt>: yields <tt>true</tt> if either operand is a QNAN or
3677 <tt>var1</tt> is greater than <tt>var2</tt>.</li>
3678 <li><tt>uge</tt>: yields <tt>true</tt> if either operand is a QNAN or
3679 <tt>var1</tt> is greater than or equal to <tt>var2</tt>.</li>
3680 <li><tt>ult</tt>: yields <tt>true</tt> if either operand is a QNAN or
3681 <tt>var1</tt> is less than <tt>var2</tt>.</li>
3682 <li><tt>ule</tt>: yields <tt>true</tt> if either operand is a QNAN or
3683 <tt>var1</tt> is less than or equal to <tt>var2</tt>.</li>
3684 <li><tt>une</tt>: yields <tt>true</tt> if either operand is a QNAN or
3685 <tt>var1</tt> is not equal to <tt>var2</tt>.</li>
3686 <li><tt>uno</tt>: yields <tt>true</tt> if either operand is a QNAN.</li>
3687 <li><tt>true</tt>: always yields <tt>true</tt>, regardless of operands.</li>
3691 <pre> <result> = fcmp oeq float 4.0, 5.0 <i>; yields: result=false</i>
3692 <result> = icmp one float 4.0, 5.0 <i>; yields: result=true</i>
3693 <result> = icmp olt float 4.0, 5.0 <i>; yields: result=true</i>
3694 <result> = icmp ueq double 1.0, 2.0 <i>; yields: result=false</i>
3698 <!-- _______________________________________________________________________ -->
3699 <div class="doc_subsubsection"> <a name="i_phi">'<tt>phi</tt>'
3700 Instruction</a> </div>
3701 <div class="doc_text">
3703 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
3705 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
3706 the SSA graph representing the function.</p>
3708 <p>The type of the incoming values is specified with the first type
3709 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
3710 as arguments, with one pair for each predecessor basic block of the
3711 current block. Only values of <a href="#t_firstclass">first class</a>
3712 type may be used as the value arguments to the PHI node. Only labels
3713 may be used as the label arguments.</p>
3714 <p>There must be no non-phi instructions between the start of a basic
3715 block and the PHI instructions: i.e. PHI instructions must be first in
3718 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the value
3719 specified by the pair corresponding to the predecessor basic block that executed
3720 just prior to the current block.</p>
3722 <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>
3725 <!-- _______________________________________________________________________ -->
3726 <div class="doc_subsubsection">
3727 <a name="i_select">'<tt>select</tt>' Instruction</a>
3730 <div class="doc_text">
3735 <result> = select i1 <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
3741 The '<tt>select</tt>' instruction is used to choose one value based on a
3742 condition, without branching.
3749 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.
3755 If the boolean condition evaluates to true, the instruction returns the first
3756 value argument; otherwise, it returns the second value argument.
3762 %X = select i1 true, i8 17, i8 42 <i>; yields i8:17</i>
3767 <!-- _______________________________________________________________________ -->
3768 <div class="doc_subsubsection">
3769 <a name="i_call">'<tt>call</tt>' Instruction</a>
3772 <div class="doc_text">
3776 <result> = [tail] call [<a href="#callingconv">cconv</a>] <ty> [<fnty>*] <fnptrval>(<param list>)
3781 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
3785 <p>This instruction requires several arguments:</p>
3789 <p>The optional "tail" marker indicates whether the callee function accesses
3790 any allocas or varargs in the caller. If the "tail" marker is present, the
3791 function call is eligible for tail call optimization. Note that calls may
3792 be marked "tail" even if they do not occur before a <a
3793 href="#i_ret"><tt>ret</tt></a> instruction.
3796 <p>The optional "cconv" marker indicates which <a href="#callingconv">calling
3797 convention</a> the call should use. If none is specified, the call defaults
3798 to using C calling conventions.
3801 <p>'<tt>ty</tt>': the type of the call instruction itself which is also
3802 the type of the return value. Functions that return no value are marked
3803 <tt><a href="#t_void">void</a></tt>.</p>
3806 <p>'<tt>fnty</tt>': shall be the signature of the pointer to function
3807 value being invoked. The argument types must match the types implied by
3808 this signature. This type can be omitted if the function is not varargs
3809 and if the function type does not return a pointer to a function.</p>
3812 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
3813 be invoked. In most cases, this is a direct function invocation, but
3814 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
3815 to function value.</p>
3818 <p>'<tt>function args</tt>': argument list whose types match the
3819 function signature argument types. All arguments must be of
3820 <a href="#t_firstclass">first class</a> type. If the function signature
3821 indicates the function accepts a variable number of arguments, the extra
3822 arguments can be specified.</p>
3828 <p>The '<tt>call</tt>' instruction is used to cause control flow to
3829 transfer to a specified function, with its incoming arguments bound to
3830 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
3831 instruction in the called function, control flow continues with the
3832 instruction after the function call, and the return value of the
3833 function is bound to the result argument. If the callee returns multiple
3834 values then the return values of the function are only accessible through
3835 the '<tt><a href="#i_getresult">getresult</a></tt>' instruction.</p>
3840 %retval = call i32 @test(i32 %argc)
3841 call i32 (i8 *, ...)* @printf(i8 * %msg, i32 12, i8 42) <i>; yields i32</i>
3842 %X = tail call i32 @foo() <i>; yields i32</i>
3843 %Y = tail call <a href="#callingconv">fastcc</a> i32 @foo() <i>; yields i32</i>
3844 call void %foo(i8 97 signext)
3846 %struct.A = type { i32, i8 }
3847 %r = call %struct.A @foo() <i>; yields { 32, i8 }</i>
3848 %gr = getresult %struct.A %r, 0 <i>; yields i32</i>
3849 %gr1 = getresult %struct.A %r, 1 <i>; yields i8</i>
3854 <!-- _______________________________________________________________________ -->
3855 <div class="doc_subsubsection">
3856 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
3859 <div class="doc_text">
3864 <resultval> = va_arg <va_list*> <arglist>, <argty>
3869 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
3870 the "variable argument" area of a function call. It is used to implement the
3871 <tt>va_arg</tt> macro in C.</p>
3875 <p>This instruction takes a <tt>va_list*</tt> value and the type of
3876 the argument. It returns a value of the specified argument type and
3877 increments the <tt>va_list</tt> to point to the next argument. The
3878 actual type of <tt>va_list</tt> is target specific.</p>
3882 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
3883 type from the specified <tt>va_list</tt> and causes the
3884 <tt>va_list</tt> to point to the next argument. For more information,
3885 see the variable argument handling <a href="#int_varargs">Intrinsic
3888 <p>It is legal for this instruction to be called in a function which does not
3889 take a variable number of arguments, for example, the <tt>vfprintf</tt>
3892 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
3893 href="#intrinsics">intrinsic function</a> because it takes a type as an
3898 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
3902 <!-- _______________________________________________________________________ -->
3903 <div class="doc_subsubsection">
3904 <a name="i_getresult">'<tt>getresult</tt>' Instruction</a>
3907 <div class="doc_text">
3911 <resultval> = getresult <type> <retval>, <index>
3916 <p> The '<tt>getresult</tt>' instruction is used to extract individual values
3917 from a '<tt><a href="#i_call">call</a></tt>'
3918 or '<tt><a href="#i_invoke">invoke</a></tt>' instruction that returns multiple
3923 <p>The '<tt>getresult</tt>' instruction takes a call or invoke value as its
3924 first argument. The value must have <a href="#t_struct">structure type</a>.
3925 The second argument is a constant unsigned index value which must be in range for
3926 the number of values returned by the call.</p>
3930 <p>The '<tt>getresult</tt>' instruction extracts the element identified by
3931 '<tt>index</tt>' from the aggregate value.</p>
3936 %struct.A = type { i32, i8 }
3938 %r = call %struct.A @foo()
3939 %gr = getresult %struct.A %r, 0 <i>; yields i32:%gr</i>
3940 %gr1 = getresult %struct.A %r, 1 <i>; yields i8:%gr1</i>
3947 <!-- *********************************************************************** -->
3948 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
3949 <!-- *********************************************************************** -->
3951 <div class="doc_text">
3953 <p>LLVM supports the notion of an "intrinsic function". These functions have
3954 well known names and semantics and are required to follow certain restrictions.
3955 Overall, these intrinsics represent an extension mechanism for the LLVM
3956 language that does not require changing all of the transformations in LLVM when
3957 adding to the language (or the bitcode reader/writer, the parser, etc...).</p>
3959 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
3960 prefix is reserved in LLVM for intrinsic names; thus, function names may not
3961 begin with this prefix. Intrinsic functions must always be external functions:
3962 you cannot define the body of intrinsic functions. Intrinsic functions may
3963 only be used in call or invoke instructions: it is illegal to take the address
3964 of an intrinsic function. Additionally, because intrinsic functions are part
3965 of the LLVM language, it is required if any are added that they be documented
3968 <p>Some intrinsic functions can be overloaded, i.e., the intrinsic represents
3969 a family of functions that perform the same operation but on different data
3970 types. Because LLVM can represent over 8 million different integer types,
3971 overloading is used commonly to allow an intrinsic function to operate on any
3972 integer type. One or more of the argument types or the result type can be
3973 overloaded to accept any integer type. Argument types may also be defined as
3974 exactly matching a previous argument's type or the result type. This allows an
3975 intrinsic function which accepts multiple arguments, but needs all of them to
3976 be of the same type, to only be overloaded with respect to a single argument or
3979 <p>Overloaded intrinsics will have the names of its overloaded argument types
3980 encoded into its function name, each preceded by a period. Only those types
3981 which are overloaded result in a name suffix. Arguments whose type is matched
3982 against another type do not. For example, the <tt>llvm.ctpop</tt> function can
3983 take an integer of any width and returns an integer of exactly the same integer
3984 width. This leads to a family of functions such as
3985 <tt>i8 @llvm.ctpop.i8(i8 %val)</tt> and <tt>i29 @llvm.ctpop.i29(i29 %val)</tt>.
3986 Only one type, the return type, is overloaded, and only one type suffix is
3987 required. Because the argument's type is matched against the return type, it
3988 does not require its own name suffix.</p>
3990 <p>To learn how to add an intrinsic function, please see the
3991 <a href="ExtendingLLVM.html">Extending LLVM Guide</a>.
3996 <!-- ======================================================================= -->
3997 <div class="doc_subsection">
3998 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
4001 <div class="doc_text">
4003 <p>Variable argument support is defined in LLVM with the <a
4004 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
4005 intrinsic functions. These functions are related to the similarly
4006 named macros defined in the <tt><stdarg.h></tt> header file.</p>
4008 <p>All of these functions operate on arguments that use a
4009 target-specific value type "<tt>va_list</tt>". The LLVM assembly
4010 language reference manual does not define what this type is, so all
4011 transformations should be prepared to handle these functions regardless of
4014 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
4015 instruction and the variable argument handling intrinsic functions are
4018 <div class="doc_code">
4020 define i32 @test(i32 %X, ...) {
4021 ; Initialize variable argument processing
4023 %ap2 = bitcast i8** %ap to i8*
4024 call void @llvm.va_start(i8* %ap2)
4026 ; Read a single integer argument
4027 %tmp = va_arg i8** %ap, i32
4029 ; Demonstrate usage of llvm.va_copy and llvm.va_end
4031 %aq2 = bitcast i8** %aq to i8*
4032 call void @llvm.va_copy(i8* %aq2, i8* %ap2)
4033 call void @llvm.va_end(i8* %aq2)
4035 ; Stop processing of arguments.
4036 call void @llvm.va_end(i8* %ap2)
4040 declare void @llvm.va_start(i8*)
4041 declare void @llvm.va_copy(i8*, i8*)
4042 declare void @llvm.va_end(i8*)
4048 <!-- _______________________________________________________________________ -->
4049 <div class="doc_subsubsection">
4050 <a name="int_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
4054 <div class="doc_text">
4056 <pre> declare void %llvm.va_start(i8* <arglist>)<br></pre>
4058 <P>The '<tt>llvm.va_start</tt>' intrinsic initializes
4059 <tt>*<arglist></tt> for subsequent use by <tt><a
4060 href="#i_va_arg">va_arg</a></tt>.</p>
4064 <P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
4068 <P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
4069 macro available in C. In a target-dependent way, it initializes the
4070 <tt>va_list</tt> element to which the argument points, so that the next call to
4071 <tt>va_arg</tt> will produce the first variable argument passed to the function.
4072 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
4073 last argument of the function as the compiler can figure that out.</p>
4077 <!-- _______________________________________________________________________ -->
4078 <div class="doc_subsubsection">
4079 <a name="int_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
4082 <div class="doc_text">
4084 <pre> declare void @llvm.va_end(i8* <arglist>)<br></pre>
4087 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt>*<arglist></tt>,
4088 which has been initialized previously with <tt><a href="#int_va_start">llvm.va_start</a></tt>
4089 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
4093 <p>The argument is a pointer to a <tt>va_list</tt> to destroy.</p>
4097 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
4098 macro available in C. In a target-dependent way, it destroys the
4099 <tt>va_list</tt> element to which the argument points. Calls to <a
4100 href="#int_va_start"><tt>llvm.va_start</tt></a> and <a href="#int_va_copy">
4101 <tt>llvm.va_copy</tt></a> must be matched exactly with calls to
4102 <tt>llvm.va_end</tt>.</p>
4106 <!-- _______________________________________________________________________ -->
4107 <div class="doc_subsubsection">
4108 <a name="int_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
4111 <div class="doc_text">
4116 declare void @llvm.va_copy(i8* <destarglist>, i8* <srcarglist>)
4121 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position
4122 from the source argument list to the destination argument list.</p>
4126 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
4127 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
4132 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt>
4133 macro available in C. In a target-dependent way, it copies the source
4134 <tt>va_list</tt> element into the destination <tt>va_list</tt> element. This
4135 intrinsic is necessary because the <tt><a href="#int_va_start">
4136 llvm.va_start</a></tt> intrinsic may be arbitrarily complex and require, for
4137 example, memory allocation.</p>
4141 <!-- ======================================================================= -->
4142 <div class="doc_subsection">
4143 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
4146 <div class="doc_text">
4149 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
4150 Collection</a> requires the implementation and generation of these intrinsics.
4151 These intrinsics allow identification of <a href="#int_gcroot">GC roots on the
4152 stack</a>, as well as garbage collector implementations that require <a
4153 href="#int_gcread">read</a> and <a href="#int_gcwrite">write</a> barriers.
4154 Front-ends for type-safe garbage collected languages should generate these
4155 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
4156 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
4159 <p>The garbage collection intrinsics only operate on objects in the generic
4160 address space (address space zero).</p>
4164 <!-- _______________________________________________________________________ -->
4165 <div class="doc_subsubsection">
4166 <a name="int_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
4169 <div class="doc_text">
4174 declare void @llvm.gcroot(i8** %ptrloc, i8* %metadata)
4179 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
4180 the code generator, and allows some metadata to be associated with it.</p>
4184 <p>The first argument specifies the address of a stack object that contains the
4185 root pointer. The second pointer (which must be either a constant or a global
4186 value address) contains the meta-data to be associated with the root.</p>
4190 <p>At runtime, a call to this intrinsics stores a null pointer into the "ptrloc"
4191 location. At compile-time, the code generator generates information to allow
4192 the runtime to find the pointer at GC safe points. The '<tt>llvm.gcroot</tt>'
4193 intrinsic may only be used in a function which <a href="#gc">specifies a GC
4199 <!-- _______________________________________________________________________ -->
4200 <div class="doc_subsubsection">
4201 <a name="int_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
4204 <div class="doc_text">
4209 declare i8* @llvm.gcread(i8* %ObjPtr, i8** %Ptr)
4214 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
4215 locations, allowing garbage collector implementations that require read
4220 <p>The second argument is the address to read from, which should be an address
4221 allocated from the garbage collector. The first object is a pointer to the
4222 start of the referenced object, if needed by the language runtime (otherwise
4227 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
4228 instruction, but may be replaced with substantially more complex code by the
4229 garbage collector runtime, as needed. The '<tt>llvm.gcread</tt>' intrinsic
4230 may only be used in a function which <a href="#gc">specifies a GC
4236 <!-- _______________________________________________________________________ -->
4237 <div class="doc_subsubsection">
4238 <a name="int_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
4241 <div class="doc_text">
4246 declare void @llvm.gcwrite(i8* %P1, i8* %Obj, i8** %P2)
4251 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
4252 locations, allowing garbage collector implementations that require write
4253 barriers (such as generational or reference counting collectors).</p>
4257 <p>The first argument is the reference to store, the second is the start of the
4258 object to store it to, and the third is the address of the field of Obj to
4259 store to. If the runtime does not require a pointer to the object, Obj may be
4264 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
4265 instruction, but may be replaced with substantially more complex code by the
4266 garbage collector runtime, as needed. The '<tt>llvm.gcwrite</tt>' intrinsic
4267 may only be used in a function which <a href="#gc">specifies a GC
4274 <!-- ======================================================================= -->
4275 <div class="doc_subsection">
4276 <a name="int_codegen">Code Generator Intrinsics</a>
4279 <div class="doc_text">
4281 These intrinsics are provided by LLVM to expose special features that may only
4282 be implemented with code generator support.
4287 <!-- _______________________________________________________________________ -->
4288 <div class="doc_subsubsection">
4289 <a name="int_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
4292 <div class="doc_text">
4296 declare i8 *@llvm.returnaddress(i32 <level>)
4302 The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
4303 target-specific value indicating the return address of the current function
4304 or one of its callers.
4310 The argument to this intrinsic indicates which function to return the address
4311 for. Zero indicates the calling function, one indicates its caller, etc. The
4312 argument is <b>required</b> to be a constant integer value.
4318 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
4319 the return address of the specified call frame, or zero if it cannot be
4320 identified. The value returned by this intrinsic is likely to be incorrect or 0
4321 for arguments other than zero, so it should only be used for debugging purposes.
4325 Note that calling this intrinsic does not prevent function inlining or other
4326 aggressive transformations, so the value returned may not be that of the obvious
4327 source-language caller.
4332 <!-- _______________________________________________________________________ -->
4333 <div class="doc_subsubsection">
4334 <a name="int_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
4337 <div class="doc_text">
4341 declare i8 *@llvm.frameaddress(i32 <level>)
4347 The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
4348 target-specific frame pointer value for the specified stack frame.
4354 The argument to this intrinsic indicates which function to return the frame
4355 pointer for. Zero indicates the calling function, one indicates its caller,
4356 etc. The argument is <b>required</b> to be a constant integer value.
4362 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
4363 the frame address of the specified call frame, or zero if it cannot be
4364 identified. The value returned by this intrinsic is likely to be incorrect or 0
4365 for arguments other than zero, so it should only be used for debugging purposes.
4369 Note that calling this intrinsic does not prevent function inlining or other
4370 aggressive transformations, so the value returned may not be that of the obvious
4371 source-language caller.
4375 <!-- _______________________________________________________________________ -->
4376 <div class="doc_subsubsection">
4377 <a name="int_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
4380 <div class="doc_text">
4384 declare i8 *@llvm.stacksave()
4390 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
4391 the function stack, for use with <a href="#int_stackrestore">
4392 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
4393 features like scoped automatic variable sized arrays in C99.
4399 This intrinsic returns a opaque pointer value that can be passed to <a
4400 href="#int_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
4401 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
4402 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
4403 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
4404 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
4405 that were allocated after the <tt>llvm.stacksave</tt> was executed.
4410 <!-- _______________________________________________________________________ -->
4411 <div class="doc_subsubsection">
4412 <a name="int_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
4415 <div class="doc_text">
4419 declare void @llvm.stackrestore(i8 * %ptr)
4425 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
4426 the function stack to the state it was in when the corresponding <a
4427 href="#int_stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
4428 useful for implementing language features like scoped automatic variable sized
4435 See the description for <a href="#int_stacksave"><tt>llvm.stacksave</tt></a>.
4441 <!-- _______________________________________________________________________ -->
4442 <div class="doc_subsubsection">
4443 <a name="int_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
4446 <div class="doc_text">
4450 declare void @llvm.prefetch(i8* <address>, i32 <rw>, i32 <locality>)
4457 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
4458 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
4460 effect on the behavior of the program but can change its performance
4467 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
4468 determining if the fetch should be for a read (0) or write (1), and
4469 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
4470 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
4471 <tt>locality</tt> arguments must be constant integers.
4477 This intrinsic does not modify the behavior of the program. In particular,
4478 prefetches cannot trap and do not produce a value. On targets that support this
4479 intrinsic, the prefetch can provide hints to the processor cache for better
4485 <!-- _______________________________________________________________________ -->
4486 <div class="doc_subsubsection">
4487 <a name="int_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
4490 <div class="doc_text">
4494 declare void @llvm.pcmarker(i32 <id>)
4501 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
4503 code to simulators and other tools. The method is target specific, but it is
4504 expected that the marker will use exported symbols to transmit the PC of the marker.
4505 The marker makes no guarantees that it will remain with any specific instruction
4506 after optimizations. It is possible that the presence of a marker will inhibit
4507 optimizations. The intended use is to be inserted after optimizations to allow
4508 correlations of simulation runs.
4514 <tt>id</tt> is a numerical id identifying the marker.
4520 This intrinsic does not modify the behavior of the program. Backends that do not
4521 support this intrinisic may ignore it.
4526 <!-- _______________________________________________________________________ -->
4527 <div class="doc_subsubsection">
4528 <a name="int_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
4531 <div class="doc_text">
4535 declare i64 @llvm.readcyclecounter( )
4542 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
4543 counter register (or similar low latency, high accuracy clocks) on those targets
4544 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
4545 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
4546 should only be used for small timings.
4552 When directly supported, reading the cycle counter should not modify any memory.
4553 Implementations are allowed to either return a application specific value or a
4554 system wide value. On backends without support, this is lowered to a constant 0.
4559 <!-- ======================================================================= -->
4560 <div class="doc_subsection">
4561 <a name="int_libc">Standard C Library Intrinsics</a>
4564 <div class="doc_text">
4566 LLVM provides intrinsics for a few important standard C library functions.
4567 These intrinsics allow source-language front-ends to pass information about the
4568 alignment of the pointer arguments to the code generator, providing opportunity
4569 for more efficient code generation.
4574 <!-- _______________________________________________________________________ -->
4575 <div class="doc_subsubsection">
4576 <a name="int_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
4579 <div class="doc_text">
4583 declare void @llvm.memcpy.i32(i8 * <dest>, i8 * <src>,
4584 i32 <len>, i32 <align>)
4585 declare void @llvm.memcpy.i64(i8 * <dest>, i8 * <src>,
4586 i64 <len>, i32 <align>)
4592 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4593 location to the destination location.
4597 Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
4598 intrinsics do not return a value, and takes an extra alignment argument.
4604 The first argument is a pointer to the destination, the second is a pointer to
4605 the source. The third argument is an integer argument
4606 specifying the number of bytes to copy, and the fourth argument is the alignment
4607 of the source and destination locations.
4611 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4612 the caller guarantees that both the source and destination pointers are aligned
4619 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
4620 location to the destination location, which are not allowed to overlap. It
4621 copies "len" bytes of memory over. If the argument is known to be aligned to
4622 some boundary, this can be specified as the fourth argument, otherwise it should
4628 <!-- _______________________________________________________________________ -->
4629 <div class="doc_subsubsection">
4630 <a name="int_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
4633 <div class="doc_text">
4637 declare void @llvm.memmove.i32(i8 * <dest>, i8 * <src>,
4638 i32 <len>, i32 <align>)
4639 declare void @llvm.memmove.i64(i8 * <dest>, i8 * <src>,
4640 i64 <len>, i32 <align>)
4646 The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
4647 location to the destination location. It is similar to the
4648 '<tt>llvm.memcpy</tt>' intrinsic but allows the two memory locations to overlap.
4652 Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
4653 intrinsics do not return a value, and takes an extra alignment argument.
4659 The first argument is a pointer to the destination, the second is a pointer to
4660 the source. The third argument is an integer argument
4661 specifying the number of bytes to copy, and the fourth argument is the alignment
4662 of the source and destination locations.
4666 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4667 the caller guarantees that the source and destination pointers are aligned to
4674 The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
4675 location to the destination location, which may overlap. It
4676 copies "len" bytes of memory over. If the argument is known to be aligned to
4677 some boundary, this can be specified as the fourth argument, otherwise it should
4683 <!-- _______________________________________________________________________ -->
4684 <div class="doc_subsubsection">
4685 <a name="int_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
4688 <div class="doc_text">
4692 declare void @llvm.memset.i32(i8 * <dest>, i8 <val>,
4693 i32 <len>, i32 <align>)
4694 declare void @llvm.memset.i64(i8 * <dest>, i8 <val>,
4695 i64 <len>, i32 <align>)
4701 The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
4706 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
4707 does not return a value, and takes an extra alignment argument.
4713 The first argument is a pointer to the destination to fill, the second is the
4714 byte value to fill it with, the third argument is an integer
4715 argument specifying the number of bytes to fill, and the fourth argument is the
4716 known alignment of destination location.
4720 If the call to this intrinisic has an alignment value that is not 0 or 1, then
4721 the caller guarantees that the destination pointer is aligned to that boundary.
4727 The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
4729 destination location. If the argument is known to be aligned to some boundary,
4730 this can be specified as the fourth argument, otherwise it should be set to 0 or
4736 <!-- _______________________________________________________________________ -->
4737 <div class="doc_subsubsection">
4738 <a name="int_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
4741 <div class="doc_text">
4744 <p>This is an overloaded intrinsic. You can use <tt>llvm.sqrt</tt> on any
4745 floating point or vector of floating point type. Not all targets support all
4748 declare float @llvm.sqrt.f32(float %Val)
4749 declare double @llvm.sqrt.f64(double %Val)
4750 declare x86_fp80 @llvm.sqrt.f80(x86_fp80 %Val)
4751 declare fp128 @llvm.sqrt.f128(fp128 %Val)
4752 declare ppc_fp128 @llvm.sqrt.ppcf128(ppc_fp128 %Val)
4758 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
4759 returning the same value as the libm '<tt>sqrt</tt>' functions would. Unlike
4760 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
4761 negative numbers other than -0.0 (which allows for better optimization, because
4762 there is no need to worry about errno being set). <tt>llvm.sqrt(-0.0)</tt> is
4763 defined to return -0.0 like IEEE sqrt.
4769 The argument and return value are floating point numbers of the same type.
4775 This function returns the sqrt of the specified operand if it is a nonnegative
4776 floating point number.
4780 <!-- _______________________________________________________________________ -->
4781 <div class="doc_subsubsection">
4782 <a name="int_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
4785 <div class="doc_text">
4788 <p>This is an overloaded intrinsic. You can use <tt>llvm.powi</tt> on any
4789 floating point or vector of floating point type. Not all targets support all
4792 declare float @llvm.powi.f32(float %Val, i32 %power)
4793 declare double @llvm.powi.f64(double %Val, i32 %power)
4794 declare x86_fp80 @llvm.powi.f80(x86_fp80 %Val, i32 %power)
4795 declare fp128 @llvm.powi.f128(fp128 %Val, i32 %power)
4796 declare ppc_fp128 @llvm.powi.ppcf128(ppc_fp128 %Val, i32 %power)
4802 The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
4803 specified (positive or negative) power. The order of evaluation of
4804 multiplications is not defined. When a vector of floating point type is
4805 used, the second argument remains a scalar integer value.
4811 The second argument is an integer power, and the first is a value to raise to
4818 This function returns the first value raised to the second power with an
4819 unspecified sequence of rounding operations.</p>
4822 <!-- _______________________________________________________________________ -->
4823 <div class="doc_subsubsection">
4824 <a name="int_sin">'<tt>llvm.sin.*</tt>' Intrinsic</a>
4827 <div class="doc_text">
4830 <p>This is an overloaded intrinsic. You can use <tt>llvm.sin</tt> on any
4831 floating point or vector of floating point type. Not all targets support all
4834 declare float @llvm.sin.f32(float %Val)
4835 declare double @llvm.sin.f64(double %Val)
4836 declare x86_fp80 @llvm.sin.f80(x86_fp80 %Val)
4837 declare fp128 @llvm.sin.f128(fp128 %Val)
4838 declare ppc_fp128 @llvm.sin.ppcf128(ppc_fp128 %Val)
4844 The '<tt>llvm.sin.*</tt>' intrinsics return the sine of the operand.
4850 The argument and return value are floating point numbers of the same type.
4856 This function returns the sine of the specified operand, returning the
4857 same values as the libm <tt>sin</tt> functions would, and handles error
4858 conditions in the same way.</p>
4861 <!-- _______________________________________________________________________ -->
4862 <div class="doc_subsubsection">
4863 <a name="int_cos">'<tt>llvm.cos.*</tt>' Intrinsic</a>
4866 <div class="doc_text">
4869 <p>This is an overloaded intrinsic. You can use <tt>llvm.cos</tt> on any
4870 floating point or vector of floating point type. Not all targets support all
4873 declare float @llvm.cos.f32(float %Val)
4874 declare double @llvm.cos.f64(double %Val)
4875 declare x86_fp80 @llvm.cos.f80(x86_fp80 %Val)
4876 declare fp128 @llvm.cos.f128(fp128 %Val)
4877 declare ppc_fp128 @llvm.cos.ppcf128(ppc_fp128 %Val)
4883 The '<tt>llvm.cos.*</tt>' intrinsics return the cosine of the operand.
4889 The argument and return value are floating point numbers of the same type.
4895 This function returns the cosine of the specified operand, returning the
4896 same values as the libm <tt>cos</tt> functions would, and handles error
4897 conditions in the same way.</p>
4900 <!-- _______________________________________________________________________ -->
4901 <div class="doc_subsubsection">
4902 <a name="int_pow">'<tt>llvm.pow.*</tt>' Intrinsic</a>
4905 <div class="doc_text">
4908 <p>This is an overloaded intrinsic. You can use <tt>llvm.pow</tt> on any
4909 floating point or vector of floating point type. Not all targets support all
4912 declare float @llvm.pow.f32(float %Val, float %Power)
4913 declare double @llvm.pow.f64(double %Val, double %Power)
4914 declare x86_fp80 @llvm.pow.f80(x86_fp80 %Val, x86_fp80 %Power)
4915 declare fp128 @llvm.pow.f128(fp128 %Val, fp128 %Power)
4916 declare ppc_fp128 @llvm.pow.ppcf128(ppc_fp128 %Val, ppc_fp128 Power)
4922 The '<tt>llvm.pow.*</tt>' intrinsics return the first operand raised to the
4923 specified (positive or negative) power.
4929 The second argument is a floating point power, and the first is a value to
4930 raise to that power.
4936 This function returns the first value raised to the second power,
4938 same values as the libm <tt>pow</tt> functions would, and handles error
4939 conditions in the same way.</p>
4943 <!-- ======================================================================= -->
4944 <div class="doc_subsection">
4945 <a name="int_manip">Bit Manipulation Intrinsics</a>
4948 <div class="doc_text">
4950 LLVM provides intrinsics for a few important bit manipulation operations.
4951 These allow efficient code generation for some algorithms.
4956 <!-- _______________________________________________________________________ -->
4957 <div class="doc_subsubsection">
4958 <a name="int_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
4961 <div class="doc_text">
4964 <p>This is an overloaded intrinsic function. You can use bswap on any integer
4965 type that is an even number of bytes (i.e. BitWidth % 16 == 0).
4967 declare i16 @llvm.bswap.i16(i16 <id>)
4968 declare i32 @llvm.bswap.i32(i32 <id>)
4969 declare i64 @llvm.bswap.i64(i64 <id>)
4975 The '<tt>llvm.bswap</tt>' family of intrinsics is used to byte swap integer
4976 values with an even number of bytes (positive multiple of 16 bits). These are
4977 useful for performing operations on data that is not in the target's native
4984 The <tt>llvm.bswap.i16</tt> intrinsic returns an i16 value that has the high
4985 and low byte of the input i16 swapped. Similarly, the <tt>llvm.bswap.i32</tt>
4986 intrinsic returns an i32 value that has the four bytes of the input i32
4987 swapped, so that if the input bytes are numbered 0, 1, 2, 3 then the returned
4988 i32 will have its bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i48</tt>,
4989 <tt>llvm.bswap.i64</tt> and other intrinsics extend this concept to
4990 additional even-byte lengths (6 bytes, 8 bytes and more, respectively).
4995 <!-- _______________________________________________________________________ -->
4996 <div class="doc_subsubsection">
4997 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
5000 <div class="doc_text">
5003 <p>This is an overloaded intrinsic. You can use llvm.ctpop on any integer bit
5004 width. Not all targets support all bit widths however.
5006 declare i8 @llvm.ctpop.i8 (i8 <src>)
5007 declare i16 @llvm.ctpop.i16(i16 <src>)
5008 declare i32 @llvm.ctpop.i32(i32 <src>)
5009 declare i64 @llvm.ctpop.i64(i64 <src>)
5010 declare i256 @llvm.ctpop.i256(i256 <src>)
5016 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
5023 The only argument is the value to be counted. The argument may be of any
5024 integer type. The return type must match the argument type.
5030 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
5034 <!-- _______________________________________________________________________ -->
5035 <div class="doc_subsubsection">
5036 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
5039 <div class="doc_text">
5042 <p>This is an overloaded intrinsic. You can use <tt>llvm.ctlz</tt> on any
5043 integer bit width. Not all targets support all bit widths however.
5045 declare i8 @llvm.ctlz.i8 (i8 <src>)
5046 declare i16 @llvm.ctlz.i16(i16 <src>)
5047 declare i32 @llvm.ctlz.i32(i32 <src>)
5048 declare i64 @llvm.ctlz.i64(i64 <src>)
5049 declare i256 @llvm.ctlz.i256(i256 <src>)
5055 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
5056 leading zeros in a variable.
5062 The only argument is the value to be counted. The argument may be of any
5063 integer type. The return type must match the argument type.
5069 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
5070 in a variable. If the src == 0 then the result is the size in bits of the type
5071 of src. For example, <tt>llvm.ctlz(i32 2) = 30</tt>.
5077 <!-- _______________________________________________________________________ -->
5078 <div class="doc_subsubsection">
5079 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
5082 <div class="doc_text">
5085 <p>This is an overloaded intrinsic. You can use <tt>llvm.cttz</tt> on any
5086 integer bit width. Not all targets support all bit widths however.
5088 declare i8 @llvm.cttz.i8 (i8 <src>)
5089 declare i16 @llvm.cttz.i16(i16 <src>)
5090 declare i32 @llvm.cttz.i32(i32 <src>)
5091 declare i64 @llvm.cttz.i64(i64 <src>)
5092 declare i256 @llvm.cttz.i256(i256 <src>)
5098 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
5105 The only argument is the value to be counted. The argument may be of any
5106 integer type. The return type must match the argument type.
5112 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
5113 in a variable. If the src == 0 then the result is the size in bits of the type
5114 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
5118 <!-- _______________________________________________________________________ -->
5119 <div class="doc_subsubsection">
5120 <a name="int_part_select">'<tt>llvm.part.select.*</tt>' Intrinsic</a>
5123 <div class="doc_text">
5126 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.select</tt>
5127 on any integer bit width.
5129 declare i17 @llvm.part.select.i17 (i17 %val, i32 %loBit, i32 %hiBit)
5130 declare i29 @llvm.part.select.i29 (i29 %val, i32 %loBit, i32 %hiBit)
5134 <p>The '<tt>llvm.part.select</tt>' family of intrinsic functions selects a
5135 range of bits from an integer value and returns them in the same bit width as
5136 the original value.</p>
5139 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5140 any bit width but they must have the same bit width. The second and third
5141 arguments must be <tt>i32</tt> type since they specify only a bit index.</p>
5144 <p>The operation of the '<tt>llvm.part.select</tt>' intrinsic has two modes
5145 of operation: forwards and reverse. If <tt>%loBit</tt> is greater than
5146 <tt>%hiBits</tt> then the intrinsic operates in reverse mode. Otherwise it
5147 operates in forward mode.</p>
5148 <p>In forward mode, this intrinsic is the equivalent of shifting <tt>%val</tt>
5149 right by <tt>%loBit</tt> bits and then ANDing it with a mask with
5150 only the <tt>%hiBit - %loBit</tt> bits set, as follows:</p>
5152 <li>The <tt>%val</tt> is shifted right (LSHR) by the number of bits specified
5153 by <tt>%loBits</tt>. This normalizes the value to the low order bits.</li>
5154 <li>The <tt>%loBits</tt> value is subtracted from the <tt>%hiBits</tt> value
5155 to determine the number of bits to retain.</li>
5156 <li>A mask of the retained bits is created by shifting a -1 value.</li>
5157 <li>The mask is ANDed with <tt>%val</tt> to produce the result.
5159 <p>In reverse mode, a similar computation is made except that the bits are
5160 returned in the reverse order. So, for example, if <tt>X</tt> has the value
5161 <tt>i16 0x0ACF (101011001111)</tt> and we apply
5162 <tt>part.select(i16 X, 8, 3)</tt> to it, we get back the value
5163 <tt>i16 0x0026 (000000100110)</tt>.</p>
5166 <div class="doc_subsubsection">
5167 <a name="int_part_set">'<tt>llvm.part.set.*</tt>' Intrinsic</a>
5170 <div class="doc_text">
5173 <p>This is an overloaded intrinsic. You can use <tt>llvm.part.set</tt>
5174 on any integer bit width.
5176 declare i17 @llvm.part.set.i17.i9 (i17 %val, i9 %repl, i32 %lo, i32 %hi)
5177 declare i29 @llvm.part.set.i29.i9 (i29 %val, i9 %repl, i32 %lo, i32 %hi)
5181 <p>The '<tt>llvm.part.set</tt>' family of intrinsic functions replaces a range
5182 of bits in an integer value with another integer value. It returns the integer
5183 with the replaced bits.</p>
5186 <p>The first argument, <tt>%val</tt> and the result may be integer types of
5187 any bit width but they must have the same bit width. <tt>%val</tt> is the value
5188 whose bits will be replaced. The second argument, <tt>%repl</tt> may be an
5189 integer of any bit width. The third and fourth arguments must be <tt>i32</tt>
5190 type since they specify only a bit index.</p>
5193 <p>The operation of the '<tt>llvm.part.set</tt>' intrinsic has two modes
5194 of operation: forwards and reverse. If <tt>%lo</tt> is greater than
5195 <tt>%hi</tt> then the intrinsic operates in reverse mode. Otherwise it
5196 operates in forward mode.</p>
5197 <p>For both modes, the <tt>%repl</tt> value is prepared for use by either
5198 truncating it down to the size of the replacement area or zero extending it
5199 up to that size.</p>
5200 <p>In forward mode, the bits between <tt>%lo</tt> and <tt>%hi</tt> (inclusive)
5201 are replaced with corresponding bits from <tt>%repl</tt>. That is the 0th bit
5202 in <tt>%repl</tt> replaces the <tt>%lo</tt>th bit in <tt>%val</tt> and etc. up
5203 to the <tt>%hi</tt>th bit.
5204 <p>In reverse mode, a similar computation is made except that the bits are
5205 reversed. That is, the <tt>0</tt>th bit in <tt>%repl</tt> replaces the
5206 <tt>%hi</tt> bit in <tt>%val</tt> and etc. down to the <tt>%lo</tt>th bit.
5209 llvm.part.set(0xFFFF, 0, 4, 7) -> 0xFF0F
5210 llvm.part.set(0xFFFF, 0, 7, 4) -> 0xFF0F
5211 llvm.part.set(0xFFFF, 1, 7, 4) -> 0xFF8F
5212 llvm.part.set(0xFFFF, F, 8, 3) -> 0xFFE7
5213 llvm.part.set(0xFFFF, 0, 3, 8) -> 0xFE07
5217 <!-- ======================================================================= -->
5218 <div class="doc_subsection">
5219 <a name="int_debugger">Debugger Intrinsics</a>
5222 <div class="doc_text">
5224 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
5225 are described in the <a
5226 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
5227 Debugging</a> document.
5232 <!-- ======================================================================= -->
5233 <div class="doc_subsection">
5234 <a name="int_eh">Exception Handling Intrinsics</a>
5237 <div class="doc_text">
5238 <p> The LLVM exception handling intrinsics (which all start with
5239 <tt>llvm.eh.</tt> prefix), are described in the <a
5240 href="ExceptionHandling.html#format_common_intrinsics">LLVM Exception
5241 Handling</a> document. </p>
5244 <!-- ======================================================================= -->
5245 <div class="doc_subsection">
5246 <a name="int_trampoline">Trampoline Intrinsic</a>
5249 <div class="doc_text">
5251 This intrinsic makes it possible to excise one parameter, marked with
5252 the <tt>nest</tt> attribute, from a function. The result is a callable
5253 function pointer lacking the nest parameter - the caller does not need
5254 to provide a value for it. Instead, the value to use is stored in
5255 advance in a "trampoline", a block of memory usually allocated
5256 on the stack, which also contains code to splice the nest value into the
5257 argument list. This is used to implement the GCC nested function address
5261 For example, if the function is
5262 <tt>i32 f(i8* nest %c, i32 %x, i32 %y)</tt> then the resulting function
5263 pointer has signature <tt>i32 (i32, i32)*</tt>. It can be created as follows:</p>
5265 %tramp = alloca [10 x i8], align 4 ; size and alignment only correct for X86
5266 %tramp1 = getelementptr [10 x i8]* %tramp, i32 0, i32 0
5267 %p = call i8* @llvm.init.trampoline( i8* %tramp1, i8* bitcast (i32 (i8* nest , i32, i32)* @f to i8*), i8* %nval )
5268 %fp = bitcast i8* %p to i32 (i32, i32)*
5270 <p>The call <tt>%val = call i32 %fp( i32 %x, i32 %y )</tt> is then equivalent
5271 to <tt>%val = call i32 %f( i8* %nval, i32 %x, i32 %y )</tt>.</p>
5274 <!-- _______________________________________________________________________ -->
5275 <div class="doc_subsubsection">
5276 <a name="int_it">'<tt>llvm.init.trampoline</tt>' Intrinsic</a>
5278 <div class="doc_text">
5281 declare i8* @llvm.init.trampoline(i8* <tramp>, i8* <func>, i8* <nval>)
5285 This fills the memory pointed to by <tt>tramp</tt> with code
5286 and returns a function pointer suitable for executing it.
5290 The <tt>llvm.init.trampoline</tt> intrinsic takes three arguments, all
5291 pointers. The <tt>tramp</tt> argument must point to a sufficiently large
5292 and sufficiently aligned block of memory; this memory is written to by the
5293 intrinsic. Note that the size and the alignment are target-specific - LLVM
5294 currently provides no portable way of determining them, so a front-end that
5295 generates this intrinsic needs to have some target-specific knowledge.
5296 The <tt>func</tt> argument must hold a function bitcast to an <tt>i8*</tt>.
5300 The block of memory pointed to by <tt>tramp</tt> is filled with target
5301 dependent code, turning it into a function. A pointer to this function is
5302 returned, but needs to be bitcast to an
5303 <a href="#int_trampoline">appropriate function pointer type</a>
5304 before being called. The new function's signature is the same as that of
5305 <tt>func</tt> with any arguments marked with the <tt>nest</tt> attribute
5306 removed. At most one such <tt>nest</tt> argument is allowed, and it must be
5307 of pointer type. Calling the new function is equivalent to calling
5308 <tt>func</tt> with the same argument list, but with <tt>nval</tt> used for the
5309 missing <tt>nest</tt> argument. If, after calling
5310 <tt>llvm.init.trampoline</tt>, the memory pointed to by <tt>tramp</tt> is
5311 modified, then the effect of any later call to the returned function pointer is
5316 <!-- ======================================================================= -->
5317 <div class="doc_subsection">
5318 <a name="int_atomics">Atomic Operations and Synchronization Intrinsics</a>
5321 <div class="doc_text">
5323 These intrinsic functions expand the "universal IR" of LLVM to represent
5324 hardware constructs for atomic operations and memory synchronization. This
5325 provides an interface to the hardware, not an interface to the programmer. It
5326 is aimed at a low enough level to allow any programming models or APIs which
5327 need atomic behaviors to map cleanly onto it. It is also modeled primarily on
5328 hardware behavior. Just as hardware provides a "universal IR" for source
5329 languages, it also provides a starting point for developing a "universal"
5330 atomic operation and synchronization IR.
5333 These do <em>not</em> form an API such as high-level threading libraries,
5334 software transaction memory systems, atomic primitives, and intrinsic
5335 functions as found in BSD, GNU libc, atomic_ops, APR, and other system and
5336 application libraries. The hardware interface provided by LLVM should allow
5337 a clean implementation of all of these APIs and parallel programming models.
5338 No one model or paradigm should be selected above others unless the hardware
5339 itself ubiquitously does so.
5344 <!-- _______________________________________________________________________ -->
5345 <div class="doc_subsubsection">
5346 <a name="int_memory_barrier">'<tt>llvm.memory.barrier</tt>' Intrinsic</a>
5348 <div class="doc_text">
5351 declare void @llvm.memory.barrier( i1 <ll>, i1 <ls>, i1 <sl>, i1 <ss>,
5357 The <tt>llvm.memory.barrier</tt> intrinsic guarantees ordering between
5358 specific pairs of memory access types.
5362 The <tt>llvm.memory.barrier</tt> intrinsic requires five boolean arguments.
5363 The first four arguments enables a specific barrier as listed below. The fith
5364 argument specifies that the barrier applies to io or device or uncached memory.
5368 <li><tt>ll</tt>: load-load barrier</li>
5369 <li><tt>ls</tt>: load-store barrier</li>
5370 <li><tt>sl</tt>: store-load barrier</li>
5371 <li><tt>ss</tt>: store-store barrier</li>
5372 <li><tt>device</tt>: barrier applies to device and uncached memory also.
5376 This intrinsic causes the system to enforce some ordering constraints upon
5377 the loads and stores of the program. This barrier does not indicate
5378 <em>when</em> any events will occur, it only enforces an <em>order</em> in
5379 which they occur. For any of the specified pairs of load and store operations
5380 (f.ex. load-load, or store-load), all of the first operations preceding the
5381 barrier will complete before any of the second operations succeeding the
5382 barrier begin. Specifically the semantics for each pairing is as follows:
5385 <li><tt>ll</tt>: All loads before the barrier must complete before any load
5386 after the barrier begins.</li>
5388 <li><tt>ls</tt>: All loads before the barrier must complete before any
5389 store after the barrier begins.</li>
5390 <li><tt>ss</tt>: All stores before the barrier must complete before any
5391 store after the barrier begins.</li>
5392 <li><tt>sl</tt>: All stores before the barrier must complete before any
5393 load after the barrier begins.</li>
5396 These semantics are applied with a logical "and" behavior when more than one
5397 is enabled in a single memory barrier intrinsic.
5400 Backends may implement stronger barriers than those requested when they do not
5401 support as fine grained a barrier as requested. Some architectures do not
5402 need all types of barriers and on such architectures, these become noops.
5409 %result1 = load i32* %ptr <i>; yields {i32}:result1 = 4</i>
5410 call void @llvm.memory.barrier( i1 false, i1 true, i1 false, i1 false )
5411 <i>; guarantee the above finishes</i>
5412 store i32 8, %ptr <i>; before this begins</i>
5416 <!-- _______________________________________________________________________ -->
5417 <div class="doc_subsubsection">
5418 <a name="int_atomic_lcs">'<tt>llvm.atomic.lcs.*</tt>' Intrinsic</a>
5420 <div class="doc_text">
5423 This is an overloaded intrinsic. You can use <tt>llvm.atomic.lcs</tt> on any
5424 integer bit width. Not all targets support all bit widths however.</p>
5427 declare i8 @llvm.atomic.lcs.i8( i8* <ptr>, i8 <cmp>, i8 <val> )
5428 declare i16 @llvm.atomic.lcs.i16( i16* <ptr>, i16 <cmp>, i16 <val> )
5429 declare i32 @llvm.atomic.lcs.i32( i32* <ptr>, i32 <cmp>, i32 <val> )
5430 declare i64 @llvm.atomic.lcs.i64( i64* <ptr>, i64 <cmp>, i64 <val> )
5435 This loads a value in memory and compares it to a given value. If they are
5436 equal, it stores a new value into the memory.
5440 The <tt>llvm.atomic.lcs</tt> intrinsic takes three arguments. The result as
5441 well as both <tt>cmp</tt> and <tt>val</tt> must be integer values with the
5442 same bit width. The <tt>ptr</tt> argument must be a pointer to a value of
5443 this integer type. While any bit width integer may be used, targets may only
5444 lower representations they support in hardware.
5449 This entire intrinsic must be executed atomically. It first loads the value
5450 in memory pointed to by <tt>ptr</tt> and compares it with the value
5451 <tt>cmp</tt>. If they are equal, <tt>val</tt> is stored into the memory. The
5452 loaded value is yielded in all cases. This provides the equivalent of an
5453 atomic compare-and-swap operation within the SSA framework.
5461 %val1 = add i32 4, 4
5462 %result1 = call i32 @llvm.atomic.lcs.i32( i32* %ptr, i32 4, %val1 )
5463 <i>; yields {i32}:result1 = 4</i>
5464 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
5465 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
5467 %val2 = add i32 1, 1
5468 %result2 = call i32 @llvm.atomic.lcs.i32( i32* %ptr, i32 5, %val2 )
5469 <i>; yields {i32}:result2 = 8</i>
5470 %stored2 = icmp eq i32 %result2, 5 <i>; yields {i1}:stored2 = false</i>
5472 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 8</i>
5476 <!-- _______________________________________________________________________ -->
5477 <div class="doc_subsubsection">
5478 <a name="int_atomic_swap">'<tt>llvm.atomic.swap.*</tt>' Intrinsic</a>
5480 <div class="doc_text">
5484 This is an overloaded intrinsic. You can use <tt>llvm.atomic.swap</tt> on any
5485 integer bit width. Not all targets support all bit widths however.</p>
5487 declare i8 @llvm.atomic.swap.i8( i8* <ptr>, i8 <val> )
5488 declare i16 @llvm.atomic.swap.i16( i16* <ptr>, i16 <val> )
5489 declare i32 @llvm.atomic.swap.i32( i32* <ptr>, i32 <val> )
5490 declare i64 @llvm.atomic.swap.i64( i64* <ptr>, i64 <val> )
5495 This intrinsic loads the value stored in memory at <tt>ptr</tt> and yields
5496 the value from memory. It then stores the value in <tt>val</tt> in the memory
5502 The <tt>llvm.atomic.ls</tt> intrinsic takes two arguments. Both the
5503 <tt>val</tt> argument and the result must be integers of the same bit width.
5504 The first argument, <tt>ptr</tt>, must be a pointer to a value of this
5505 integer type. The targets may only lower integer representations they
5510 This intrinsic loads the value pointed to by <tt>ptr</tt>, yields it, and
5511 stores <tt>val</tt> back into <tt>ptr</tt> atomically. This provides the
5512 equivalent of an atomic swap operation within the SSA framework.
5520 %val1 = add i32 4, 4
5521 %result1 = call i32 @llvm.atomic.swap.i32( i32* %ptr, i32 %val1 )
5522 <i>; yields {i32}:result1 = 4</i>
5523 %stored1 = icmp eq i32 %result1, 4 <i>; yields {i1}:stored1 = true</i>
5524 %memval1 = load i32* %ptr <i>; yields {i32}:memval1 = 8</i>
5526 %val2 = add i32 1, 1
5527 %result2 = call i32 @llvm.atomic.swap.i32( i32* %ptr, i32 %val2 )
5528 <i>; yields {i32}:result2 = 8</i>
5530 %stored2 = icmp eq i32 %result2, 8 <i>; yields {i1}:stored2 = true</i>
5531 %memval2 = load i32* %ptr <i>; yields {i32}:memval2 = 2</i>
5535 <!-- _______________________________________________________________________ -->
5536 <div class="doc_subsubsection">
5537 <a name="int_atomic_las">'<tt>llvm.atomic.las.*</tt>' Intrinsic</a>
5540 <div class="doc_text">
5543 This is an overloaded intrinsic. You can use <tt>llvm.atomic.las</tt> on any
5544 integer bit width. Not all targets support all bit widths however.</p>
5546 declare i8 @llvm.atomic.las.i8.( i8* <ptr>, i8 <delta> )
5547 declare i16 @llvm.atomic.las.i16.( i16* <ptr>, i16 <delta> )
5548 declare i32 @llvm.atomic.las.i32.( i32* <ptr>, i32 <delta> )
5549 declare i64 @llvm.atomic.las.i64.( i64* <ptr>, i64 <delta> )
5554 This intrinsic adds <tt>delta</tt> to the value stored in memory at
5555 <tt>ptr</tt>. It yields the original value at <tt>ptr</tt>.
5560 The intrinsic takes two arguments, the first a pointer to an integer value
5561 and the second an integer value. The result is also an integer value. These
5562 integer types can have any bit width, but they must all have the same bit
5563 width. The targets may only lower integer representations they support.
5567 This intrinsic does a series of operations atomically. It first loads the
5568 value stored at <tt>ptr</tt>. It then adds <tt>delta</tt>, stores the result
5569 to <tt>ptr</tt>. It yields the original value stored at <tt>ptr</tt>.
5576 %result1 = call i32 @llvm.atomic.las.i32( i32* %ptr, i32 4 )
5577 <i>; yields {i32}:result1 = 4</i>
5578 %result2 = call i32 @llvm.atomic.las.i32( i32* %ptr, i32 2 )
5579 <i>; yields {i32}:result2 = 8</i>
5580 %result3 = call i32 @llvm.atomic.las.i32( i32* %ptr, i32 5 )
5581 <i>; yields {i32}:result3 = 10</i>
5582 %memval = load i32* %ptr <i>; yields {i32}:memval1 = 15</i>
5587 <!-- ======================================================================= -->
5588 <div class="doc_subsection">
5589 <a name="int_general">General Intrinsics</a>
5592 <div class="doc_text">
5593 <p> This class of intrinsics is designed to be generic and has
5594 no specific purpose. </p>
5597 <!-- _______________________________________________________________________ -->
5598 <div class="doc_subsubsection">
5599 <a name="int_var_annotation">'<tt>llvm.var.annotation</tt>' Intrinsic</a>
5602 <div class="doc_text">
5606 declare void @llvm.var.annotation(i8* <val>, i8* <str>, i8* <str>, i32 <int> )
5612 The '<tt>llvm.var.annotation</tt>' intrinsic
5618 The first argument is a pointer to a value, the second is a pointer to a
5619 global string, the third is a pointer to a global string which is the source
5620 file name, and the last argument is the line number.
5626 This intrinsic allows annotation of local variables with arbitrary strings.
5627 This can be useful for special purpose optimizations that want to look for these
5628 annotations. These have no other defined use, they are ignored by code
5629 generation and optimization.
5633 <!-- _______________________________________________________________________ -->
5634 <div class="doc_subsubsection">
5635 <a name="int_annotation">'<tt>llvm.annotation.*</tt>' Intrinsic</a>
5638 <div class="doc_text">
5641 <p>This is an overloaded intrinsic. You can use '<tt>llvm.annotation</tt>' on
5642 any integer bit width.
5645 declare i8 @llvm.annotation.i8(i8 <val>, i8* <str>, i8* <str>, i32 <int> )
5646 declare i16 @llvm.annotation.i16(i16 <val>, i8* <str>, i8* <str>, i32 <int> )
5647 declare i32 @llvm.annotation.i32(i32 <val>, i8* <str>, i8* <str>, i32 <int> )
5648 declare i64 @llvm.annotation.i64(i64 <val>, i8* <str>, i8* <str>, i32 <int> )
5649 declare i256 @llvm.annotation.i256(i256 <val>, i8* <str>, i8* <str>, i32 <int> )
5655 The '<tt>llvm.annotation</tt>' intrinsic.
5661 The first argument is an integer value (result of some expression),
5662 the second is a pointer to a global string, the third is a pointer to a global
5663 string which is the source file name, and the last argument is the line number.
5664 It returns the value of the first argument.
5670 This intrinsic allows annotations to be put on arbitrary expressions
5671 with arbitrary strings. This can be useful for special purpose optimizations
5672 that want to look for these annotations. These have no other defined use, they
5673 are ignored by code generation and optimization.
5676 <!-- _______________________________________________________________________ -->
5677 <div class="doc_subsubsection">
5678 <a name="int_trap">'<tt>llvm.trap</tt>' Intrinsic</a>
5681 <div class="doc_text">
5685 declare void @llvm.trap()
5691 The '<tt>llvm.trap</tt>' intrinsic
5703 This intrinsics is lowered to the target dependent trap instruction. If the
5704 target does not have a trap instruction, this intrinsic will be lowered to the
5705 call of the abort() function.
5709 <!-- *********************************************************************** -->
5712 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
5713 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
5714 <a href="http://validator.w3.org/check/referer"><img
5715 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!"></a>
5717 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
5718 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
5719 Last modified: $Date$