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="#moduleasm">Module-Level Inline Assembly</a></li>
30 <li><a href="#typesystem">Type System</a>
32 <li><a href="#t_primitive">Primitive Types</a>
34 <li><a href="#t_classifications">Type Classifications</a></li>
37 <li><a href="#t_derived">Derived Types</a>
39 <li><a href="#t_array">Array Type</a></li>
40 <li><a href="#t_function">Function Type</a></li>
41 <li><a href="#t_pointer">Pointer Type</a></li>
42 <li><a href="#t_struct">Structure Type</a></li>
43 <li><a href="#t_packed">Packed Type</a></li>
44 <li><a href="#t_opaque">Opaque Type</a></li>
49 <li><a href="#constants">Constants</a>
51 <li><a href="#simpleconstants">Simple Constants</a>
52 <li><a href="#aggregateconstants">Aggregate Constants</a>
53 <li><a href="#globalconstants">Global Variable and Function Addresses</a>
54 <li><a href="#undefvalues">Undefined Values</a>
55 <li><a href="#constantexprs">Constant Expressions</a>
58 <li><a href="#othervalues">Other Values</a>
60 <li><a href="#inlineasm">Inline Assembler Expressions</a>
63 <li><a href="#instref">Instruction Reference</a>
65 <li><a href="#terminators">Terminator Instructions</a>
67 <li><a href="#i_ret">'<tt>ret</tt>' Instruction</a></li>
68 <li><a href="#i_br">'<tt>br</tt>' Instruction</a></li>
69 <li><a href="#i_switch">'<tt>switch</tt>' Instruction</a></li>
70 <li><a href="#i_invoke">'<tt>invoke</tt>' Instruction</a></li>
71 <li><a href="#i_unwind">'<tt>unwind</tt>' Instruction</a></li>
72 <li><a href="#i_unreachable">'<tt>unreachable</tt>' Instruction</a></li>
75 <li><a href="#binaryops">Binary Operations</a>
77 <li><a href="#i_add">'<tt>add</tt>' Instruction</a></li>
78 <li><a href="#i_sub">'<tt>sub</tt>' Instruction</a></li>
79 <li><a href="#i_mul">'<tt>mul</tt>' Instruction</a></li>
80 <li><a href="#i_udiv">'<tt>udiv</tt>' Instruction</a></li>
81 <li><a href="#i_sdiv">'<tt>sdiv</tt>' Instruction</a></li>
82 <li><a href="#i_fdiv">'<tt>fdiv</tt>' Instruction</a></li>
83 <li><a href="#i_rem">'<tt>rem</tt>' Instruction</a></li>
84 <li><a href="#i_setcc">'<tt>set<i>cc</i></tt>' Instructions</a></li>
87 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
89 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
90 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
91 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
92 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
93 <li><a href="#i_shr">'<tt>shr</tt>' Instruction</a></li>
96 <li><a href="#vectorops">Vector Operations</a>
98 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
99 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
100 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
101 <li><a href="#i_vsetint">'<tt>vsetint</tt>' Instruction</a></li>
102 <li><a href="#i_vsetfp">'<tt>vsetfp</tt>' Instruction</a></li>
103 <li><a href="#i_vselect">'<tt>vselect</tt>' Instruction</a></li>
106 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
108 <li><a href="#i_malloc">'<tt>malloc</tt>' Instruction</a></li>
109 <li><a href="#i_free">'<tt>free</tt>' Instruction</a></li>
110 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
111 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
112 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
113 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
116 <li><a href="#otherops">Other Operations</a>
118 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
119 <li><a href="#i_cast">'<tt>cast .. to</tt>' Instruction</a></li>
120 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
121 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
122 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
127 <li><a href="#intrinsics">Intrinsic Functions</a>
129 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
131 <li><a href="#i_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
132 <li><a href="#i_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
133 <li><a href="#i_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
136 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
138 <li><a href="#i_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
139 <li><a href="#i_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
140 <li><a href="#i_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
143 <li><a href="#int_codegen">Code Generator Intrinsics</a>
145 <li><a href="#i_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
146 <li><a href="#i_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
147 <li><a href="#i_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
148 <li><a href="#i_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
149 <li><a href="#i_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
150 <li><a href="#i_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
151 <li><a href="#i_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
154 <li><a href="#int_libc">Standard C Library Intrinsics</a>
156 <li><a href="#i_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
157 <li><a href="#i_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
158 <li><a href="#i_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
159 <li><a href="#i_isunordered">'<tt>llvm.isunordered.*</tt>' Intrinsic</a></li>
160 <li><a href="#i_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
161 <li><a href="#i_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
164 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
166 <li><a href="#i_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
167 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
168 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
169 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
172 <li><a href="#int_debugger">Debugger intrinsics</a></li>
177 <div class="doc_author">
178 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
179 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
182 <!-- *********************************************************************** -->
183 <div class="doc_section"> <a name="abstract">Abstract </a></div>
184 <!-- *********************************************************************** -->
186 <div class="doc_text">
187 <p>This document is a reference manual for the LLVM assembly language.
188 LLVM is an SSA based representation that provides type safety,
189 low-level operations, flexibility, and the capability of representing
190 'all' high-level languages cleanly. It is the common code
191 representation used throughout all phases of the LLVM compilation
195 <!-- *********************************************************************** -->
196 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
197 <!-- *********************************************************************** -->
199 <div class="doc_text">
201 <p>The LLVM code representation is designed to be used in three
202 different forms: as an in-memory compiler IR, as an on-disk bytecode
203 representation (suitable for fast loading by a Just-In-Time compiler),
204 and as a human readable assembly language representation. This allows
205 LLVM to provide a powerful intermediate representation for efficient
206 compiler transformations and analysis, while providing a natural means
207 to debug and visualize the transformations. The three different forms
208 of LLVM are all equivalent. This document describes the human readable
209 representation and notation.</p>
211 <p>The LLVM representation aims to be light-weight and low-level
212 while being expressive, typed, and extensible at the same time. It
213 aims to be a "universal IR" of sorts, by being at a low enough level
214 that high-level ideas may be cleanly mapped to it (similar to how
215 microprocessors are "universal IR's", allowing many source languages to
216 be mapped to them). By providing type information, LLVM can be used as
217 the target of optimizations: for example, through pointer analysis, it
218 can be proven that a C automatic variable is never accessed outside of
219 the current function... allowing it to be promoted to a simple SSA
220 value instead of a memory location.</p>
224 <!-- _______________________________________________________________________ -->
225 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
227 <div class="doc_text">
229 <p>It is important to note that this document describes 'well formed'
230 LLVM assembly language. There is a difference between what the parser
231 accepts and what is considered 'well formed'. For example, the
232 following instruction is syntactically okay, but not well formed:</p>
235 %x = <a href="#i_add">add</a> int 1, %x
238 <p>...because the definition of <tt>%x</tt> does not dominate all of
239 its uses. The LLVM infrastructure provides a verification pass that may
240 be used to verify that an LLVM module is well formed. This pass is
241 automatically run by the parser after parsing input assembly and by
242 the optimizer before it outputs bytecode. The violations pointed out
243 by the verifier pass indicate bugs in transformation passes or input to
246 <!-- Describe the typesetting conventions here. --> </div>
248 <!-- *********************************************************************** -->
249 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
250 <!-- *********************************************************************** -->
252 <div class="doc_text">
254 <p>LLVM uses three different forms of identifiers, for different
258 <li>Named values are represented as a string of characters with a '%' prefix.
259 For example, %foo, %DivisionByZero, %a.really.long.identifier. The actual
260 regular expression used is '<tt>%[a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
261 Identifiers which require other characters in their names can be surrounded
262 with quotes. In this way, anything except a <tt>"</tt> character can be used
265 <li>Unnamed values are represented as an unsigned numeric value with a '%'
266 prefix. For example, %12, %2, %44.</li>
268 <li>Constants, which are described in a <a href="#constants">section about
269 constants</a>, below.</li>
272 <p>LLVM requires that values start with a '%' sign for two reasons: Compilers
273 don't need to worry about name clashes with reserved words, and the set of
274 reserved words may be expanded in the future without penalty. Additionally,
275 unnamed identifiers allow a compiler to quickly come up with a temporary
276 variable without having to avoid symbol table conflicts.</p>
278 <p>Reserved words in LLVM are very similar to reserved words in other
279 languages. There are keywords for different opcodes ('<tt><a
280 href="#i_add">add</a></tt>', '<tt><a href="#i_cast">cast</a></tt>', '<tt><a
281 href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
282 href="#t_void">void</a></tt>', '<tt><a href="#t_uint">uint</a></tt>', etc...),
283 and others. These reserved words cannot conflict with variable names, because
284 none of them start with a '%' character.</p>
286 <p>Here is an example of LLVM code to multiply the integer variable
287 '<tt>%X</tt>' by 8:</p>
292 %result = <a href="#i_mul">mul</a> uint %X, 8
295 <p>After strength reduction:</p>
298 %result = <a href="#i_shl">shl</a> uint %X, ubyte 3
301 <p>And the hard way:</p>
304 <a href="#i_add">add</a> uint %X, %X <i>; yields {uint}:%0</i>
305 <a href="#i_add">add</a> uint %0, %0 <i>; yields {uint}:%1</i>
306 %result = <a href="#i_add">add</a> uint %1, %1
309 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
310 important lexical features of LLVM:</p>
314 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
317 <li>Unnamed temporaries are created when the result of a computation is not
318 assigned to a named value.</li>
320 <li>Unnamed temporaries are numbered sequentially</li>
324 <p>...and it also shows a convention that we follow in this document. When
325 demonstrating instructions, we will follow an instruction with a comment that
326 defines the type and name of value produced. Comments are shown in italic
331 <!-- *********************************************************************** -->
332 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
333 <!-- *********************************************************************** -->
335 <!-- ======================================================================= -->
336 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
339 <div class="doc_text">
341 <p>LLVM programs are composed of "Module"s, each of which is a
342 translation unit of the input programs. Each module consists of
343 functions, global variables, and symbol table entries. Modules may be
344 combined together with the LLVM linker, which merges function (and
345 global variable) definitions, resolves forward declarations, and merges
346 symbol table entries. Here is an example of the "hello world" module:</p>
348 <pre><i>; Declare the string constant as a global constant...</i>
349 <a href="#identifiers">%.LC0</a> = <a href="#linkage_internal">internal</a> <a
350 href="#globalvars">constant</a> <a href="#t_array">[13 x sbyte]</a> c"hello world\0A\00" <i>; [13 x sbyte]*</i>
352 <i>; External declaration of the puts function</i>
353 <a href="#functionstructure">declare</a> int %puts(sbyte*) <i>; int(sbyte*)* </i>
355 <i>; Global variable / Function body section separator</i>
358 <i>; Definition of main function</i>
359 int %main() { <i>; int()* </i>
360 <i>; Convert [13x sbyte]* to sbyte *...</i>
362 href="#i_getelementptr">getelementptr</a> [13 x sbyte]* %.LC0, long 0, long 0 <i>; sbyte*</i>
364 <i>; Call puts function to write out the string to stdout...</i>
366 href="#i_call">call</a> int %puts(sbyte* %cast210) <i>; int</i>
368 href="#i_ret">ret</a> int 0<br>}<br></pre>
370 <p>This example is made up of a <a href="#globalvars">global variable</a>
371 named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
372 function, and a <a href="#functionstructure">function definition</a>
373 for "<tt>main</tt>".</p>
375 <p>In general, a module is made up of a list of global values,
376 where both functions and global variables are global values. Global values are
377 represented by a pointer to a memory location (in this case, a pointer to an
378 array of char, and a pointer to a function), and have one of the following <a
379 href="#linkage">linkage types</a>.</p>
381 <p>Due to a limitation in the current LLVM assembly parser (it is limited by
382 one-token lookahead), modules are split into two pieces by the "implementation"
383 keyword. Global variable prototypes and definitions must occur before the
384 keyword, and function definitions must occur after it. Function prototypes may
385 occur either before or after it. In the future, the implementation keyword may
386 become a noop, if the parser gets smarter.</p>
390 <!-- ======================================================================= -->
391 <div class="doc_subsection">
392 <a name="linkage">Linkage Types</a>
395 <div class="doc_text">
398 All Global Variables and Functions have one of the following types of linkage:
403 <dt><tt><b><a name="linkage_internal">internal</a></b></tt> </dt>
405 <dd>Global values with internal linkage are only directly accessible by
406 objects in the current module. In particular, linking code into a module with
407 an internal global value may cause the internal to be renamed as necessary to
408 avoid collisions. Because the symbol is internal to the module, all
409 references can be updated. This corresponds to the notion of the
410 '<tt>static</tt>' keyword in C, or the idea of "anonymous namespaces" in C++.
413 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
415 <dd>"<tt>linkonce</tt>" linkage is similar to <tt>internal</tt> linkage, with
416 the twist that linking together two modules defining the same
417 <tt>linkonce</tt> globals will cause one of the globals to be discarded. This
418 is typically used to implement inline functions. Unreferenced
419 <tt>linkonce</tt> globals are allowed to be discarded.
422 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
424 <dd>"<tt>weak</tt>" linkage is exactly the same as <tt>linkonce</tt> linkage,
425 except that unreferenced <tt>weak</tt> globals may not be discarded. This is
426 used to implement constructs in C such as "<tt>int X;</tt>" at global scope.
429 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
431 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
432 pointer to array type. When two global variables with appending linkage are
433 linked together, the two global arrays are appended together. This is the
434 LLVM, typesafe, equivalent of having the system linker append together
435 "sections" with identical names when .o files are linked.
438 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
440 <dd>If none of the above identifiers are used, the global is externally
441 visible, meaning that it participates in linkage and can be used to resolve
442 external symbol references.
445 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt>: </dt>
447 <dd>"<tt>extern_weak</tt>" TBD
451 The next two types of linkage are targeted for Microsoft Windows platform
452 only. They are designed to support importing (exporting) symbols from (to)
456 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt>: </dt>
458 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
459 or variable via a global pointer to a pointer that is set up by the DLL
460 exporting the symbol. On Microsoft Windows targets, the pointer name is
461 formed by combining <code>_imp__</code> and the function or variable name.
464 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt>: </dt>
466 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
467 pointer to a pointer in a DLL, so that it can be referenced with the
468 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
469 name is formed by combining <code>_imp__</code> and the function or variable
475 <p><a name="linkage_external">For example, since the "<tt>.LC0</tt>"
476 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
477 variable and was linked with this one, one of the two would be renamed,
478 preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
479 external (i.e., lacking any linkage declarations), they are accessible
480 outside of the current module. It is illegal for a function <i>declaration</i>
481 to have any linkage type other than "externally visible".</a></p>
485 <!-- ======================================================================= -->
486 <div class="doc_subsection">
487 <a name="callingconv">Calling Conventions</a>
490 <div class="doc_text">
492 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
493 and <a href="#i_invoke">invokes</a> can all have an optional calling convention
494 specified for the call. The calling convention of any pair of dynamic
495 caller/callee must match, or the behavior of the program is undefined. The
496 following calling conventions are supported by LLVM, and more may be added in
500 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
502 <dd>This calling convention (the default if no other calling convention is
503 specified) matches the target C calling conventions. This calling convention
504 supports varargs function calls and tolerates some mismatch in the declared
505 prototype and implemented declaration of the function (as does normal C).
508 <dt><b>"<tt>csretcc</tt>" - The C struct return calling convention</b>:</dt>
510 <dd>This calling convention matches the target C calling conventions, except
511 that functions with this convention are required to take a pointer as their
512 first argument, and the return type of the function must be void. This is
513 used for C functions that return aggregates by-value. In this case, the
514 function has been transformed to take a pointer to the struct as the first
515 argument to the function. For targets where the ABI specifies specific
516 behavior for structure-return calls, the calling convention can be used to
517 distinguish between struct return functions and other functions that take a
518 pointer to a struct as the first argument.
521 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
523 <dd>This calling convention attempts to make calls as fast as possible
524 (e.g. by passing things in registers). This calling convention allows the
525 target to use whatever tricks it wants to produce fast code for the target,
526 without having to conform to an externally specified ABI. Implementations of
527 this convention should allow arbitrary tail call optimization to be supported.
528 This calling convention does not support varargs and requires the prototype of
529 all callees to exactly match the prototype of the function definition.
532 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
534 <dd>This calling convention attempts to make code in the caller as efficient
535 as possible under the assumption that the call is not commonly executed. As
536 such, these calls often preserve all registers so that the call does not break
537 any live ranges in the caller side. This calling convention does not support
538 varargs and requires the prototype of all callees to exactly match the
539 prototype of the function definition.
542 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
544 <dd>Any calling convention may be specified by number, allowing
545 target-specific calling conventions to be used. Target specific calling
546 conventions start at 64.
550 <p>More calling conventions can be added/defined on an as-needed basis, to
551 support pascal conventions or any other well-known target-independent
556 <!-- ======================================================================= -->
557 <div class="doc_subsection">
558 <a name="globalvars">Global Variables</a>
561 <div class="doc_text">
563 <p>Global variables define regions of memory allocated at compilation time
564 instead of run-time. Global variables may optionally be initialized, may have
565 an explicit section to be placed in, and may
566 have an optional explicit alignment specified. A
567 variable may be defined as a global "constant," which indicates that the
568 contents of the variable will <b>never</b> be modified (enabling better
569 optimization, allowing the global data to be placed in the read-only section of
570 an executable, etc). Note that variables that need runtime initialization
571 cannot be marked "constant" as there is a store to the variable.</p>
574 LLVM explicitly allows <em>declarations</em> of global variables to be marked
575 constant, even if the final definition of the global is not. This capability
576 can be used to enable slightly better optimization of the program, but requires
577 the language definition to guarantee that optimizations based on the
578 'constantness' are valid for the translation units that do not include the
582 <p>As SSA values, global variables define pointer values that are in
583 scope (i.e. they dominate) all basic blocks in the program. Global
584 variables always define a pointer to their "content" type because they
585 describe a region of memory, and all memory objects in LLVM are
586 accessed through pointers.</p>
588 <p>LLVM allows an explicit section to be specified for globals. If the target
589 supports it, it will emit globals to the section specified.</p>
591 <p>An explicit alignment may be specified for a global. If not present, or if
592 the alignment is set to zero, the alignment of the global is set by the target
593 to whatever it feels convenient. If an explicit alignment is specified, the
594 global is forced to have at least that much alignment. All alignments must be
600 <!-- ======================================================================= -->
601 <div class="doc_subsection">
602 <a name="functionstructure">Functions</a>
605 <div class="doc_text">
607 <p>LLVM function definitions consist of an optional <a href="#linkage">linkage
608 type</a>, an optional <a href="#callingconv">calling convention</a>, a return
609 type, a function name, a (possibly empty) argument list, an optional section,
610 an optional alignment, an opening curly brace,
611 a list of basic blocks, and a closing curly brace. LLVM function declarations
612 are defined with the "<tt>declare</tt>" keyword, an optional <a
613 href="#callingconv">calling convention</a>, a return type, a function name,
614 a possibly empty list of arguments, and an optional alignment.</p>
616 <p>A function definition contains a list of basic blocks, forming the CFG for
617 the function. Each basic block may optionally start with a label (giving the
618 basic block a symbol table entry), contains a list of instructions, and ends
619 with a <a href="#terminators">terminator</a> instruction (such as a branch or
620 function return).</p>
622 <p>The first basic block in a program is special in two ways: it is immediately
623 executed on entrance to the function, and it is not allowed to have predecessor
624 basic blocks (i.e. there can not be any branches to the entry block of a
625 function). Because the block can have no predecessors, it also cannot have any
626 <a href="#i_phi">PHI nodes</a>.</p>
628 <p>LLVM functions are identified by their name and type signature. Hence, two
629 functions with the same name but different parameter lists or return values are
630 considered different functions, and LLVM will resolve references to each
633 <p>LLVM allows an explicit section to be specified for functions. If the target
634 supports it, it will emit functions to the section specified.</p>
636 <p>An explicit alignment may be specified for a function. If not present, or if
637 the alignment is set to zero, the alignment of the function is set by the target
638 to whatever it feels convenient. If an explicit alignment is specified, the
639 function is forced to have at least that much alignment. All alignments must be
644 <!-- ======================================================================= -->
645 <div class="doc_subsection">
646 <a name="moduleasm">Module-Level Inline Assembly</a>
649 <div class="doc_text">
651 Modules may contain "module-level inline asm" blocks, which corresponds to the
652 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
653 LLVM and treated as a single unit, but may be separated in the .ll file if
654 desired. The syntax is very simple:
657 <div class="doc_code"><pre>
658 module asm "inline asm code goes here"
659 module asm "more can go here"
662 <p>The strings can contain any character by escaping non-printable characters.
663 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
668 The inline asm code is simply printed to the machine code .s file when
669 assembly code is generated.
674 <!-- *********************************************************************** -->
675 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
676 <!-- *********************************************************************** -->
678 <div class="doc_text">
680 <p>The LLVM type system is one of the most important features of the
681 intermediate representation. Being typed enables a number of
682 optimizations to be performed on the IR directly, without having to do
683 extra analyses on the side before the transformation. A strong type
684 system makes it easier to read the generated code and enables novel
685 analyses and transformations that are not feasible to perform on normal
686 three address code representations.</p>
690 <!-- ======================================================================= -->
691 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
692 <div class="doc_text">
693 <p>The primitive types are the fundamental building blocks of the LLVM
694 system. The current set of primitive types is as follows:</p>
696 <table class="layout">
701 <tr><th>Type</th><th>Description</th></tr>
702 <tr><td><tt>void</tt></td><td>No value</td></tr>
703 <tr><td><tt>ubyte</tt></td><td>Unsigned 8-bit value</td></tr>
704 <tr><td><tt>ushort</tt></td><td>Unsigned 16-bit value</td></tr>
705 <tr><td><tt>uint</tt></td><td>Unsigned 32-bit value</td></tr>
706 <tr><td><tt>ulong</tt></td><td>Unsigned 64-bit value</td></tr>
707 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
708 <tr><td><tt>label</tt></td><td>Branch destination</td></tr>
715 <tr><th>Type</th><th>Description</th></tr>
716 <tr><td><tt>bool</tt></td><td>True or False value</td></tr>
717 <tr><td><tt>sbyte</tt></td><td>Signed 8-bit value</td></tr>
718 <tr><td><tt>short</tt></td><td>Signed 16-bit value</td></tr>
719 <tr><td><tt>int</tt></td><td>Signed 32-bit value</td></tr>
720 <tr><td><tt>long</tt></td><td>Signed 64-bit value</td></tr>
721 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
729 <!-- _______________________________________________________________________ -->
730 <div class="doc_subsubsection"> <a name="t_classifications">Type
731 Classifications</a> </div>
732 <div class="doc_text">
733 <p>These different primitive types fall into a few useful
736 <table border="1" cellspacing="0" cellpadding="4">
738 <tr><th>Classification</th><th>Types</th></tr>
740 <td><a name="t_signed">signed</a></td>
741 <td><tt>sbyte, short, int, long, float, double</tt></td>
744 <td><a name="t_unsigned">unsigned</a></td>
745 <td><tt>ubyte, ushort, uint, ulong</tt></td>
748 <td><a name="t_integer">integer</a></td>
749 <td><tt>ubyte, sbyte, ushort, short, uint, int, ulong, long</tt></td>
752 <td><a name="t_integral">integral</a></td>
753 <td><tt>bool, ubyte, sbyte, ushort, short, uint, int, ulong, long</tt>
757 <td><a name="t_floating">floating point</a></td>
758 <td><tt>float, double</tt></td>
761 <td><a name="t_firstclass">first class</a></td>
762 <td><tt>bool, ubyte, sbyte, ushort, short, uint, int, ulong, long,<br>
763 float, double, <a href="#t_pointer">pointer</a>,
764 <a href="#t_packed">packed</a></tt></td>
769 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
770 most important. Values of these types are the only ones which can be
771 produced by instructions, passed as arguments, or used as operands to
772 instructions. This means that all structures and arrays must be
773 manipulated either by pointer or by component.</p>
776 <!-- ======================================================================= -->
777 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
779 <div class="doc_text">
781 <p>The real power in LLVM comes from the derived types in the system.
782 This is what allows a programmer to represent arrays, functions,
783 pointers, and other useful types. Note that these derived types may be
784 recursive: For example, it is possible to have a two dimensional array.</p>
788 <!-- _______________________________________________________________________ -->
789 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
791 <div class="doc_text">
795 <p>The array type is a very simple derived type that arranges elements
796 sequentially in memory. The array type requires a size (number of
797 elements) and an underlying data type.</p>
802 [<# elements> x <elementtype>]
805 <p>The number of elements is a constant integer value; elementtype may
806 be any type with a size.</p>
809 <table class="layout">
812 <tt>[40 x int ]</tt><br/>
813 <tt>[41 x int ]</tt><br/>
814 <tt>[40 x uint]</tt><br/>
817 Array of 40 integer values.<br/>
818 Array of 41 integer values.<br/>
819 Array of 40 unsigned integer values.<br/>
823 <p>Here are some examples of multidimensional arrays:</p>
824 <table class="layout">
827 <tt>[3 x [4 x int]]</tt><br/>
828 <tt>[12 x [10 x float]]</tt><br/>
829 <tt>[2 x [3 x [4 x uint]]]</tt><br/>
832 3x4 array of integer values.<br/>
833 12x10 array of single precision floating point values.<br/>
834 2x3x4 array of unsigned integer values.<br/>
839 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
840 length array. Normally, accesses past the end of an array are undefined in
841 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
842 As a special case, however, zero length arrays are recognized to be variable
843 length. This allows implementation of 'pascal style arrays' with the LLVM
844 type "{ int, [0 x float]}", for example.</p>
848 <!-- _______________________________________________________________________ -->
849 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
850 <div class="doc_text">
852 <p>The function type can be thought of as a function signature. It
853 consists of a return type and a list of formal parameter types.
854 Function types are usually used to build virtual function tables
855 (which are structures of pointers to functions), for indirect function
856 calls, and when defining a function.</p>
858 The return type of a function type cannot be an aggregate type.
861 <pre> <returntype> (<parameter list>)<br></pre>
862 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
863 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
864 which indicates that the function takes a variable number of arguments.
865 Variable argument functions can access their arguments with the <a
866 href="#int_varargs">variable argument handling intrinsic</a> functions.</p>
868 <table class="layout">
871 <tt>int (int)</tt> <br/>
872 <tt>float (int, int *) *</tt><br/>
873 <tt>int (sbyte *, ...)</tt><br/>
876 function taking an <tt>int</tt>, returning an <tt>int</tt><br/>
877 <a href="#t_pointer">Pointer</a> to a function that takes an
878 <tt>int</tt> and a <a href="#t_pointer">pointer</a> to <tt>int</tt>,
879 returning <tt>float</tt>.<br/>
880 A vararg function that takes at least one <a href="#t_pointer">pointer</a>
881 to <tt>sbyte</tt> (signed char in C), which returns an integer. This is
882 the signature for <tt>printf</tt> in LLVM.<br/>
888 <!-- _______________________________________________________________________ -->
889 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
890 <div class="doc_text">
892 <p>The structure type is used to represent a collection of data members
893 together in memory. The packing of the field types is defined to match
894 the ABI of the underlying processor. The elements of a structure may
895 be any type that has a size.</p>
896 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
897 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
898 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
901 <pre> { <type list> }<br></pre>
903 <table class="layout">
906 <tt>{ int, int, int }</tt><br/>
907 <tt>{ float, int (int) * }</tt><br/>
910 a triple of three <tt>int</tt> values<br/>
911 A pair, where the first element is a <tt>float</tt> and the second element
912 is a <a href="#t_pointer">pointer</a> to a <a href="#t_function">function</a>
913 that takes an <tt>int</tt>, returning an <tt>int</tt>.<br/>
919 <!-- _______________________________________________________________________ -->
920 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
921 <div class="doc_text">
923 <p>As in many languages, the pointer type represents a pointer or
924 reference to another object, which must live in memory.</p>
926 <pre> <type> *<br></pre>
928 <table class="layout">
931 <tt>[4x int]*</tt><br/>
932 <tt>int (int *) *</tt><br/>
935 A <a href="#t_pointer">pointer</a> to <a href="#t_array">array</a> of
936 four <tt>int</tt> values<br/>
937 A <a href="#t_pointer">pointer</a> to a <a
938 href="#t_function">function</a> that takes an <tt>int*</tt>, returning an
945 <!-- _______________________________________________________________________ -->
946 <div class="doc_subsubsection"> <a name="t_packed">Packed Type</a> </div>
947 <div class="doc_text">
951 <p>A packed type is a simple derived type that represents a vector
952 of elements. Packed types are used when multiple primitive data
953 are operated in parallel using a single instruction (SIMD).
954 A packed type requires a size (number of
955 elements) and an underlying primitive data type. Vectors must have a power
956 of two length (1, 2, 4, 8, 16 ...). Packed types are
957 considered <a href="#t_firstclass">first class</a>.</p>
962 < <# elements> x <elementtype> >
965 <p>The number of elements is a constant integer value; elementtype may
966 be any integral or floating point type.</p>
970 <table class="layout">
973 <tt><4 x int></tt><br/>
974 <tt><8 x float></tt><br/>
975 <tt><2 x uint></tt><br/>
978 Packed vector of 4 integer values.<br/>
979 Packed vector of 8 floating-point values.<br/>
980 Packed vector of 2 unsigned integer values.<br/>
986 <!-- _______________________________________________________________________ -->
987 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
988 <div class="doc_text">
992 <p>Opaque types are used to represent unknown types in the system. This
993 corresponds (for example) to the C notion of a foward declared structure type.
994 In LLVM, opaque types can eventually be resolved to any type (not just a
1005 <table class="layout">
1011 An opaque type.<br/>
1018 <!-- *********************************************************************** -->
1019 <div class="doc_section"> <a name="constants">Constants</a> </div>
1020 <!-- *********************************************************************** -->
1022 <div class="doc_text">
1024 <p>LLVM has several different basic types of constants. This section describes
1025 them all and their syntax.</p>
1029 <!-- ======================================================================= -->
1030 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1032 <div class="doc_text">
1035 <dt><b>Boolean constants</b></dt>
1037 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1038 constants of the <tt><a href="#t_primitive">bool</a></tt> type.
1041 <dt><b>Integer constants</b></dt>
1043 <dd>Standard integers (such as '4') are constants of the <a
1044 href="#t_integer">integer</a> type. Negative numbers may be used with signed
1048 <dt><b>Floating point constants</b></dt>
1050 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1051 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1052 notation (see below). Floating point constants must have a <a
1053 href="#t_floating">floating point</a> type. </dd>
1055 <dt><b>Null pointer constants</b></dt>
1057 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1058 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1062 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1063 of floating point constants. For example, the form '<tt>double
1064 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1065 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1066 (and the only time that they are generated by the disassembler) is when a
1067 floating point constant must be emitted but it cannot be represented as a
1068 decimal floating point number. For example, NaN's, infinities, and other
1069 special values are represented in their IEEE hexadecimal format so that
1070 assembly and disassembly do not cause any bits to change in the constants.</p>
1074 <!-- ======================================================================= -->
1075 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1078 <div class="doc_text">
1079 <p>Aggregate constants arise from aggregation of simple constants
1080 and smaller aggregate constants.</p>
1083 <dt><b>Structure constants</b></dt>
1085 <dd>Structure constants are represented with notation similar to structure
1086 type definitions (a comma separated list of elements, surrounded by braces
1087 (<tt>{}</tt>)). For example: "<tt>{ int 4, float 17.0, int* %G }</tt>",
1088 where "<tt>%G</tt>" is declared as "<tt>%G = external global int</tt>". Structure constants
1089 must have <a href="#t_struct">structure type</a>, and the number and
1090 types of elements must match those specified by the type.
1093 <dt><b>Array constants</b></dt>
1095 <dd>Array constants are represented with notation similar to array type
1096 definitions (a comma separated list of elements, surrounded by square brackets
1097 (<tt>[]</tt>)). For example: "<tt>[ int 42, int 11, int 74 ]</tt>". Array
1098 constants must have <a href="#t_array">array type</a>, and the number and
1099 types of elements must match those specified by the type.
1102 <dt><b>Packed constants</b></dt>
1104 <dd>Packed constants are represented with notation similar to packed type
1105 definitions (a comma separated list of elements, surrounded by
1106 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< int 42,
1107 int 11, int 74, int 100 ></tt>". Packed constants must have <a
1108 href="#t_packed">packed type</a>, and the number and types of elements must
1109 match those specified by the type.
1112 <dt><b>Zero initialization</b></dt>
1114 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1115 value to zero of <em>any</em> type, including scalar and aggregate types.
1116 This is often used to avoid having to print large zero initializers (e.g. for
1117 large arrays) and is always exactly equivalent to using explicit zero
1124 <!-- ======================================================================= -->
1125 <div class="doc_subsection">
1126 <a name="globalconstants">Global Variable and Function Addresses</a>
1129 <div class="doc_text">
1131 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1132 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1133 constants. These constants are explicitly referenced when the <a
1134 href="#identifiers">identifier for the global</a> is used and always have <a
1135 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1141 %Z = global [2 x int*] [ int* %X, int* %Y ]
1146 <!-- ======================================================================= -->
1147 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1148 <div class="doc_text">
1149 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1150 no specific value. Undefined values may be of any type and be used anywhere
1151 a constant is permitted.</p>
1153 <p>Undefined values indicate to the compiler that the program is well defined
1154 no matter what value is used, giving the compiler more freedom to optimize.
1158 <!-- ======================================================================= -->
1159 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1162 <div class="doc_text">
1164 <p>Constant expressions are used to allow expressions involving other constants
1165 to be used as constants. Constant expressions may be of any <a
1166 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1167 that does not have side effects (e.g. load and call are not supported). The
1168 following is the syntax for constant expressions:</p>
1171 <dt><b><tt>cast ( CST to TYPE )</tt></b></dt>
1173 <dd>Cast a constant to another type.</dd>
1175 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1177 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1178 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1179 instruction, the index list may have zero or more indexes, which are required
1180 to make sense for the type of "CSTPTR".</dd>
1182 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1184 <dd>Perform the <a href="#i_select">select operation</a> on
1187 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1189 <dd>Perform the <a href="#i_extractelement">extractelement
1190 operation</a> on constants.
1192 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1194 <dd>Perform the <a href="#i_insertelement">insertelement
1195 operation</a> on constants.
1198 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
1200 <dd>Perform the <a href="#i_shufflevector">shufflevector
1201 operation</a> on constants.
1203 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1205 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1206 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1207 binary</a> operations. The constraints on operands are the same as those for
1208 the corresponding instruction (e.g. no bitwise operations on floating point
1209 values are allowed).</dd>
1213 <!-- *********************************************************************** -->
1214 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1215 <!-- *********************************************************************** -->
1217 <!-- ======================================================================= -->
1218 <div class="doc_subsection">
1219 <a name="inlineasm">Inline Assembler Expressions</a>
1222 <div class="doc_text">
1225 LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
1226 Module-Level Inline Assembly</a>) through the use of a special value. This
1227 value represents the inline assembler as a string (containing the instructions
1228 to emit), a list of operand constraints (stored as a string), and a flag that
1229 indicates whether or not the inline asm expression has side effects. An example
1230 inline assembler expression is:
1234 int(int) asm "bswap $0", "=r,r"
1238 Inline assembler expressions may <b>only</b> be used as the callee operand of
1239 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
1243 %X = call int asm "<a href="#i_bswap">bswap</a> $0", "=r,r"(int %Y)
1247 Inline asms with side effects not visible in the constraint list must be marked
1248 as having side effects. This is done through the use of the
1249 '<tt>sideeffect</tt>' keyword, like so:
1253 call void asm sideeffect "eieio", ""()
1256 <p>TODO: The format of the asm and constraints string still need to be
1257 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1258 need to be documented).
1263 <!-- *********************************************************************** -->
1264 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1265 <!-- *********************************************************************** -->
1267 <div class="doc_text">
1269 <p>The LLVM instruction set consists of several different
1270 classifications of instructions: <a href="#terminators">terminator
1271 instructions</a>, <a href="#binaryops">binary instructions</a>,
1272 <a href="#bitwiseops">bitwise binary instructions</a>, <a
1273 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1274 instructions</a>.</p>
1278 <!-- ======================================================================= -->
1279 <div class="doc_subsection"> <a name="terminators">Terminator
1280 Instructions</a> </div>
1282 <div class="doc_text">
1284 <p>As mentioned <a href="#functionstructure">previously</a>, every
1285 basic block in a program ends with a "Terminator" instruction, which
1286 indicates which block should be executed after the current block is
1287 finished. These terminator instructions typically yield a '<tt>void</tt>'
1288 value: they produce control flow, not values (the one exception being
1289 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1290 <p>There are six different terminator instructions: the '<a
1291 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1292 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1293 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1294 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1295 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1299 <!-- _______________________________________________________________________ -->
1300 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1301 Instruction</a> </div>
1302 <div class="doc_text">
1304 <pre> ret <type> <value> <i>; Return a value from a non-void function</i>
1305 ret void <i>; Return from void function</i>
1308 <p>The '<tt>ret</tt>' instruction is used to return control flow (and a
1309 value) from a function back to the caller.</p>
1310 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
1311 returns a value and then causes control flow, and one that just causes
1312 control flow to occur.</p>
1314 <p>The '<tt>ret</tt>' instruction may return any '<a
1315 href="#t_firstclass">first class</a>' type. Notice that a function is
1316 not <a href="#wellformed">well formed</a> if there exists a '<tt>ret</tt>'
1317 instruction inside of the function that returns a value that does not
1318 match the return type of the function.</p>
1320 <p>When the '<tt>ret</tt>' instruction is executed, control flow
1321 returns back to the calling function's context. If the caller is a "<a
1322 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1323 the instruction after the call. If the caller was an "<a
1324 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1325 at the beginning of the "normal" destination block. If the instruction
1326 returns a value, that value shall set the call or invoke instruction's
1329 <pre> ret int 5 <i>; Return an integer value of 5</i>
1330 ret void <i>; Return from a void function</i>
1333 <!-- _______________________________________________________________________ -->
1334 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1335 <div class="doc_text">
1337 <pre> br bool <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1340 <p>The '<tt>br</tt>' instruction is used to cause control flow to
1341 transfer to a different basic block in the current function. There are
1342 two forms of this instruction, corresponding to a conditional branch
1343 and an unconditional branch.</p>
1345 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1346 single '<tt>bool</tt>' value and two '<tt>label</tt>' values. The
1347 unconditional form of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>'
1348 value as a target.</p>
1350 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>bool</tt>'
1351 argument is evaluated. If the value is <tt>true</tt>, control flows
1352 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1353 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1355 <pre>Test:<br> %cond = <a href="#i_setcc">seteq</a> int %a, %b<br> br bool %cond, label %IfEqual, label %IfUnequal<br>IfEqual:<br> <a
1356 href="#i_ret">ret</a> int 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> int 0<br></pre>
1358 <!-- _______________________________________________________________________ -->
1359 <div class="doc_subsubsection">
1360 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
1363 <div class="doc_text">
1367 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
1372 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
1373 several different places. It is a generalization of the '<tt>br</tt>'
1374 instruction, allowing a branch to occur to one of many possible
1380 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
1381 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
1382 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
1383 table is not allowed to contain duplicate constant entries.</p>
1387 <p>The <tt>switch</tt> instruction specifies a table of values and
1388 destinations. When the '<tt>switch</tt>' instruction is executed, this
1389 table is searched for the given value. If the value is found, control flow is
1390 transfered to the corresponding destination; otherwise, control flow is
1391 transfered to the default destination.</p>
1393 <h5>Implementation:</h5>
1395 <p>Depending on properties of the target machine and the particular
1396 <tt>switch</tt> instruction, this instruction may be code generated in different
1397 ways. For example, it could be generated as a series of chained conditional
1398 branches or with a lookup table.</p>
1403 <i>; Emulate a conditional br instruction</i>
1404 %Val = <a href="#i_cast">cast</a> bool %value to int
1405 switch int %Val, label %truedest [int 0, label %falsedest ]
1407 <i>; Emulate an unconditional br instruction</i>
1408 switch uint 0, label %dest [ ]
1410 <i>; Implement a jump table:</i>
1411 switch uint %val, label %otherwise [ uint 0, label %onzero
1412 uint 1, label %onone
1413 uint 2, label %ontwo ]
1417 <!-- _______________________________________________________________________ -->
1418 <div class="doc_subsubsection">
1419 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
1422 <div class="doc_text">
1427 <result> = invoke [<a href="#callingconv">cconv</a>] <ptr to function ty> %<function ptr val>(<function args>)
1428 to label <normal label> unwind label <exception label>
1433 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
1434 function, with the possibility of control flow transfer to either the
1435 '<tt>normal</tt>' label or the
1436 '<tt>exception</tt>' label. If the callee function returns with the
1437 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
1438 "normal" label. If the callee (or any indirect callees) returns with the "<a
1439 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
1440 continued at the dynamically nearest "exception" label.</p>
1444 <p>This instruction requires several arguments:</p>
1448 The optional "cconv" marker indicates which <a href="callingconv">calling
1449 convention</a> the call should use. If none is specified, the call defaults
1450 to using C calling conventions.
1452 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
1453 function value being invoked. In most cases, this is a direct function
1454 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
1455 an arbitrary pointer to function value.
1458 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
1459 function to be invoked. </li>
1461 <li>'<tt>function args</tt>': argument list whose types match the function
1462 signature argument types. If the function signature indicates the function
1463 accepts a variable number of arguments, the extra arguments can be
1466 <li>'<tt>normal label</tt>': the label reached when the called function
1467 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
1469 <li>'<tt>exception label</tt>': the label reached when a callee returns with
1470 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
1476 <p>This instruction is designed to operate as a standard '<tt><a
1477 href="#i_call">call</a></tt>' instruction in most regards. The primary
1478 difference is that it establishes an association with a label, which is used by
1479 the runtime library to unwind the stack.</p>
1481 <p>This instruction is used in languages with destructors to ensure that proper
1482 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
1483 exception. Additionally, this is important for implementation of
1484 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
1488 %retval = invoke int %Test(int 15) to label %Continue
1489 unwind label %TestCleanup <i>; {int}:retval set</i>
1490 %retval = invoke <a href="#callingconv">coldcc</a> int %Test(int 15) to label %Continue
1491 unwind label %TestCleanup <i>; {int}:retval set</i>
1496 <!-- _______________________________________________________________________ -->
1498 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
1499 Instruction</a> </div>
1501 <div class="doc_text">
1510 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
1511 at the first callee in the dynamic call stack which used an <a
1512 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
1513 primarily used to implement exception handling.</p>
1517 <p>The '<tt>unwind</tt>' intrinsic causes execution of the current function to
1518 immediately halt. The dynamic call stack is then searched for the first <a
1519 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
1520 execution continues at the "exceptional" destination block specified by the
1521 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
1522 dynamic call chain, undefined behavior results.</p>
1525 <!-- _______________________________________________________________________ -->
1527 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
1528 Instruction</a> </div>
1530 <div class="doc_text">
1539 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
1540 instruction is used to inform the optimizer that a particular portion of the
1541 code is not reachable. This can be used to indicate that the code after a
1542 no-return function cannot be reached, and other facts.</p>
1546 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
1551 <!-- ======================================================================= -->
1552 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
1553 <div class="doc_text">
1554 <p>Binary operators are used to do most of the computation in a
1555 program. They require two operands, execute an operation on them, and
1556 produce a single value. The operands might represent
1557 multiple data, as is the case with the <a href="#t_packed">packed</a> data type.
1558 The result value of a binary operator is not
1559 necessarily the same type as its operands.</p>
1560 <p>There are several different binary operators:</p>
1562 <!-- _______________________________________________________________________ -->
1563 <div class="doc_subsubsection"> <a name="i_add">'<tt>add</tt>'
1564 Instruction</a> </div>
1565 <div class="doc_text">
1567 <pre> <result> = add <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1570 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
1572 <p>The two arguments to the '<tt>add</tt>' instruction must be either <a
1573 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a> values.
1574 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1575 Both arguments must have identical types.</p>
1577 <p>The value produced is the integer or floating point sum of the two
1580 <pre> <result> = add int 4, %var <i>; yields {int}:result = 4 + %var</i>
1583 <!-- _______________________________________________________________________ -->
1584 <div class="doc_subsubsection"> <a name="i_sub">'<tt>sub</tt>'
1585 Instruction</a> </div>
1586 <div class="doc_text">
1588 <pre> <result> = sub <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1591 <p>The '<tt>sub</tt>' instruction returns the difference of its two
1593 <p>Note that the '<tt>sub</tt>' instruction is used to represent the '<tt>neg</tt>'
1594 instruction present in most other intermediate representations.</p>
1596 <p>The two arguments to the '<tt>sub</tt>' instruction must be either <a
1597 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1599 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1600 Both arguments must have identical types.</p>
1602 <p>The value produced is the integer or floating point difference of
1603 the two operands.</p>
1605 <pre> <result> = sub int 4, %var <i>; yields {int}:result = 4 - %var</i>
1606 <result> = sub int 0, %val <i>; yields {int}:result = -%var</i>
1609 <!-- _______________________________________________________________________ -->
1610 <div class="doc_subsubsection"> <a name="i_mul">'<tt>mul</tt>'
1611 Instruction</a> </div>
1612 <div class="doc_text">
1614 <pre> <result> = mul <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1617 <p>The '<tt>mul</tt>' instruction returns the product of its two
1620 <p>The two arguments to the '<tt>mul</tt>' instruction must be either <a
1621 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1623 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1624 Both arguments must have identical types.</p>
1626 <p>The value produced is the integer or floating point product of the
1628 <p>There is no signed vs unsigned multiplication. The appropriate
1629 action is taken based on the type of the operand.</p>
1631 <pre> <result> = mul int 4, %var <i>; yields {int}:result = 4 * %var</i>
1634 <!-- _______________________________________________________________________ -->
1635 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
1637 <div class="doc_text">
1639 <pre> <result> = udiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1642 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two
1645 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
1646 <a href="#t_integer">integer</a> values. Both arguments must have identical
1647 types. This instruction can also take <a href="#t_packed">packed</a> versions
1648 of the values in which case the elements must be integers.</p>
1650 <p>The value produced is the unsigned integer quotient of the two operands. This
1651 instruction always performs an unsigned division operation, regardless of
1652 whether the arguments are unsigned or not.</p>
1654 <pre> <result> = udiv uint 4, %var <i>; yields {uint}:result = 4 / %var</i>
1657 <!-- _______________________________________________________________________ -->
1658 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
1660 <div class="doc_text">
1662 <pre> <result> = sdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1665 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
1668 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
1669 <a href="#t_integer">integer</a> values. Both arguments must have identical
1670 types. This instruction can also take <a href="#t_packed">packed</a> versions
1671 of the values in which case the elements must be integers.</p>
1673 <p>The value produced is the signed integer quotient of the two operands. This
1674 instruction always performs a signed division operation, regardless of whether
1675 the arguments are signed or not.</p>
1677 <pre> <result> = sdiv int 4, %var <i>; yields {int}:result = 4 / %var</i>
1680 <!-- _______________________________________________________________________ -->
1681 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
1682 Instruction</a> </div>
1683 <div class="doc_text">
1685 <pre> <result> = fdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1688 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
1691 <p>The two arguments to the '<tt>div</tt>' instruction must be
1692 <a href="#t_floating">floating point</a> values. Both arguments must have
1693 identical types. This instruction can also take <a href="#t_packed">packed</a>
1694 versions of the values in which case the elements must be floating point.</p>
1696 <p>The value produced is the floating point quotient of the two operands.</p>
1698 <pre> <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
1701 <!-- _______________________________________________________________________ -->
1702 <div class="doc_subsubsection"> <a name="i_rem">'<tt>rem</tt>'
1703 Instruction</a> </div>
1704 <div class="doc_text">
1706 <pre> <result> = rem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1709 <p>The '<tt>rem</tt>' instruction returns the remainder from the
1710 division of its two operands.</p>
1712 <p>The two arguments to the '<tt>rem</tt>' instruction must be either <a
1713 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1715 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1716 Both arguments must have identical types.</p>
1718 <p>This returns the <i>remainder</i> of a division (where the result
1719 has the same sign as the divisor), not the <i>modulus</i> (where the
1720 result has the same sign as the dividend) of a value. For more
1721 information about the difference, see <a
1722 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
1725 <pre> <result> = rem int 4, %var <i>; yields {int}:result = 4 % %var</i>
1729 <!-- _______________________________________________________________________ -->
1730 <div class="doc_subsubsection"> <a name="i_setcc">'<tt>set<i>cc</i></tt>'
1731 Instructions</a> </div>
1732 <div class="doc_text">
1734 <pre> <result> = seteq <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1735 <result> = setne <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1736 <result> = setlt <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1737 <result> = setgt <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1738 <result> = setle <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1739 <result> = setge <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1742 <p>The '<tt>set<i>cc</i></tt>' family of instructions returns a boolean
1743 value based on a comparison of their two operands.</p>
1745 <p>The two arguments to the '<tt>set<i>cc</i></tt>' instructions must
1746 be of <a href="#t_firstclass">first class</a> type (it is not possible
1747 to compare '<tt>label</tt>'s, '<tt>array</tt>'s, '<tt>structure</tt>'
1748 or '<tt>void</tt>' values, etc...). Both arguments must have identical
1751 <p>The '<tt>seteq</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1752 value if both operands are equal.<br>
1753 The '<tt>setne</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1754 value if both operands are unequal.<br>
1755 The '<tt>setlt</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1756 value if the first operand is less than the second operand.<br>
1757 The '<tt>setgt</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1758 value if the first operand is greater than the second operand.<br>
1759 The '<tt>setle</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1760 value if the first operand is less than or equal to the second operand.<br>
1761 The '<tt>setge</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1762 value if the first operand is greater than or equal to the second
1765 <pre> <result> = seteq int 4, 5 <i>; yields {bool}:result = false</i>
1766 <result> = setne float 4, 5 <i>; yields {bool}:result = true</i>
1767 <result> = setlt uint 4, 5 <i>; yields {bool}:result = true</i>
1768 <result> = setgt sbyte 4, 5 <i>; yields {bool}:result = false</i>
1769 <result> = setle sbyte 4, 5 <i>; yields {bool}:result = true</i>
1770 <result> = setge sbyte 4, 5 <i>; yields {bool}:result = false</i>
1774 <!-- ======================================================================= -->
1775 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
1776 Operations</a> </div>
1777 <div class="doc_text">
1778 <p>Bitwise binary operators are used to do various forms of
1779 bit-twiddling in a program. They are generally very efficient
1780 instructions and can commonly be strength reduced from other
1781 instructions. They require two operands, execute an operation on them,
1782 and produce a single value. The resulting value of the bitwise binary
1783 operators is always the same type as its first operand.</p>
1785 <!-- _______________________________________________________________________ -->
1786 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
1787 Instruction</a> </div>
1788 <div class="doc_text">
1790 <pre> <result> = and <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1793 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
1794 its two operands.</p>
1796 <p>The two arguments to the '<tt>and</tt>' instruction must be <a
1797 href="#t_integral">integral</a> values. Both arguments must have
1798 identical types.</p>
1800 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
1802 <div style="align: center">
1803 <table border="1" cellspacing="0" cellpadding="4">
1834 <pre> <result> = and int 4, %var <i>; yields {int}:result = 4 & %var</i>
1835 <result> = and int 15, 40 <i>; yields {int}:result = 8</i>
1836 <result> = and int 4, 8 <i>; yields {int}:result = 0</i>
1839 <!-- _______________________________________________________________________ -->
1840 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
1841 <div class="doc_text">
1843 <pre> <result> = or <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1846 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
1847 or of its two operands.</p>
1849 <p>The two arguments to the '<tt>or</tt>' instruction must be <a
1850 href="#t_integral">integral</a> values. Both arguments must have
1851 identical types.</p>
1853 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
1855 <div style="align: center">
1856 <table border="1" cellspacing="0" cellpadding="4">
1887 <pre> <result> = or int 4, %var <i>; yields {int}:result = 4 | %var</i>
1888 <result> = or int 15, 40 <i>; yields {int}:result = 47</i>
1889 <result> = or int 4, 8 <i>; yields {int}:result = 12</i>
1892 <!-- _______________________________________________________________________ -->
1893 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
1894 Instruction</a> </div>
1895 <div class="doc_text">
1897 <pre> <result> = xor <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1900 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
1901 or of its two operands. The <tt>xor</tt> is used to implement the
1902 "one's complement" operation, which is the "~" operator in C.</p>
1904 <p>The two arguments to the '<tt>xor</tt>' instruction must be <a
1905 href="#t_integral">integral</a> values. Both arguments must have
1906 identical types.</p>
1908 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
1910 <div style="align: center">
1911 <table border="1" cellspacing="0" cellpadding="4">
1943 <pre> <result> = xor int 4, %var <i>; yields {int}:result = 4 ^ %var</i>
1944 <result> = xor int 15, 40 <i>; yields {int}:result = 39</i>
1945 <result> = xor int 4, 8 <i>; yields {int}:result = 12</i>
1946 <result> = xor int %V, -1 <i>; yields {int}:result = ~%V</i>
1949 <!-- _______________________________________________________________________ -->
1950 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
1951 Instruction</a> </div>
1952 <div class="doc_text">
1954 <pre> <result> = shl <ty> <var1>, ubyte <var2> <i>; yields {ty}:result</i>
1957 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
1958 the left a specified number of bits.</p>
1960 <p>The first argument to the '<tt>shl</tt>' instruction must be an <a
1961 href="#t_integer">integer</a> type. The second argument must be an '<tt>ubyte</tt>'
1964 <p>The value produced is <tt>var1</tt> * 2<sup><tt>var2</tt></sup>.</p>
1966 <pre> <result> = shl int 4, ubyte %var <i>; yields {int}:result = 4 << %var</i>
1967 <result> = shl int 4, ubyte 2 <i>; yields {int}:result = 16</i>
1968 <result> = shl int 1, ubyte 10 <i>; yields {int}:result = 1024</i>
1971 <!-- _______________________________________________________________________ -->
1972 <div class="doc_subsubsection"> <a name="i_shr">'<tt>shr</tt>'
1973 Instruction</a> </div>
1974 <div class="doc_text">
1976 <pre> <result> = shr <ty> <var1>, ubyte <var2> <i>; yields {ty}:result</i>
1979 <p>The '<tt>shr</tt>' instruction returns the first operand shifted to
1980 the right a specified number of bits.</p>
1982 <p>The first argument to the '<tt>shr</tt>' instruction must be an <a
1983 href="#t_integer">integer</a> type. The second argument must be an '<tt>ubyte</tt>'
1986 <p>If the first argument is a <a href="#t_signed">signed</a> type, the
1987 most significant bit is duplicated in the newly free'd bit positions.
1988 If the first argument is unsigned, zero bits shall fill the empty
1991 <pre> <result> = shr int 4, ubyte %var <i>; yields {int}:result = 4 >> %var</i>
1992 <result> = shr uint 4, ubyte 1 <i>; yields {uint}:result = 2</i>
1993 <result> = shr int 4, ubyte 2 <i>; yields {int}:result = 1</i>
1994 <result> = shr sbyte 4, ubyte 3 <i>; yields {sbyte}:result = 0</i>
1995 <result> = shr sbyte -2, ubyte 1 <i>; yields {sbyte}:result = -1</i>
1999 <!-- ======================================================================= -->
2000 <div class="doc_subsection">
2001 <a name="vectorops">Vector Operations</a>
2004 <div class="doc_text">
2006 <p>LLVM supports several instructions to represent vector operations in a
2007 target-independent manner. This instructions cover the element-access and
2008 vector-specific operations needed to process vectors effectively. While LLVM
2009 does directly support these vector operations, many sophisticated algorithms
2010 will want to use target-specific intrinsics to take full advantage of a specific
2015 <!-- _______________________________________________________________________ -->
2016 <div class="doc_subsubsection">
2017 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2020 <div class="doc_text">
2025 <result> = extractelement <n x <ty>> <val>, uint <idx> <i>; yields <ty></i>
2031 The '<tt>extractelement</tt>' instruction extracts a single scalar
2032 element from a packed vector at a specified index.
2039 The first operand of an '<tt>extractelement</tt>' instruction is a
2040 value of <a href="#t_packed">packed</a> type. The second operand is
2041 an index indicating the position from which to extract the element.
2042 The index may be a variable.</p>
2047 The result is a scalar of the same type as the element type of
2048 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2049 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2050 results are undefined.
2056 %result = extractelement <4 x int> %vec, uint 0 <i>; yields int</i>
2061 <!-- _______________________________________________________________________ -->
2062 <div class="doc_subsubsection">
2063 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
2066 <div class="doc_text">
2071 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, uint <idx> <i>; yields <n x <ty>></i>
2077 The '<tt>insertelement</tt>' instruction inserts a scalar
2078 element into a packed vector at a specified index.
2085 The first operand of an '<tt>insertelement</tt>' instruction is a
2086 value of <a href="#t_packed">packed</a> type. The second operand is a
2087 scalar value whose type must equal the element type of the first
2088 operand. The third operand is an index indicating the position at
2089 which to insert the value. The index may be a variable.</p>
2094 The result is a packed vector of the same type as <tt>val</tt>. Its
2095 element values are those of <tt>val</tt> except at position
2096 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
2097 exceeds the length of <tt>val</tt>, the results are undefined.
2103 %result = insertelement <4 x int> %vec, int 1, uint 0 <i>; yields <4 x int></i>
2107 <!-- _______________________________________________________________________ -->
2108 <div class="doc_subsubsection">
2109 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
2112 <div class="doc_text">
2117 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <n x uint> <mask> <i>; yields <n x <ty>></i>
2123 The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
2124 from two input vectors, returning a vector of the same type.
2130 The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
2131 with types that match each other and types that match the result of the
2132 instruction. The third argument is a shuffle mask, which has the same number
2133 of elements as the other vector type, but whose element type is always 'uint'.
2137 The shuffle mask operand is required to be a constant vector with either
2138 constant integer or undef values.
2144 The elements of the two input vectors are numbered from left to right across
2145 both of the vectors. The shuffle mask operand specifies, for each element of
2146 the result vector, which element of the two input registers the result element
2147 gets. The element selector may be undef (meaning "don't care") and the second
2148 operand may be undef if performing a shuffle from only one vector.
2154 %result = shufflevector <4 x int> %v1, <4 x int> %v2,
2155 <4 x uint> <uint 0, uint 4, uint 1, uint 5> <i>; yields <4 x int></i>
2156 %result = shufflevector <4 x int> %v1, <4 x int> undef,
2157 <4 x uint> <uint 0, uint 1, uint 2, uint 3> <i>; yields <4 x int></i> - Identity shuffle.
2162 <!-- _______________________________________________________________________ -->
2163 <div class="doc_subsubsection"> <a name="i_vsetint">'<tt>vsetint</tt>'
2164 Instruction</a> </div>
2165 <div class="doc_text">
2167 <pre><result> = vsetint <op>, <n x <ty>> <var1>, <var2> <i>; yields <n x bool></i>
2172 <p>The '<tt>vsetint</tt>' instruction takes two integer vectors and
2173 returns a vector of boolean values representing, at each position, the
2174 result of the comparison between the values at that position in the
2179 <p>The arguments to a '<tt>vsetint</tt>' instruction are a comparison
2180 operation and two value arguments. The value arguments must be of <a
2181 href="#t_integral">integral</a> <a href="#t_packed">packed</a> type,
2182 and they must have identical types. The operation argument must be
2183 one of <tt>eq</tt>, <tt>ne</tt>, <tt>slt</tt>, <tt>sgt</tt>,
2184 <tt>sle</tt>, <tt>sge</tt>, <tt>ult</tt>, <tt>ugt</tt>, <tt>ule</tt>,
2185 <tt>uge</tt>, <tt>true</tt>, and <tt>false</tt>. The result is a
2186 packed <tt>bool</tt> value with the same length as each operand.</p>
2190 <p>The following table shows the semantics of '<tt>vsetint</tt>'. For
2191 each position of the result, the comparison is done on the
2192 corresponding positions of the two value arguments. Note that the
2193 signedness of the comparison depends on the comparison opcode and
2194 <i>not</i> on the signedness of the value operands. E.g., <tt>vsetint
2195 slt <4 x unsigned> %x, %y</tt> does an elementwise <i>signed</i>
2196 comparison of <tt>%x</tt> and <tt>%y</tt>.</p>
2198 <table border="1" cellspacing="0" cellpadding="4">
2200 <tr><th>Operation</th><th>Result is true iff</th><th>Comparison is</th></tr>
2201 <tr><td><tt>eq</tt></td><td>var1 == var2</td><td>--</td></tr>
2202 <tr><td><tt>ne</tt></td><td>var1 != var2</td><td>--</td></tr>
2203 <tr><td><tt>slt</tt></td><td>var1 < var2</td><td>signed</td></tr>
2204 <tr><td><tt>sgt</tt></td><td>var1 > var2</td><td>signed</td></tr>
2205 <tr><td><tt>sle</tt></td><td>var1 <= var2</td><td>signed</td></tr>
2206 <tr><td><tt>sge</tt></td><td>var1 >= var2</td><td>signed</td></tr>
2207 <tr><td><tt>ult</tt></td><td>var1 < var2</td><td>unsigned</td></tr>
2208 <tr><td><tt>ugt</tt></td><td>var1 > var2</td><td>unsigned</td></tr>
2209 <tr><td><tt>ule</tt></td><td>var1 <= var2</td><td>unsigned</td></tr>
2210 <tr><td><tt>uge</tt></td><td>var1 >= var2</td><td>unsigned</td></tr>
2211 <tr><td><tt>true</tt></td><td>always</td><td>--</td></tr>
2212 <tr><td><tt>false</tt></td><td>never</td><td>--</td></tr>
2217 <pre> <result> = vsetint eq <2 x int> <int 0, int 1>, <int 1, int 0> <i>; yields {<2 x bool>}:result = false, false</i>
2218 <result> = vsetint ne <2 x int> <int 0, int 1>, <int 1, int 0> <i>; yields {<2 x bool>}:result = true, true</i>
2219 <result> = vsetint slt <2 x int> <int 0, int 1>, <int 1, int 0> <i>; yields {<2 x bool>}:result = true, false</i>
2220 <result> = vsetint sgt <2 x int> <int 0, int 1>, <int 1, int 0> <i>; yields {<2 x bool>}:result = false, true</i>
2221 <result> = vsetint sle <2 x int> <int 0, int 1>, <int 1, int 0> <i>; yields {<2 x bool>}:result = true, false</i>
2222 <result> = vsetint sge <2 x int> <int 0, int 1>, <int 1, int 0> <i>; yields {<2 x bool>}:result = false, true</i>
2226 <!-- _______________________________________________________________________ -->
2227 <div class="doc_subsubsection"> <a name="i_vsetfp">'<tt>vsetfp</tt>'
2228 Instruction</a> </div>
2229 <div class="doc_text">
2231 <pre><result> = vsetfp <op>, <n x <ty>> <var1>, <var2> <i>; yields <n x bool></i>
2236 <p>The '<tt>vsetfp</tt>' instruction takes two floating point vector
2237 arguments and returns a vector of boolean values representing, at each
2238 position, the result of the comparison between the values at that
2239 position in the two operands.</p>
2243 <p>The arguments to a '<tt>vsetfp</tt>' instruction are a comparison
2244 operation and two value arguments. The value arguments must be of <a
2245 href="t_floating">floating point</a> <a href="#t_packed">packed</a>
2246 type, and they must have identical types. The operation argument must
2247 be one of <tt>eq</tt>, <tt>ne</tt>, <tt>lt</tt>, <tt>gt</tt>,
2248 <tt>le</tt>, <tt>ge</tt>, <tt>oeq</tt>, <tt>one</tt>, <tt>olt</tt>,
2249 <tt>ogt</tt>, <tt>ole</tt>, <tt>oge</tt>, <tt>ueq</tt>, <tt>une</tt>,
2250 <tt>ult</tt>, <tt>ugt</tt>, <tt>ule</tt>, <tt>uge</tt>, <tt>o</tt>,
2251 <tt>u</tt>, <tt>true</tt>, and <tt>false</tt>. The result is a packed
2252 <tt>bool</tt> value with the same length as each operand.</p>
2256 <p>The following table shows the semantics of '<tt>vsetfp</tt>' for
2257 floating point types. If either operand is a floating point Not a
2258 Number (NaN) value, the operation is unordered, and the value in the
2259 first column below is produced at that position. Otherwise, the
2260 operation is ordered, and the value in the second column is
2263 <table border="1" cellspacing="0" cellpadding="4">
2265 <tr><th>Operation</th><th>If unordered<th>Otherwise true iff</th></tr>
2266 <tr><td><tt>eq</tt></td><td>undefined</td><td>var1 == var2</td></tr>
2267 <tr><td><tt>ne</tt></td><td>undefined</td><td>var1 != var2</td></tr>
2268 <tr><td><tt>lt</tt></td><td>undefined</td><td>var1 < var2</td></tr>
2269 <tr><td><tt>gt</tt></td><td>undefined</td><td>var1 > var2</td></tr>
2270 <tr><td><tt>le</tt></td><td>undefined</td><td>var1 <= var2</td></tr>
2271 <tr><td><tt>ge</tt></td><td>undefined</td><td>var1 >= var2</td></tr>
2272 <tr><td><tt>oeq</tt></td><td>false</td><td>var1 == var2</td></tr>
2273 <tr><td><tt>one</tt></td><td>false</td><td>var1 != var2</td></tr>
2274 <tr><td><tt>olt</tt></td><td>false</td><td>var1 < var2</td></tr>
2275 <tr><td><tt>ogt</tt></td><td>false</td><td>var1 > var2</td></tr>
2276 <tr><td><tt>ole</tt></td><td>false</td><td>var1 <= var2</td></tr>
2277 <tr><td><tt>oge</tt></td><td>false</td><td>var1 >= var2</td></tr>
2278 <tr><td><tt>ueq</tt></td><td>true</td><td>var1 == var2</td></tr>
2279 <tr><td><tt>une</tt></td><td>true</td><td>var1 != var2</td></tr>
2280 <tr><td><tt>ult</tt></td><td>true</td><td>var1 < var2</td></tr>
2281 <tr><td><tt>ugt</tt></td><td>true</td><td>var1 > var2</td></tr>
2282 <tr><td><tt>ule</tt></td><td>true</td><td>var1 <= var2</td></tr>
2283 <tr><td><tt>uge</tt></td><td>true</td><td>var1 >= var2</td></tr>
2284 <tr><td><tt>o</tt></td><td>false</td><td>always</td></tr>
2285 <tr><td><tt>u</tt></td><td>true</td><td>never</td></tr>
2286 <tr><td><tt>true</tt></td><td>true</td><td>always</td></tr>
2287 <tr><td><tt>false</tt></td><td>false</td><td>never</td></tr>
2292 <pre> <result> = vsetfp eq <2 x float> <float 0.0, float 1.0>, <float 1.0, float 0.0> <i>; yields {<2 x bool>}:result = false, false</i>
2293 <result> = vsetfp ne <2 x float> <float 0.0, float 1.0>, <float 1.0, float 0.0> <i>; yields {<2 x bool>}:result = true, true</i>
2294 <result> = vsetfp lt <2 x float> <float 0.0, float 1.0>, <float 1.0, float 0.0> <i>; yields {<2 x bool>}:result = true, false</i>
2295 <result> = vsetfp gt <2 x float> <float 0.0, float 1.0>, <float 1.0, float 0.0> <i>; yields {<2 x bool>}:result = false, true</i>
2296 <result> = vsetfp le <2 x float> <float 0.0, float 1.0>, <float 1.0, float 0.0> <i>; yields {<2 x bool>}:result = true, false</i>
2297 <result> = vsetfp ge <2 x float> <float 0.0, float 1.0>, <float 1.0, float 0.0> <i>; yields {<2 x bool>}:result = false, true</i>
2301 <!-- _______________________________________________________________________ -->
2302 <div class="doc_subsubsection">
2303 <a name="i_vselect">'<tt>vselect</tt>' Instruction</a>
2306 <div class="doc_text">
2311 <result> = vselect <n x bool> <cond>, <n x <ty>> <val1>, <n x <ty>> <val2> <i>; yields <n x <ty>></i>
2317 The '<tt>vselect</tt>' instruction chooses one value at each position
2318 of a vector based on a condition.
2325 The '<tt>vselect</tt>' instruction requires a <a
2326 href="#t_packed">packed</a> <tt>bool</tt> value indicating the
2327 condition at each vector position, and two values of the same packed
2328 type. All three operands must have the same length. The type of the
2329 result is the same as the type of the two value operands.</p>
2334 At each position where the <tt>bool</tt> vector is true, that position
2335 of the result gets its value from the first value argument; otherwise,
2336 it gets its value from the second value argument.
2342 %X = vselect bool <2 x bool> <bool true, bool false>, <2 x ubyte> <ubyte 17, ubyte 17>,
2343 <2 x ubyte> <ubyte 42, ubyte 42> <i>; yields <2 x ubyte>:17, 42</i>
2349 <!-- ======================================================================= -->
2350 <div class="doc_subsection">
2351 <a name="memoryops">Memory Access and Addressing Operations</a>
2354 <div class="doc_text">
2356 <p>A key design point of an SSA-based representation is how it
2357 represents memory. In LLVM, no memory locations are in SSA form, which
2358 makes things very simple. This section describes how to read, write,
2359 allocate, and free memory in LLVM.</p>
2363 <!-- _______________________________________________________________________ -->
2364 <div class="doc_subsubsection">
2365 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
2368 <div class="doc_text">
2373 <result> = malloc <type>[, uint <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2378 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
2379 heap and returns a pointer to it.</p>
2383 <p>The '<tt>malloc</tt>' instruction allocates
2384 <tt>sizeof(<type>)*NumElements</tt>
2385 bytes of memory from the operating system and returns a pointer of the
2386 appropriate type to the program. If "NumElements" is specified, it is the
2387 number of elements allocated. If an alignment is specified, the value result
2388 of the allocation is guaranteed to be aligned to at least that boundary. If
2389 not specified, or if zero, the target can choose to align the allocation on any
2390 convenient boundary.</p>
2392 <p>'<tt>type</tt>' must be a sized type.</p>
2396 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
2397 a pointer is returned.</p>
2402 %array = malloc [4 x ubyte ] <i>; yields {[%4 x ubyte]*}:array</i>
2404 %size = <a href="#i_add">add</a> uint 2, 2 <i>; yields {uint}:size = uint 4</i>
2405 %array1 = malloc ubyte, uint 4 <i>; yields {ubyte*}:array1</i>
2406 %array2 = malloc [12 x ubyte], uint %size <i>; yields {[12 x ubyte]*}:array2</i>
2407 %array3 = malloc int, uint 4, align 1024 <i>; yields {int*}:array3</i>
2408 %array4 = malloc int, align 1024 <i>; yields {int*}:array4</i>
2412 <!-- _______________________________________________________________________ -->
2413 <div class="doc_subsubsection">
2414 <a name="i_free">'<tt>free</tt>' Instruction</a>
2417 <div class="doc_text">
2422 free <type> <value> <i>; yields {void}</i>
2427 <p>The '<tt>free</tt>' instruction returns memory back to the unused
2428 memory heap to be reallocated in the future.</p>
2432 <p>'<tt>value</tt>' shall be a pointer value that points to a value
2433 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
2438 <p>Access to the memory pointed to by the pointer is no longer defined
2439 after this instruction executes.</p>
2444 %array = <a href="#i_malloc">malloc</a> [4 x ubyte] <i>; yields {[4 x ubyte]*}:array</i>
2445 free [4 x ubyte]* %array
2449 <!-- _______________________________________________________________________ -->
2450 <div class="doc_subsubsection">
2451 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
2454 <div class="doc_text">
2459 <result> = alloca <type>[, uint <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2464 <p>The '<tt>alloca</tt>' instruction allocates memory on the current
2465 stack frame of the procedure that is live until the current function
2466 returns to its caller.</p>
2470 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
2471 bytes of memory on the runtime stack, returning a pointer of the
2472 appropriate type to the program. If "NumElements" is specified, it is the
2473 number of elements allocated. If an alignment is specified, the value result
2474 of the allocation is guaranteed to be aligned to at least that boundary. If
2475 not specified, or if zero, the target can choose to align the allocation on any
2476 convenient boundary.</p>
2478 <p>'<tt>type</tt>' may be any sized type.</p>
2482 <p>Memory is allocated; a pointer is returned. '<tt>alloca</tt>'d
2483 memory is automatically released when the function returns. The '<tt>alloca</tt>'
2484 instruction is commonly used to represent automatic variables that must
2485 have an address available. When the function returns (either with the <tt><a
2486 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
2487 instructions), the memory is reclaimed.</p>
2492 %ptr = alloca int <i>; yields {int*}:ptr</i>
2493 %ptr = alloca int, uint 4 <i>; yields {int*}:ptr</i>
2494 %ptr = alloca int, uint 4, align 1024 <i>; yields {int*}:ptr</i>
2495 %ptr = alloca int, align 1024 <i>; yields {int*}:ptr</i>
2499 <!-- _______________________________________________________________________ -->
2500 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
2501 Instruction</a> </div>
2502 <div class="doc_text">
2504 <pre> <result> = load <ty>* <pointer><br> <result> = volatile load <ty>* <pointer><br></pre>
2506 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
2508 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
2509 address from which to load. The pointer must point to a <a
2510 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
2511 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
2512 the number or order of execution of this <tt>load</tt> with other
2513 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
2516 <p>The location of memory pointed to is loaded.</p>
2518 <pre> %ptr = <a href="#i_alloca">alloca</a> int <i>; yields {int*}:ptr</i>
2520 href="#i_store">store</a> int 3, int* %ptr <i>; yields {void}</i>
2521 %val = load int* %ptr <i>; yields {int}:val = int 3</i>
2524 <!-- _______________________________________________________________________ -->
2525 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
2526 Instruction</a> </div>
2528 <pre> store <ty> <value>, <ty>* <pointer> <i>; yields {void}</i>
2529 volatile store <ty> <value>, <ty>* <pointer> <i>; yields {void}</i>
2532 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
2534 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
2535 to store and an address in which to store it. The type of the '<tt><pointer></tt>'
2536 operand must be a pointer to the type of the '<tt><value></tt>'
2537 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
2538 optimizer is not allowed to modify the number or order of execution of
2539 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
2540 href="#i_store">store</a></tt> instructions.</p>
2542 <p>The contents of memory are updated to contain '<tt><value></tt>'
2543 at the location specified by the '<tt><pointer></tt>' operand.</p>
2545 <pre> %ptr = <a href="#i_alloca">alloca</a> int <i>; yields {int*}:ptr</i>
2547 href="#i_store">store</a> int 3, int* %ptr <i>; yields {void}</i>
2548 %val = load int* %ptr <i>; yields {int}:val = int 3</i>
2550 <!-- _______________________________________________________________________ -->
2551 <div class="doc_subsubsection">
2552 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
2555 <div class="doc_text">
2558 <result> = getelementptr <ty>* <ptrval>{, <ty> <idx>}*
2564 The '<tt>getelementptr</tt>' instruction is used to get the address of a
2565 subelement of an aggregate data structure.</p>
2569 <p>This instruction takes a list of integer constants that indicate what
2570 elements of the aggregate object to index to. The actual types of the arguments
2571 provided depend on the type of the first pointer argument. The
2572 '<tt>getelementptr</tt>' instruction is used to index down through the type
2573 levels of a structure or to a specific index in an array. When indexing into a
2574 structure, only <tt>uint</tt>
2575 integer constants are allowed. When indexing into an array or pointer,
2576 <tt>int</tt> and <tt>long</tt> indexes are allowed of any sign.</p>
2578 <p>For example, let's consider a C code fragment and how it gets
2579 compiled to LLVM:</p>
2593 int *foo(struct ST *s) {
2594 return &s[1].Z.B[5][13];
2598 <p>The LLVM code generated by the GCC frontend is:</p>
2601 %RT = type { sbyte, [10 x [20 x int]], sbyte }
2602 %ST = type { int, double, %RT }
2606 int* %foo(%ST* %s) {
2608 %reg = getelementptr %ST* %s, int 1, uint 2, uint 1, int 5, int 13
2615 <p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
2616 on the pointer type that is being indexed into. <a href="#t_pointer">Pointer</a>
2617 and <a href="#t_array">array</a> types require <tt>uint</tt>, <tt>int</tt>,
2618 <tt>ulong</tt>, or <tt>long</tt> values, and <a href="#t_struct">structure</a>
2619 types require <tt>uint</tt> <b>constants</b>.</p>
2621 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
2622 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ int, double, %RT
2623 }</tt>' type, a structure. The second index indexes into the third element of
2624 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ sbyte, [10 x [20 x int]],
2625 sbyte }</tt>' type, another structure. The third index indexes into the second
2626 element of the structure, yielding a '<tt>[10 x [20 x int]]</tt>' type, an
2627 array. The two dimensions of the array are subscripted into, yielding an
2628 '<tt>int</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
2629 to this element, thus computing a value of '<tt>int*</tt>' type.</p>
2631 <p>Note that it is perfectly legal to index partially through a
2632 structure, returning a pointer to an inner element. Because of this,
2633 the LLVM code for the given testcase is equivalent to:</p>
2636 int* %foo(%ST* %s) {
2637 %t1 = getelementptr %ST* %s, int 1 <i>; yields %ST*:%t1</i>
2638 %t2 = getelementptr %ST* %t1, int 0, uint 2 <i>; yields %RT*:%t2</i>
2639 %t3 = getelementptr %RT* %t2, int 0, uint 1 <i>; yields [10 x [20 x int]]*:%t3</i>
2640 %t4 = getelementptr [10 x [20 x int]]* %t3, int 0, int 5 <i>; yields [20 x int]*:%t4</i>
2641 %t5 = getelementptr [20 x int]* %t4, int 0, int 13 <i>; yields int*:%t5</i>
2646 <p>Note that it is undefined to access an array out of bounds: array and
2647 pointer indexes must always be within the defined bounds of the array type.
2648 The one exception for this rules is zero length arrays. These arrays are
2649 defined to be accessible as variable length arrays, which requires access
2650 beyond the zero'th element.</p>
2652 <p>The getelementptr instruction is often confusing. For some more insight
2653 into how it works, see <a href="GetElementPtr.html">the getelementptr
2659 <i>; yields [12 x ubyte]*:aptr</i>
2660 %aptr = getelementptr {int, [12 x ubyte]}* %sptr, long 0, uint 1
2664 <!-- ======================================================================= -->
2665 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
2666 <div class="doc_text">
2667 <p>The instructions in this category are the "miscellaneous"
2668 instructions, which defy better classification.</p>
2670 <!-- _______________________________________________________________________ -->
2671 <div class="doc_subsubsection"> <a name="i_phi">'<tt>phi</tt>'
2672 Instruction</a> </div>
2673 <div class="doc_text">
2675 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
2677 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
2678 the SSA graph representing the function.</p>
2680 <p>The type of the incoming values are specified with the first type
2681 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
2682 as arguments, with one pair for each predecessor basic block of the
2683 current block. Only values of <a href="#t_firstclass">first class</a>
2684 type may be used as the value arguments to the PHI node. Only labels
2685 may be used as the label arguments.</p>
2686 <p>There must be no non-phi instructions between the start of a basic
2687 block and the PHI instructions: i.e. PHI instructions must be first in
2690 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the
2691 value specified by the parameter, depending on which basic block we
2692 came from in the last <a href="#terminators">terminator</a> instruction.</p>
2694 <pre>Loop: ; Infinite loop that counts from 0 on up...<br> %indvar = phi uint [ 0, %LoopHeader ], [ %nextindvar, %Loop ]<br> %nextindvar = add uint %indvar, 1<br> br label %Loop<br></pre>
2697 <!-- _______________________________________________________________________ -->
2698 <div class="doc_subsubsection">
2699 <a name="i_cast">'<tt>cast .. to</tt>' Instruction</a>
2702 <div class="doc_text">
2707 <result> = cast <ty> <value> to <ty2> <i>; yields ty2</i>
2713 The '<tt>cast</tt>' instruction is used as the primitive means to convert
2714 integers to floating point, change data type sizes, and break type safety (by
2722 The '<tt>cast</tt>' instruction takes a value to cast, which must be a first
2723 class value, and a type to cast it to, which must also be a <a
2724 href="#t_firstclass">first class</a> type.
2730 This instruction follows the C rules for explicit casts when determining how the
2731 data being cast must change to fit in its new container.
2735 When casting to bool, any value that would be considered true in the context of
2736 a C '<tt>if</tt>' condition is converted to the boolean '<tt>true</tt>' values,
2737 all else are '<tt>false</tt>'.
2741 When extending an integral value from a type of one signness to another (for
2742 example '<tt>sbyte</tt>' to '<tt>ulong</tt>'), the value is sign-extended if the
2743 <b>source</b> value is signed, and zero-extended if the source value is
2744 unsigned. <tt>bool</tt> values are always zero extended into either zero or
2751 %X = cast int 257 to ubyte <i>; yields ubyte:1</i>
2752 %Y = cast int 123 to bool <i>; yields bool:true</i>
2756 <!-- _______________________________________________________________________ -->
2757 <div class="doc_subsubsection">
2758 <a name="i_select">'<tt>select</tt>' Instruction</a>
2761 <div class="doc_text">
2766 <result> = select bool <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
2772 The '<tt>select</tt>' instruction is used to choose one value based on a
2773 condition, without branching.
2780 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.
2786 If the boolean condition evaluates to true, the instruction returns the first
2787 value argument; otherwise, it returns the second value argument.
2793 %X = select bool true, ubyte 17, ubyte 42 <i>; yields ubyte:17</i>
2798 <!-- _______________________________________________________________________ -->
2799 <div class="doc_subsubsection">
2800 <a name="i_call">'<tt>call</tt>' Instruction</a>
2803 <div class="doc_text">
2807 <result> = [tail] call [<a href="#callingconv">cconv</a>] <ty>* <fnptrval>(<param list>)
2812 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
2816 <p>This instruction requires several arguments:</p>
2820 <p>The optional "tail" marker indicates whether the callee function accesses
2821 any allocas or varargs in the caller. If the "tail" marker is present, the
2822 function call is eligible for tail call optimization. Note that calls may
2823 be marked "tail" even if they do not occur before a <a
2824 href="#i_ret"><tt>ret</tt></a> instruction.
2827 <p>The optional "cconv" marker indicates which <a href="callingconv">calling
2828 convention</a> the call should use. If none is specified, the call defaults
2829 to using C calling conventions.
2832 <p>'<tt>ty</tt>': shall be the signature of the pointer to function value
2833 being invoked. The argument types must match the types implied by this
2834 signature. This type can be omitted if the function is not varargs and
2835 if the function type does not return a pointer to a function.</p>
2838 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
2839 be invoked. In most cases, this is a direct function invocation, but
2840 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
2841 to function value.</p>
2844 <p>'<tt>function args</tt>': argument list whose types match the
2845 function signature argument types. All arguments must be of
2846 <a href="#t_firstclass">first class</a> type. If the function signature
2847 indicates the function accepts a variable number of arguments, the extra
2848 arguments can be specified.</p>
2854 <p>The '<tt>call</tt>' instruction is used to cause control flow to
2855 transfer to a specified function, with its incoming arguments bound to
2856 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
2857 instruction in the called function, control flow continues with the
2858 instruction after the function call, and the return value of the
2859 function is bound to the result argument. This is a simpler case of
2860 the <a href="#i_invoke">invoke</a> instruction.</p>
2865 %retval = call int %test(int %argc)
2866 call int(sbyte*, ...) *%printf(sbyte* %msg, int 12, sbyte 42);
2867 %X = tail call int %foo()
2868 %Y = tail call <a href="#callingconv">fastcc</a> int %foo()
2873 <!-- _______________________________________________________________________ -->
2874 <div class="doc_subsubsection">
2875 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
2878 <div class="doc_text">
2883 <resultval> = va_arg <va_list*> <arglist>, <argty>
2888 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
2889 the "variable argument" area of a function call. It is used to implement the
2890 <tt>va_arg</tt> macro in C.</p>
2894 <p>This instruction takes a <tt>va_list*</tt> value and the type of
2895 the argument. It returns a value of the specified argument type and
2896 increments the <tt>va_list</tt> to point to the next argument. Again, the
2897 actual type of <tt>va_list</tt> is target specific.</p>
2901 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
2902 type from the specified <tt>va_list</tt> and causes the
2903 <tt>va_list</tt> to point to the next argument. For more information,
2904 see the variable argument handling <a href="#int_varargs">Intrinsic
2907 <p>It is legal for this instruction to be called in a function which does not
2908 take a variable number of arguments, for example, the <tt>vfprintf</tt>
2911 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
2912 href="#intrinsics">intrinsic function</a> because it takes a type as an
2917 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
2921 <!-- *********************************************************************** -->
2922 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
2923 <!-- *********************************************************************** -->
2925 <div class="doc_text">
2927 <p>LLVM supports the notion of an "intrinsic function". These functions have
2928 well known names and semantics and are required to follow certain
2929 restrictions. Overall, these instructions represent an extension mechanism for
2930 the LLVM language that does not require changing all of the transformations in
2931 LLVM to add to the language (or the bytecode reader/writer, the parser,
2934 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
2935 prefix is reserved in LLVM for intrinsic names; thus, functions may not be named
2936 this. Intrinsic functions must always be external functions: you cannot define
2937 the body of intrinsic functions. Intrinsic functions may only be used in call
2938 or invoke instructions: it is illegal to take the address of an intrinsic
2939 function. Additionally, because intrinsic functions are part of the LLVM
2940 language, it is required that they all be documented here if any are added.</p>
2943 <p>To learn how to add an intrinsic function, please see the <a
2944 href="ExtendingLLVM.html">Extending LLVM Guide</a>.
2949 <!-- ======================================================================= -->
2950 <div class="doc_subsection">
2951 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
2954 <div class="doc_text">
2956 <p>Variable argument support is defined in LLVM with the <a
2957 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
2958 intrinsic functions. These functions are related to the similarly
2959 named macros defined in the <tt><stdarg.h></tt> header file.</p>
2961 <p>All of these functions operate on arguments that use a
2962 target-specific value type "<tt>va_list</tt>". The LLVM assembly
2963 language reference manual does not define what this type is, so all
2964 transformations should be prepared to handle intrinsics with any type
2967 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
2968 instruction and the variable argument handling intrinsic functions are
2972 int %test(int %X, ...) {
2973 ; Initialize variable argument processing
2975 call void %<a href="#i_va_start">llvm.va_start</a>(sbyte** %ap)
2977 ; Read a single integer argument
2978 %tmp = va_arg sbyte** %ap, int
2980 ; Demonstrate usage of llvm.va_copy and llvm.va_end
2982 call void %<a href="#i_va_copy">llvm.va_copy</a>(sbyte** %aq, sbyte** %ap)
2983 call void %<a href="#i_va_end">llvm.va_end</a>(sbyte** %aq)
2985 ; Stop processing of arguments.
2986 call void %<a href="#i_va_end">llvm.va_end</a>(sbyte** %ap)
2992 <!-- _______________________________________________________________________ -->
2993 <div class="doc_subsubsection">
2994 <a name="i_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
2998 <div class="doc_text">
3000 <pre> declare void %llvm.va_start(<va_list>* <arglist>)<br></pre>
3002 <P>The '<tt>llvm.va_start</tt>' intrinsic initializes
3003 <tt>*<arglist></tt> for subsequent use by <tt><a
3004 href="#i_va_arg">va_arg</a></tt>.</p>
3008 <P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
3012 <P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
3013 macro available in C. In a target-dependent way, it initializes the
3014 <tt>va_list</tt> element the argument points to, so that the next call to
3015 <tt>va_arg</tt> will produce the first variable argument passed to the function.
3016 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
3017 last argument of the function, the compiler can figure that out.</p>
3021 <!-- _______________________________________________________________________ -->
3022 <div class="doc_subsubsection">
3023 <a name="i_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
3026 <div class="doc_text">
3028 <pre> declare void %llvm.va_end(<va_list*> <arglist>)<br></pre>
3030 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt><arglist></tt>
3031 which has been initialized previously with <tt><a href="#i_va_start">llvm.va_start</a></tt>
3032 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
3034 <p>The argument is a <tt>va_list</tt> to destroy.</p>
3036 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
3037 macro available in C. In a target-dependent way, it destroys the <tt>va_list</tt>.
3038 Calls to <a href="#i_va_start"><tt>llvm.va_start</tt></a> and <a
3039 href="#i_va_copy"><tt>llvm.va_copy</tt></a> must be matched exactly
3040 with calls to <tt>llvm.va_end</tt>.</p>
3043 <!-- _______________________________________________________________________ -->
3044 <div class="doc_subsubsection">
3045 <a name="i_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
3048 <div class="doc_text">
3053 declare void %llvm.va_copy(<va_list>* <destarglist>,
3054 <va_list>* <srcarglist>)
3059 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position from
3060 the source argument list to the destination argument list.</p>
3064 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
3065 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
3070 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt> macro
3071 available in C. In a target-dependent way, it copies the source
3072 <tt>va_list</tt> element into the destination list. This intrinsic is necessary
3073 because the <tt><a href="i_va_begin">llvm.va_begin</a></tt> intrinsic may be
3074 arbitrarily complex and require memory allocation, for example.</p>
3078 <!-- ======================================================================= -->
3079 <div class="doc_subsection">
3080 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
3083 <div class="doc_text">
3086 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
3087 Collection</a> requires the implementation and generation of these intrinsics.
3088 These intrinsics allow identification of <a href="#i_gcroot">GC roots on the
3089 stack</a>, as well as garbage collector implementations that require <a
3090 href="#i_gcread">read</a> and <a href="#i_gcwrite">write</a> barriers.
3091 Front-ends for type-safe garbage collected languages should generate these
3092 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
3093 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
3097 <!-- _______________________________________________________________________ -->
3098 <div class="doc_subsubsection">
3099 <a name="i_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
3102 <div class="doc_text">
3107 declare void %llvm.gcroot(<ty>** %ptrloc, <ty2>* %metadata)
3112 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
3113 the code generator, and allows some metadata to be associated with it.</p>
3117 <p>The first argument specifies the address of a stack object that contains the
3118 root pointer. The second pointer (which must be either a constant or a global
3119 value address) contains the meta-data to be associated with the root.</p>
3123 <p>At runtime, a call to this intrinsics stores a null pointer into the "ptrloc"
3124 location. At compile-time, the code generator generates information to allow
3125 the runtime to find the pointer at GC safe points.
3131 <!-- _______________________________________________________________________ -->
3132 <div class="doc_subsubsection">
3133 <a name="i_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
3136 <div class="doc_text">
3141 declare sbyte* %llvm.gcread(sbyte* %ObjPtr, sbyte** %Ptr)
3146 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
3147 locations, allowing garbage collector implementations that require read
3152 <p>The second argument is the address to read from, which should be an address
3153 allocated from the garbage collector. The first object is a pointer to the
3154 start of the referenced object, if needed by the language runtime (otherwise
3159 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
3160 instruction, but may be replaced with substantially more complex code by the
3161 garbage collector runtime, as needed.</p>
3166 <!-- _______________________________________________________________________ -->
3167 <div class="doc_subsubsection">
3168 <a name="i_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
3171 <div class="doc_text">
3176 declare void %llvm.gcwrite(sbyte* %P1, sbyte* %Obj, sbyte** %P2)
3181 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
3182 locations, allowing garbage collector implementations that require write
3183 barriers (such as generational or reference counting collectors).</p>
3187 <p>The first argument is the reference to store, the second is the start of the
3188 object to store it to, and the third is the address of the field of Obj to
3189 store to. If the runtime does not require a pointer to the object, Obj may be
3194 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
3195 instruction, but may be replaced with substantially more complex code by the
3196 garbage collector runtime, as needed.</p>
3202 <!-- ======================================================================= -->
3203 <div class="doc_subsection">
3204 <a name="int_codegen">Code Generator Intrinsics</a>
3207 <div class="doc_text">
3209 These intrinsics are provided by LLVM to expose special features that may only
3210 be implemented with code generator support.
3215 <!-- _______________________________________________________________________ -->
3216 <div class="doc_subsubsection">
3217 <a name="i_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
3220 <div class="doc_text">
3224 declare sbyte *%llvm.returnaddress(uint <level>)
3230 The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
3231 target-specific value indicating the return address of the current function
3232 or one of its callers.
3238 The argument to this intrinsic indicates which function to return the address
3239 for. Zero indicates the calling function, one indicates its caller, etc. The
3240 argument is <b>required</b> to be a constant integer value.
3246 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
3247 the return address of the specified call frame, or zero if it cannot be
3248 identified. The value returned by this intrinsic is likely to be incorrect or 0
3249 for arguments other than zero, so it should only be used for debugging purposes.
3253 Note that calling this intrinsic does not prevent function inlining or other
3254 aggressive transformations, so the value returned may not be that of the obvious
3255 source-language caller.
3260 <!-- _______________________________________________________________________ -->
3261 <div class="doc_subsubsection">
3262 <a name="i_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
3265 <div class="doc_text">
3269 declare sbyte *%llvm.frameaddress(uint <level>)
3275 The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
3276 target-specific frame pointer value for the specified stack frame.
3282 The argument to this intrinsic indicates which function to return the frame
3283 pointer for. Zero indicates the calling function, one indicates its caller,
3284 etc. The argument is <b>required</b> to be a constant integer value.
3290 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
3291 the frame address of the specified call frame, or zero if it cannot be
3292 identified. The value returned by this intrinsic is likely to be incorrect or 0
3293 for arguments other than zero, so it should only be used for debugging purposes.
3297 Note that calling this intrinsic does not prevent function inlining or other
3298 aggressive transformations, so the value returned may not be that of the obvious
3299 source-language caller.
3303 <!-- _______________________________________________________________________ -->
3304 <div class="doc_subsubsection">
3305 <a name="i_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
3308 <div class="doc_text">
3312 declare sbyte *%llvm.stacksave()
3318 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
3319 the function stack, for use with <a href="#i_stackrestore">
3320 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
3321 features like scoped automatic variable sized arrays in C99.
3327 This intrinsic returns a opaque pointer value that can be passed to <a
3328 href="#i_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
3329 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
3330 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
3331 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
3332 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
3333 that were allocated after the <tt>llvm.stacksave</tt> was executed.
3338 <!-- _______________________________________________________________________ -->
3339 <div class="doc_subsubsection">
3340 <a name="i_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
3343 <div class="doc_text">
3347 declare void %llvm.stackrestore(sbyte* %ptr)
3353 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
3354 the function stack to the state it was in when the corresponding <a
3355 href="#llvm.stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
3356 useful for implementing language features like scoped automatic variable sized
3363 See the description for <a href="#i_stacksave"><tt>llvm.stacksave</tt></a>.
3369 <!-- _______________________________________________________________________ -->
3370 <div class="doc_subsubsection">
3371 <a name="i_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
3374 <div class="doc_text">
3378 declare void %llvm.prefetch(sbyte * <address>,
3379 uint <rw>, uint <locality>)
3386 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
3387 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
3389 effect on the behavior of the program but can change its performance
3396 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
3397 determining if the fetch should be for a read (0) or write (1), and
3398 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
3399 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
3400 <tt>locality</tt> arguments must be constant integers.
3406 This intrinsic does not modify the behavior of the program. In particular,
3407 prefetches cannot trap and do not produce a value. On targets that support this
3408 intrinsic, the prefetch can provide hints to the processor cache for better
3414 <!-- _______________________________________________________________________ -->
3415 <div class="doc_subsubsection">
3416 <a name="i_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
3419 <div class="doc_text">
3423 declare void %llvm.pcmarker( uint <id> )
3430 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
3432 code to simulators and other tools. The method is target specific, but it is
3433 expected that the marker will use exported symbols to transmit the PC of the marker.
3434 The marker makes no guarantees that it will remain with any specific instruction
3435 after optimizations. It is possible that the presence of a marker will inhibit
3436 optimizations. The intended use is to be inserted after optimizations to allow
3437 correlations of simulation runs.
3443 <tt>id</tt> is a numerical id identifying the marker.
3449 This intrinsic does not modify the behavior of the program. Backends that do not
3450 support this intrinisic may ignore it.
3455 <!-- _______________________________________________________________________ -->
3456 <div class="doc_subsubsection">
3457 <a name="i_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
3460 <div class="doc_text">
3464 declare ulong %llvm.readcyclecounter( )
3471 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
3472 counter register (or similar low latency, high accuracy clocks) on those targets
3473 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
3474 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
3475 should only be used for small timings.
3481 When directly supported, reading the cycle counter should not modify any memory.
3482 Implementations are allowed to either return a application specific value or a
3483 system wide value. On backends without support, this is lowered to a constant 0.
3488 <!-- ======================================================================= -->
3489 <div class="doc_subsection">
3490 <a name="int_libc">Standard C Library Intrinsics</a>
3493 <div class="doc_text">
3495 LLVM provides intrinsics for a few important standard C library functions.
3496 These intrinsics allow source-language front-ends to pass information about the
3497 alignment of the pointer arguments to the code generator, providing opportunity
3498 for more efficient code generation.
3503 <!-- _______________________________________________________________________ -->
3504 <div class="doc_subsubsection">
3505 <a name="i_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
3508 <div class="doc_text">
3512 declare void %llvm.memcpy.i32(sbyte* <dest>, sbyte* <src>,
3513 uint <len>, uint <align>)
3514 declare void %llvm.memcpy.i64(sbyte* <dest>, sbyte* <src>,
3515 ulong <len>, uint <align>)
3521 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
3522 location to the destination location.
3526 Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
3527 intrinsics do not return a value, and takes an extra alignment argument.
3533 The first argument is a pointer to the destination, the second is a pointer to
3534 the source. The third argument is an integer argument
3535 specifying the number of bytes to copy, and the fourth argument is the alignment
3536 of the source and destination locations.
3540 If the call to this intrinisic has an alignment value that is not 0 or 1, then
3541 the caller guarantees that both the source and destination pointers are aligned
3548 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
3549 location to the destination location, which are not allowed to overlap. It
3550 copies "len" bytes of memory over. If the argument is known to be aligned to
3551 some boundary, this can be specified as the fourth argument, otherwise it should
3557 <!-- _______________________________________________________________________ -->
3558 <div class="doc_subsubsection">
3559 <a name="i_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
3562 <div class="doc_text">
3566 declare void %llvm.memmove.i32(sbyte* <dest>, sbyte* <src>,
3567 uint <len>, uint <align>)
3568 declare void %llvm.memmove.i64(sbyte* <dest>, sbyte* <src>,
3569 ulong <len>, uint <align>)
3575 The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
3576 location to the destination location. It is similar to the
3577 '<tt>llvm.memcmp</tt>' intrinsic but allows the two memory locations to overlap.
3581 Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
3582 intrinsics do not return a value, and takes an extra alignment argument.
3588 The first argument is a pointer to the destination, the second is a pointer to
3589 the source. The third argument is an integer argument
3590 specifying the number of bytes to copy, and the fourth argument is the alignment
3591 of the source and destination locations.
3595 If the call to this intrinisic has an alignment value that is not 0 or 1, then
3596 the caller guarantees that the source and destination pointers are aligned to
3603 The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
3604 location to the destination location, which may overlap. It
3605 copies "len" bytes of memory over. If the argument is known to be aligned to
3606 some boundary, this can be specified as the fourth argument, otherwise it should
3612 <!-- _______________________________________________________________________ -->
3613 <div class="doc_subsubsection">
3614 <a name="i_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
3617 <div class="doc_text">
3621 declare void %llvm.memset.i32(sbyte* <dest>, ubyte <val>,
3622 uint <len>, uint <align>)
3623 declare void %llvm.memset.i64(sbyte* <dest>, ubyte <val>,
3624 ulong <len>, uint <align>)
3630 The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
3635 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
3636 does not return a value, and takes an extra alignment argument.
3642 The first argument is a pointer to the destination to fill, the second is the
3643 byte value to fill it with, the third argument is an integer
3644 argument specifying the number of bytes to fill, and the fourth argument is the
3645 known alignment of destination location.
3649 If the call to this intrinisic has an alignment value that is not 0 or 1, then
3650 the caller guarantees that the destination pointer is aligned to that boundary.
3656 The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
3658 destination location. If the argument is known to be aligned to some boundary,
3659 this can be specified as the fourth argument, otherwise it should be set to 0 or
3665 <!-- _______________________________________________________________________ -->
3666 <div class="doc_subsubsection">
3667 <a name="i_isunordered">'<tt>llvm.isunordered.*</tt>' Intrinsic</a>
3670 <div class="doc_text">
3674 declare bool %llvm.isunordered.f32(float Val1, float Val2)
3675 declare bool %llvm.isunordered.f64(double Val1, double Val2)
3681 The '<tt>llvm.isunordered</tt>' intrinsics return true if either or both of the
3682 specified floating point values is a NAN.
3688 The arguments are floating point numbers of the same type.
3694 If either or both of the arguments is a SNAN or QNAN, it returns true, otherwise
3700 <!-- _______________________________________________________________________ -->
3701 <div class="doc_subsubsection">
3702 <a name="i_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
3705 <div class="doc_text">
3709 declare float %llvm.sqrt.f32(float %Val)
3710 declare double %llvm.sqrt.f64(double %Val)
3716 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
3717 returning the same value as the libm '<tt>sqrt</tt>' function would. Unlike
3718 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
3719 negative numbers (which allows for better optimization).
3725 The argument and return value are floating point numbers of the same type.
3731 This function returns the sqrt of the specified operand if it is a positive
3732 floating point number.
3736 <!-- _______________________________________________________________________ -->
3737 <div class="doc_subsubsection">
3738 <a name="i_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
3741 <div class="doc_text">
3745 declare float %llvm.powi.f32(float %Val, int %power)
3746 declare double %llvm.powi.f64(double %Val, int %power)
3752 The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
3753 specified (positive or negative) power. The order of evaluation of
3754 multiplications is not defined.
3760 The second argument is an integer power, and the first is a value to raise to
3767 This function returns the first value raised to the second power with an
3768 unspecified sequence of rounding operations.</p>
3772 <!-- ======================================================================= -->
3773 <div class="doc_subsection">
3774 <a name="int_manip">Bit Manipulation Intrinsics</a>
3777 <div class="doc_text">
3779 LLVM provides intrinsics for a few important bit manipulation operations.
3780 These allow efficient code generation for some algorithms.
3785 <!-- _______________________________________________________________________ -->
3786 <div class="doc_subsubsection">
3787 <a name="i_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
3790 <div class="doc_text">
3794 declare ushort %llvm.bswap.i16(ushort <id>)
3795 declare uint %llvm.bswap.i32(uint <id>)
3796 declare ulong %llvm.bswap.i64(ulong <id>)
3802 The '<tt>llvm.bwsap</tt>' family of intrinsics is used to byteswap a 16, 32 or
3803 64 bit quantity. These are useful for performing operations on data that is not
3804 in the target's native byte order.
3810 The <tt>llvm.bswap.16</tt> intrinsic returns a ushort value that has the high and low
3811 byte of the input ushort swapped. Similarly, the <tt>llvm.bswap.i32</tt> intrinsic
3812 returns a uint value that has the four bytes of the input uint swapped, so that
3813 if the input bytes are numbered 0, 1, 2, 3 then the returned uint will have its
3814 bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i64</tt> intrinsic extends this concept
3820 <!-- _______________________________________________________________________ -->
3821 <div class="doc_subsubsection">
3822 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
3825 <div class="doc_text">
3829 declare ubyte %llvm.ctpop.i8 (ubyte <src>)
3830 declare ushort %llvm.ctpop.i16(ushort <src>)
3831 declare uint %llvm.ctpop.i32(uint <src>)
3832 declare ulong %llvm.ctpop.i64(ulong <src>)
3838 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
3845 The only argument is the value to be counted. The argument may be of any
3846 unsigned integer type. The return type must match the argument type.
3852 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
3856 <!-- _______________________________________________________________________ -->
3857 <div class="doc_subsubsection">
3858 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
3861 <div class="doc_text">
3865 declare ubyte %llvm.ctlz.i8 (ubyte <src>)
3866 declare ushort %llvm.ctlz.i16(ushort <src>)
3867 declare uint %llvm.ctlz.i32(uint <src>)
3868 declare ulong %llvm.ctlz.i64(ulong <src>)
3874 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
3875 leading zeros in a variable.
3881 The only argument is the value to be counted. The argument may be of any
3882 unsigned integer type. The return type must match the argument type.
3888 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
3889 in a variable. If the src == 0 then the result is the size in bits of the type
3890 of src. For example, <tt>llvm.ctlz(int 2) = 30</tt>.
3896 <!-- _______________________________________________________________________ -->
3897 <div class="doc_subsubsection">
3898 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
3901 <div class="doc_text">
3905 declare ubyte %llvm.cttz.i8 (ubyte <src>)
3906 declare ushort %llvm.cttz.i16(ushort <src>)
3907 declare uint %llvm.cttz.i32(uint <src>)
3908 declare ulong %llvm.cttz.i64(ulong <src>)
3914 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
3921 The only argument is the value to be counted. The argument may be of any
3922 unsigned integer type. The return type must match the argument type.
3928 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
3929 in a variable. If the src == 0 then the result is the size in bits of the type
3930 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
3934 <!-- ======================================================================= -->
3935 <div class="doc_subsection">
3936 <a name="int_debugger">Debugger Intrinsics</a>
3939 <div class="doc_text">
3941 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
3942 are described in the <a
3943 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
3944 Debugging</a> document.
3949 <!-- *********************************************************************** -->
3952 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
3953 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
3954 <a href="http://validator.w3.org/check/referer"><img
3955 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a>
3957 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
3958 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
3959 Last modified: $Date$