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_urem">'<tt>urem</tt>' Instruction</a></li>
84 <li><a href="#i_srem">'<tt>srem</tt>' Instruction</a></li>
85 <li><a href="#i_frem">'<tt>frem</tt>' Instruction</a></li>
86 <li><a href="#i_setcc">'<tt>set<i>cc</i></tt>' Instructions</a></li>
89 <li><a href="#bitwiseops">Bitwise Binary Operations</a>
91 <li><a href="#i_and">'<tt>and</tt>' Instruction</a></li>
92 <li><a href="#i_or">'<tt>or</tt>' Instruction</a></li>
93 <li><a href="#i_xor">'<tt>xor</tt>' Instruction</a></li>
94 <li><a href="#i_shl">'<tt>shl</tt>' Instruction</a></li>
95 <li><a href="#i_shr">'<tt>shr</tt>' Instruction</a></li>
98 <li><a href="#vectorops">Vector Operations</a>
100 <li><a href="#i_extractelement">'<tt>extractelement</tt>' Instruction</a></li>
101 <li><a href="#i_insertelement">'<tt>insertelement</tt>' Instruction</a></li>
102 <li><a href="#i_shufflevector">'<tt>shufflevector</tt>' Instruction</a></li>
103 <li><a href="#i_vsetint">'<tt>vsetint</tt>' Instruction</a></li>
104 <li><a href="#i_vsetfp">'<tt>vsetfp</tt>' Instruction</a></li>
105 <li><a href="#i_vselect">'<tt>vselect</tt>' Instruction</a></li>
108 <li><a href="#memoryops">Memory Access and Addressing Operations</a>
110 <li><a href="#i_malloc">'<tt>malloc</tt>' Instruction</a></li>
111 <li><a href="#i_free">'<tt>free</tt>' Instruction</a></li>
112 <li><a href="#i_alloca">'<tt>alloca</tt>' Instruction</a></li>
113 <li><a href="#i_load">'<tt>load</tt>' Instruction</a></li>
114 <li><a href="#i_store">'<tt>store</tt>' Instruction</a></li>
115 <li><a href="#i_getelementptr">'<tt>getelementptr</tt>' Instruction</a></li>
118 <li><a href="#otherops">Other Operations</a>
120 <li><a href="#i_phi">'<tt>phi</tt>' Instruction</a></li>
121 <li><a href="#i_cast">'<tt>cast .. to</tt>' Instruction</a></li>
122 <li><a href="#i_select">'<tt>select</tt>' Instruction</a></li>
123 <li><a href="#i_call">'<tt>call</tt>' Instruction</a></li>
124 <li><a href="#i_va_arg">'<tt>va_arg</tt>' Instruction</a></li>
129 <li><a href="#intrinsics">Intrinsic Functions</a>
131 <li><a href="#int_varargs">Variable Argument Handling Intrinsics</a>
133 <li><a href="#i_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a></li>
134 <li><a href="#i_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a></li>
135 <li><a href="#i_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a></li>
138 <li><a href="#int_gc">Accurate Garbage Collection Intrinsics</a>
140 <li><a href="#i_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a></li>
141 <li><a href="#i_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a></li>
142 <li><a href="#i_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a></li>
145 <li><a href="#int_codegen">Code Generator Intrinsics</a>
147 <li><a href="#i_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a></li>
148 <li><a href="#i_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a></li>
149 <li><a href="#i_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a></li>
150 <li><a href="#i_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a></li>
151 <li><a href="#i_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a></li>
152 <li><a href="#i_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a></li>
153 <li><a href="#i_readcyclecounter"><tt>llvm.readcyclecounter</tt>' Intrinsic</a></li>
156 <li><a href="#int_libc">Standard C Library Intrinsics</a>
158 <li><a href="#i_memcpy">'<tt>llvm.memcpy.*</tt>' Intrinsic</a></li>
159 <li><a href="#i_memmove">'<tt>llvm.memmove.*</tt>' Intrinsic</a></li>
160 <li><a href="#i_memset">'<tt>llvm.memset.*</tt>' Intrinsic</a></li>
161 <li><a href="#i_isunordered">'<tt>llvm.isunordered.*</tt>' Intrinsic</a></li>
162 <li><a href="#i_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a></li>
163 <li><a href="#i_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a></li>
166 <li><a href="#int_manip">Bit Manipulation Intrinsics</a>
168 <li><a href="#i_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a></li>
169 <li><a href="#int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic </a></li>
170 <li><a href="#int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic </a></li>
171 <li><a href="#int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic </a></li>
174 <li><a href="#int_debugger">Debugger intrinsics</a></li>
179 <div class="doc_author">
180 <p>Written by <a href="mailto:sabre@nondot.org">Chris Lattner</a>
181 and <a href="mailto:vadve@cs.uiuc.edu">Vikram Adve</a></p>
184 <!-- *********************************************************************** -->
185 <div class="doc_section"> <a name="abstract">Abstract </a></div>
186 <!-- *********************************************************************** -->
188 <div class="doc_text">
189 <p>This document is a reference manual for the LLVM assembly language.
190 LLVM is an SSA based representation that provides type safety,
191 low-level operations, flexibility, and the capability of representing
192 'all' high-level languages cleanly. It is the common code
193 representation used throughout all phases of the LLVM compilation
197 <!-- *********************************************************************** -->
198 <div class="doc_section"> <a name="introduction">Introduction</a> </div>
199 <!-- *********************************************************************** -->
201 <div class="doc_text">
203 <p>The LLVM code representation is designed to be used in three
204 different forms: as an in-memory compiler IR, as an on-disk bytecode
205 representation (suitable for fast loading by a Just-In-Time compiler),
206 and as a human readable assembly language representation. This allows
207 LLVM to provide a powerful intermediate representation for efficient
208 compiler transformations and analysis, while providing a natural means
209 to debug and visualize the transformations. The three different forms
210 of LLVM are all equivalent. This document describes the human readable
211 representation and notation.</p>
213 <p>The LLVM representation aims to be light-weight and low-level
214 while being expressive, typed, and extensible at the same time. It
215 aims to be a "universal IR" of sorts, by being at a low enough level
216 that high-level ideas may be cleanly mapped to it (similar to how
217 microprocessors are "universal IR's", allowing many source languages to
218 be mapped to them). By providing type information, LLVM can be used as
219 the target of optimizations: for example, through pointer analysis, it
220 can be proven that a C automatic variable is never accessed outside of
221 the current function... allowing it to be promoted to a simple SSA
222 value instead of a memory location.</p>
226 <!-- _______________________________________________________________________ -->
227 <div class="doc_subsubsection"> <a name="wellformed">Well-Formedness</a> </div>
229 <div class="doc_text">
231 <p>It is important to note that this document describes 'well formed'
232 LLVM assembly language. There is a difference between what the parser
233 accepts and what is considered 'well formed'. For example, the
234 following instruction is syntactically okay, but not well formed:</p>
237 %x = <a href="#i_add">add</a> int 1, %x
240 <p>...because the definition of <tt>%x</tt> does not dominate all of
241 its uses. The LLVM infrastructure provides a verification pass that may
242 be used to verify that an LLVM module is well formed. This pass is
243 automatically run by the parser after parsing input assembly and by
244 the optimizer before it outputs bytecode. The violations pointed out
245 by the verifier pass indicate bugs in transformation passes or input to
248 <!-- Describe the typesetting conventions here. --> </div>
250 <!-- *********************************************************************** -->
251 <div class="doc_section"> <a name="identifiers">Identifiers</a> </div>
252 <!-- *********************************************************************** -->
254 <div class="doc_text">
256 <p>LLVM uses three different forms of identifiers, for different
260 <li>Named values are represented as a string of characters with a '%' prefix.
261 For example, %foo, %DivisionByZero, %a.really.long.identifier. The actual
262 regular expression used is '<tt>%[a-zA-Z$._][a-zA-Z$._0-9]*</tt>'.
263 Identifiers which require other characters in their names can be surrounded
264 with quotes. In this way, anything except a <tt>"</tt> character can be used
267 <li>Unnamed values are represented as an unsigned numeric value with a '%'
268 prefix. For example, %12, %2, %44.</li>
270 <li>Constants, which are described in a <a href="#constants">section about
271 constants</a>, below.</li>
274 <p>LLVM requires that values start with a '%' sign for two reasons: Compilers
275 don't need to worry about name clashes with reserved words, and the set of
276 reserved words may be expanded in the future without penalty. Additionally,
277 unnamed identifiers allow a compiler to quickly come up with a temporary
278 variable without having to avoid symbol table conflicts.</p>
280 <p>Reserved words in LLVM are very similar to reserved words in other
281 languages. There are keywords for different opcodes ('<tt><a
282 href="#i_add">add</a></tt>', '<tt><a href="#i_cast">cast</a></tt>', '<tt><a
283 href="#i_ret">ret</a></tt>', etc...), for primitive type names ('<tt><a
284 href="#t_void">void</a></tt>', '<tt><a href="#t_uint">uint</a></tt>', etc...),
285 and others. These reserved words cannot conflict with variable names, because
286 none of them start with a '%' character.</p>
288 <p>Here is an example of LLVM code to multiply the integer variable
289 '<tt>%X</tt>' by 8:</p>
294 %result = <a href="#i_mul">mul</a> uint %X, 8
297 <p>After strength reduction:</p>
300 %result = <a href="#i_shl">shl</a> uint %X, ubyte 3
303 <p>And the hard way:</p>
306 <a href="#i_add">add</a> uint %X, %X <i>; yields {uint}:%0</i>
307 <a href="#i_add">add</a> uint %0, %0 <i>; yields {uint}:%1</i>
308 %result = <a href="#i_add">add</a> uint %1, %1
311 <p>This last way of multiplying <tt>%X</tt> by 8 illustrates several
312 important lexical features of LLVM:</p>
316 <li>Comments are delimited with a '<tt>;</tt>' and go until the end of
319 <li>Unnamed temporaries are created when the result of a computation is not
320 assigned to a named value.</li>
322 <li>Unnamed temporaries are numbered sequentially</li>
326 <p>...and it also shows a convention that we follow in this document. When
327 demonstrating instructions, we will follow an instruction with a comment that
328 defines the type and name of value produced. Comments are shown in italic
333 <!-- *********************************************************************** -->
334 <div class="doc_section"> <a name="highlevel">High Level Structure</a> </div>
335 <!-- *********************************************************************** -->
337 <!-- ======================================================================= -->
338 <div class="doc_subsection"> <a name="modulestructure">Module Structure</a>
341 <div class="doc_text">
343 <p>LLVM programs are composed of "Module"s, each of which is a
344 translation unit of the input programs. Each module consists of
345 functions, global variables, and symbol table entries. Modules may be
346 combined together with the LLVM linker, which merges function (and
347 global variable) definitions, resolves forward declarations, and merges
348 symbol table entries. Here is an example of the "hello world" module:</p>
350 <pre><i>; Declare the string constant as a global constant...</i>
351 <a href="#identifiers">%.LC0</a> = <a href="#linkage_internal">internal</a> <a
352 href="#globalvars">constant</a> <a href="#t_array">[13 x sbyte]</a> c"hello world\0A\00" <i>; [13 x sbyte]*</i>
354 <i>; External declaration of the puts function</i>
355 <a href="#functionstructure">declare</a> int %puts(sbyte*) <i>; int(sbyte*)* </i>
357 <i>; Global variable / Function body section separator</i>
360 <i>; Definition of main function</i>
361 int %main() { <i>; int()* </i>
362 <i>; Convert [13x sbyte]* to sbyte *...</i>
364 href="#i_getelementptr">getelementptr</a> [13 x sbyte]* %.LC0, long 0, long 0 <i>; sbyte*</i>
366 <i>; Call puts function to write out the string to stdout...</i>
368 href="#i_call">call</a> int %puts(sbyte* %cast210) <i>; int</i>
370 href="#i_ret">ret</a> int 0<br>}<br></pre>
372 <p>This example is made up of a <a href="#globalvars">global variable</a>
373 named "<tt>.LC0</tt>", an external declaration of the "<tt>puts</tt>"
374 function, and a <a href="#functionstructure">function definition</a>
375 for "<tt>main</tt>".</p>
377 <p>In general, a module is made up of a list of global values,
378 where both functions and global variables are global values. Global values are
379 represented by a pointer to a memory location (in this case, a pointer to an
380 array of char, and a pointer to a function), and have one of the following <a
381 href="#linkage">linkage types</a>.</p>
383 <p>Due to a limitation in the current LLVM assembly parser (it is limited by
384 one-token lookahead), modules are split into two pieces by the "implementation"
385 keyword. Global variable prototypes and definitions must occur before the
386 keyword, and function definitions must occur after it. Function prototypes may
387 occur either before or after it. In the future, the implementation keyword may
388 become a noop, if the parser gets smarter.</p>
392 <!-- ======================================================================= -->
393 <div class="doc_subsection">
394 <a name="linkage">Linkage Types</a>
397 <div class="doc_text">
400 All Global Variables and Functions have one of the following types of linkage:
405 <dt><tt><b><a name="linkage_internal">internal</a></b></tt> </dt>
407 <dd>Global values with internal linkage are only directly accessible by
408 objects in the current module. In particular, linking code into a module with
409 an internal global value may cause the internal to be renamed as necessary to
410 avoid collisions. Because the symbol is internal to the module, all
411 references can be updated. This corresponds to the notion of the
412 '<tt>static</tt>' keyword in C, or the idea of "anonymous namespaces" in C++.
415 <dt><tt><b><a name="linkage_linkonce">linkonce</a></b></tt>: </dt>
417 <dd>"<tt>linkonce</tt>" linkage is similar to <tt>internal</tt> linkage, with
418 the twist that linking together two modules defining the same
419 <tt>linkonce</tt> globals will cause one of the globals to be discarded. This
420 is typically used to implement inline functions. Unreferenced
421 <tt>linkonce</tt> globals are allowed to be discarded.
424 <dt><tt><b><a name="linkage_weak">weak</a></b></tt>: </dt>
426 <dd>"<tt>weak</tt>" linkage is exactly the same as <tt>linkonce</tt> linkage,
427 except that unreferenced <tt>weak</tt> globals may not be discarded. This is
428 used to implement constructs in C such as "<tt>int X;</tt>" at global scope.
431 <dt><tt><b><a name="linkage_appending">appending</a></b></tt>: </dt>
433 <dd>"<tt>appending</tt>" linkage may only be applied to global variables of
434 pointer to array type. When two global variables with appending linkage are
435 linked together, the two global arrays are appended together. This is the
436 LLVM, typesafe, equivalent of having the system linker append together
437 "sections" with identical names when .o files are linked.
440 <dt><tt><b><a name="linkage_external">externally visible</a></b></tt>:</dt>
442 <dd>If none of the above identifiers are used, the global is externally
443 visible, meaning that it participates in linkage and can be used to resolve
444 external symbol references.
447 <dt><tt><b><a name="linkage_externweak">extern_weak</a></b></tt>: </dt>
449 <dd>"<tt>extern_weak</tt>" TBD
453 The next two types of linkage are targeted for Microsoft Windows platform
454 only. They are designed to support importing (exporting) symbols from (to)
458 <dt><tt><b><a name="linkage_dllimport">dllimport</a></b></tt>: </dt>
460 <dd>"<tt>dllimport</tt>" linkage causes the compiler to reference a function
461 or variable via a global pointer to a pointer that is set up by the DLL
462 exporting the symbol. On Microsoft Windows targets, the pointer name is
463 formed by combining <code>_imp__</code> and the function or variable name.
466 <dt><tt><b><a name="linkage_dllexport">dllexport</a></b></tt>: </dt>
468 <dd>"<tt>dllexport</tt>" linkage causes the compiler to provide a global
469 pointer to a pointer in a DLL, so that it can be referenced with the
470 <tt>dllimport</tt> attribute. On Microsoft Windows targets, the pointer
471 name is formed by combining <code>_imp__</code> and the function or variable
477 <p><a name="linkage_external">For example, since the "<tt>.LC0</tt>"
478 variable is defined to be internal, if another module defined a "<tt>.LC0</tt>"
479 variable and was linked with this one, one of the two would be renamed,
480 preventing a collision. Since "<tt>main</tt>" and "<tt>puts</tt>" are
481 external (i.e., lacking any linkage declarations), they are accessible
482 outside of the current module. It is illegal for a function <i>declaration</i>
483 to have any linkage type other than "externally visible".</a></p>
487 <!-- ======================================================================= -->
488 <div class="doc_subsection">
489 <a name="callingconv">Calling Conventions</a>
492 <div class="doc_text">
494 <p>LLVM <a href="#functionstructure">functions</a>, <a href="#i_call">calls</a>
495 and <a href="#i_invoke">invokes</a> can all have an optional calling convention
496 specified for the call. The calling convention of any pair of dynamic
497 caller/callee must match, or the behavior of the program is undefined. The
498 following calling conventions are supported by LLVM, and more may be added in
502 <dt><b>"<tt>ccc</tt>" - The C calling convention</b>:</dt>
504 <dd>This calling convention (the default if no other calling convention is
505 specified) matches the target C calling conventions. This calling convention
506 supports varargs function calls and tolerates some mismatch in the declared
507 prototype and implemented declaration of the function (as does normal C).
510 <dt><b>"<tt>csretcc</tt>" - The C struct return calling convention</b>:</dt>
512 <dd>This calling convention matches the target C calling conventions, except
513 that functions with this convention are required to take a pointer as their
514 first argument, and the return type of the function must be void. This is
515 used for C functions that return aggregates by-value. In this case, the
516 function has been transformed to take a pointer to the struct as the first
517 argument to the function. For targets where the ABI specifies specific
518 behavior for structure-return calls, the calling convention can be used to
519 distinguish between struct return functions and other functions that take a
520 pointer to a struct as the first argument.
523 <dt><b>"<tt>fastcc</tt>" - The fast calling convention</b>:</dt>
525 <dd>This calling convention attempts to make calls as fast as possible
526 (e.g. by passing things in registers). This calling convention allows the
527 target to use whatever tricks it wants to produce fast code for the target,
528 without having to conform to an externally specified ABI. Implementations of
529 this convention should allow arbitrary tail call optimization to be supported.
530 This calling convention does not support varargs and requires the prototype of
531 all callees to exactly match the prototype of the function definition.
534 <dt><b>"<tt>coldcc</tt>" - The cold calling convention</b>:</dt>
536 <dd>This calling convention attempts to make code in the caller as efficient
537 as possible under the assumption that the call is not commonly executed. As
538 such, these calls often preserve all registers so that the call does not break
539 any live ranges in the caller side. This calling convention does not support
540 varargs and requires the prototype of all callees to exactly match the
541 prototype of the function definition.
544 <dt><b>"<tt>cc <<em>n</em>></tt>" - Numbered convention</b>:</dt>
546 <dd>Any calling convention may be specified by number, allowing
547 target-specific calling conventions to be used. Target specific calling
548 conventions start at 64.
552 <p>More calling conventions can be added/defined on an as-needed basis, to
553 support pascal conventions or any other well-known target-independent
558 <!-- ======================================================================= -->
559 <div class="doc_subsection">
560 <a name="globalvars">Global Variables</a>
563 <div class="doc_text">
565 <p>Global variables define regions of memory allocated at compilation time
566 instead of run-time. Global variables may optionally be initialized, may have
567 an explicit section to be placed in, and may
568 have an optional explicit alignment specified. A
569 variable may be defined as a global "constant," which indicates that the
570 contents of the variable will <b>never</b> be modified (enabling better
571 optimization, allowing the global data to be placed in the read-only section of
572 an executable, etc). Note that variables that need runtime initialization
573 cannot be marked "constant" as there is a store to the variable.</p>
576 LLVM explicitly allows <em>declarations</em> of global variables to be marked
577 constant, even if the final definition of the global is not. This capability
578 can be used to enable slightly better optimization of the program, but requires
579 the language definition to guarantee that optimizations based on the
580 'constantness' are valid for the translation units that do not include the
584 <p>As SSA values, global variables define pointer values that are in
585 scope (i.e. they dominate) all basic blocks in the program. Global
586 variables always define a pointer to their "content" type because they
587 describe a region of memory, and all memory objects in LLVM are
588 accessed through pointers.</p>
590 <p>LLVM allows an explicit section to be specified for globals. If the target
591 supports it, it will emit globals to the section specified.</p>
593 <p>An explicit alignment may be specified for a global. If not present, or if
594 the alignment is set to zero, the alignment of the global is set by the target
595 to whatever it feels convenient. If an explicit alignment is specified, the
596 global is forced to have at least that much alignment. All alignments must be
602 <!-- ======================================================================= -->
603 <div class="doc_subsection">
604 <a name="functionstructure">Functions</a>
607 <div class="doc_text">
609 <p>LLVM function definitions consist of an optional <a href="#linkage">linkage
610 type</a>, an optional <a href="#callingconv">calling convention</a>, a return
611 type, a function name, a (possibly empty) argument list, an optional section,
612 an optional alignment, an opening curly brace,
613 a list of basic blocks, and a closing curly brace. LLVM function declarations
614 are defined with the "<tt>declare</tt>" keyword, an optional <a
615 href="#callingconv">calling convention</a>, a return type, a function name,
616 a possibly empty list of arguments, and an optional alignment.</p>
618 <p>A function definition contains a list of basic blocks, forming the CFG for
619 the function. Each basic block may optionally start with a label (giving the
620 basic block a symbol table entry), contains a list of instructions, and ends
621 with a <a href="#terminators">terminator</a> instruction (such as a branch or
622 function return).</p>
624 <p>The first basic block in a program is special in two ways: it is immediately
625 executed on entrance to the function, and it is not allowed to have predecessor
626 basic blocks (i.e. there can not be any branches to the entry block of a
627 function). Because the block can have no predecessors, it also cannot have any
628 <a href="#i_phi">PHI nodes</a>.</p>
630 <p>LLVM functions are identified by their name and type signature. Hence, two
631 functions with the same name but different parameter lists or return values are
632 considered different functions, and LLVM will resolve references to each
635 <p>LLVM allows an explicit section to be specified for functions. If the target
636 supports it, it will emit functions to the section specified.</p>
638 <p>An explicit alignment may be specified for a function. If not present, or if
639 the alignment is set to zero, the alignment of the function is set by the target
640 to whatever it feels convenient. If an explicit alignment is specified, the
641 function is forced to have at least that much alignment. All alignments must be
646 <!-- ======================================================================= -->
647 <div class="doc_subsection">
648 <a name="moduleasm">Module-Level Inline Assembly</a>
651 <div class="doc_text">
653 Modules may contain "module-level inline asm" blocks, which corresponds to the
654 GCC "file scope inline asm" blocks. These blocks are internally concatenated by
655 LLVM and treated as a single unit, but may be separated in the .ll file if
656 desired. The syntax is very simple:
659 <div class="doc_code"><pre>
660 module asm "inline asm code goes here"
661 module asm "more can go here"
664 <p>The strings can contain any character by escaping non-printable characters.
665 The escape sequence used is simply "\xx" where "xx" is the two digit hex code
670 The inline asm code is simply printed to the machine code .s file when
671 assembly code is generated.
676 <!-- *********************************************************************** -->
677 <div class="doc_section"> <a name="typesystem">Type System</a> </div>
678 <!-- *********************************************************************** -->
680 <div class="doc_text">
682 <p>The LLVM type system is one of the most important features of the
683 intermediate representation. Being typed enables a number of
684 optimizations to be performed on the IR directly, without having to do
685 extra analyses on the side before the transformation. A strong type
686 system makes it easier to read the generated code and enables novel
687 analyses and transformations that are not feasible to perform on normal
688 three address code representations.</p>
692 <!-- ======================================================================= -->
693 <div class="doc_subsection"> <a name="t_primitive">Primitive Types</a> </div>
694 <div class="doc_text">
695 <p>The primitive types are the fundamental building blocks of the LLVM
696 system. The current set of primitive types is as follows:</p>
698 <table class="layout">
703 <tr><th>Type</th><th>Description</th></tr>
704 <tr><td><tt>void</tt></td><td>No value</td></tr>
705 <tr><td><tt>ubyte</tt></td><td>Unsigned 8-bit value</td></tr>
706 <tr><td><tt>ushort</tt></td><td>Unsigned 16-bit value</td></tr>
707 <tr><td><tt>uint</tt></td><td>Unsigned 32-bit value</td></tr>
708 <tr><td><tt>ulong</tt></td><td>Unsigned 64-bit value</td></tr>
709 <tr><td><tt>float</tt></td><td>32-bit floating point value</td></tr>
710 <tr><td><tt>label</tt></td><td>Branch destination</td></tr>
717 <tr><th>Type</th><th>Description</th></tr>
718 <tr><td><tt>bool</tt></td><td>True or False value</td></tr>
719 <tr><td><tt>sbyte</tt></td><td>Signed 8-bit value</td></tr>
720 <tr><td><tt>short</tt></td><td>Signed 16-bit value</td></tr>
721 <tr><td><tt>int</tt></td><td>Signed 32-bit value</td></tr>
722 <tr><td><tt>long</tt></td><td>Signed 64-bit value</td></tr>
723 <tr><td><tt>double</tt></td><td>64-bit floating point value</td></tr>
731 <!-- _______________________________________________________________________ -->
732 <div class="doc_subsubsection"> <a name="t_classifications">Type
733 Classifications</a> </div>
734 <div class="doc_text">
735 <p>These different primitive types fall into a few useful
738 <table border="1" cellspacing="0" cellpadding="4">
740 <tr><th>Classification</th><th>Types</th></tr>
742 <td><a name="t_signed">signed</a></td>
743 <td><tt>sbyte, short, int, long, float, double</tt></td>
746 <td><a name="t_unsigned">unsigned</a></td>
747 <td><tt>ubyte, ushort, uint, ulong</tt></td>
750 <td><a name="t_integer">integer</a></td>
751 <td><tt>ubyte, sbyte, ushort, short, uint, int, ulong, long</tt></td>
754 <td><a name="t_integral">integral</a></td>
755 <td><tt>bool, ubyte, sbyte, ushort, short, uint, int, ulong, long</tt>
759 <td><a name="t_floating">floating point</a></td>
760 <td><tt>float, double</tt></td>
763 <td><a name="t_firstclass">first class</a></td>
764 <td><tt>bool, ubyte, sbyte, ushort, short, uint, int, ulong, long,<br>
765 float, double, <a href="#t_pointer">pointer</a>,
766 <a href="#t_packed">packed</a></tt></td>
771 <p>The <a href="#t_firstclass">first class</a> types are perhaps the
772 most important. Values of these types are the only ones which can be
773 produced by instructions, passed as arguments, or used as operands to
774 instructions. This means that all structures and arrays must be
775 manipulated either by pointer or by component.</p>
778 <!-- ======================================================================= -->
779 <div class="doc_subsection"> <a name="t_derived">Derived Types</a> </div>
781 <div class="doc_text">
783 <p>The real power in LLVM comes from the derived types in the system.
784 This is what allows a programmer to represent arrays, functions,
785 pointers, and other useful types. Note that these derived types may be
786 recursive: For example, it is possible to have a two dimensional array.</p>
790 <!-- _______________________________________________________________________ -->
791 <div class="doc_subsubsection"> <a name="t_array">Array Type</a> </div>
793 <div class="doc_text">
797 <p>The array type is a very simple derived type that arranges elements
798 sequentially in memory. The array type requires a size (number of
799 elements) and an underlying data type.</p>
804 [<# elements> x <elementtype>]
807 <p>The number of elements is a constant integer value; elementtype may
808 be any type with a size.</p>
811 <table class="layout">
814 <tt>[40 x int ]</tt><br/>
815 <tt>[41 x int ]</tt><br/>
816 <tt>[40 x uint]</tt><br/>
819 Array of 40 integer values.<br/>
820 Array of 41 integer values.<br/>
821 Array of 40 unsigned integer values.<br/>
825 <p>Here are some examples of multidimensional arrays:</p>
826 <table class="layout">
829 <tt>[3 x [4 x int]]</tt><br/>
830 <tt>[12 x [10 x float]]</tt><br/>
831 <tt>[2 x [3 x [4 x uint]]]</tt><br/>
834 3x4 array of integer values.<br/>
835 12x10 array of single precision floating point values.<br/>
836 2x3x4 array of unsigned integer values.<br/>
841 <p>Note that 'variable sized arrays' can be implemented in LLVM with a zero
842 length array. Normally, accesses past the end of an array are undefined in
843 LLVM (e.g. it is illegal to access the 5th element of a 3 element array).
844 As a special case, however, zero length arrays are recognized to be variable
845 length. This allows implementation of 'pascal style arrays' with the LLVM
846 type "{ int, [0 x float]}", for example.</p>
850 <!-- _______________________________________________________________________ -->
851 <div class="doc_subsubsection"> <a name="t_function">Function Type</a> </div>
852 <div class="doc_text">
854 <p>The function type can be thought of as a function signature. It
855 consists of a return type and a list of formal parameter types.
856 Function types are usually used to build virtual function tables
857 (which are structures of pointers to functions), for indirect function
858 calls, and when defining a function.</p>
860 The return type of a function type cannot be an aggregate type.
863 <pre> <returntype> (<parameter list>)<br></pre>
864 <p>...where '<tt><parameter list></tt>' is a comma-separated list of type
865 specifiers. Optionally, the parameter list may include a type <tt>...</tt>,
866 which indicates that the function takes a variable number of arguments.
867 Variable argument functions can access their arguments with the <a
868 href="#int_varargs">variable argument handling intrinsic</a> functions.</p>
870 <table class="layout">
873 <tt>int (int)</tt> <br/>
874 <tt>float (int, int *) *</tt><br/>
875 <tt>int (sbyte *, ...)</tt><br/>
878 function taking an <tt>int</tt>, returning an <tt>int</tt><br/>
879 <a href="#t_pointer">Pointer</a> to a function that takes an
880 <tt>int</tt> and a <a href="#t_pointer">pointer</a> to <tt>int</tt>,
881 returning <tt>float</tt>.<br/>
882 A vararg function that takes at least one <a href="#t_pointer">pointer</a>
883 to <tt>sbyte</tt> (signed char in C), which returns an integer. This is
884 the signature for <tt>printf</tt> in LLVM.<br/>
890 <!-- _______________________________________________________________________ -->
891 <div class="doc_subsubsection"> <a name="t_struct">Structure Type</a> </div>
892 <div class="doc_text">
894 <p>The structure type is used to represent a collection of data members
895 together in memory. The packing of the field types is defined to match
896 the ABI of the underlying processor. The elements of a structure may
897 be any type that has a size.</p>
898 <p>Structures are accessed using '<tt><a href="#i_load">load</a></tt>
899 and '<tt><a href="#i_store">store</a></tt>' by getting a pointer to a
900 field with the '<tt><a href="#i_getelementptr">getelementptr</a></tt>'
903 <pre> { <type list> }<br></pre>
905 <table class="layout">
908 <tt>{ int, int, int }</tt><br/>
909 <tt>{ float, int (int) * }</tt><br/>
912 a triple of three <tt>int</tt> values<br/>
913 A pair, where the first element is a <tt>float</tt> and the second element
914 is a <a href="#t_pointer">pointer</a> to a <a href="#t_function">function</a>
915 that takes an <tt>int</tt>, returning an <tt>int</tt>.<br/>
921 <!-- _______________________________________________________________________ -->
922 <div class="doc_subsubsection"> <a name="t_pointer">Pointer Type</a> </div>
923 <div class="doc_text">
925 <p>As in many languages, the pointer type represents a pointer or
926 reference to another object, which must live in memory.</p>
928 <pre> <type> *<br></pre>
930 <table class="layout">
933 <tt>[4x int]*</tt><br/>
934 <tt>int (int *) *</tt><br/>
937 A <a href="#t_pointer">pointer</a> to <a href="#t_array">array</a> of
938 four <tt>int</tt> values<br/>
939 A <a href="#t_pointer">pointer</a> to a <a
940 href="#t_function">function</a> that takes an <tt>int*</tt>, returning an
947 <!-- _______________________________________________________________________ -->
948 <div class="doc_subsubsection"> <a name="t_packed">Packed Type</a> </div>
949 <div class="doc_text">
953 <p>A packed type is a simple derived type that represents a vector
954 of elements. Packed types are used when multiple primitive data
955 are operated in parallel using a single instruction (SIMD).
956 A packed type requires a size (number of
957 elements) and an underlying primitive data type. Vectors must have a power
958 of two length (1, 2, 4, 8, 16 ...). Packed types are
959 considered <a href="#t_firstclass">first class</a>.</p>
964 < <# elements> x <elementtype> >
967 <p>The number of elements is a constant integer value; elementtype may
968 be any integral or floating point type.</p>
972 <table class="layout">
975 <tt><4 x int></tt><br/>
976 <tt><8 x float></tt><br/>
977 <tt><2 x uint></tt><br/>
980 Packed vector of 4 integer values.<br/>
981 Packed vector of 8 floating-point values.<br/>
982 Packed vector of 2 unsigned integer values.<br/>
988 <!-- _______________________________________________________________________ -->
989 <div class="doc_subsubsection"> <a name="t_opaque">Opaque Type</a> </div>
990 <div class="doc_text">
994 <p>Opaque types are used to represent unknown types in the system. This
995 corresponds (for example) to the C notion of a foward declared structure type.
996 In LLVM, opaque types can eventually be resolved to any type (not just a
1007 <table class="layout">
1013 An opaque type.<br/>
1020 <!-- *********************************************************************** -->
1021 <div class="doc_section"> <a name="constants">Constants</a> </div>
1022 <!-- *********************************************************************** -->
1024 <div class="doc_text">
1026 <p>LLVM has several different basic types of constants. This section describes
1027 them all and their syntax.</p>
1031 <!-- ======================================================================= -->
1032 <div class="doc_subsection"><a name="simpleconstants">Simple Constants</a></div>
1034 <div class="doc_text">
1037 <dt><b>Boolean constants</b></dt>
1039 <dd>The two strings '<tt>true</tt>' and '<tt>false</tt>' are both valid
1040 constants of the <tt><a href="#t_primitive">bool</a></tt> type.
1043 <dt><b>Integer constants</b></dt>
1045 <dd>Standard integers (such as '4') are constants of the <a
1046 href="#t_integer">integer</a> type. Negative numbers may be used with signed
1050 <dt><b>Floating point constants</b></dt>
1052 <dd>Floating point constants use standard decimal notation (e.g. 123.421),
1053 exponential notation (e.g. 1.23421e+2), or a more precise hexadecimal
1054 notation (see below). Floating point constants must have a <a
1055 href="#t_floating">floating point</a> type. </dd>
1057 <dt><b>Null pointer constants</b></dt>
1059 <dd>The identifier '<tt>null</tt>' is recognized as a null pointer constant
1060 and must be of <a href="#t_pointer">pointer type</a>.</dd>
1064 <p>The one non-intuitive notation for constants is the optional hexadecimal form
1065 of floating point constants. For example, the form '<tt>double
1066 0x432ff973cafa8000</tt>' is equivalent to (but harder to read than) '<tt>double
1067 4.5e+15</tt>'. The only time hexadecimal floating point constants are required
1068 (and the only time that they are generated by the disassembler) is when a
1069 floating point constant must be emitted but it cannot be represented as a
1070 decimal floating point number. For example, NaN's, infinities, and other
1071 special values are represented in their IEEE hexadecimal format so that
1072 assembly and disassembly do not cause any bits to change in the constants.</p>
1076 <!-- ======================================================================= -->
1077 <div class="doc_subsection"><a name="aggregateconstants">Aggregate Constants</a>
1080 <div class="doc_text">
1081 <p>Aggregate constants arise from aggregation of simple constants
1082 and smaller aggregate constants.</p>
1085 <dt><b>Structure constants</b></dt>
1087 <dd>Structure constants are represented with notation similar to structure
1088 type definitions (a comma separated list of elements, surrounded by braces
1089 (<tt>{}</tt>)). For example: "<tt>{ int 4, float 17.0, int* %G }</tt>",
1090 where "<tt>%G</tt>" is declared as "<tt>%G = external global int</tt>". Structure constants
1091 must have <a href="#t_struct">structure type</a>, and the number and
1092 types of elements must match those specified by the type.
1095 <dt><b>Array constants</b></dt>
1097 <dd>Array constants are represented with notation similar to array type
1098 definitions (a comma separated list of elements, surrounded by square brackets
1099 (<tt>[]</tt>)). For example: "<tt>[ int 42, int 11, int 74 ]</tt>". Array
1100 constants must have <a href="#t_array">array type</a>, and the number and
1101 types of elements must match those specified by the type.
1104 <dt><b>Packed constants</b></dt>
1106 <dd>Packed constants are represented with notation similar to packed type
1107 definitions (a comma separated list of elements, surrounded by
1108 less-than/greater-than's (<tt><></tt>)). For example: "<tt>< int 42,
1109 int 11, int 74, int 100 ></tt>". Packed constants must have <a
1110 href="#t_packed">packed type</a>, and the number and types of elements must
1111 match those specified by the type.
1114 <dt><b>Zero initialization</b></dt>
1116 <dd>The string '<tt>zeroinitializer</tt>' can be used to zero initialize a
1117 value to zero of <em>any</em> type, including scalar and aggregate types.
1118 This is often used to avoid having to print large zero initializers (e.g. for
1119 large arrays) and is always exactly equivalent to using explicit zero
1126 <!-- ======================================================================= -->
1127 <div class="doc_subsection">
1128 <a name="globalconstants">Global Variable and Function Addresses</a>
1131 <div class="doc_text">
1133 <p>The addresses of <a href="#globalvars">global variables</a> and <a
1134 href="#functionstructure">functions</a> are always implicitly valid (link-time)
1135 constants. These constants are explicitly referenced when the <a
1136 href="#identifiers">identifier for the global</a> is used and always have <a
1137 href="#t_pointer">pointer</a> type. For example, the following is a legal LLVM
1143 %Z = global [2 x int*] [ int* %X, int* %Y ]
1148 <!-- ======================================================================= -->
1149 <div class="doc_subsection"><a name="undefvalues">Undefined Values</a></div>
1150 <div class="doc_text">
1151 <p>The string '<tt>undef</tt>' is recognized as a type-less constant that has
1152 no specific value. Undefined values may be of any type and be used anywhere
1153 a constant is permitted.</p>
1155 <p>Undefined values indicate to the compiler that the program is well defined
1156 no matter what value is used, giving the compiler more freedom to optimize.
1160 <!-- ======================================================================= -->
1161 <div class="doc_subsection"><a name="constantexprs">Constant Expressions</a>
1164 <div class="doc_text">
1166 <p>Constant expressions are used to allow expressions involving other constants
1167 to be used as constants. Constant expressions may be of any <a
1168 href="#t_firstclass">first class</a> type and may involve any LLVM operation
1169 that does not have side effects (e.g. load and call are not supported). The
1170 following is the syntax for constant expressions:</p>
1173 <dt><b><tt>cast ( CST to TYPE )</tt></b></dt>
1175 <dd>Cast a constant to another type.</dd>
1177 <dt><b><tt>getelementptr ( CSTPTR, IDX0, IDX1, ... )</tt></b></dt>
1179 <dd>Perform the <a href="#i_getelementptr">getelementptr operation</a> on
1180 constants. As with the <a href="#i_getelementptr">getelementptr</a>
1181 instruction, the index list may have zero or more indexes, which are required
1182 to make sense for the type of "CSTPTR".</dd>
1184 <dt><b><tt>select ( COND, VAL1, VAL2 )</tt></b></dt>
1186 <dd>Perform the <a href="#i_select">select operation</a> on
1189 <dt><b><tt>extractelement ( VAL, IDX )</tt></b></dt>
1191 <dd>Perform the <a href="#i_extractelement">extractelement
1192 operation</a> on constants.
1194 <dt><b><tt>insertelement ( VAL, ELT, IDX )</tt></b></dt>
1196 <dd>Perform the <a href="#i_insertelement">insertelement
1197 operation</a> on constants.
1200 <dt><b><tt>shufflevector ( VEC1, VEC2, IDXMASK )</tt></b></dt>
1202 <dd>Perform the <a href="#i_shufflevector">shufflevector
1203 operation</a> on constants.
1205 <dt><b><tt>OPCODE ( LHS, RHS )</tt></b></dt>
1207 <dd>Perform the specified operation of the LHS and RHS constants. OPCODE may
1208 be any of the <a href="#binaryops">binary</a> or <a href="#bitwiseops">bitwise
1209 binary</a> operations. The constraints on operands are the same as those for
1210 the corresponding instruction (e.g. no bitwise operations on floating point
1211 values are allowed).</dd>
1215 <!-- *********************************************************************** -->
1216 <div class="doc_section"> <a name="othervalues">Other Values</a> </div>
1217 <!-- *********************************************************************** -->
1219 <!-- ======================================================================= -->
1220 <div class="doc_subsection">
1221 <a name="inlineasm">Inline Assembler Expressions</a>
1224 <div class="doc_text">
1227 LLVM supports inline assembler expressions (as opposed to <a href="#moduleasm">
1228 Module-Level Inline Assembly</a>) through the use of a special value. This
1229 value represents the inline assembler as a string (containing the instructions
1230 to emit), a list of operand constraints (stored as a string), and a flag that
1231 indicates whether or not the inline asm expression has side effects. An example
1232 inline assembler expression is:
1236 int(int) asm "bswap $0", "=r,r"
1240 Inline assembler expressions may <b>only</b> be used as the callee operand of
1241 a <a href="#i_call"><tt>call</tt> instruction</a>. Thus, typically we have:
1245 %X = call int asm "<a href="#i_bswap">bswap</a> $0", "=r,r"(int %Y)
1249 Inline asms with side effects not visible in the constraint list must be marked
1250 as having side effects. This is done through the use of the
1251 '<tt>sideeffect</tt>' keyword, like so:
1255 call void asm sideeffect "eieio", ""()
1258 <p>TODO: The format of the asm and constraints string still need to be
1259 documented here. Constraints on what can be done (e.g. duplication, moving, etc
1260 need to be documented).
1265 <!-- *********************************************************************** -->
1266 <div class="doc_section"> <a name="instref">Instruction Reference</a> </div>
1267 <!-- *********************************************************************** -->
1269 <div class="doc_text">
1271 <p>The LLVM instruction set consists of several different
1272 classifications of instructions: <a href="#terminators">terminator
1273 instructions</a>, <a href="#binaryops">binary instructions</a>,
1274 <a href="#bitwiseops">bitwise binary instructions</a>, <a
1275 href="#memoryops">memory instructions</a>, and <a href="#otherops">other
1276 instructions</a>.</p>
1280 <!-- ======================================================================= -->
1281 <div class="doc_subsection"> <a name="terminators">Terminator
1282 Instructions</a> </div>
1284 <div class="doc_text">
1286 <p>As mentioned <a href="#functionstructure">previously</a>, every
1287 basic block in a program ends with a "Terminator" instruction, which
1288 indicates which block should be executed after the current block is
1289 finished. These terminator instructions typically yield a '<tt>void</tt>'
1290 value: they produce control flow, not values (the one exception being
1291 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction).</p>
1292 <p>There are six different terminator instructions: the '<a
1293 href="#i_ret"><tt>ret</tt></a>' instruction, the '<a href="#i_br"><tt>br</tt></a>'
1294 instruction, the '<a href="#i_switch"><tt>switch</tt></a>' instruction,
1295 the '<a href="#i_invoke"><tt>invoke</tt></a>' instruction, the '<a
1296 href="#i_unwind"><tt>unwind</tt></a>' instruction, and the '<a
1297 href="#i_unreachable"><tt>unreachable</tt></a>' instruction.</p>
1301 <!-- _______________________________________________________________________ -->
1302 <div class="doc_subsubsection"> <a name="i_ret">'<tt>ret</tt>'
1303 Instruction</a> </div>
1304 <div class="doc_text">
1306 <pre> ret <type> <value> <i>; Return a value from a non-void function</i>
1307 ret void <i>; Return from void function</i>
1310 <p>The '<tt>ret</tt>' instruction is used to return control flow (and a
1311 value) from a function back to the caller.</p>
1312 <p>There are two forms of the '<tt>ret</tt>' instruction: one that
1313 returns a value and then causes control flow, and one that just causes
1314 control flow to occur.</p>
1316 <p>The '<tt>ret</tt>' instruction may return any '<a
1317 href="#t_firstclass">first class</a>' type. Notice that a function is
1318 not <a href="#wellformed">well formed</a> if there exists a '<tt>ret</tt>'
1319 instruction inside of the function that returns a value that does not
1320 match the return type of the function.</p>
1322 <p>When the '<tt>ret</tt>' instruction is executed, control flow
1323 returns back to the calling function's context. If the caller is a "<a
1324 href="#i_call"><tt>call</tt></a>" instruction, execution continues at
1325 the instruction after the call. If the caller was an "<a
1326 href="#i_invoke"><tt>invoke</tt></a>" instruction, execution continues
1327 at the beginning of the "normal" destination block. If the instruction
1328 returns a value, that value shall set the call or invoke instruction's
1331 <pre> ret int 5 <i>; Return an integer value of 5</i>
1332 ret void <i>; Return from a void function</i>
1335 <!-- _______________________________________________________________________ -->
1336 <div class="doc_subsubsection"> <a name="i_br">'<tt>br</tt>' Instruction</a> </div>
1337 <div class="doc_text">
1339 <pre> br bool <cond>, label <iftrue>, label <iffalse><br> br label <dest> <i>; Unconditional branch</i>
1342 <p>The '<tt>br</tt>' instruction is used to cause control flow to
1343 transfer to a different basic block in the current function. There are
1344 two forms of this instruction, corresponding to a conditional branch
1345 and an unconditional branch.</p>
1347 <p>The conditional branch form of the '<tt>br</tt>' instruction takes a
1348 single '<tt>bool</tt>' value and two '<tt>label</tt>' values. The
1349 unconditional form of the '<tt>br</tt>' instruction takes a single '<tt>label</tt>'
1350 value as a target.</p>
1352 <p>Upon execution of a conditional '<tt>br</tt>' instruction, the '<tt>bool</tt>'
1353 argument is evaluated. If the value is <tt>true</tt>, control flows
1354 to the '<tt>iftrue</tt>' <tt>label</tt> argument. If "cond" is <tt>false</tt>,
1355 control flows to the '<tt>iffalse</tt>' <tt>label</tt> argument.</p>
1357 <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
1358 href="#i_ret">ret</a> int 1<br>IfUnequal:<br> <a href="#i_ret">ret</a> int 0<br></pre>
1360 <!-- _______________________________________________________________________ -->
1361 <div class="doc_subsubsection">
1362 <a name="i_switch">'<tt>switch</tt>' Instruction</a>
1365 <div class="doc_text">
1369 switch <intty> <value>, label <defaultdest> [ <intty> <val>, label <dest> ... ]
1374 <p>The '<tt>switch</tt>' instruction is used to transfer control flow to one of
1375 several different places. It is a generalization of the '<tt>br</tt>'
1376 instruction, allowing a branch to occur to one of many possible
1382 <p>The '<tt>switch</tt>' instruction uses three parameters: an integer
1383 comparison value '<tt>value</tt>', a default '<tt>label</tt>' destination, and
1384 an array of pairs of comparison value constants and '<tt>label</tt>'s. The
1385 table is not allowed to contain duplicate constant entries.</p>
1389 <p>The <tt>switch</tt> instruction specifies a table of values and
1390 destinations. When the '<tt>switch</tt>' instruction is executed, this
1391 table is searched for the given value. If the value is found, control flow is
1392 transfered to the corresponding destination; otherwise, control flow is
1393 transfered to the default destination.</p>
1395 <h5>Implementation:</h5>
1397 <p>Depending on properties of the target machine and the particular
1398 <tt>switch</tt> instruction, this instruction may be code generated in different
1399 ways. For example, it could be generated as a series of chained conditional
1400 branches or with a lookup table.</p>
1405 <i>; Emulate a conditional br instruction</i>
1406 %Val = <a href="#i_cast">cast</a> bool %value to int
1407 switch int %Val, label %truedest [int 0, label %falsedest ]
1409 <i>; Emulate an unconditional br instruction</i>
1410 switch uint 0, label %dest [ ]
1412 <i>; Implement a jump table:</i>
1413 switch uint %val, label %otherwise [ uint 0, label %onzero
1414 uint 1, label %onone
1415 uint 2, label %ontwo ]
1419 <!-- _______________________________________________________________________ -->
1420 <div class="doc_subsubsection">
1421 <a name="i_invoke">'<tt>invoke</tt>' Instruction</a>
1424 <div class="doc_text">
1429 <result> = invoke [<a href="#callingconv">cconv</a>] <ptr to function ty> %<function ptr val>(<function args>)
1430 to label <normal label> unwind label <exception label>
1435 <p>The '<tt>invoke</tt>' instruction causes control to transfer to a specified
1436 function, with the possibility of control flow transfer to either the
1437 '<tt>normal</tt>' label or the
1438 '<tt>exception</tt>' label. If the callee function returns with the
1439 "<tt><a href="#i_ret">ret</a></tt>" instruction, control flow will return to the
1440 "normal" label. If the callee (or any indirect callees) returns with the "<a
1441 href="#i_unwind"><tt>unwind</tt></a>" instruction, control is interrupted and
1442 continued at the dynamically nearest "exception" label.</p>
1446 <p>This instruction requires several arguments:</p>
1450 The optional "cconv" marker indicates which <a href="callingconv">calling
1451 convention</a> the call should use. If none is specified, the call defaults
1452 to using C calling conventions.
1454 <li>'<tt>ptr to function ty</tt>': shall be the signature of the pointer to
1455 function value being invoked. In most cases, this is a direct function
1456 invocation, but indirect <tt>invoke</tt>s are just as possible, branching off
1457 an arbitrary pointer to function value.
1460 <li>'<tt>function ptr val</tt>': An LLVM value containing a pointer to a
1461 function to be invoked. </li>
1463 <li>'<tt>function args</tt>': argument list whose types match the function
1464 signature argument types. If the function signature indicates the function
1465 accepts a variable number of arguments, the extra arguments can be
1468 <li>'<tt>normal label</tt>': the label reached when the called function
1469 executes a '<tt><a href="#i_ret">ret</a></tt>' instruction. </li>
1471 <li>'<tt>exception label</tt>': the label reached when a callee returns with
1472 the <a href="#i_unwind"><tt>unwind</tt></a> instruction. </li>
1478 <p>This instruction is designed to operate as a standard '<tt><a
1479 href="#i_call">call</a></tt>' instruction in most regards. The primary
1480 difference is that it establishes an association with a label, which is used by
1481 the runtime library to unwind the stack.</p>
1483 <p>This instruction is used in languages with destructors to ensure that proper
1484 cleanup is performed in the case of either a <tt>longjmp</tt> or a thrown
1485 exception. Additionally, this is important for implementation of
1486 '<tt>catch</tt>' clauses in high-level languages that support them.</p>
1490 %retval = invoke int %Test(int 15) to label %Continue
1491 unwind label %TestCleanup <i>; {int}:retval set</i>
1492 %retval = invoke <a href="#callingconv">coldcc</a> int %Test(int 15) to label %Continue
1493 unwind label %TestCleanup <i>; {int}:retval set</i>
1498 <!-- _______________________________________________________________________ -->
1500 <div class="doc_subsubsection"> <a name="i_unwind">'<tt>unwind</tt>'
1501 Instruction</a> </div>
1503 <div class="doc_text">
1512 <p>The '<tt>unwind</tt>' instruction unwinds the stack, continuing control flow
1513 at the first callee in the dynamic call stack which used an <a
1514 href="#i_invoke"><tt>invoke</tt></a> instruction to perform the call. This is
1515 primarily used to implement exception handling.</p>
1519 <p>The '<tt>unwind</tt>' intrinsic causes execution of the current function to
1520 immediately halt. The dynamic call stack is then searched for the first <a
1521 href="#i_invoke"><tt>invoke</tt></a> instruction on the call stack. Once found,
1522 execution continues at the "exceptional" destination block specified by the
1523 <tt>invoke</tt> instruction. If there is no <tt>invoke</tt> instruction in the
1524 dynamic call chain, undefined behavior results.</p>
1527 <!-- _______________________________________________________________________ -->
1529 <div class="doc_subsubsection"> <a name="i_unreachable">'<tt>unreachable</tt>'
1530 Instruction</a> </div>
1532 <div class="doc_text">
1541 <p>The '<tt>unreachable</tt>' instruction has no defined semantics. This
1542 instruction is used to inform the optimizer that a particular portion of the
1543 code is not reachable. This can be used to indicate that the code after a
1544 no-return function cannot be reached, and other facts.</p>
1548 <p>The '<tt>unreachable</tt>' instruction has no defined semantics.</p>
1553 <!-- ======================================================================= -->
1554 <div class="doc_subsection"> <a name="binaryops">Binary Operations</a> </div>
1555 <div class="doc_text">
1556 <p>Binary operators are used to do most of the computation in a
1557 program. They require two operands, execute an operation on them, and
1558 produce a single value. The operands might represent
1559 multiple data, as is the case with the <a href="#t_packed">packed</a> data type.
1560 The result value of a binary operator is not
1561 necessarily the same type as its operands.</p>
1562 <p>There are several different binary operators:</p>
1564 <!-- _______________________________________________________________________ -->
1565 <div class="doc_subsubsection"> <a name="i_add">'<tt>add</tt>'
1566 Instruction</a> </div>
1567 <div class="doc_text">
1569 <pre> <result> = add <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1572 <p>The '<tt>add</tt>' instruction returns the sum of its two operands.</p>
1574 <p>The two arguments to the '<tt>add</tt>' instruction must be either <a
1575 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a> values.
1576 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1577 Both arguments must have identical types.</p>
1579 <p>The value produced is the integer or floating point sum of the two
1582 <pre> <result> = add int 4, %var <i>; yields {int}:result = 4 + %var</i>
1585 <!-- _______________________________________________________________________ -->
1586 <div class="doc_subsubsection"> <a name="i_sub">'<tt>sub</tt>'
1587 Instruction</a> </div>
1588 <div class="doc_text">
1590 <pre> <result> = sub <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1593 <p>The '<tt>sub</tt>' instruction returns the difference of its two
1595 <p>Note that the '<tt>sub</tt>' instruction is used to represent the '<tt>neg</tt>'
1596 instruction present in most other intermediate representations.</p>
1598 <p>The two arguments to the '<tt>sub</tt>' instruction must be either <a
1599 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1601 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1602 Both arguments must have identical types.</p>
1604 <p>The value produced is the integer or floating point difference of
1605 the two operands.</p>
1607 <pre> <result> = sub int 4, %var <i>; yields {int}:result = 4 - %var</i>
1608 <result> = sub int 0, %val <i>; yields {int}:result = -%var</i>
1611 <!-- _______________________________________________________________________ -->
1612 <div class="doc_subsubsection"> <a name="i_mul">'<tt>mul</tt>'
1613 Instruction</a> </div>
1614 <div class="doc_text">
1616 <pre> <result> = mul <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1619 <p>The '<tt>mul</tt>' instruction returns the product of its two
1622 <p>The two arguments to the '<tt>mul</tt>' instruction must be either <a
1623 href="#t_integer">integer</a> or <a href="#t_floating">floating point</a>
1625 This instruction can also take <a href="#t_packed">packed</a> versions of the values.
1626 Both arguments must have identical types.</p>
1628 <p>The value produced is the integer or floating point product of the
1630 <p>There is no signed vs unsigned multiplication. The appropriate
1631 action is taken based on the type of the operand.</p>
1633 <pre> <result> = mul int 4, %var <i>; yields {int}:result = 4 * %var</i>
1636 <!-- _______________________________________________________________________ -->
1637 <div class="doc_subsubsection"> <a name="i_udiv">'<tt>udiv</tt>' Instruction
1639 <div class="doc_text">
1641 <pre> <result> = udiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1644 <p>The '<tt>udiv</tt>' instruction returns the quotient of its two
1647 <p>The two arguments to the '<tt>udiv</tt>' instruction must be
1648 <a href="#t_integer">integer</a> values. Both arguments must have identical
1649 types. This instruction can also take <a href="#t_packed">packed</a> versions
1650 of the values in which case the elements must be integers.</p>
1652 <p>The value produced is the unsigned integer quotient of the two operands. This
1653 instruction always performs an unsigned division operation, regardless of
1654 whether the arguments are unsigned or not.</p>
1656 <pre> <result> = udiv uint 4, %var <i>; yields {uint}:result = 4 / %var</i>
1659 <!-- _______________________________________________________________________ -->
1660 <div class="doc_subsubsection"> <a name="i_sdiv">'<tt>sdiv</tt>' Instruction
1662 <div class="doc_text">
1664 <pre> <result> = sdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1667 <p>The '<tt>sdiv</tt>' instruction returns the quotient of its two
1670 <p>The two arguments to the '<tt>sdiv</tt>' instruction must be
1671 <a href="#t_integer">integer</a> values. Both arguments must have identical
1672 types. This instruction can also take <a href="#t_packed">packed</a> versions
1673 of the values in which case the elements must be integers.</p>
1675 <p>The value produced is the signed integer quotient of the two operands. This
1676 instruction always performs a signed division operation, regardless of whether
1677 the arguments are signed or not.</p>
1679 <pre> <result> = sdiv int 4, %var <i>; yields {int}:result = 4 / %var</i>
1682 <!-- _______________________________________________________________________ -->
1683 <div class="doc_subsubsection"> <a name="i_fdiv">'<tt>fdiv</tt>'
1684 Instruction</a> </div>
1685 <div class="doc_text">
1687 <pre> <result> = fdiv <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1690 <p>The '<tt>fdiv</tt>' instruction returns the quotient of its two
1693 <p>The two arguments to the '<tt>div</tt>' instruction must be
1694 <a href="#t_floating">floating point</a> values. Both arguments must have
1695 identical types. This instruction can also take <a href="#t_packed">packed</a>
1696 versions of the values in which case the elements must be floating point.</p>
1698 <p>The value produced is the floating point quotient of the two operands.</p>
1700 <pre> <result> = fdiv float 4.0, %var <i>; yields {float}:result = 4.0 / %var</i>
1703 <!-- _______________________________________________________________________ -->
1704 <div class="doc_subsubsection"> <a name="i_urem">'<tt>urem</tt>' Instruction</a>
1706 <div class="doc_text">
1708 <pre> <result> = urem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1711 <p>The '<tt>urem</tt>' instruction returns the remainder from the
1712 unsigned division of its two arguments.</p>
1714 <p>The two arguments to the '<tt>urem</tt>' instruction must be
1715 <a href="#t_integer">integer</a> values. Both arguments must have identical
1718 <p>This instruction returns the unsigned integer <i>remainder</i> of a division.
1719 This instruction always performs an unsigned division to get the remainder,
1720 regardless of whether the arguments are unsigned or not.</p>
1722 <pre> <result> = urem uint 4, %var <i>; yields {uint}:result = 4 % %var</i>
1726 <!-- _______________________________________________________________________ -->
1727 <div class="doc_subsubsection"> <a name="i_srem">'<tt>srem</tt>'
1728 Instruction</a> </div>
1729 <div class="doc_text">
1731 <pre> <result> = srem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1734 <p>The '<tt>srem</tt>' instruction returns the remainder from the
1735 signed division of its two operands.</p>
1737 <p>The two arguments to the '<tt>srem</tt>' instruction must be
1738 <a href="#t_integer">integer</a> values. Both arguments must have identical
1741 <p>This instruction returns the <i>remainder</i> of a division (where the result
1742 has the same sign as the divisor), not the <i>modulus</i> (where the
1743 result has the same sign as the dividend) of a value. For more
1744 information about the difference, see <a
1745 href="http://mathforum.org/dr.math/problems/anne.4.28.99.html">The
1748 <pre> <result> = srem int 4, %var <i>; yields {int}:result = 4 % %var</i>
1752 <!-- _______________________________________________________________________ -->
1753 <div class="doc_subsubsection"> <a name="i_frem">'<tt>frem</tt>'
1754 Instruction</a> </div>
1755 <div class="doc_text">
1757 <pre> <result> = frem <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1760 <p>The '<tt>frem</tt>' instruction returns the remainder from the
1761 division of its two operands.</p>
1763 <p>The two arguments to the '<tt>frem</tt>' instruction must be
1764 <a href="#t_floating">floating point</a> values. Both arguments must have
1765 identical types.</p>
1767 <p>This instruction returns the <i>remainder</i> of a division.</p>
1769 <pre> <result> = frem float 4.0, %var <i>; yields {float}:result = 4.0 % %var</i>
1773 <!-- _______________________________________________________________________ -->
1774 <div class="doc_subsubsection"> <a name="i_setcc">'<tt>set<i>cc</i></tt>'
1775 Instructions</a> </div>
1776 <div class="doc_text">
1778 <pre> <result> = seteq <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1779 <result> = setne <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1780 <result> = setlt <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1781 <result> = setgt <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1782 <result> = setle <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1783 <result> = setge <ty> <var1>, <var2> <i>; yields {bool}:result</i>
1786 <p>The '<tt>set<i>cc</i></tt>' family of instructions returns a boolean
1787 value based on a comparison of their two operands.</p>
1789 <p>The two arguments to the '<tt>set<i>cc</i></tt>' instructions must
1790 be of <a href="#t_firstclass">first class</a> type (it is not possible
1791 to compare '<tt>label</tt>'s, '<tt>array</tt>'s, '<tt>structure</tt>'
1792 or '<tt>void</tt>' values, etc...). Both arguments must have identical
1795 <p>The '<tt>seteq</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1796 value if both operands are equal.<br>
1797 The '<tt>setne</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1798 value if both operands are unequal.<br>
1799 The '<tt>setlt</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1800 value if the first operand is less than the second operand.<br>
1801 The '<tt>setgt</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1802 value if the first operand is greater than the second operand.<br>
1803 The '<tt>setle</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1804 value if the first operand is less than or equal to the second operand.<br>
1805 The '<tt>setge</tt>' instruction yields a <tt>true</tt> '<tt>bool</tt>'
1806 value if the first operand is greater than or equal to the second
1809 <pre> <result> = seteq int 4, 5 <i>; yields {bool}:result = false</i>
1810 <result> = setne float 4, 5 <i>; yields {bool}:result = true</i>
1811 <result> = setlt uint 4, 5 <i>; yields {bool}:result = true</i>
1812 <result> = setgt sbyte 4, 5 <i>; yields {bool}:result = false</i>
1813 <result> = setle sbyte 4, 5 <i>; yields {bool}:result = true</i>
1814 <result> = setge sbyte 4, 5 <i>; yields {bool}:result = false</i>
1818 <!-- ======================================================================= -->
1819 <div class="doc_subsection"> <a name="bitwiseops">Bitwise Binary
1820 Operations</a> </div>
1821 <div class="doc_text">
1822 <p>Bitwise binary operators are used to do various forms of
1823 bit-twiddling in a program. They are generally very efficient
1824 instructions and can commonly be strength reduced from other
1825 instructions. They require two operands, execute an operation on them,
1826 and produce a single value. The resulting value of the bitwise binary
1827 operators is always the same type as its first operand.</p>
1829 <!-- _______________________________________________________________________ -->
1830 <div class="doc_subsubsection"> <a name="i_and">'<tt>and</tt>'
1831 Instruction</a> </div>
1832 <div class="doc_text">
1834 <pre> <result> = and <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1837 <p>The '<tt>and</tt>' instruction returns the bitwise logical and of
1838 its two operands.</p>
1840 <p>The two arguments to the '<tt>and</tt>' instruction must be <a
1841 href="#t_integral">integral</a> values. Both arguments must have
1842 identical types.</p>
1844 <p>The truth table used for the '<tt>and</tt>' instruction is:</p>
1846 <div style="align: center">
1847 <table border="1" cellspacing="0" cellpadding="4">
1878 <pre> <result> = and int 4, %var <i>; yields {int}:result = 4 & %var</i>
1879 <result> = and int 15, 40 <i>; yields {int}:result = 8</i>
1880 <result> = and int 4, 8 <i>; yields {int}:result = 0</i>
1883 <!-- _______________________________________________________________________ -->
1884 <div class="doc_subsubsection"> <a name="i_or">'<tt>or</tt>' Instruction</a> </div>
1885 <div class="doc_text">
1887 <pre> <result> = or <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1890 <p>The '<tt>or</tt>' instruction returns the bitwise logical inclusive
1891 or of its two operands.</p>
1893 <p>The two arguments to the '<tt>or</tt>' instruction must be <a
1894 href="#t_integral">integral</a> values. Both arguments must have
1895 identical types.</p>
1897 <p>The truth table used for the '<tt>or</tt>' instruction is:</p>
1899 <div style="align: center">
1900 <table border="1" cellspacing="0" cellpadding="4">
1931 <pre> <result> = or int 4, %var <i>; yields {int}:result = 4 | %var</i>
1932 <result> = or int 15, 40 <i>; yields {int}:result = 47</i>
1933 <result> = or int 4, 8 <i>; yields {int}:result = 12</i>
1936 <!-- _______________________________________________________________________ -->
1937 <div class="doc_subsubsection"> <a name="i_xor">'<tt>xor</tt>'
1938 Instruction</a> </div>
1939 <div class="doc_text">
1941 <pre> <result> = xor <ty> <var1>, <var2> <i>; yields {ty}:result</i>
1944 <p>The '<tt>xor</tt>' instruction returns the bitwise logical exclusive
1945 or of its two operands. The <tt>xor</tt> is used to implement the
1946 "one's complement" operation, which is the "~" operator in C.</p>
1948 <p>The two arguments to the '<tt>xor</tt>' instruction must be <a
1949 href="#t_integral">integral</a> values. Both arguments must have
1950 identical types.</p>
1952 <p>The truth table used for the '<tt>xor</tt>' instruction is:</p>
1954 <div style="align: center">
1955 <table border="1" cellspacing="0" cellpadding="4">
1987 <pre> <result> = xor int 4, %var <i>; yields {int}:result = 4 ^ %var</i>
1988 <result> = xor int 15, 40 <i>; yields {int}:result = 39</i>
1989 <result> = xor int 4, 8 <i>; yields {int}:result = 12</i>
1990 <result> = xor int %V, -1 <i>; yields {int}:result = ~%V</i>
1993 <!-- _______________________________________________________________________ -->
1994 <div class="doc_subsubsection"> <a name="i_shl">'<tt>shl</tt>'
1995 Instruction</a> </div>
1996 <div class="doc_text">
1998 <pre> <result> = shl <ty> <var1>, ubyte <var2> <i>; yields {ty}:result</i>
2001 <p>The '<tt>shl</tt>' instruction returns the first operand shifted to
2002 the left a specified number of bits.</p>
2004 <p>The first argument to the '<tt>shl</tt>' instruction must be an <a
2005 href="#t_integer">integer</a> type. The second argument must be an '<tt>ubyte</tt>'
2008 <p>The value produced is <tt>var1</tt> * 2<sup><tt>var2</tt></sup>.</p>
2010 <pre> <result> = shl int 4, ubyte %var <i>; yields {int}:result = 4 << %var</i>
2011 <result> = shl int 4, ubyte 2 <i>; yields {int}:result = 16</i>
2012 <result> = shl int 1, ubyte 10 <i>; yields {int}:result = 1024</i>
2015 <!-- _______________________________________________________________________ -->
2016 <div class="doc_subsubsection"> <a name="i_shr">'<tt>shr</tt>'
2017 Instruction</a> </div>
2018 <div class="doc_text">
2020 <pre> <result> = shr <ty> <var1>, ubyte <var2> <i>; yields {ty}:result</i>
2023 <p>The '<tt>shr</tt>' instruction returns the first operand shifted to
2024 the right a specified number of bits.</p>
2026 <p>The first argument to the '<tt>shr</tt>' instruction must be an <a
2027 href="#t_integer">integer</a> type. The second argument must be an '<tt>ubyte</tt>'
2030 <p>If the first argument is a <a href="#t_signed">signed</a> type, the
2031 most significant bit is duplicated in the newly free'd bit positions.
2032 If the first argument is unsigned, zero bits shall fill the empty
2035 <pre> <result> = shr int 4, ubyte %var <i>; yields {int}:result = 4 >> %var</i>
2036 <result> = shr uint 4, ubyte 1 <i>; yields {uint}:result = 2</i>
2037 <result> = shr int 4, ubyte 2 <i>; yields {int}:result = 1</i>
2038 <result> = shr sbyte 4, ubyte 3 <i>; yields {sbyte}:result = 0</i>
2039 <result> = shr sbyte -2, ubyte 1 <i>; yields {sbyte}:result = -1</i>
2043 <!-- ======================================================================= -->
2044 <div class="doc_subsection">
2045 <a name="vectorops">Vector Operations</a>
2048 <div class="doc_text">
2050 <p>LLVM supports several instructions to represent vector operations in a
2051 target-independent manner. This instructions cover the element-access and
2052 vector-specific operations needed to process vectors effectively. While LLVM
2053 does directly support these vector operations, many sophisticated algorithms
2054 will want to use target-specific intrinsics to take full advantage of a specific
2059 <!-- _______________________________________________________________________ -->
2060 <div class="doc_subsubsection">
2061 <a name="i_extractelement">'<tt>extractelement</tt>' Instruction</a>
2064 <div class="doc_text">
2069 <result> = extractelement <n x <ty>> <val>, uint <idx> <i>; yields <ty></i>
2075 The '<tt>extractelement</tt>' instruction extracts a single scalar
2076 element from a packed vector at a specified index.
2083 The first operand of an '<tt>extractelement</tt>' instruction is a
2084 value of <a href="#t_packed">packed</a> type. The second operand is
2085 an index indicating the position from which to extract the element.
2086 The index may be a variable.</p>
2091 The result is a scalar of the same type as the element type of
2092 <tt>val</tt>. Its value is the value at position <tt>idx</tt> of
2093 <tt>val</tt>. If <tt>idx</tt> exceeds the length of <tt>val</tt>, the
2094 results are undefined.
2100 %result = extractelement <4 x int> %vec, uint 0 <i>; yields int</i>
2105 <!-- _______________________________________________________________________ -->
2106 <div class="doc_subsubsection">
2107 <a name="i_insertelement">'<tt>insertelement</tt>' Instruction</a>
2110 <div class="doc_text">
2115 <result> = insertelement <n x <ty>> <val>, <ty> <elt>, uint <idx> <i>; yields <n x <ty>></i>
2121 The '<tt>insertelement</tt>' instruction inserts a scalar
2122 element into a packed vector at a specified index.
2129 The first operand of an '<tt>insertelement</tt>' instruction is a
2130 value of <a href="#t_packed">packed</a> type. The second operand is a
2131 scalar value whose type must equal the element type of the first
2132 operand. The third operand is an index indicating the position at
2133 which to insert the value. The index may be a variable.</p>
2138 The result is a packed vector of the same type as <tt>val</tt>. Its
2139 element values are those of <tt>val</tt> except at position
2140 <tt>idx</tt>, where it gets the value <tt>elt</tt>. If <tt>idx</tt>
2141 exceeds the length of <tt>val</tt>, the results are undefined.
2147 %result = insertelement <4 x int> %vec, int 1, uint 0 <i>; yields <4 x int></i>
2151 <!-- _______________________________________________________________________ -->
2152 <div class="doc_subsubsection">
2153 <a name="i_shufflevector">'<tt>shufflevector</tt>' Instruction</a>
2156 <div class="doc_text">
2161 <result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <n x uint> <mask> <i>; yields <n x <ty>></i>
2167 The '<tt>shufflevector</tt>' instruction constructs a permutation of elements
2168 from two input vectors, returning a vector of the same type.
2174 The first two operands of a '<tt>shufflevector</tt>' instruction are vectors
2175 with types that match each other and types that match the result of the
2176 instruction. The third argument is a shuffle mask, which has the same number
2177 of elements as the other vector type, but whose element type is always 'uint'.
2181 The shuffle mask operand is required to be a constant vector with either
2182 constant integer or undef values.
2188 The elements of the two input vectors are numbered from left to right across
2189 both of the vectors. The shuffle mask operand specifies, for each element of
2190 the result vector, which element of the two input registers the result element
2191 gets. The element selector may be undef (meaning "don't care") and the second
2192 operand may be undef if performing a shuffle from only one vector.
2198 %result = shufflevector <4 x int> %v1, <4 x int> %v2,
2199 <4 x uint> <uint 0, uint 4, uint 1, uint 5> <i>; yields <4 x int></i>
2200 %result = shufflevector <4 x int> %v1, <4 x int> undef,
2201 <4 x uint> <uint 0, uint 1, uint 2, uint 3> <i>; yields <4 x int></i> - Identity shuffle.
2206 <!-- _______________________________________________________________________ -->
2207 <div class="doc_subsubsection"> <a name="i_vsetint">'<tt>vsetint</tt>'
2208 Instruction</a> </div>
2209 <div class="doc_text">
2211 <pre><result> = vsetint <op>, <n x <ty>> <var1>, <var2> <i>; yields <n x bool></i>
2216 <p>The '<tt>vsetint</tt>' instruction takes two integer vectors and
2217 returns a vector of boolean values representing, at each position, the
2218 result of the comparison between the values at that position in the
2223 <p>The arguments to a '<tt>vsetint</tt>' instruction are a comparison
2224 operation and two value arguments. The value arguments must be of <a
2225 href="#t_integral">integral</a> <a href="#t_packed">packed</a> type,
2226 and they must have identical types. The operation argument must be
2227 one of <tt>eq</tt>, <tt>ne</tt>, <tt>slt</tt>, <tt>sgt</tt>,
2228 <tt>sle</tt>, <tt>sge</tt>, <tt>ult</tt>, <tt>ugt</tt>, <tt>ule</tt>,
2229 <tt>uge</tt>, <tt>true</tt>, and <tt>false</tt>. The result is a
2230 packed <tt>bool</tt> value with the same length as each operand.</p>
2234 <p>The following table shows the semantics of '<tt>vsetint</tt>'. For
2235 each position of the result, the comparison is done on the
2236 corresponding positions of the two value arguments. Note that the
2237 signedness of the comparison depends on the comparison opcode and
2238 <i>not</i> on the signedness of the value operands. E.g., <tt>vsetint
2239 slt <4 x unsigned> %x, %y</tt> does an elementwise <i>signed</i>
2240 comparison of <tt>%x</tt> and <tt>%y</tt>.</p>
2242 <table border="1" cellspacing="0" cellpadding="4">
2244 <tr><th>Operation</th><th>Result is true iff</th><th>Comparison is</th></tr>
2245 <tr><td><tt>eq</tt></td><td>var1 == var2</td><td>--</td></tr>
2246 <tr><td><tt>ne</tt></td><td>var1 != var2</td><td>--</td></tr>
2247 <tr><td><tt>slt</tt></td><td>var1 < var2</td><td>signed</td></tr>
2248 <tr><td><tt>sgt</tt></td><td>var1 > var2</td><td>signed</td></tr>
2249 <tr><td><tt>sle</tt></td><td>var1 <= var2</td><td>signed</td></tr>
2250 <tr><td><tt>sge</tt></td><td>var1 >= var2</td><td>signed</td></tr>
2251 <tr><td><tt>ult</tt></td><td>var1 < var2</td><td>unsigned</td></tr>
2252 <tr><td><tt>ugt</tt></td><td>var1 > var2</td><td>unsigned</td></tr>
2253 <tr><td><tt>ule</tt></td><td>var1 <= var2</td><td>unsigned</td></tr>
2254 <tr><td><tt>uge</tt></td><td>var1 >= var2</td><td>unsigned</td></tr>
2255 <tr><td><tt>true</tt></td><td>always</td><td>--</td></tr>
2256 <tr><td><tt>false</tt></td><td>never</td><td>--</td></tr>
2261 <pre> <result> = vsetint eq <2 x int> <int 0, int 1>, <int 1, int 0> <i>; yields {<2 x bool>}:result = false, false</i>
2262 <result> = vsetint ne <2 x int> <int 0, int 1>, <int 1, int 0> <i>; yields {<2 x bool>}:result = true, true</i>
2263 <result> = vsetint slt <2 x int> <int 0, int 1>, <int 1, int 0> <i>; yields {<2 x bool>}:result = true, false</i>
2264 <result> = vsetint sgt <2 x int> <int 0, int 1>, <int 1, int 0> <i>; yields {<2 x bool>}:result = false, true</i>
2265 <result> = vsetint sle <2 x int> <int 0, int 1>, <int 1, int 0> <i>; yields {<2 x bool>}:result = true, false</i>
2266 <result> = vsetint sge <2 x int> <int 0, int 1>, <int 1, int 0> <i>; yields {<2 x bool>}:result = false, true</i>
2270 <!-- _______________________________________________________________________ -->
2271 <div class="doc_subsubsection"> <a name="i_vsetfp">'<tt>vsetfp</tt>'
2272 Instruction</a> </div>
2273 <div class="doc_text">
2275 <pre><result> = vsetfp <op>, <n x <ty>> <var1>, <var2> <i>; yields <n x bool></i>
2280 <p>The '<tt>vsetfp</tt>' instruction takes two floating point vector
2281 arguments and returns a vector of boolean values representing, at each
2282 position, the result of the comparison between the values at that
2283 position in the two operands.</p>
2287 <p>The arguments to a '<tt>vsetfp</tt>' instruction are a comparison
2288 operation and two value arguments. The value arguments must be of <a
2289 href="t_floating">floating point</a> <a href="#t_packed">packed</a>
2290 type, and they must have identical types. The operation argument must
2291 be one of <tt>eq</tt>, <tt>ne</tt>, <tt>lt</tt>, <tt>gt</tt>,
2292 <tt>le</tt>, <tt>ge</tt>, <tt>oeq</tt>, <tt>one</tt>, <tt>olt</tt>,
2293 <tt>ogt</tt>, <tt>ole</tt>, <tt>oge</tt>, <tt>ueq</tt>, <tt>une</tt>,
2294 <tt>ult</tt>, <tt>ugt</tt>, <tt>ule</tt>, <tt>uge</tt>, <tt>o</tt>,
2295 <tt>u</tt>, <tt>true</tt>, and <tt>false</tt>. The result is a packed
2296 <tt>bool</tt> value with the same length as each operand.</p>
2300 <p>The following table shows the semantics of '<tt>vsetfp</tt>' for
2301 floating point types. If either operand is a floating point Not a
2302 Number (NaN) value, the operation is unordered, and the value in the
2303 first column below is produced at that position. Otherwise, the
2304 operation is ordered, and the value in the second column is
2307 <table border="1" cellspacing="0" cellpadding="4">
2309 <tr><th>Operation</th><th>If unordered<th>Otherwise true iff</th></tr>
2310 <tr><td><tt>eq</tt></td><td>undefined</td><td>var1 == var2</td></tr>
2311 <tr><td><tt>ne</tt></td><td>undefined</td><td>var1 != var2</td></tr>
2312 <tr><td><tt>lt</tt></td><td>undefined</td><td>var1 < var2</td></tr>
2313 <tr><td><tt>gt</tt></td><td>undefined</td><td>var1 > var2</td></tr>
2314 <tr><td><tt>le</tt></td><td>undefined</td><td>var1 <= var2</td></tr>
2315 <tr><td><tt>ge</tt></td><td>undefined</td><td>var1 >= var2</td></tr>
2316 <tr><td><tt>oeq</tt></td><td>false</td><td>var1 == var2</td></tr>
2317 <tr><td><tt>one</tt></td><td>false</td><td>var1 != var2</td></tr>
2318 <tr><td><tt>olt</tt></td><td>false</td><td>var1 < var2</td></tr>
2319 <tr><td><tt>ogt</tt></td><td>false</td><td>var1 > var2</td></tr>
2320 <tr><td><tt>ole</tt></td><td>false</td><td>var1 <= var2</td></tr>
2321 <tr><td><tt>oge</tt></td><td>false</td><td>var1 >= var2</td></tr>
2322 <tr><td><tt>ueq</tt></td><td>true</td><td>var1 == var2</td></tr>
2323 <tr><td><tt>une</tt></td><td>true</td><td>var1 != var2</td></tr>
2324 <tr><td><tt>ult</tt></td><td>true</td><td>var1 < var2</td></tr>
2325 <tr><td><tt>ugt</tt></td><td>true</td><td>var1 > var2</td></tr>
2326 <tr><td><tt>ule</tt></td><td>true</td><td>var1 <= var2</td></tr>
2327 <tr><td><tt>uge</tt></td><td>true</td><td>var1 >= var2</td></tr>
2328 <tr><td><tt>o</tt></td><td>false</td><td>always</td></tr>
2329 <tr><td><tt>u</tt></td><td>true</td><td>never</td></tr>
2330 <tr><td><tt>true</tt></td><td>true</td><td>always</td></tr>
2331 <tr><td><tt>false</tt></td><td>false</td><td>never</td></tr>
2336 <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>
2337 <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>
2338 <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>
2339 <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>
2340 <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>
2341 <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>
2345 <!-- _______________________________________________________________________ -->
2346 <div class="doc_subsubsection">
2347 <a name="i_vselect">'<tt>vselect</tt>' Instruction</a>
2350 <div class="doc_text">
2355 <result> = vselect <n x bool> <cond>, <n x <ty>> <val1>, <n x <ty>> <val2> <i>; yields <n x <ty>></i>
2361 The '<tt>vselect</tt>' instruction chooses one value at each position
2362 of a vector based on a condition.
2369 The '<tt>vselect</tt>' instruction requires a <a
2370 href="#t_packed">packed</a> <tt>bool</tt> value indicating the
2371 condition at each vector position, and two values of the same packed
2372 type. All three operands must have the same length. The type of the
2373 result is the same as the type of the two value operands.</p>
2378 At each position where the <tt>bool</tt> vector is true, that position
2379 of the result gets its value from the first value argument; otherwise,
2380 it gets its value from the second value argument.
2386 %X = vselect bool <2 x bool> <bool true, bool false>, <2 x ubyte> <ubyte 17, ubyte 17>,
2387 <2 x ubyte> <ubyte 42, ubyte 42> <i>; yields <2 x ubyte>:17, 42</i>
2393 <!-- ======================================================================= -->
2394 <div class="doc_subsection">
2395 <a name="memoryops">Memory Access and Addressing Operations</a>
2398 <div class="doc_text">
2400 <p>A key design point of an SSA-based representation is how it
2401 represents memory. In LLVM, no memory locations are in SSA form, which
2402 makes things very simple. This section describes how to read, write,
2403 allocate, and free memory in LLVM.</p>
2407 <!-- _______________________________________________________________________ -->
2408 <div class="doc_subsubsection">
2409 <a name="i_malloc">'<tt>malloc</tt>' Instruction</a>
2412 <div class="doc_text">
2417 <result> = malloc <type>[, uint <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2422 <p>The '<tt>malloc</tt>' instruction allocates memory from the system
2423 heap and returns a pointer to it.</p>
2427 <p>The '<tt>malloc</tt>' instruction allocates
2428 <tt>sizeof(<type>)*NumElements</tt>
2429 bytes of memory from the operating system and returns a pointer of the
2430 appropriate type to the program. If "NumElements" is specified, it is the
2431 number of elements allocated. If an alignment is specified, the value result
2432 of the allocation is guaranteed to be aligned to at least that boundary. If
2433 not specified, or if zero, the target can choose to align the allocation on any
2434 convenient boundary.</p>
2436 <p>'<tt>type</tt>' must be a sized type.</p>
2440 <p>Memory is allocated using the system "<tt>malloc</tt>" function, and
2441 a pointer is returned.</p>
2446 %array = malloc [4 x ubyte ] <i>; yields {[%4 x ubyte]*}:array</i>
2448 %size = <a href="#i_add">add</a> uint 2, 2 <i>; yields {uint}:size = uint 4</i>
2449 %array1 = malloc ubyte, uint 4 <i>; yields {ubyte*}:array1</i>
2450 %array2 = malloc [12 x ubyte], uint %size <i>; yields {[12 x ubyte]*}:array2</i>
2451 %array3 = malloc int, uint 4, align 1024 <i>; yields {int*}:array3</i>
2452 %array4 = malloc int, align 1024 <i>; yields {int*}:array4</i>
2456 <!-- _______________________________________________________________________ -->
2457 <div class="doc_subsubsection">
2458 <a name="i_free">'<tt>free</tt>' Instruction</a>
2461 <div class="doc_text">
2466 free <type> <value> <i>; yields {void}</i>
2471 <p>The '<tt>free</tt>' instruction returns memory back to the unused
2472 memory heap to be reallocated in the future.</p>
2476 <p>'<tt>value</tt>' shall be a pointer value that points to a value
2477 that was allocated with the '<tt><a href="#i_malloc">malloc</a></tt>'
2482 <p>Access to the memory pointed to by the pointer is no longer defined
2483 after this instruction executes.</p>
2488 %array = <a href="#i_malloc">malloc</a> [4 x ubyte] <i>; yields {[4 x ubyte]*}:array</i>
2489 free [4 x ubyte]* %array
2493 <!-- _______________________________________________________________________ -->
2494 <div class="doc_subsubsection">
2495 <a name="i_alloca">'<tt>alloca</tt>' Instruction</a>
2498 <div class="doc_text">
2503 <result> = alloca <type>[, uint <NumElements>][, align <alignment>] <i>; yields {type*}:result</i>
2508 <p>The '<tt>alloca</tt>' instruction allocates memory on the current
2509 stack frame of the procedure that is live until the current function
2510 returns to its caller.</p>
2514 <p>The '<tt>alloca</tt>' instruction allocates <tt>sizeof(<type>)*NumElements</tt>
2515 bytes of memory on the runtime stack, returning a pointer of the
2516 appropriate type to the program. If "NumElements" is specified, it is the
2517 number of elements allocated. If an alignment is specified, the value result
2518 of the allocation is guaranteed to be aligned to at least that boundary. If
2519 not specified, or if zero, the target can choose to align the allocation on any
2520 convenient boundary.</p>
2522 <p>'<tt>type</tt>' may be any sized type.</p>
2526 <p>Memory is allocated; a pointer is returned. '<tt>alloca</tt>'d
2527 memory is automatically released when the function returns. The '<tt>alloca</tt>'
2528 instruction is commonly used to represent automatic variables that must
2529 have an address available. When the function returns (either with the <tt><a
2530 href="#i_ret">ret</a></tt> or <tt><a href="#i_unwind">unwind</a></tt>
2531 instructions), the memory is reclaimed.</p>
2536 %ptr = alloca int <i>; yields {int*}:ptr</i>
2537 %ptr = alloca int, uint 4 <i>; yields {int*}:ptr</i>
2538 %ptr = alloca int, uint 4, align 1024 <i>; yields {int*}:ptr</i>
2539 %ptr = alloca int, align 1024 <i>; yields {int*}:ptr</i>
2543 <!-- _______________________________________________________________________ -->
2544 <div class="doc_subsubsection"> <a name="i_load">'<tt>load</tt>'
2545 Instruction</a> </div>
2546 <div class="doc_text">
2548 <pre> <result> = load <ty>* <pointer><br> <result> = volatile load <ty>* <pointer><br></pre>
2550 <p>The '<tt>load</tt>' instruction is used to read from memory.</p>
2552 <p>The argument to the '<tt>load</tt>' instruction specifies the memory
2553 address from which to load. The pointer must point to a <a
2554 href="#t_firstclass">first class</a> type. If the <tt>load</tt> is
2555 marked as <tt>volatile</tt>, then the optimizer is not allowed to modify
2556 the number or order of execution of this <tt>load</tt> with other
2557 volatile <tt>load</tt> and <tt><a href="#i_store">store</a></tt>
2560 <p>The location of memory pointed to is loaded.</p>
2562 <pre> %ptr = <a href="#i_alloca">alloca</a> int <i>; yields {int*}:ptr</i>
2564 href="#i_store">store</a> int 3, int* %ptr <i>; yields {void}</i>
2565 %val = load int* %ptr <i>; yields {int}:val = int 3</i>
2568 <!-- _______________________________________________________________________ -->
2569 <div class="doc_subsubsection"> <a name="i_store">'<tt>store</tt>'
2570 Instruction</a> </div>
2572 <pre> store <ty> <value>, <ty>* <pointer> <i>; yields {void}</i>
2573 volatile store <ty> <value>, <ty>* <pointer> <i>; yields {void}</i>
2576 <p>The '<tt>store</tt>' instruction is used to write to memory.</p>
2578 <p>There are two arguments to the '<tt>store</tt>' instruction: a value
2579 to store and an address in which to store it. The type of the '<tt><pointer></tt>'
2580 operand must be a pointer to the type of the '<tt><value></tt>'
2581 operand. If the <tt>store</tt> is marked as <tt>volatile</tt>, then the
2582 optimizer is not allowed to modify the number or order of execution of
2583 this <tt>store</tt> with other volatile <tt>load</tt> and <tt><a
2584 href="#i_store">store</a></tt> instructions.</p>
2586 <p>The contents of memory are updated to contain '<tt><value></tt>'
2587 at the location specified by the '<tt><pointer></tt>' operand.</p>
2589 <pre> %ptr = <a href="#i_alloca">alloca</a> int <i>; yields {int*}:ptr</i>
2591 href="#i_store">store</a> int 3, int* %ptr <i>; yields {void}</i>
2592 %val = load int* %ptr <i>; yields {int}:val = int 3</i>
2594 <!-- _______________________________________________________________________ -->
2595 <div class="doc_subsubsection">
2596 <a name="i_getelementptr">'<tt>getelementptr</tt>' Instruction</a>
2599 <div class="doc_text">
2602 <result> = getelementptr <ty>* <ptrval>{, <ty> <idx>}*
2608 The '<tt>getelementptr</tt>' instruction is used to get the address of a
2609 subelement of an aggregate data structure.</p>
2613 <p>This instruction takes a list of integer constants that indicate what
2614 elements of the aggregate object to index to. The actual types of the arguments
2615 provided depend on the type of the first pointer argument. The
2616 '<tt>getelementptr</tt>' instruction is used to index down through the type
2617 levels of a structure or to a specific index in an array. When indexing into a
2618 structure, only <tt>uint</tt>
2619 integer constants are allowed. When indexing into an array or pointer,
2620 <tt>int</tt> and <tt>long</tt> indexes are allowed of any sign.</p>
2622 <p>For example, let's consider a C code fragment and how it gets
2623 compiled to LLVM:</p>
2637 int *foo(struct ST *s) {
2638 return &s[1].Z.B[5][13];
2642 <p>The LLVM code generated by the GCC frontend is:</p>
2645 %RT = type { sbyte, [10 x [20 x int]], sbyte }
2646 %ST = type { int, double, %RT }
2650 int* %foo(%ST* %s) {
2652 %reg = getelementptr %ST* %s, int 1, uint 2, uint 1, int 5, int 13
2659 <p>The index types specified for the '<tt>getelementptr</tt>' instruction depend
2660 on the pointer type that is being indexed into. <a href="#t_pointer">Pointer</a>
2661 and <a href="#t_array">array</a> types require <tt>uint</tt>, <tt>int</tt>,
2662 <tt>ulong</tt>, or <tt>long</tt> values, and <a href="#t_struct">structure</a>
2663 types require <tt>uint</tt> <b>constants</b>.</p>
2665 <p>In the example above, the first index is indexing into the '<tt>%ST*</tt>'
2666 type, which is a pointer, yielding a '<tt>%ST</tt>' = '<tt>{ int, double, %RT
2667 }</tt>' type, a structure. The second index indexes into the third element of
2668 the structure, yielding a '<tt>%RT</tt>' = '<tt>{ sbyte, [10 x [20 x int]],
2669 sbyte }</tt>' type, another structure. The third index indexes into the second
2670 element of the structure, yielding a '<tt>[10 x [20 x int]]</tt>' type, an
2671 array. The two dimensions of the array are subscripted into, yielding an
2672 '<tt>int</tt>' type. The '<tt>getelementptr</tt>' instruction returns a pointer
2673 to this element, thus computing a value of '<tt>int*</tt>' type.</p>
2675 <p>Note that it is perfectly legal to index partially through a
2676 structure, returning a pointer to an inner element. Because of this,
2677 the LLVM code for the given testcase is equivalent to:</p>
2680 int* %foo(%ST* %s) {
2681 %t1 = getelementptr %ST* %s, int 1 <i>; yields %ST*:%t1</i>
2682 %t2 = getelementptr %ST* %t1, int 0, uint 2 <i>; yields %RT*:%t2</i>
2683 %t3 = getelementptr %RT* %t2, int 0, uint 1 <i>; yields [10 x [20 x int]]*:%t3</i>
2684 %t4 = getelementptr [10 x [20 x int]]* %t3, int 0, int 5 <i>; yields [20 x int]*:%t4</i>
2685 %t5 = getelementptr [20 x int]* %t4, int 0, int 13 <i>; yields int*:%t5</i>
2690 <p>Note that it is undefined to access an array out of bounds: array and
2691 pointer indexes must always be within the defined bounds of the array type.
2692 The one exception for this rules is zero length arrays. These arrays are
2693 defined to be accessible as variable length arrays, which requires access
2694 beyond the zero'th element.</p>
2696 <p>The getelementptr instruction is often confusing. For some more insight
2697 into how it works, see <a href="GetElementPtr.html">the getelementptr
2703 <i>; yields [12 x ubyte]*:aptr</i>
2704 %aptr = getelementptr {int, [12 x ubyte]}* %sptr, long 0, uint 1
2708 <!-- ======================================================================= -->
2709 <div class="doc_subsection"> <a name="otherops">Other Operations</a> </div>
2710 <div class="doc_text">
2711 <p>The instructions in this category are the "miscellaneous"
2712 instructions, which defy better classification.</p>
2714 <!-- _______________________________________________________________________ -->
2715 <div class="doc_subsubsection"> <a name="i_phi">'<tt>phi</tt>'
2716 Instruction</a> </div>
2717 <div class="doc_text">
2719 <pre> <result> = phi <ty> [ <val0>, <label0>], ...<br></pre>
2721 <p>The '<tt>phi</tt>' instruction is used to implement the φ node in
2722 the SSA graph representing the function.</p>
2724 <p>The type of the incoming values are specified with the first type
2725 field. After this, the '<tt>phi</tt>' instruction takes a list of pairs
2726 as arguments, with one pair for each predecessor basic block of the
2727 current block. Only values of <a href="#t_firstclass">first class</a>
2728 type may be used as the value arguments to the PHI node. Only labels
2729 may be used as the label arguments.</p>
2730 <p>There must be no non-phi instructions between the start of a basic
2731 block and the PHI instructions: i.e. PHI instructions must be first in
2734 <p>At runtime, the '<tt>phi</tt>' instruction logically takes on the
2735 value specified by the parameter, depending on which basic block we
2736 came from in the last <a href="#terminators">terminator</a> instruction.</p>
2738 <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>
2741 <!-- _______________________________________________________________________ -->
2742 <div class="doc_subsubsection">
2743 <a name="i_cast">'<tt>cast .. to</tt>' Instruction</a>
2746 <div class="doc_text">
2751 <result> = cast <ty> <value> to <ty2> <i>; yields ty2</i>
2757 The '<tt>cast</tt>' instruction is used as the primitive means to convert
2758 integers to floating point, change data type sizes, and break type safety (by
2766 The '<tt>cast</tt>' instruction takes a value to cast, which must be a first
2767 class value, and a type to cast it to, which must also be a <a
2768 href="#t_firstclass">first class</a> type.
2774 This instruction follows the C rules for explicit casts when determining how the
2775 data being cast must change to fit in its new container.
2779 When casting to bool, any value that would be considered true in the context of
2780 a C '<tt>if</tt>' condition is converted to the boolean '<tt>true</tt>' values,
2781 all else are '<tt>false</tt>'.
2785 When extending an integral value from a type of one signness to another (for
2786 example '<tt>sbyte</tt>' to '<tt>ulong</tt>'), the value is sign-extended if the
2787 <b>source</b> value is signed, and zero-extended if the source value is
2788 unsigned. <tt>bool</tt> values are always zero extended into either zero or
2795 %X = cast int 257 to ubyte <i>; yields ubyte:1</i>
2796 %Y = cast int 123 to bool <i>; yields bool:true</i>
2800 <!-- _______________________________________________________________________ -->
2801 <div class="doc_subsubsection">
2802 <a name="i_select">'<tt>select</tt>' Instruction</a>
2805 <div class="doc_text">
2810 <result> = select bool <cond>, <ty> <val1>, <ty> <val2> <i>; yields ty</i>
2816 The '<tt>select</tt>' instruction is used to choose one value based on a
2817 condition, without branching.
2824 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.
2830 If the boolean condition evaluates to true, the instruction returns the first
2831 value argument; otherwise, it returns the second value argument.
2837 %X = select bool true, ubyte 17, ubyte 42 <i>; yields ubyte:17</i>
2842 <!-- _______________________________________________________________________ -->
2843 <div class="doc_subsubsection">
2844 <a name="i_call">'<tt>call</tt>' Instruction</a>
2847 <div class="doc_text">
2851 <result> = [tail] call [<a href="#callingconv">cconv</a>] <ty>* <fnptrval>(<param list>)
2856 <p>The '<tt>call</tt>' instruction represents a simple function call.</p>
2860 <p>This instruction requires several arguments:</p>
2864 <p>The optional "tail" marker indicates whether the callee function accesses
2865 any allocas or varargs in the caller. If the "tail" marker is present, the
2866 function call is eligible for tail call optimization. Note that calls may
2867 be marked "tail" even if they do not occur before a <a
2868 href="#i_ret"><tt>ret</tt></a> instruction.
2871 <p>The optional "cconv" marker indicates which <a href="callingconv">calling
2872 convention</a> the call should use. If none is specified, the call defaults
2873 to using C calling conventions.
2876 <p>'<tt>ty</tt>': shall be the signature of the pointer to function value
2877 being invoked. The argument types must match the types implied by this
2878 signature. This type can be omitted if the function is not varargs and
2879 if the function type does not return a pointer to a function.</p>
2882 <p>'<tt>fnptrval</tt>': An LLVM value containing a pointer to a function to
2883 be invoked. In most cases, this is a direct function invocation, but
2884 indirect <tt>call</tt>s are just as possible, calling an arbitrary pointer
2885 to function value.</p>
2888 <p>'<tt>function args</tt>': argument list whose types match the
2889 function signature argument types. All arguments must be of
2890 <a href="#t_firstclass">first class</a> type. If the function signature
2891 indicates the function accepts a variable number of arguments, the extra
2892 arguments can be specified.</p>
2898 <p>The '<tt>call</tt>' instruction is used to cause control flow to
2899 transfer to a specified function, with its incoming arguments bound to
2900 the specified values. Upon a '<tt><a href="#i_ret">ret</a></tt>'
2901 instruction in the called function, control flow continues with the
2902 instruction after the function call, and the return value of the
2903 function is bound to the result argument. This is a simpler case of
2904 the <a href="#i_invoke">invoke</a> instruction.</p>
2909 %retval = call int %test(int %argc)
2910 call int(sbyte*, ...) *%printf(sbyte* %msg, int 12, sbyte 42);
2911 %X = tail call int %foo()
2912 %Y = tail call <a href="#callingconv">fastcc</a> int %foo()
2917 <!-- _______________________________________________________________________ -->
2918 <div class="doc_subsubsection">
2919 <a name="i_va_arg">'<tt>va_arg</tt>' Instruction</a>
2922 <div class="doc_text">
2927 <resultval> = va_arg <va_list*> <arglist>, <argty>
2932 <p>The '<tt>va_arg</tt>' instruction is used to access arguments passed through
2933 the "variable argument" area of a function call. It is used to implement the
2934 <tt>va_arg</tt> macro in C.</p>
2938 <p>This instruction takes a <tt>va_list*</tt> value and the type of
2939 the argument. It returns a value of the specified argument type and
2940 increments the <tt>va_list</tt> to point to the next argument. Again, the
2941 actual type of <tt>va_list</tt> is target specific.</p>
2945 <p>The '<tt>va_arg</tt>' instruction loads an argument of the specified
2946 type from the specified <tt>va_list</tt> and causes the
2947 <tt>va_list</tt> to point to the next argument. For more information,
2948 see the variable argument handling <a href="#int_varargs">Intrinsic
2951 <p>It is legal for this instruction to be called in a function which does not
2952 take a variable number of arguments, for example, the <tt>vfprintf</tt>
2955 <p><tt>va_arg</tt> is an LLVM instruction instead of an <a
2956 href="#intrinsics">intrinsic function</a> because it takes a type as an
2961 <p>See the <a href="#int_varargs">variable argument processing</a> section.</p>
2965 <!-- *********************************************************************** -->
2966 <div class="doc_section"> <a name="intrinsics">Intrinsic Functions</a> </div>
2967 <!-- *********************************************************************** -->
2969 <div class="doc_text">
2971 <p>LLVM supports the notion of an "intrinsic function". These functions have
2972 well known names and semantics and are required to follow certain
2973 restrictions. Overall, these instructions represent an extension mechanism for
2974 the LLVM language that does not require changing all of the transformations in
2975 LLVM to add to the language (or the bytecode reader/writer, the parser,
2978 <p>Intrinsic function names must all start with an "<tt>llvm.</tt>" prefix. This
2979 prefix is reserved in LLVM for intrinsic names; thus, functions may not be named
2980 this. Intrinsic functions must always be external functions: you cannot define
2981 the body of intrinsic functions. Intrinsic functions may only be used in call
2982 or invoke instructions: it is illegal to take the address of an intrinsic
2983 function. Additionally, because intrinsic functions are part of the LLVM
2984 language, it is required that they all be documented here if any are added.</p>
2987 <p>To learn how to add an intrinsic function, please see the <a
2988 href="ExtendingLLVM.html">Extending LLVM Guide</a>.
2993 <!-- ======================================================================= -->
2994 <div class="doc_subsection">
2995 <a name="int_varargs">Variable Argument Handling Intrinsics</a>
2998 <div class="doc_text">
3000 <p>Variable argument support is defined in LLVM with the <a
3001 href="#i_va_arg"><tt>va_arg</tt></a> instruction and these three
3002 intrinsic functions. These functions are related to the similarly
3003 named macros defined in the <tt><stdarg.h></tt> header file.</p>
3005 <p>All of these functions operate on arguments that use a
3006 target-specific value type "<tt>va_list</tt>". The LLVM assembly
3007 language reference manual does not define what this type is, so all
3008 transformations should be prepared to handle intrinsics with any type
3011 <p>This example shows how the <a href="#i_va_arg"><tt>va_arg</tt></a>
3012 instruction and the variable argument handling intrinsic functions are
3016 int %test(int %X, ...) {
3017 ; Initialize variable argument processing
3019 call void %<a href="#i_va_start">llvm.va_start</a>(sbyte** %ap)
3021 ; Read a single integer argument
3022 %tmp = va_arg sbyte** %ap, int
3024 ; Demonstrate usage of llvm.va_copy and llvm.va_end
3026 call void %<a href="#i_va_copy">llvm.va_copy</a>(sbyte** %aq, sbyte** %ap)
3027 call void %<a href="#i_va_end">llvm.va_end</a>(sbyte** %aq)
3029 ; Stop processing of arguments.
3030 call void %<a href="#i_va_end">llvm.va_end</a>(sbyte** %ap)
3036 <!-- _______________________________________________________________________ -->
3037 <div class="doc_subsubsection">
3038 <a name="i_va_start">'<tt>llvm.va_start</tt>' Intrinsic</a>
3042 <div class="doc_text">
3044 <pre> declare void %llvm.va_start(<va_list>* <arglist>)<br></pre>
3046 <P>The '<tt>llvm.va_start</tt>' intrinsic initializes
3047 <tt>*<arglist></tt> for subsequent use by <tt><a
3048 href="#i_va_arg">va_arg</a></tt>.</p>
3052 <P>The argument is a pointer to a <tt>va_list</tt> element to initialize.</p>
3056 <P>The '<tt>llvm.va_start</tt>' intrinsic works just like the <tt>va_start</tt>
3057 macro available in C. In a target-dependent way, it initializes the
3058 <tt>va_list</tt> element the argument points to, so that the next call to
3059 <tt>va_arg</tt> will produce the first variable argument passed to the function.
3060 Unlike the C <tt>va_start</tt> macro, this intrinsic does not need to know the
3061 last argument of the function, the compiler can figure that out.</p>
3065 <!-- _______________________________________________________________________ -->
3066 <div class="doc_subsubsection">
3067 <a name="i_va_end">'<tt>llvm.va_end</tt>' Intrinsic</a>
3070 <div class="doc_text">
3072 <pre> declare void %llvm.va_end(<va_list*> <arglist>)<br></pre>
3074 <p>The '<tt>llvm.va_end</tt>' intrinsic destroys <tt><arglist></tt>
3075 which has been initialized previously with <tt><a href="#i_va_start">llvm.va_start</a></tt>
3076 or <tt><a href="#i_va_copy">llvm.va_copy</a></tt>.</p>
3078 <p>The argument is a <tt>va_list</tt> to destroy.</p>
3080 <p>The '<tt>llvm.va_end</tt>' intrinsic works just like the <tt>va_end</tt>
3081 macro available in C. In a target-dependent way, it destroys the <tt>va_list</tt>.
3082 Calls to <a href="#i_va_start"><tt>llvm.va_start</tt></a> and <a
3083 href="#i_va_copy"><tt>llvm.va_copy</tt></a> must be matched exactly
3084 with calls to <tt>llvm.va_end</tt>.</p>
3087 <!-- _______________________________________________________________________ -->
3088 <div class="doc_subsubsection">
3089 <a name="i_va_copy">'<tt>llvm.va_copy</tt>' Intrinsic</a>
3092 <div class="doc_text">
3097 declare void %llvm.va_copy(<va_list>* <destarglist>,
3098 <va_list>* <srcarglist>)
3103 <p>The '<tt>llvm.va_copy</tt>' intrinsic copies the current argument position from
3104 the source argument list to the destination argument list.</p>
3108 <p>The first argument is a pointer to a <tt>va_list</tt> element to initialize.
3109 The second argument is a pointer to a <tt>va_list</tt> element to copy from.</p>
3114 <p>The '<tt>llvm.va_copy</tt>' intrinsic works just like the <tt>va_copy</tt> macro
3115 available in C. In a target-dependent way, it copies the source
3116 <tt>va_list</tt> element into the destination list. This intrinsic is necessary
3117 because the <tt><a href="i_va_begin">llvm.va_begin</a></tt> intrinsic may be
3118 arbitrarily complex and require memory allocation, for example.</p>
3122 <!-- ======================================================================= -->
3123 <div class="doc_subsection">
3124 <a name="int_gc">Accurate Garbage Collection Intrinsics</a>
3127 <div class="doc_text">
3130 LLVM support for <a href="GarbageCollection.html">Accurate Garbage
3131 Collection</a> requires the implementation and generation of these intrinsics.
3132 These intrinsics allow identification of <a href="#i_gcroot">GC roots on the
3133 stack</a>, as well as garbage collector implementations that require <a
3134 href="#i_gcread">read</a> and <a href="#i_gcwrite">write</a> barriers.
3135 Front-ends for type-safe garbage collected languages should generate these
3136 intrinsics to make use of the LLVM garbage collectors. For more details, see <a
3137 href="GarbageCollection.html">Accurate Garbage Collection with LLVM</a>.
3141 <!-- _______________________________________________________________________ -->
3142 <div class="doc_subsubsection">
3143 <a name="i_gcroot">'<tt>llvm.gcroot</tt>' Intrinsic</a>
3146 <div class="doc_text">
3151 declare void %llvm.gcroot(<ty>** %ptrloc, <ty2>* %metadata)
3156 <p>The '<tt>llvm.gcroot</tt>' intrinsic declares the existence of a GC root to
3157 the code generator, and allows some metadata to be associated with it.</p>
3161 <p>The first argument specifies the address of a stack object that contains the
3162 root pointer. The second pointer (which must be either a constant or a global
3163 value address) contains the meta-data to be associated with the root.</p>
3167 <p>At runtime, a call to this intrinsics stores a null pointer into the "ptrloc"
3168 location. At compile-time, the code generator generates information to allow
3169 the runtime to find the pointer at GC safe points.
3175 <!-- _______________________________________________________________________ -->
3176 <div class="doc_subsubsection">
3177 <a name="i_gcread">'<tt>llvm.gcread</tt>' Intrinsic</a>
3180 <div class="doc_text">
3185 declare sbyte* %llvm.gcread(sbyte* %ObjPtr, sbyte** %Ptr)
3190 <p>The '<tt>llvm.gcread</tt>' intrinsic identifies reads of references from heap
3191 locations, allowing garbage collector implementations that require read
3196 <p>The second argument is the address to read from, which should be an address
3197 allocated from the garbage collector. The first object is a pointer to the
3198 start of the referenced object, if needed by the language runtime (otherwise
3203 <p>The '<tt>llvm.gcread</tt>' intrinsic has the same semantics as a load
3204 instruction, but may be replaced with substantially more complex code by the
3205 garbage collector runtime, as needed.</p>
3210 <!-- _______________________________________________________________________ -->
3211 <div class="doc_subsubsection">
3212 <a name="i_gcwrite">'<tt>llvm.gcwrite</tt>' Intrinsic</a>
3215 <div class="doc_text">
3220 declare void %llvm.gcwrite(sbyte* %P1, sbyte* %Obj, sbyte** %P2)
3225 <p>The '<tt>llvm.gcwrite</tt>' intrinsic identifies writes of references to heap
3226 locations, allowing garbage collector implementations that require write
3227 barriers (such as generational or reference counting collectors).</p>
3231 <p>The first argument is the reference to store, the second is the start of the
3232 object to store it to, and the third is the address of the field of Obj to
3233 store to. If the runtime does not require a pointer to the object, Obj may be
3238 <p>The '<tt>llvm.gcwrite</tt>' intrinsic has the same semantics as a store
3239 instruction, but may be replaced with substantially more complex code by the
3240 garbage collector runtime, as needed.</p>
3246 <!-- ======================================================================= -->
3247 <div class="doc_subsection">
3248 <a name="int_codegen">Code Generator Intrinsics</a>
3251 <div class="doc_text">
3253 These intrinsics are provided by LLVM to expose special features that may only
3254 be implemented with code generator support.
3259 <!-- _______________________________________________________________________ -->
3260 <div class="doc_subsubsection">
3261 <a name="i_returnaddress">'<tt>llvm.returnaddress</tt>' Intrinsic</a>
3264 <div class="doc_text">
3268 declare sbyte *%llvm.returnaddress(uint <level>)
3274 The '<tt>llvm.returnaddress</tt>' intrinsic attempts to compute a
3275 target-specific value indicating the return address of the current function
3276 or one of its callers.
3282 The argument to this intrinsic indicates which function to return the address
3283 for. Zero indicates the calling function, one indicates its caller, etc. The
3284 argument is <b>required</b> to be a constant integer value.
3290 The '<tt>llvm.returnaddress</tt>' intrinsic either returns a pointer indicating
3291 the return 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.
3304 <!-- _______________________________________________________________________ -->
3305 <div class="doc_subsubsection">
3306 <a name="i_frameaddress">'<tt>llvm.frameaddress</tt>' Intrinsic</a>
3309 <div class="doc_text">
3313 declare sbyte *%llvm.frameaddress(uint <level>)
3319 The '<tt>llvm.frameaddress</tt>' intrinsic attempts to return the
3320 target-specific frame pointer value for the specified stack frame.
3326 The argument to this intrinsic indicates which function to return the frame
3327 pointer for. Zero indicates the calling function, one indicates its caller,
3328 etc. The argument is <b>required</b> to be a constant integer value.
3334 The '<tt>llvm.frameaddress</tt>' intrinsic either returns a pointer indicating
3335 the frame address of the specified call frame, or zero if it cannot be
3336 identified. The value returned by this intrinsic is likely to be incorrect or 0
3337 for arguments other than zero, so it should only be used for debugging purposes.
3341 Note that calling this intrinsic does not prevent function inlining or other
3342 aggressive transformations, so the value returned may not be that of the obvious
3343 source-language caller.
3347 <!-- _______________________________________________________________________ -->
3348 <div class="doc_subsubsection">
3349 <a name="i_stacksave">'<tt>llvm.stacksave</tt>' Intrinsic</a>
3352 <div class="doc_text">
3356 declare sbyte *%llvm.stacksave()
3362 The '<tt>llvm.stacksave</tt>' intrinsic is used to remember the current state of
3363 the function stack, for use with <a href="#i_stackrestore">
3364 <tt>llvm.stackrestore</tt></a>. This is useful for implementing language
3365 features like scoped automatic variable sized arrays in C99.
3371 This intrinsic returns a opaque pointer value that can be passed to <a
3372 href="#i_stackrestore"><tt>llvm.stackrestore</tt></a>. When an
3373 <tt>llvm.stackrestore</tt> intrinsic is executed with a value saved from
3374 <tt>llvm.stacksave</tt>, it effectively restores the state of the stack to the
3375 state it was in when the <tt>llvm.stacksave</tt> intrinsic executed. In
3376 practice, this pops any <a href="#i_alloca">alloca</a> blocks from the stack
3377 that were allocated after the <tt>llvm.stacksave</tt> was executed.
3382 <!-- _______________________________________________________________________ -->
3383 <div class="doc_subsubsection">
3384 <a name="i_stackrestore">'<tt>llvm.stackrestore</tt>' Intrinsic</a>
3387 <div class="doc_text">
3391 declare void %llvm.stackrestore(sbyte* %ptr)
3397 The '<tt>llvm.stackrestore</tt>' intrinsic is used to restore the state of
3398 the function stack to the state it was in when the corresponding <a
3399 href="#llvm.stacksave"><tt>llvm.stacksave</tt></a> intrinsic executed. This is
3400 useful for implementing language features like scoped automatic variable sized
3407 See the description for <a href="#i_stacksave"><tt>llvm.stacksave</tt></a>.
3413 <!-- _______________________________________________________________________ -->
3414 <div class="doc_subsubsection">
3415 <a name="i_prefetch">'<tt>llvm.prefetch</tt>' Intrinsic</a>
3418 <div class="doc_text">
3422 declare void %llvm.prefetch(sbyte * <address>,
3423 uint <rw>, uint <locality>)
3430 The '<tt>llvm.prefetch</tt>' intrinsic is a hint to the code generator to insert
3431 a prefetch instruction if supported; otherwise, it is a noop. Prefetches have
3433 effect on the behavior of the program but can change its performance
3440 <tt>address</tt> is the address to be prefetched, <tt>rw</tt> is the specifier
3441 determining if the fetch should be for a read (0) or write (1), and
3442 <tt>locality</tt> is a temporal locality specifier ranging from (0) - no
3443 locality, to (3) - extremely local keep in cache. The <tt>rw</tt> and
3444 <tt>locality</tt> arguments must be constant integers.
3450 This intrinsic does not modify the behavior of the program. In particular,
3451 prefetches cannot trap and do not produce a value. On targets that support this
3452 intrinsic, the prefetch can provide hints to the processor cache for better
3458 <!-- _______________________________________________________________________ -->
3459 <div class="doc_subsubsection">
3460 <a name="i_pcmarker">'<tt>llvm.pcmarker</tt>' Intrinsic</a>
3463 <div class="doc_text">
3467 declare void %llvm.pcmarker( uint <id> )
3474 The '<tt>llvm.pcmarker</tt>' intrinsic is a method to export a Program Counter
3476 code to simulators and other tools. The method is target specific, but it is
3477 expected that the marker will use exported symbols to transmit the PC of the marker.
3478 The marker makes no guarantees that it will remain with any specific instruction
3479 after optimizations. It is possible that the presence of a marker will inhibit
3480 optimizations. The intended use is to be inserted after optimizations to allow
3481 correlations of simulation runs.
3487 <tt>id</tt> is a numerical id identifying the marker.
3493 This intrinsic does not modify the behavior of the program. Backends that do not
3494 support this intrinisic may ignore it.
3499 <!-- _______________________________________________________________________ -->
3500 <div class="doc_subsubsection">
3501 <a name="i_readcyclecounter">'<tt>llvm.readcyclecounter</tt>' Intrinsic</a>
3504 <div class="doc_text">
3508 declare ulong %llvm.readcyclecounter( )
3515 The '<tt>llvm.readcyclecounter</tt>' intrinsic provides access to the cycle
3516 counter register (or similar low latency, high accuracy clocks) on those targets
3517 that support it. On X86, it should map to RDTSC. On Alpha, it should map to RPCC.
3518 As the backing counters overflow quickly (on the order of 9 seconds on alpha), this
3519 should only be used for small timings.
3525 When directly supported, reading the cycle counter should not modify any memory.
3526 Implementations are allowed to either return a application specific value or a
3527 system wide value. On backends without support, this is lowered to a constant 0.
3532 <!-- ======================================================================= -->
3533 <div class="doc_subsection">
3534 <a name="int_libc">Standard C Library Intrinsics</a>
3537 <div class="doc_text">
3539 LLVM provides intrinsics for a few important standard C library functions.
3540 These intrinsics allow source-language front-ends to pass information about the
3541 alignment of the pointer arguments to the code generator, providing opportunity
3542 for more efficient code generation.
3547 <!-- _______________________________________________________________________ -->
3548 <div class="doc_subsubsection">
3549 <a name="i_memcpy">'<tt>llvm.memcpy</tt>' Intrinsic</a>
3552 <div class="doc_text">
3556 declare void %llvm.memcpy.i32(sbyte* <dest>, sbyte* <src>,
3557 uint <len>, uint <align>)
3558 declare void %llvm.memcpy.i64(sbyte* <dest>, sbyte* <src>,
3559 ulong <len>, uint <align>)
3565 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
3566 location to the destination location.
3570 Note that, unlike the standard libc function, the <tt>llvm.memcpy.*</tt>
3571 intrinsics do not return a value, and takes an extra alignment argument.
3577 The first argument is a pointer to the destination, the second is a pointer to
3578 the source. The third argument is an integer argument
3579 specifying the number of bytes to copy, and the fourth argument is the alignment
3580 of the source and destination locations.
3584 If the call to this intrinisic has an alignment value that is not 0 or 1, then
3585 the caller guarantees that both the source and destination pointers are aligned
3592 The '<tt>llvm.memcpy.*</tt>' intrinsics copy a block of memory from the source
3593 location to the destination location, which are not allowed to overlap. It
3594 copies "len" bytes of memory over. If the argument is known to be aligned to
3595 some boundary, this can be specified as the fourth argument, otherwise it should
3601 <!-- _______________________________________________________________________ -->
3602 <div class="doc_subsubsection">
3603 <a name="i_memmove">'<tt>llvm.memmove</tt>' Intrinsic</a>
3606 <div class="doc_text">
3610 declare void %llvm.memmove.i32(sbyte* <dest>, sbyte* <src>,
3611 uint <len>, uint <align>)
3612 declare void %llvm.memmove.i64(sbyte* <dest>, sbyte* <src>,
3613 ulong <len>, uint <align>)
3619 The '<tt>llvm.memmove.*</tt>' intrinsics move a block of memory from the source
3620 location to the destination location. It is similar to the
3621 '<tt>llvm.memcmp</tt>' intrinsic but allows the two memory locations to overlap.
3625 Note that, unlike the standard libc function, the <tt>llvm.memmove.*</tt>
3626 intrinsics do not return a value, and takes an extra alignment argument.
3632 The first argument is a pointer to the destination, the second is a pointer to
3633 the source. The third argument is an integer argument
3634 specifying the number of bytes to copy, and the fourth argument is the alignment
3635 of the source and destination locations.
3639 If the call to this intrinisic has an alignment value that is not 0 or 1, then
3640 the caller guarantees that the source and destination pointers are aligned to
3647 The '<tt>llvm.memmove.*</tt>' intrinsics copy a block of memory from the source
3648 location to the destination location, which may overlap. It
3649 copies "len" bytes of memory over. If the argument is known to be aligned to
3650 some boundary, this can be specified as the fourth argument, otherwise it should
3656 <!-- _______________________________________________________________________ -->
3657 <div class="doc_subsubsection">
3658 <a name="i_memset">'<tt>llvm.memset.*</tt>' Intrinsics</a>
3661 <div class="doc_text">
3665 declare void %llvm.memset.i32(sbyte* <dest>, ubyte <val>,
3666 uint <len>, uint <align>)
3667 declare void %llvm.memset.i64(sbyte* <dest>, ubyte <val>,
3668 ulong <len>, uint <align>)
3674 The '<tt>llvm.memset.*</tt>' intrinsics fill a block of memory with a particular
3679 Note that, unlike the standard libc function, the <tt>llvm.memset</tt> intrinsic
3680 does not return a value, and takes an extra alignment argument.
3686 The first argument is a pointer to the destination to fill, the second is the
3687 byte value to fill it with, the third argument is an integer
3688 argument specifying the number of bytes to fill, and the fourth argument is the
3689 known alignment of destination location.
3693 If the call to this intrinisic has an alignment value that is not 0 or 1, then
3694 the caller guarantees that the destination pointer is aligned to that boundary.
3700 The '<tt>llvm.memset.*</tt>' intrinsics fill "len" bytes of memory starting at
3702 destination location. If the argument is known to be aligned to some boundary,
3703 this can be specified as the fourth argument, otherwise it should be set to 0 or
3709 <!-- _______________________________________________________________________ -->
3710 <div class="doc_subsubsection">
3711 <a name="i_isunordered">'<tt>llvm.isunordered.*</tt>' Intrinsic</a>
3714 <div class="doc_text">
3718 declare bool %llvm.isunordered.f32(float Val1, float Val2)
3719 declare bool %llvm.isunordered.f64(double Val1, double Val2)
3725 The '<tt>llvm.isunordered</tt>' intrinsics return true if either or both of the
3726 specified floating point values is a NAN.
3732 The arguments are floating point numbers of the same type.
3738 If either or both of the arguments is a SNAN or QNAN, it returns true, otherwise
3744 <!-- _______________________________________________________________________ -->
3745 <div class="doc_subsubsection">
3746 <a name="i_sqrt">'<tt>llvm.sqrt.*</tt>' Intrinsic</a>
3749 <div class="doc_text">
3753 declare float %llvm.sqrt.f32(float %Val)
3754 declare double %llvm.sqrt.f64(double %Val)
3760 The '<tt>llvm.sqrt</tt>' intrinsics return the sqrt of the specified operand,
3761 returning the same value as the libm '<tt>sqrt</tt>' function would. Unlike
3762 <tt>sqrt</tt> in libm, however, <tt>llvm.sqrt</tt> has undefined behavior for
3763 negative numbers (which allows for better optimization).
3769 The argument and return value are floating point numbers of the same type.
3775 This function returns the sqrt of the specified operand if it is a positive
3776 floating point number.
3780 <!-- _______________________________________________________________________ -->
3781 <div class="doc_subsubsection">
3782 <a name="i_powi">'<tt>llvm.powi.*</tt>' Intrinsic</a>
3785 <div class="doc_text">
3789 declare float %llvm.powi.f32(float %Val, int %power)
3790 declare double %llvm.powi.f64(double %Val, int %power)
3796 The '<tt>llvm.powi.*</tt>' intrinsics return the first operand raised to the
3797 specified (positive or negative) power. The order of evaluation of
3798 multiplications is not defined.
3804 The second argument is an integer power, and the first is a value to raise to
3811 This function returns the first value raised to the second power with an
3812 unspecified sequence of rounding operations.</p>
3816 <!-- ======================================================================= -->
3817 <div class="doc_subsection">
3818 <a name="int_manip">Bit Manipulation Intrinsics</a>
3821 <div class="doc_text">
3823 LLVM provides intrinsics for a few important bit manipulation operations.
3824 These allow efficient code generation for some algorithms.
3829 <!-- _______________________________________________________________________ -->
3830 <div class="doc_subsubsection">
3831 <a name="i_bswap">'<tt>llvm.bswap.*</tt>' Intrinsics</a>
3834 <div class="doc_text">
3838 declare ushort %llvm.bswap.i16(ushort <id>)
3839 declare uint %llvm.bswap.i32(uint <id>)
3840 declare ulong %llvm.bswap.i64(ulong <id>)
3846 The '<tt>llvm.bwsap</tt>' family of intrinsics is used to byteswap a 16, 32 or
3847 64 bit quantity. These are useful for performing operations on data that is not
3848 in the target's native byte order.
3854 The <tt>llvm.bswap.16</tt> intrinsic returns a ushort value that has the high and low
3855 byte of the input ushort swapped. Similarly, the <tt>llvm.bswap.i32</tt> intrinsic
3856 returns a uint value that has the four bytes of the input uint swapped, so that
3857 if the input bytes are numbered 0, 1, 2, 3 then the returned uint will have its
3858 bytes in 3, 2, 1, 0 order. The <tt>llvm.bswap.i64</tt> intrinsic extends this concept
3864 <!-- _______________________________________________________________________ -->
3865 <div class="doc_subsubsection">
3866 <a name="int_ctpop">'<tt>llvm.ctpop.*</tt>' Intrinsic</a>
3869 <div class="doc_text">
3873 declare ubyte %llvm.ctpop.i8 (ubyte <src>)
3874 declare ushort %llvm.ctpop.i16(ushort <src>)
3875 declare uint %llvm.ctpop.i32(uint <src>)
3876 declare ulong %llvm.ctpop.i64(ulong <src>)
3882 The '<tt>llvm.ctpop</tt>' family of intrinsics counts the number of bits set in a
3889 The only argument is the value to be counted. The argument may be of any
3890 unsigned integer type. The return type must match the argument type.
3896 The '<tt>llvm.ctpop</tt>' intrinsic counts the 1's in a variable.
3900 <!-- _______________________________________________________________________ -->
3901 <div class="doc_subsubsection">
3902 <a name="int_ctlz">'<tt>llvm.ctlz.*</tt>' Intrinsic</a>
3905 <div class="doc_text">
3909 declare ubyte %llvm.ctlz.i8 (ubyte <src>)
3910 declare ushort %llvm.ctlz.i16(ushort <src>)
3911 declare uint %llvm.ctlz.i32(uint <src>)
3912 declare ulong %llvm.ctlz.i64(ulong <src>)
3918 The '<tt>llvm.ctlz</tt>' family of intrinsic functions counts the number of
3919 leading zeros in a variable.
3925 The only argument is the value to be counted. The argument may be of any
3926 unsigned integer type. The return type must match the argument type.
3932 The '<tt>llvm.ctlz</tt>' intrinsic counts the leading (most significant) zeros
3933 in a variable. If the src == 0 then the result is the size in bits of the type
3934 of src. For example, <tt>llvm.ctlz(int 2) = 30</tt>.
3940 <!-- _______________________________________________________________________ -->
3941 <div class="doc_subsubsection">
3942 <a name="int_cttz">'<tt>llvm.cttz.*</tt>' Intrinsic</a>
3945 <div class="doc_text">
3949 declare ubyte %llvm.cttz.i8 (ubyte <src>)
3950 declare ushort %llvm.cttz.i16(ushort <src>)
3951 declare uint %llvm.cttz.i32(uint <src>)
3952 declare ulong %llvm.cttz.i64(ulong <src>)
3958 The '<tt>llvm.cttz</tt>' family of intrinsic functions counts the number of
3965 The only argument is the value to be counted. The argument may be of any
3966 unsigned integer type. The return type must match the argument type.
3972 The '<tt>llvm.cttz</tt>' intrinsic counts the trailing (least significant) zeros
3973 in a variable. If the src == 0 then the result is the size in bits of the type
3974 of src. For example, <tt>llvm.cttz(2) = 1</tt>.
3978 <!-- ======================================================================= -->
3979 <div class="doc_subsection">
3980 <a name="int_debugger">Debugger Intrinsics</a>
3983 <div class="doc_text">
3985 The LLVM debugger intrinsics (which all start with <tt>llvm.dbg.</tt> prefix),
3986 are described in the <a
3987 href="SourceLevelDebugging.html#format_common_intrinsics">LLVM Source Level
3988 Debugging</a> document.
3993 <!-- *********************************************************************** -->
3996 <a href="http://jigsaw.w3.org/css-validator/check/referer"><img
3997 src="http://jigsaw.w3.org/css-validator/images/vcss" alt="Valid CSS!"></a>
3998 <a href="http://validator.w3.org/check/referer"><img
3999 src="http://www.w3.org/Icons/valid-html401" alt="Valid HTML 4.01!" /></a>
4001 <a href="mailto:sabre@nondot.org">Chris Lattner</a><br>
4002 <a href="http://llvm.org">The LLVM Compiler Infrastructure</a><br>
4003 Last modified: $Date$